Electromagnetism

Oscar Reula1

1 Facultad de Matemáticas, Astronomía Física y Computación, Univ. Nac. Córdoba.

2
Contents

I First Course 13
1 Maxwell’s Equations: An overview 15
1.1 Maxwell’s equations in vacuum . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.1 Maxwell’s evolution equations . . . . . . . . . . . . . . . . . . . . . . . 16
1.1.2 The Constraint Equations . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.2 Initial Value Formulation II: Sources . . . . . . . . . . . . . . . . . . . . . . . 32
1.2.1 Charge Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.2.2 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . 34

2 Energy and Momentum of the Electromagnetic Field. 39

2.1 The Electromagnetic Field Energy . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 The Electromagnetic Field Momentum . . . . . . . . . . . . . . . . . . . . . . 42

3 The Symmetries of Maxwell’s Equations 45

3.1 Time Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Space Translations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Discrete Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.1 Time Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.2 Space Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 Galilean Transformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.1 The origin of the constant c on Maxwell’s equations . . . . . . . . . . . 54
3.6 Other Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4 Stationary Solutions: Electrostatics 57

4.1 Stationary and Static Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.1 The General Solution for Isolates Systems . . . . . . . . . . . . . . . . 58
4.2.2 The Weak Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.3 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.4 Existence of the Potential Function . . . . . . . . . . . . . . . . . . . . 62
4.2.5 Weak Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 The principle of superposition . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.2 Example: [Method of Images] . . . . . . . . . . . . . . . . . . . . . . . 70

3
4 CONTENTS

4.4 Capacities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5 Static Problems – Constructing Interesting Solutions 77

5.1 Method of Separation of Variables in Cartesian Coordinates . . . . . . . . . . 77
5.2 Method of Separation of Variables in Spherical Coordinates . . . . . . . . . . . 82
5.2.1 Associated Legendre Functions and Spherical Harmonics . . . . . . . . 89
5.3 Application: Multipole moments of a static configuration . . . . . . . . . . . . 94
5.3.1 The Addition Theorem for Spherical Harmonics . . . . . . . . . . . . . 96

6 Eigenfunction Expansions 101

6.1 The General Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2 The Finite Dimensional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.3 The Infinite Dimensional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3.1 The L2 norm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.3.2 Point-wise Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7 The Theory of Green functions 107

7.1 The Dirichlet Green function when ∂V is the union of two concentric spheres. 108
7.2 Some examples using the spherical Green functions . . . . . . . . . . . . . . . 112
7.2.1 Empty sphere at potential V (θ, ϕ) . . . . . . . . . . . . . . . . . . . . . 112
7.2.2 Homogeneously charged ring inside a sphere at zero potential . . . . . . 113
7.2.3 Uniformly charge rod inside a grounded sphere . . . . . . . . . . . . . . 114
7.3 Construction of Green Functions Using Eigenfunctions. . . . . . . . . . . . . . 115

8 Dielectrics 119
8.1 The Nature of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.2 A Microscopic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.2.1 Matching conditions for dielectrics . . . . . . . . . . . . . . . . . . . . 127
8.3 The Electrostatic Energy of Dielectrics . . . . . . . . . . . . . . . . . . . . . . 131

9 Stationary Solutions: Magnetostatics 133

9.1 The General Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
9.1.1 Isolated systems of currents . . . . . . . . . . . . . . . . . . . . . . . . 134
9.2 Boundary Conditions - Super conductors . . . . . . . . . . . . . . . . . . . . . 137
9.3 The magnetic scalar potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
9.3.1 Wires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.4 Non-Simply Connected Super-Conductors . . . . . . . . . . . . . . . . . . . . 149
9.5 Multipolar expansion of the magnetostatics field . . . . . . . . . . . . . . . . . 153

10 The Energy of the Magnetostatic Field 157

10.1 The Energy of Current Line Distributions. . . . . . . . . . . . . . . . . . . . . 159
10.2 Inductances and Magnetic Fluxes. . . . . . . . . . . . . . . . . . . . . . . . . . 160
CONTENTS 5

11 Magnetic Materials 163

11.0.1 Matching conditions on material boundaries . . . . . . . . . . . . . . . 164
11.1 Constant Magnetization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
11.2 Permanent Magnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
11.3 Generalized Forces on Charged Conductors and Circuits . . . . . . . . . . . . 170
11.4 The Energy of Magnetic Materials . . . . . . . . . . . . . . . . . . . . . . . . . 173

12 Examination Questions 1 177

II Second Course 181

13 The Symmetries of Maxwell’s Equations Continued 183
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

14 Special Relativity 189

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
14.1.1 The Aristotelian viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . 191
14.1.2 The Galilean viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
14.1.3 There comes the light! . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
14.1.4 The Einstein viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . 197
14.2 The geometry of Minkowski space-time. . . . . . . . . . . . . . . . . . . . . . . 199
14.2.1 4-vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
14.2.2 Reference frame descriptions. . . . . . . . . . . . . . . . . . . . . . . . 200
14.2.3 The proper time of a trajectory. . . . . . . . . . . . . . . . . . . . . . . 206
14.2.4 The trajectories of free particles and a variational principle. . . . . . . 206
14.2.5 The size of objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
14.2.6 Volume contraction and densities . . . . . . . . . . . . . . . . . . . . . 210
14.3 Relativistic Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
14.3.1 Doppler shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
14.3.2 Aberration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
14.4 Energy-Momentum 4-vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

15 Relativistic Electrodynamics 221

15.1 Maxwell’s tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
15.2 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
15.3 Invariants of Maxwell Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
15.4 The Energy-Momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
15.4.1 Energy conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
15.4.2 Momentum Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.4.3 Angular Momentum conservation . . . . . . . . . . . . . . . . . . . . . 233
15.5 Killing Vector Fields in Minkowski space-time. . . . . . . . . . . . . . . . . . . 234
15.5.1 Conformal Killing Vector Fields. . . . . . . . . . . . . . . . . . . . . . . 235
6 CONTENTS

16 Vector Potential - Connection 237

16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
16.2 Gauge equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
16.2.1 Lorentz Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
16.2.2 Coulomb Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
16.2.3 Temporal Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

17 Variational Formalisms 243

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
17.2 Variational Dynamics of a Charged Particle . . . . . . . . . . . . . . . . . . . 243
17.3 Variational Dynamics of Maxwell’s Fields . . . . . . . . . . . . . . . . . . . . . 245

18 Plane Waves 249

18.1 Definition and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
18.2 Invariants and Energy Momentum Tensor . . . . . . . . . . . . . . . . . . . . 252
18.3 Gauge Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
18.4 Helicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
18.5 Observer dependent expressions . . . . . . . . . . . . . . . . . . . . . . . . . . 254
18.6 Covariant Boundary Conditions for Superconductors and Total Reflection . . . 254
18.7 Monochromatic Waves and Fourier Decomposition . . . . . . . . . . . . . . . . 258

19 Resonant Cavities and Wave Guides 261

19.1 Resonant Cavities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
19.1.1 Monochromatic solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 262
19.1.2 Rectangular hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
19.2 Wave Guides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
19.2.1 Case 1: Transversal (T) waves, Ez = Bz = 0. . . . . . . . . . . . . . . . 268
19.2.2 Case 2: Transverse Magnetic (TM) waves, Bz = 0. . . . . . . . . . . . . 269
19.2.3 Case 3: Transverse Electric (TE) waves, Ez = 0. . . . . . . . . . . . . . 271
19.2.4 Energy Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
19.3 Cavities with constant section . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
19.3.1 T modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
19.3.2 TE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
19.3.3 TM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

20 Wave Propagation in Continuum Media 275

20.1 The polarizability model for time dependent fields. . . . . . . . . . . . . . . . 275
20.2 Plane Waves in Homogeneous and Isotropic Media. . . . . . . . . . . . . . . . 279
20.3 Reflection and Refraction across interfaces, the Perpendicular Case. . . . . . . 282
20.4 Reflection and Refraction, the General Case. . . . . . . . . . . . . . . . . . . . 284
20.4.1 Electric field perpendicular to the incidence plane . . . . . . . . . . . . 287
20.4.2 Electric field at the incidence plane . . . . . . . . . . . . . . . . . . . . 288
20.4.3 reflection Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
20.4.4 Total Polarization Angle or Brewster angle . . . . . . . . . . . . . . . . 289
CONTENTS 7

20.4.5 Total reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

20.5 Skin effect and the surface impedance of metals . . . . . . . . . . . . . . . . . 290
20.6 Energy losses in cavities and guides . . . . . . . . . . . . . . . . . . . . . . . . 293

21 Wave Packets, Kramers–Kroning relations and Causality 295

21.1 Propagation of wave packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
21.1.1 Non-dispersive Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
21.1.2 Dispersive Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
21.2 Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
21.3 Kramers–Kroning Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
21.4 Einstein’s Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

22 Electromagnetic Radiation produced by charge currents. 307

22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
22.2 Asymptotic behavior of the fields . . . . . . . . . . . . . . . . . . . . . . . . . 310
22.2.1 Large Space Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
22.2.2 Large Null Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
22.2.3 The nature of the approximation . . . . . . . . . . . . . . . . . . . . . 312
22.2.4 The power output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
22.2.5 Quadrupole, and Magnetic Moment contributions . . . . . . . . . . . . 315
22.2.6 The intermediate radiation zone . . . . . . . . . . . . . . . . . . . . . . 316
22.3 Spectral decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
22.4 Lienard–Wiechert potentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
22.4.1 Energy Momentum Tensor . . . . . . . . . . . . . . . . . . . . . . . . . 326
22.4.2 Power radiated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
22.4.3 Co–linear acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
22.4.4 Centripetal acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . 330
22.5 Angular distribution of radiation . . . . . . . . . . . . . . . . . . . . . . . . . 330
22.5.1 Co-linear motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
22.5.2 Circular motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
22.5.3 Alternative deduction for the potential . . . . . . . . . . . . . . . . . . 334

23 Quasi static fields, different time scales 339

23.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
23.1.1 Example: The lowest decay of a magnetic field on a conducting sphere. 341
23.1.2 Example: The magnetic polarizability of an isotropic conducting sphere. 341
23.1.3 Example: The skin effect on a wire . . . . . . . . . . . . . . . . . . . . 341

24 Examination Questions 2 343

8 CONTENTS
List of Figures

1.1 Set of points which influence the event (t, ~x) . . . . . . . . . . . . . . . . . . . 23

1.2 The map between solutions and initial data sets . . . . . . . . . . . . . . . . . 30
1.3 Map between true solutions and constrained initial data sets . . . . . . . . . . 31

3.1 A 90 degree rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.2 A wave seen by two observers . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.1 A loop with zero circulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.2 A pill box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 Geometrical setting for theorem 4.2 . . . . . . . . . . . . . . . . . . . . . . . 67
4.4 Point like charge outside a conducting sphere . . . . . . . . . . . . . . . . . . 71
4.5 Contour plot on the z = 0 plane of the potential of two charges as in the problem. 72
4.6 A configuration of conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.1 A conducting box with only its upper face at non-zero potential. . . . . . . . . 78
5.2 The value of the potential at z=c with 10 Fourier modes . . . . . . . . . . . . 80
5.3 The value of the potential at y=b/2 (0.5) . . . . . . . . . . . . . . . . . . . . . 81
5.4 The value of the potential at y=b/2 (5) for a wide box . . . . . . . . . . . . . 81
5.5 The value of the potential at y=5 with only 3 Fourier modes . . . . . . . . . . 81
5.6 Potential at cusps with 9 coefficients. . . . . . . . . . . . . . . . . . . . . . . . 87
5.7 Potential with 20 coefficients. . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.8 Absolute value of the real part of Y20 (θ, φ) . . . . . . . . . . . . . . . . . . . . 91
5.9 Absolute value of the real part of Y21 (θ, φ) . . . . . . . . . . . . . . . . . . . . 91
5.10 Absolute value of the real part of Y30 (θ, φ) . . . . . . . . . . . . . . . . . . . . 92
5.11 Absolute value of the real part of Y31 (θ, φ) . . . . . . . . . . . . . . . . . . . . 92
5.12 The relation among angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

7.1 Two concentric spheres and a point charge in between . . . . . . . . . . . . . . 109

7.2 The interface problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.3 Homogeneously charged ring inside a sphere at zero potential . . . . . . . . . . 114
7.4 Uniformly charge rod inside a grounded sphere . . . . . . . . . . . . . . . . . . 114

8.1 Charged sphere hanging form a spring . . . . . . . . . . . . . . . . . . . . . . 120

8.2 Charged sphere hanging form a spring . . . . . . . . . . . . . . . . . . . . . . 121
8.3 Individual molecule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.4 Molecules in a solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

9
10 LIST OF FIGURES

8.5 Charge at the center of a dielectric sphere . . . . . . . . . . . . . . . . . . . . 127

8.6 Pill box in a dielectric material . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.7 Loop in a dielectric material . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.8 Dielectric sphere on a constant external field . . . . . . . . . . . . . . . . . . . 129
8.9 Dielectric material in an external field . . . . . . . . . . . . . . . . . . . . . . . 131

9.1 Circular Current Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

9.2 Super conducting sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.3 Magnetic field: dipole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
9.4 Integrating the electric potential . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.5 Integrating the electric potential along another path . . . . . . . . . . . . . . . 143
9.6 Electric potential: path independence . . . . . . . . . . . . . . . . . . . . . . . 144
9.7 Magnetic potential: path dependence . . . . . . . . . . . . . . . . . . . . . . . 145
9.8 A current wire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
9.9 Vector potential of a wire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.10 Magnetic potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
9.11 Magnetic potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.12 Superconducting ring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.13 Magnetic Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
9.14 Inductances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
9.15 Circuit and area differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.16 The area differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

10.1 Piece of surface with superficial current distribution. . . . . . . . . . . . . . . 159

10.2 A wire distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
10.3 Boundary for V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

11.1 Hysteresis curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

11.2 Intersection of hysteresis curve and linear relation . . . . . . . . . . . . . . . . 170

13.1 The Lorentz map. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

13.2 How a square is transformed. . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

14.1 Describing the event date. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

14.2 Idealized model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
14.3 Events history of a hair. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
14.4 Aristotelian view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
14.5 Space in the Aristotelian view . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
14.6 Galilean view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
14.7 Different causal relations among events: events A and B are space-related,
events A and B are light-related, events A and D future-time-related, and
event A and E past-time-related. . . . . . . . . . . . . . . . . . . . . . . . . . 198
14.8 Sum of future directed time-like vectors . . . . . . . . . . . . . . . . . . . . . . 200
14.9 Different unit time-like vectors in a given coordinate system. . . . . . . . . . . 203
14.10Different frames and simultaneity hyper-surfaces . . . . . . . . . . . . . . . . . 203
LIST OF FIGURES 11

14.11Curve variations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

14.12Chalk description and sizes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
14.13Volume contraction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
14.14Null planes and unit vectors. The number of planes the unit vector pierced
through by the unit vector is proportional to the frequency the observer moving
along it will measure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
14.15Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
14.16Source-receiver Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
14.17Aberration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
14.18Relativistic Aberration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
14.19Summing vectors at different times. . . . . . . . . . . . . . . . . . . . . . . . . 220
14.20Scattering situation, the sums are done before and after all interactions have
finished. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

15.1 Energy conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

15.2 Moving region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.3 Unicity of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

18.1 Level surfaces of s := kµ xµ + s0 for a non-null vector k. . . . . . . . . . . . . 250

18.2 Null plane waves s := kµ xµ + s0 . . . . . . . . . . . . . . . . . . . . . . . . . . 251
18.3 The helicity plane, one dimension suppressed u ≡ ũ := u + αk. . . . . . . . . 253
18.4 Space view of plane waves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
18.5 Plane wave reflexion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

19.1 A resonant cavity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262

19.2 A wave guide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

20.1 Relation between ε and n in the complex plane. . . . . . . . . . . . . . . . . . 281

20.2 The incidence plane, 3 dimensional view. . . . . . . . . . . . . . . . . . . . . . 285
20.3 The incidence plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
20.4 EI and ET for the case where the electric field is tangent to the interface. Case
n′
n
< 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
20.5 EI and ET for the case where the electric field is tangent to the interface. Case
n′
n
> 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
20.6 Reflection coefficient as a function of the incidence angle for the case n′ > n. . 291
20.7 Reflection coefficient as a function of the incidence for the case n′ < n. . . . . 291
20.8 EI and ET for a metal in the case where the incident electric field is tangent
to the interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

21.1 Integration path for t̃ < 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

21.2 Integration path for t̃ > 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
12 LIST OF FIGURES

21.3 Measured extinction curves of F and M color centers in a KBr crystal at dif-
ferent temperatures (a); the corresponding changes of the refractive index are
computed via Kramers-Kronig relation (b). The curves correspond to 1) 250
K, 2) 170 K, 3) 100 K, and 4) 35 K, with peaks becoming sharper at low
temperature. Adapted from Peiponen et al. (1984). . . . . . . . . . . . . . . . 304
21.4 Path deformation for Einstein Causality argument. . . . . . . . . . . . . . . . 306

22.1 The Current Integral. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310

22.2 First approximation, the integration region is at the surface t = t − |~xc | . . . . . 313
22.3 Second approximation, the integration region is now along a null plane tangent
to the cone. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
22.4 Dipolar antenna configuration. . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
22.5 Dipole radiation. Power radiated by solid angle. . . . . . . . . . . . . . . . . . 319
22.6 Quadrupolar Radiation, case a = 1, b = 1. Power radiated by solid angle. . . . 320
22.7 Quadrupolar Radiation, case a = 1, b = −1. Power radiated by solid angle. . . 320
22.8 Lienard–Wiechert setting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
22.9 Different cuts where the flux integration can be carried out. . . . . . . . . . . 328
22.10Power radiated by solid angle. Approximation for small angle for γ = 2. . . . . 331
22.11Power radiated by solid angle. Approximation for small angle for γ = 10. . . . 332
22.12Power radiated by solid angle in the co-linear case, β = 0.5. . . . . . . . . . . . 332
22.13Power radiated by solid angle in the co-linear case, β = 0.9. . . . . . . . . . . . 333
22.14Power radiated by solid angle, velocity and acceleration co-linear, β = 0.4 . . . 333
22.15Power radiated by solid angle, velocity and acceleration co-linear, β = 0.9. . . 334
22.16Angles definition for the circular motion. . . . . . . . . . . . . . . . . . . . . . 335
22.17Power radiated by solid angle, case velocity perpendicular to acceleration, β =
0.5, φ = 0 (blue) and φ = π/2 (violet). . . . . . . . . . . . . . . . . . . . . . . 335
22.18Power radiated by solid angle, case velocity perpendicular to acceleration, β =
0.92 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
22.19Power radiated by solid angle, case velocity perpendicular to acceleration, β =
0.92, cut to see the φ dependence. . . . . . . . . . . . . . . . . . . . . . . . . . 336
 Part I

First Course

13
Chapter 1

Maxwell’s Equations: An overview

In this chapter we shall study general aspects of Maxwell’s equations and their solutions.
In particular we shall concentrate on their dynamical properties. For simplicity, but also
to stress that the existence of the electromagnetic fields is independent of the existence of
sources, that is, that they are an entity in their own right, we shall first consider these
equations in vacuum. For them we shall consider their initial value formulation, showing
that the prescription of the electric and magnetic field vectors at a given time, say t = 0,
~ ~x) = F~ (~x) and B(0,
E(0, ~ ~x) = G(~ ~ x) determines uniquely a solution for all future and past
times. To accomplish this we shall first make a detour into the wave equation and using the
time and space translation invariance of that equation, write down the general solution for it,
as a function of its values at a given initial time. We shall them use that formula to obtain the
general solution to Maxwell’s equations. Second we shall define the energy of electromagnetic
fields, and use its conservation in time to show that the solution obtained is the only one with
the prescribed values at t = 0.
In the last part of the chapter we shall consider sources for Maxwell equations, - from a
microscopical point of view - and depending on the nature of them, discuss to what extent
the results already found for the vacuum are still valid.

1.1 Maxwell’s equations in vacuum

In this case the equation are given by:

~
∂E ~ ∧B
= c∇ ~ (1.1)
∂t
~
∂B ~ ∧E
= −c∇ ~ (1.2)
∂t
~ ·E
∇ ~ =0 (1.3)

~ ·B
∇ ~ =0 (1.4)
There are several remarks to be made:

15
16 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

1. The first two equations include time derivatives, in fact tell us what is the instantaneous
~ and B
time evolution of E ~ as a function of the values of their space derivatives at the
present time. They shall be called the evolution equations.
2. The other two are equations relating different space derivatives of the fields at the same
time, that is they constrain the possible value the electric and magnetic fields can have
at any given time, and so shall be called the constraint equations.
3. The equations are linear, that is if (E ~ 1, B
~ 1 ) and (E
~ 2, B
~ 2 ) are two solutions, then (E
~1 +
αE~ 2, B
~ 1 + αB
~ 2 ), where α is any constant, is also a solution. This property is very useful
since it allows finding complicated solutions as sum of simpler ones.
4. There is an asymmetric under interchange of E ~ and B ~ on the evolution equations, [a
sign difference between 1.1 and 1.2]. This asymmetric on the evolution equations is
crucial: Without it they would be inconsistent, not just because they would give a
different evolution than the one observed, but because they would – for generic values
of E~ and B~ at any given initial time – give no evolution at all! We shall come back to
this point later in this chapter. This asymmetry between E ~ and B~ is different from the
one that appears in the constraint equations when sources are present, reflecting the
fact that no magnetic charges have been observed in nature.
5. In the equations there is a constant, c, which has the dimensions of a speed. The
presence of this constant has deep consequences on our present conception of space and
time, but we shall discuss that problem later in the book.
6. If we count the unknowns we see we have six, the three components of E ~ and three of
~
B. But we have 8 equations, 6 evolutions + 2 constraints. So, in order for the system
not to be over-determined two of them should be consequences of the other six. This is
indeed the case, but in a very subtle way. We shall see this later in the chapter.

1.1.1 Maxwell’s evolution equations

We shall forget the last remark above and consider, for the moment, just the two evolution
equations, 1.1 and 1.2.
Fix an initial time t = 0, say 1 , and assume we are given there the values of E ~ and
~ ~ ~ ~ ~
B ; E(0, ~x) = F (~x), B(0, ~x) = G(~x). Using the evolution equations one could device the
following procedure to solve them: we could compute the values of E~ and B
~ an instant later,
∆t. Indeed, using a Taylor expansion in t, and equation (1.1) we have,

~
~
E(∆t, ~ ~x) + ∂ E (t, ~x) |t=0 ·∆t + O((∆t)2 )
~x) = E(0,
∂t
~
= E(0, ~x) + c∆t∇~ ∧ B(0,
~ ~x) + O((∆t)2 ),

and correspondingly
1
The origin of time is irrelevant for this discussion as will become apparent in chapter II.
1.1. MAXWELL’S EQUATIONS IN VACUUM 17

~
B(∆t, ~ ~x) − c∆t∇
~x) = B(0, ~ ∧ E(0,
~ ~x) + O((∆t)2 ).

The symbol O((∆t)2 ) means the presence of extra terms, which are bounded and go to
zero as (∆t)2 when ∆t → 0.
Once we have the values of E ~ and B ~ at t = ∆t (to that given order) we can take their
space derivatives at t + ∆t and repeat the above argument to obtain the values of E ~ and B ~
at t = 2∆t and so on until obtaining a “solution" for a given interval of time T .
Of course the solution is only an approximated one, but we could think –in analogy with
ordinary differential equations– of improving on it by taking smaller and smaller ∆’s and
more and more of them as to keep reaching T in the last step.
One would hope, in the limit ∆t → 0, to obtain a unique solution to the evolution equations
~ ~x) and B(0,
corresponding to the given initial data, E(0, ~ ~x) and which depends continuously
under arbitrary variations on that initial data. This argument is in general misleading, for in
each step we must take spatial derivatives. If we trace the formulas to the bare dependence on
~ and B
initial data we see that if we are taking n steps to reach T , then the values of E ~ there
depend on n space derivatives of the initial data. So in the limit ∆t → 0 it would depend on
an infinity of derivatives of the initial data values. One could have then the situation were
very bumpy, but tiny variations, in, say, the 1079th. derivative of E~ would significantly affect
the value of E~ at a later time. One would consider such a situation very unphysical and
should be prepare to through away any theory with such a pathology. It can be shown that
for electromagnetism this awkward situation does not appear, and that this happens because
very precise cancellations occur due to the asymmetry already mentioned of the evolution
equations 2 .
For the problem at hand one does not need to use an argument along the above lines to
show the existence of solutions, for one can find the general solution to the problem, which
we shall display in the proof of the next theorem.

Theorem 1.1 (The Cauchy Problem for electromagnetism) Given (F~ (~x), G(~ ~ x)),
3 3 ~ ~x), B(t,
smooth vectorial functions in space (lR ). There exists a unique smooth solution (E(t, ~ ~x))
to Maxwell’s vacuum evolution equations, 1.1 and 1.2, satisfying

~ ~x), B(0,
(E(0, ~ ~x)) = (F~ (~x), G(~
~ x))

Furthermore the solution depends continuously on the initial data.

2
With the opposite sign on one of Maxwell’s equations, that is, with the equations symmetric under
interchange of E ~ and B,~ one can construct examples of initial data for which, one has a solution for all
times and a nearby arbitrarily close has a solution only for a limited time span, as small as one wishes. The
mathematical theory which justify the existence argument given above for electromagnetism, and for all other
classical theories of physics, is the theory of symmetric hyperbolic systems and is one of the greatest scientific
achievements of this century.
3
“smooth" here means an infinite differentiable function. We shall assume - when possible - all our functions
to be smooth, not only for simplicity, but also because physically we can not distinguish between an function
say 3 times differentiable and another, say, 7 times differentiable.
18 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

This class of theorem 4 is without doubt one of the more important and basic results of all
physics. It gives ground to our causality conceptions about physical phenomena, and so to the
experimental method in sciences. It says that if we know “the present", that is the initial data
at a given time we can predict the future (the corresponding unique solution to that data) and
that it is a continuous function of the present, that is, if we change things at present by an
small amount the change this will in turn produce in the future shall be correspondingly small.
In particular measurement errors at present, if sufficiently small, would result in small errors
on our predictions about the future. About the experimental method as basic to sciences, it
says that if we carefully prepare an experiment, we can then predict its outcome.
Proof: As already mentioned the proof of this theorem will not follow the argument given
earlier in the chapter, for this would require a fair amount of previous mathematical steps
which are not the subject of this course and which although very interesting would take us
into a long detour. We shall follow instead a much more direct approach, which, although
would take us into some detours -but always within the main subject of the course-, it would
give us also, as a by product, the general solution to Maxwell’s equations! This method will
have the disadvantage that it would be only applicable to the vacuum equations and to some
very particular type of sources, but not to the most general type of problem encountered in
electromagnetism, let alone the rest of physics!
We shall split the proof into two Lemmas. We shall first find a solution to the problem
and second show that this is the solution, e.i. show uniqueness.

Lemma 1.1 Existence

Given (F~ (~x), G(~
~ x)) smooth vector functions in lR3 , then:

~ ~x) = ∂
E(t, (tMct (F~ (~x))) + ctMct (∇ ~ ∧ G(~
~ x))
∂t Z
t
~
− ∇( ~ · F~ (~x))]cdt̃)
[ct̃Mct̃ (∇
0

~ ~x) = ∂ ~ x))) − ctMct (∇ ~ ∧ F~ (~x))

B(t, (tMct (G(~
∂t Z
t
~
− ∇( ~ · G(~
[ct̃Mct̃ (∇ ~ x))]cdt̃),
0

satisfy Maxwell’s evolution equations, 1.1 and 1.2 and furthermore;

~ ~x) = F~ (~x)
E(0,
~ ~x) = G(~
B(0, ~ x)

In the above formulae:

4
The corresponding type of theorem in classical mechanics is the one which asserts the existence and
uniqueness for all times to certain Lagrangian systems. In this case the mathematical theory underneath
these results is the one of ordinary differential equations.
1.1. MAXWELL’S EQUATIONS IN VACUUM 19

1 Z
Mt (f (~x)) =
f (~x + tn̂) dΩ,
4π Ω
where Ω is the sphere of unit radius, that is,

1 π 2π
Z Z
Mt (f (~x)) = f (x + t cos θ sin ϕ, y + t cos θ cos ϕ, z + t sin θ) sin θdθdϕ.
4π 0 0

Proof: This Lemma can be proven by brute force, namely applying directly the equations to
the proposed solution, but here me propose another method which will teach us some things
about the wave equation. Taking a time derivative of (1.1) and using (1.2) we obtain,

1 ∂2E~ 1 ~ ∂B ~
= ∇ ∧
c2 ∂t2 c ∂t
~
= −∇ ∧ (∇~ ∧ E)
~
= ∆E~ − ∇(
~ ∇ ~ · E),
~

where in the third step we have used the vectorial calculus identity:

~ ∧ (∇
∇ ~ ∧V
~ ) = ∇(
~ ∇~ ·V
~ ) − (∇
~ · ∇)
~ V~
.

Exercise: Show the above identity.

At this point we could use equation 1.3 to drop the second term in the r.h.s. of the above
equation, but since at the moment we are only solving for the evolution equations, we prefer
to eliminate it by taking another time derivative.
~
Taking it and calling Y~ = ∂ E we obtain,∂t

1 ∂ 2 Y~
= ∆Y~ − ∇(
~ ∇~ · (c∇
~ ∧ B))
~
c2 ∂t2
= ∆Y~
~ to substitute in the second
where in the first step we have use the evolution equation for E
term on the r.h.s. Y~ by c∇ ~ ∧ B,
~ and in the second step the vectorial calculus identity
~ ~ ~
∇ · (∇ ∧ A) = 0.

Exercise: Show the above identity.

We see then that Y~ satisfies the wave equation,

1 ∂2
( 2 2 − ∆)Y~ = 0. (1.5)
c ∂t
20 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

If we express Y~ and the above equation in Cartesian coordinates it is easy to see that each
individual component of Y~ , Y k , satisfies the scalar wave equation:

1 ∂2 ij ∂2 k 1 ∂2 ∂2 ∂2 ∂2
( − δ )Y = ( − − − )Y k = 0,
c2 ∂t2 ∂xi ∂xj c2 ∂t2 ∂x2 ∂y 2 ∂z 2
where by Y k we are denoting the Cartesian component of Y~ along the axis, ~xk , and
(
ij 1 i=j
δ =
0 i=
6 j
denotes the Euclidean metric of lR3 and we are using the Einstein’s notation, by which repeated
indices, one up one down, imply summation.
The scalar wave equation is the small brother of Maxwell’s equation and we shall have the
opportunity to encounter it many times during this course, so it is convenient to make the
following:

Detour into the wave equation

The wave equation shares many property with Maxwell’s equations. Besides having also
the constant c appearing on it, and been linear, [that is linear combinations of solutions are
also solutions], it is invariant under time and space translations 5 . That is, if ϕ(t, ~x) is a
solution, then ϕT (t, ~x) = ϕ(t − T, ~x) and φ~x0 (t, ~x) = φ(t, ~x − ~x0 ), where T is a constant
number and ~x0 a constant vector, are also solutions.

Exercise: Show that this invariance plus linearity implies that if φ is a smooth solution, then
∂φ ~ also satisfy the wave equation.
and ∇φ
∂t

To see that the wave equation has these invariances consider the wave equation applied
to φT (t, ~x). To apply it we need to compute ∂φ
∂t
T
in terms of φ,

∂φT (t, ~x) ∂φ(u, ~x) ∂u ∂φ

= = (u, ~x)
∂t ∂u ∂t ∂u
where we have defined u ≡ t − T . Since this change does not affect the space dependences we
get,

1 ∂2 1 ∂2
−(∆)φ T (t, ~
x ) = ( − ∆)φ(u, ~x) = 0,
c2 ∂t2 c2 ∂u2
proving that φT is also a solution. The case of space translations is similar using the trick of
aligning one of the coordinate axis with the space translation under consideration.

Exercise: Do the space-translation case.

We shall make use of this invariance to find the general solution to the wave equation.
5
We shall study this type of invariance for Maxwell’s equations in the next chapter.
1.1. MAXWELL’S EQUATIONS IN VACUUM 21

The general solution to the wave equation

In spherical coordinates the wave equation becomes:

1 ∂2φ 1 ∂2 1 ∂ ∂φ 1 ∂2φ
− (rφ) + (sin θ ) + =0
c2 ∂t2 r ∂r 2 r 2 sin θ ∂θ ∂θ r sin2 θ ∂φ2

Exercise: Show this by direct calculation using the chain rule.

So if we look for spherically symmetric solutions, that is solutions which only depend on
t and r the equation reduces to:

1 ∂ 2 φ 1 ∂ 2 (rφ)
− =0
c2 ∂t2 r ∂r 2
which can be further simplified, by defining. ψ(t, r) = rφ(t, r), to

1 ∂2ψ ∂2ψ
− 2 = 0,
c2 ∂t2 ∂r
that is, a one dimensional wave equation. A particular solution to this equation is ψ(t, r) :=
F (ct − r), where F is any sufficiently smooth function.

Exercise: Show that ψ(t, r) := G(ct + r), with G an arbitrary smooth function is also a
solution.

Thus we obtain a solution to the three dimensional wave equation, φ(t, r) = F (ct−r)r
, at all
points besides the origin, which represents a spherical wave going away from the coordinate
axis and meanwhile decaying in intensity as 1r if F is taken to be of compact support. Note
that this solution will, in general, be unbounded at the origin. This can be arranged by
instead choosing alternatively the solution φ(t, r) = F (ct−r)−F
r
(ct+r)
. The problem at the origin
will not be important for our present application.
Because of the translational invariance we can take any coordinate origin we please and
get a solution of this type. Of course their sum would also be a solution and so we conclude
that

X F (ct − |~x − ~xj |)

φ(t, ~x) = φ̃xj
j |~x − ~xj |

is also a solution, where φ̃xj is a weight constant for each “symmetry center". One can picture
this type of solutions by throwing a hand full of marbles into a lake, roughly, the weight factors
there are related to the actual weight of each marble. The picture is not totally correct, for
those waves decay at a different rate with the distance.
There is no reason not to consider a smooth distribution of these symmetry centers and
so passing from the sum to an integral we conclude that
22 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

Z
φ̃(~x′ )F (ct − |~x − ~x′ |) 3 ′
φ(t, ~x) = d ~x ,
lR3 |~x − ~x′ |
is also a solution to the wave equation.
Using spherical coordinates centered at ~x, ~x′ = ~x + rn̂, so that d3~x′ = dΩ2 r 2 dr the integral
becomes,

φ̃(~x + rn̂)F (ct − r)

Z
φ(t, ~x) = 3
dΩ2 r 2 dr,
lR r
We now further specialize this solution by first taking as F ,
(
1
0≤s≤ε
Fε (s) = ε4π
0 s ≤ 0, ε ≤ s,

noticing that when ε → 0 the function becomes more and more concentrated at s = 0 while
its integral remains finite,
1
Z ∞
Fε (s)ds = .
0 4π
Thus, in the above integral the values which contribute are more and more concentrated at
the values of r for which ct − r is near zero, namely r = ct. Taking such a limit, we reach the
conclusion that the following is also a solution,

ct
Z
φ(t, ~x) = φ̃(~x + ctn̂)dΩ
4π Ω
:= ctMct (φ̃).

where now the integral is on the sphere of unit directions Ω.

So far we have found, using the invariance under space translation, a large set of solutions:

Lemma 1.2 Given any function φ̃ on ℜ3 ,

φ(t, ~x) = ctMct (φ̃).

is a solution to the wave equation.

But these are still very particular ones, for at t = 0 they all vanish. To enlarge that set we
consider the bigger set obtained by adding to the ones found the sets of all its time derivatives.
We now claim:
Any solution of the wave equation can be written as

∂
φ(t, ~x) = tMct (φ1 (~x)) + (tMct (φ0 (~x))),
∂t
∂φ
noticing that ∂t
(t, ~x) |t=0 = φ1 (~x) and φ(0, ~x) = φ0 (~x).
1.1. MAXWELL’S EQUATIONS IN VACUUM 23

We have already shown that the first term is a solution. The second term is also a solution
for time translational invariance plus linearity implies that any time derivative of a solution
is also a solution.

Exercise: Show the statement about the values of φ0 (~x) and φ1 (~x).

But any solution of the wave equation can be expressed in the above form, since any two
solutions which coincide and have the same time derivatives at t = 0, necessarily also coincide
in the whole space. We shall not proof that this statement, since, although the proof is not
difficult, it is very similar to the one we shall give for Maxwell’s equations, and we shall not
need it in what follows. Notice that since all solutions have this form we can “classify" them
by listing the values and the values of their time derivatives at a t = const. surface. These
values constitute the initial data set of the equation, and we have just asserted that to each
element on this set, that is, a pair (φ0 (~x), φ1 (~x)), there corresponds a unique solution to the
equation.
The form of the general solution we have found underlines an important property of the
wave equation which it is also shared by Maxwell’s equations, namely that the value of a
solution at a point ~x at time t depends only on the value of the solution at the points ~x′ at
t = 0 such that |~x − ~x′ | = ct, that is, the waves travel at the speed c.
(t, ~
x)

|~ x′ =ct
x−~

t=0

Figure 1.1: Set of points which influence the event (t, ~x)

The diagram above represents points of space and time, usually called events, horizontal
displacements mean displacements along space, vertical ones along time. The points drawn
represent those points in space time which can influence the value of any solution of the wave
equation at the point (t, ~x). It is clearly a cone,
C − (t, ~x) = {(t′ , ~x′ ) | |~x − ~x′ | = c(t − t′ )}

Exercise: Let (φ0 (~x), φ1 (~x)) of compact support, show that given any point ~x0 there exists
a time t0 such that for all t > t0 , φ(t, ~x0 ) = 0.
24 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

Exercise: Find all the points in space-time which can be influenced from data given at
(t = 0, ~x0 ).

We finish this detour into the wave equation giving a direct proof of Lemma 1.2:
Given any smooth function φ̃(~x) : R3 → R, the function R3 × R → R,
φ(t, ~x) := t Mt (φ̃(~x))
satisfies the wave equation with initial data (φ(0, ~x) = 0, ∂t φ(t, ~x)|t=0 = φ̃(~x)).
Proof:
First notice that

1
Z
∂t Mt (φ̃(~x)) = ∂t φ̃(~x + tn̂) dΩ
4π ZS 2
1 ~ φ̃(~x + tn̂) · n̂ dΩ
= ∇
4π SZ2
1 ~ φ̃(~x + tn̂) · n̂ t2 dΩ
= ∇ (1.6)
4πt S
2 2

The last term above is just an integral on a sphere of radius t of the gradient of a function
normal to the surface, so we can use Gauss theorem to conclude that:

1
Z
∂t Mt (φ̃(~x)) = ∆φ̃(~x + tn̂) dV
4πt2 Bt (~x)
1
Z
= ∆x φ̃(~x + tn̂) dV
4πt2 Bt (~x)
1
Z tZ
= ∆x φ̃(~x + τ n̂) dΩ τ 2 dτ
4πt2 0 S 2
1 t
Z
= 2 ∆x Mτ (φ̃(~x) τ 2 dτ (1.7)
t 0
where Bt (~x) is a ball of radius t centered at ~x, in the second line we change to the Laplacian
with respect to ~x (just a coordinate change), and in the third line we had written the integral
in spherical coordinates. Finally we re-write the expression in terms of spherical means. Thus,
multiplying by t2 and taking a derivative we get,

∂t (t2 ∂t Mt (φ̃(~x)) = t2 ∆Mt (φ̃(~x))

or, in terms of the solution, Mt (φ̃(~x)) = φ(t, ~x)/t,

t∆φ(t, ~x)) = ∂t (t2 ∂t (φ(t, ~x)/t)),

but then,
t∆φ(t, ~x)) = ∂t (t2 (∂t φ(t, ~x)/t − φ(t, ~x)/t2 )
= ∂t (t∂t φ(t, ~x) − φ(t, ~x))
∂ 2 φ(t, ~x)
= t
∂t2
1.1. MAXWELL’S EQUATIONS IN VACUUM 25

from which the proposition follows.

The wave equation with sources,

1 ∂ 2 φ(t, ~x)
− ∆φ(t, ~x) = 4πf (t, ~x)
c2 ∂t2
has the general solution,

∂
φ(t, ~x) = (tMct (φ0 (~x))) + tMct (φ1 (~x))
∂t Z
t
+ 4πc2 t̃Mct̃ (f (t − t̃, ~x))dt̃, (1.8)
0

∂φ
with φ(0, ~x) = φ0 (~x), ∂t
(t, ~x) |t=0 = φ1 (~x).

Exercise: Show that the inhomogeneous term can be re-written, via a change of variables,
as
Z tc Z
4π rf (t − r/c, ~x + rn̂)drdΩ (1.9)
0 Ω
Draw the integration region.

Exercise: Check that in one dimension, the situation is different. Namely that the inhomo-
geneous solution to the corresponding wave equation,
1
∂tt φ − ∂rr φ = 4πf (t, r) (1.10)
c2
is,
Z tZ r+ct̃
φI = 2πc f (t̃, r̃) dr̃dt̃ (1.11)
0 r−ct̃

Draw the integration region.

Exercise: NOT AT ALL EASY! We have seen that the first two terms are solutions to the
homogeneous (e.i. without sources) equations, with the correct initial data. Show that the
third term is a solution to the inhomogeneous (e.i. with sources) wave equation with zero
initial data.

Supplementary discussion: Duhamel’s principle

The previous formula can be also deduced from what it is called Duhamel’s principle. This
very general principle applies to the case when one has an equation of the form,

du
= Au,
dt
26 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

where u is a vector in some space and A a linear operator independent of t. The solutions
are then of the form,
u(t) = Pst (φ(s))
That is, this is the unique solution with initial data φ(s) at time t = s, where the operator
Pst satisfies the equation,

d t
P (φ(s)) = APst (φ(s)), with initial condition Ptt = id
dt s
We can use this solution representation to find the solutions to the inhomogeneous equation,

du
= Au + f,
dt
with homogeneous initial data. They are given by,
Z t
v(t) = Pst (f (s)) ds. (1.12)
0

Indeed,

d t d t
Z
v = Ptt (f (t)) + P (f (s)) ds
dt 0 dt s
Z t
= f (t) + APst (f (s)) ds
0
Z t
= f (t) + A Pst (f (s)) ds
0
= f (t) + Av(t).

For the case at hand, we can write the wave equation as a first order system as,
! ! ! !
∂ φ 0 1 φ 0
= 2 +
∂t π c ∆ 0 π 4πc2 f
For the homogeneous equation we have,
! !
φ(t) ∂t (tMct (φ(0))) + tMct (π(0))
u(t) = P0t (u(0)) = = .
π(t) ∂t (∂t (tMct (φ(0))) + tMct (π(0)))

Thus, applying the above result we get, (the source f has to be replaced for π),
Z t Z t
2 2
φ(t) = 4πc (t − s)Mc(t−s) (f (s)) ds = 4πc τ Mcτ (f (t − τ )) dτ,
0 0

and we get the above result, (1.8).

End of detour
1.1. MAXWELL’S EQUATIONS IN VACUUM 27

We now continue with the proof of our lemma. Since each Cartesian component of Y~ (t, ~x)
satisfies the wave equation we can readily write down some solutions:
∂
Y~ (t, ~x) = (tMct (Y~ 0 (~x))) + tMct (Y~ 1 (~x))
∂t
Y~ (t, ~x)|
with Y~ 0 (~x) ≡ Y~ (0, ~x) and Y~ 1 (~x) ≡ ∂∂t t=0
Using now the evolution equations, 1.1, 1.2, we obtain,

~
∂E
Y~ 0 (~x) = Y~ (0, ~x) = (t, ~x) |t=0 = c∇ ~ ∧ B(t,
~ ~x) |t=0
∂t
= c∇~ ∧ G(~~ x),
∂ Y~ ∂2E ~ ~
~ ∧ ∂ B (t, ~x) |t=0
Y~ 1 (~x) = (t, ~x) |t=0 = (t, ~
x ) | t=0 = c ∇
∂t ∂t2 ∂t
2~ ~ ∧ E(t,
~ ~x) |t=0 = −c ∇ 2 ~ ∧∇ ~ ∧ F~ (~x),
= −c ∇ ∧ ∇
And so,
∂ ~ ∧ G(~
Y~ (t, ~x) = (ctMct (∇ ~ x))) − c2 tMct (∇
~ ∧∇
~ ∧ F~ (~x)).
∂t
But by definition, Z t
~ ~x) = E(0,
E(t, ~ ~x) + Y~ (t̃, ~x)dt̃,
0
and so,
~ ~x) = F~ (~x) + ctMct (∇
E(t, ~ ∧ G)(~
~ x))
Z t
− c ~ ∧∇
ct̃Mct̃ (∇ ~ ∧ F~ (~x))dt̃.
0

~ ~x) we integrate in time equation 1.2,

To obtain B(t,

Z t
~ ~x) = G(~
B(t, ~ x) − c ~ ∧ E(
∇ ~ t̃, ~x)dt̃
0
Z t
~ x) − c
= G(~ ~ ∧ F~ (~x) − ct̃Mct̃ (∇
[∇ ~ ∧∇
~ ∧ G(~
~ x))
0
Z t̃
− c ˜ ˜(∇
ct̃M ~ ∧∇
~ ∧∇ ˜ t̃
~ F~ (~x))dt̃]d
ct̃
0

The following two exercises complete the proof of the lemma.

Exercise: Show, using the identity ∇ ~ ∧∇~ ∧ F~ (~x) = −∆F~ (~x) + ∇(

~ ∇~ · F~ (~x)) that the above
~
formula for E(t, ~x) can be further reduced to:

~ ~x) = ∂
E(t, (tMct (F~ (~x))) + ctMct (∇ ~ ∧ G(~
~ x))
∂t Z
t
~
− ∇( ~ · F~ (~x))]cdt̃)
[ct̃Mct̃ (∇ (1.13)
0
28 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

Hint: Use that t̃Mct̃ (F~ (~x)) satisfies the wave equation.

Exercise: Show, using the identity ∇ ~ ∧∇ ~ ∧∇

~ ∧ F~ (~x) = −∆(∇
~ ∧ F~ (~x)), that the above
~ ~x) can be further reduced to:
formula for B(t,

~ ~x) = ∂ ~ x))) − ctMct (∇ ~ ∧ F~ (~x))

B(t, (tMct (G(~
∂t Z
t
~
− ∇( ~ · G(~
[ct̃Mct̃ (∇ ~ x))]cdt̃). (1.14)
0

Alternatively, run again the above procedure using this time Y~ = ∂ ~

∂t
B.

To complete the proof of the theorem we must prove now that the general solution we
have obtained are the only possible.

Lemma 1.3 Uniqueness

~ 1 (t, ~x), B
Let (E ~ 1 , (t, ~x)) and (E
~ 2 (t, ~x), B
~ 2 (t, ~x)) be two smooth solutions of Maxwell’s evo-
lution 1.1 and 1.2, such that:

1. Their initial data coincide at some t = t0 , that is

~ 1 (t0 , ~x), B
(E ~ 1 (t0 , ~x)) = (E
~ 2 (t0 , ~x), B
~ 2 (t0 , ~x)).

2. Their difference decay to zero sufficiently fast at large distances 6 .

Then they are identical.

Proof: To see this we consider their differences:

~ 1 (t, ~x) − E
E(t, ~x) = E ~ 2 (t, ~x),
~ 1 (t, ~x) − B
B(t, ~x) = B ~ 2 (t, ~x).

Since Maxwell’s equations are linear these differences are also a solution, and
~ 0 , ~x) = B(t
E(t ~ 0 , ~x) = 0.

So, because the linearity of the equations, the proof reduces to show that the only solution
with vanishing initial data is the zero solution. If these fields decay sufficiently fast at infinity,
then the following integral is finite:
1
Z
E(t) := ~ ~x) · E(t,
{E(t, ~ ~x) + B(t,
~ ~x) · B(t,
~ ~x)}d3~x.
2 lR3
6
This last requirement can be lifted away using an argument involving the finite velocity of propagation of
the electromagnetic waves, but the proof would become too complicated.
1.1. MAXWELL’S EQUATIONS IN VACUUM 29

Taking a time derivative and using the evolution equations we find:

d
Z
~ ~x) · ∇
E(t) = c 3 {E(t, ~ ∧ B(t,
~ ~x) − B(t,
~ ~x) · ∇
~ ∧ E(t,
~ ~x)}d3~x.
dt lR
~ · (V
Using now the vectorial calculus identity: ∇ ~ ∧W
~)=W
~ · (∇
~ ∧V
~)−V
~ · (∇
~ ∧W
~ ),
and Gauss theorem, we see that

d
Z
E(t) = lim ~ ~x) ∧ B(t,
−c(E(t, ~ ~x)) · n̂ dS
dt r→∞ S 2 (r)

where the integral is over a sphere of radius r, and n̂ is the outward unit normal to it. The
limit is taken after making the integral.
~ and B
If the field differences E ~ decay sufficiently fast at infinity, then in the limit the
surface integral becomes zero and so E(t) is conserved, E(t) = E(t0 ). But at t = t0 E ~ and B ~
were zero, so that E(t0 ) = 0 and so we conclude that E(t) = 0 ∀t. Since E(t) is the integral of
a sum of positive definite terms we conclude that each one of them must vanish point wise 7
and so we have
~ ~x) = B(t,
E(t, ~ ~x) = 0

This concludes the proof of the lemma and so of the theorem.

Exercise: Prove a similar Lemma for the wave equation using as the energy functional,

1 ∂φ
Z
E(t) = ~ · ∇φ}d
{( )2 + ∇φ ~ 3
~x.
2 lR 3 ∂t

One asks oneself what is behind such a neat proof of uniqueness? Usually simplicity
and beauty are correlated with some deep aspect of nature. In this case it happens that the
integral which so conveniently helped us to prove uniqueness is in fact the total energy carried
~ ~x), B(t,
by the electromagnetic configuration (E(t, ~ ~x)), and the fact that it is constant in time
is nothing else than energy conservation. The surface integral is nothing else but the energy
by unit of time radiated away from the region, and its integrand, S ~ = cE ~ ∧ B,
~ called the
Poynting vector, is the flux of energy.

Exercise: Give an argument, using the fact that the solutions propagate at finite speed to
show, first the if the initial data vanishes in, say, inside ball of radius R, then after some
interval δt it vanishes in a ball of radius R − cδt, and second that the requirement of the fields
decaying at infinity of the previous result is superfluous.

~ ~x), B(t,
What have we found so far? Consider the set, S, of all pairs of vectors (E(t, ~ ~x)),
defined at all points of space and for all times. This set has a subset, SE, which consists of
all pairs which satisfies the evolution Maxwell’s equations. If we evaluate those pairs in SE
7
Since we are assuming all fields are smooth.
30 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

at t = 0, we see that we have a map, Φ, from SE to the set of “free initial data", F ID, that
is the set of all pairs of vector (F~ (~x), G(~
~ x)) defined at point of lR3 , the map given by

~ ~x), B(t,
Φ((E(t, ~ ~x))) = (E(0,
~ ~x), B(0,
~ ~x)) = (F~ (~x), G(~
~ x)).

What we have shown so far is that this map is one to one, and so invertible. That is, each
free initial data pair (F~ (~x), G(~
~ x)) gives rice to a unique solution to the evolution equation.
Solutions are uniquely characterized by their initial data. To speak of a solution or of its
initial data is completely equivalent.

S
FID

SE

Figure 1.2: The map between solutions and initial data sets

1.1.2 The Constraint Equations

What happens with the other pair of Maxwell’s equations, namely the constraint equations?
In analogy with the picture above the solutions of the constraint equations form a subset, SC,
~ ~x), B(t,
of S, namely the subset of all pairs (E(t, ~ ~x)) such that ∇ ~ · E(t,
~ ~x) = ∇ ~ · B(t,
~ ~x) = 0.
The solution to the whole set of Maxwell’s equation is clearly the intersection P = SE∩SC.
In this sense the question is now: How big is P ? 8 Can we characterize P as a subset of F ID?
Note that the obvious subset of F ID for this characterization is ID := {Set of all pairs
~ x)) such that, ∇
(F (~x), G(~ ~ · F~ (~x) = ∇
~ · G(~
~ x) = 0}, for at least the intersection can not be
bigger than this. We express the answer to this question in the following theorem:

Theorem 1.2 P is uniquely characterized by ID. That is, if the initial data satisfies the con-
straint equations, then the solution to the evolution equations they give rise to, automatically
satisfy the constraint equations for all times.
1.1. MAXWELL’S EQUATIONS IN VACUUM 31

F ID
ID

P SE SC

Figure 1.3: Map between true solutions and constrained initial data sets

Proof: Assume we are given any pair in ID, that is a pair (F~ (~x), G(~
~ x)) such that ∇
~ · F~ (~x) =
~ · G(~
∇ ~ x) = 0. We want to prove that the solution to the evolution equations this pair gives
~ ~x), B(t,
rise to,9 (E(t, ~ ~x)) satisfies the constraints equations for all times.
Taking a time derivative to each one of the constraint equations and using the evolution
equations we have,

∂ ~ ~ ~ · ∂ E(t,
~ ~x) = c∇
~ · (∇
~ ∧ B(t,
~ ~x)) = 0,
(∇ · E(t, ~x)) = ∇
∂t ∂t
∂ ~ ~ ~ · ∂ B(t,
~ ~x) = −c∇~ · (∇
~ ∧ E(t,
~ ~x)) = 0,
(∇ · B(t, ~x)) = ∇
∂t ∂t
~ · (∇
where we have used the vector calculus identity, ∇ ~ ∧ A)
~ = 0. Therefore,

~ · E(t,
∇ ~ ~x) = ∇
~ · E(0,
~ ~x) = ∇
~ · F~ (~x) = 0,
~ · B(t,
∇ ~ ~x) = ∇
~ · B(0,
~ ~x) = ∇
~ · G(~
~ x) = 0

Remark: From this result we see that the name “constraint equations" is justified. These
equations relate different components of the initial data, that is, it is not possible to prescribe
arbitrarily the three components of the electric and the three components magnetic field as
initial data for Maxwell’s equations. At most one could prescribe two for each one of them,
the rest been determined from the ones given. But this counting of freely given components
is only approximate, as the following exercise shows.

Exercise: Let V~ (~x) and W ~ (~x) two arbitrary vector functions in lR3 . Show that the pair
~ ∧V
(∇ ~ (~x), ∇
~ ∧W
~ (~x)) is in ID.
8
P is not empty because the zero pair bellow to it.
9
Whose existence and uniqueness has been already shown.
32 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

The above example also shows that there are plenty of elements of P . These solutions
represents “pure radiation fields", since they do not arise from any source whatsoever.

Exercise: Show by direct calculation that (E(t, ~ ~x), B(t,

~ ~x)), as given by (1.13,1.14) have
zero divergence if their initial data satisfies the constraints.

Exercise: Find expressions (1.13,1.14) again, this time using that the divergences cancell.
Hint: Check that now both E,~ and B~ satisfy the wave equation and write the general solutions
for them from the initial data. You will need to use equations (1.1,1.2).

1.2 Initial Value Formulation II: Sources

What happens with Maxwell’s equations when sources are present?
In this case the equations are:

~
∂E ~ ∧ B(t,
(t, ~x) = c∇ ~ ~x) − 4π J(t,
~ ~x) (1.15)
∂t

~
∂B ~ ∧ E(t,
(t, ~x) = −c∇ ~ ~x) (1.16)
∂t

~ · E(t,
∇ ~ ~x) = 4πρ(t, ~x) (1.17)

~ · B(t,
∇ ~ ~x) = 0, (1.18)

~ ~x) its current.

where ρ(t, ~x) represents a charge distribution density and J(t,
Remark: Only a pair of equations, 1.15, 1.17, changes, the other two, 1.16, and 1.18, remain
unchanged. We shall see later that this splitting of the equations is also natural, as we shall
see later when we study the four dimensional covariance of the theory. It can be seen that the
two “unaltered" equations are integrability conditions for the existence of a four dimensional
vector potential. Although such a potential is considered just an “auxiliary tool" in dealing
with the equations of classical electromagnetism, it becomes a key ingredient when coupling
electromagnetism to quantum fields.

1.2.1 Charge Conservation

The two equations having sources, 1.15, and 1.17, imply that the total charge must be constant
in time, even more, they imply a “continuity" equation for the sources, namely ∂ρ +∇~ · J~ = 0.
∂t
1.2. INITIAL VALUE FORMULATION II: SOURCES 33

To see this take the divergence of 1.1 and substract from it the time derivative of 1.17,
∂ ~ ~ ~ · ( ∂ E)
~
0 = (∇ · E) − ∇
∂t ∂t
∂ρ ~ ~ ∧B ~ − 4π J)
~
= 4π − ∇ · (c∇
∂t
∂ρ ~ ~
= 4π( + ∇ · J),
∂t
where in the second step we have used the vector identity, ∇~ · (∇
~ ∧ A) ≡ 0.
Thus if this relation between the charge density and the charge current is not satisfied,
then there can not be solutions to Maxwell’s equations.
The continuity equation has a simple interpretation, it just says that charge can not
just disappear from a given volume, if it decreases there, it has to be because it leaves the
boundaries of that volume as a current thought it boundaries. Indeed taking a fix volume, V ,
the total charge enclosed on it is,
Z
QV (t) = ρ(t, ~x)d~x3 ,
V

and so its time derivative is,

dQV ∂ρ(t, ~x) 3
Z
(t) = d~x
dt VZ ∂t
= − ∇ ~ · J(t,
~ ~x)d~x3
ZV
= − J~ · n̂dS 2 .
∂V

where after using the continuity equation we have used Stoke’s theorem. ∂V denotes the
boundary of V , and n̂ its outer normal. Thus, the sources of Maxwell’s equations can not be
anything we like, they must have a “material" entity in the sense that if their amount change
in a given volume then a correlated flux of the quantity must be present at the boundary of
it. In particular we have global charge conservation:

Lemma 1.4 If the sources have compact support along space directions, then the total charge
is constant.

Indeed, defining the total charge as,

Z
Q(t) ≡ 3
ρ(t, ~x)d~x3 ,
lR
which is well defined since ρ(t, ~x) is assumed of compact support, we have,
dQ
Z
(t) = −limr→∞ J~ · n̂dS 2 = 0,
dt S(r)

~ ~x) is also of compact support to

where r is the radial coordinate and we are using that J(t,
set the last integral to zero.
34 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

~ ~x), and ρ(0, ~x) = ρ0 (~x) are given then we also have ρ(t, ~x), indeed the
Note that if J(t,
continuity equation asserts,
Z t
ρ(t, ~x) = ρ0 (~x) − ~ · J(
∇ ~ t̃, ~x)dt̃.
0

This fact is sometimes useful in constructing consistent sources for Maxwell’s equations.

1.2.2 Existence and Uniqueness

For most physical applications the above system, 1.15, 1.16, 1.17, and 1.18, is incomplete and
at most it can only be taken as a first approximation.
~ ~x)
The reason for this incompleteness is that the above system assumes that ρ(t, ~x) and J(t,
are given before hand, that is before solving Maxwell’s equations, but this is not usually the
case, since the electromagnetic field influence, and in many cases very strongly, the motion
of their sources. Thus the above equations have to be supplemented with extra equations
for the motion of the sources which would take into account the influence on such a motion
of the electromagnetic fields been generated. In general one is then left with the task of
solving the complete system of equations, that is the case of magneto-hydro-dynamics. But
there are also very important cases where one can solve the source equations for very generic
~ B),
electromagnetic fields and so get a pair (ρ(t, ~x, E, ~ J(t,
~ ~x, E,
~ B))
~ which can then be plugged
into Maxwell’s equations giving now modified equations for E ~ and B.
~ This is for instance the
case of polarization phenomena. 10
Nevertheless we shall consider now the above system, for it can be thought of a valid first
approximation for situations where sources are not strongly affected by the electromagnetic
fields, this is the case in magneto hydrodynamics of very heavy particles me ≪ 1, or when
charges are driven by much stronger forces (which could be also electromagnetic, but not
contemplated in the system) as in the case of particles moving in a synchrotron. For this
ideal case we shall also prove theorems similar to the ones for the vacuum case, but now it
pays to do it in the reverse order as the one used before.

Theorem 1.3 Let ρ(t, ~x), J(t, ~ ~x)) satisfy the continuity equation, and let the initial data
(F~ (~x), G(~
~ x)) satisfies the constraint equations, ∇
~ · F~ (~x) = 4πρ(0, ~x), ∇
~ · G(~
~ x) = 0. Then
the solution to the evolution equations it generates (if it exist) satisfies the constraint equa-
tions for all times.

Proof: Taking a time derivative of the constraint equations, using the evolution equations,
~ · (∇
and the identity, ∇ ~ ∧ A)
~ = 0, we obtain,

10
The sources in general depend on the electromagnetic fields at past times, so in general the new equations
become integro–differential equations. They become local only when Fourier transformed.
1.2. INITIAL VALUE FORMULATION II: SOURCES 35

∂ ~ ~ ~
(∇ · E − 4πρ) = ∇~ · ∂ E − 4π ∂ρ
∂t ∂t ∂t
~ · (c∇
= ∇ ~ ∧B ~ − 4π ∂ρ
~ − 4π J)
∂t
~ · J~ + )∂ρ
= −4π(∇
∂t
= 0
∂ ~ ~ ~
~ · ∂B
(∇ · B) = ∇
∂t ∂t
~
= ∇ · (−c∇~ ∧ E)~
= 0.

Thus
~ · E(t,
∇ ~ ~x) − 4πρ(t, ~x) = ∇
~ · E(0,
~ ~x) − 4πρ(0, ~x)
~ · F~ (0, ~x) − 4πρ(0, ~x)
= ∇
= 0,

and,
~ · B(t,
∇ ~ ~x) = ∇
~ · G(~
~ x) = 0.
Remarks
1. The equation of continuity is a key ingredient here. If it were not satisfied by the sources
then Maxwell’s equations would be inconsistent, namely P = Φ

2. The above proof does not use anything about the character of the equations for the
sources or their nature, so the conclusion of the theorem are general, as long as the
solution exists, as is the necessity of the continuity equation to hold for consistency.

Theorem 1.4 (Existence) Given (F~ (~x), G(~

~ x)), and (ρ0 (t, ~x), J(t,
~ ~x)) smooth and such that

~ · F~ (~x) = 4πρ(0, ~x), ∇

∇ ~ · G(~
~ x) = 0.

Then

~ ~x) = ∂
E(t, (tMct (F~ (~x))) + ctMct (∇ ~ ∧ G(~
~ x))
∂t
∂ J~(t − t̃, ~x)
Z t
− 4π ~
t̃Mct̃ (c2 ∇ρ(t − t̃, ~x) + )dt̃
0 ∂t
~ ~x) = ∂ (tMct (G(~
B(t, ~ x))) − ctMct (∇ ~ ∧ F~ (~x))
∂t Z
t
+ 4πc t̃Mct̃ (∇ ~ ∧ J(t
~ − t̃, ~x))dt̃
0

satisfy Maxwell’s equations with sources

36 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW

~ ~x))
(ρ(t, ~x), J(t,
and initial data
~ ~x) = F~ (~x),
E(0, ~ ~x) = G(~
B(0, ~ x).

~ and using the evolution

Proof: Taking another time derivative of the evolution equation for B
~ we get,
equation for E

~
∂2B ~ ∧ (∇
= −c2 ∇ ~ ∧ B)
~ + 4πc∇~ ∧ J~
∂t2
= +c2 (∆B~ − ∇(
~ ∇~ · B))
~ + 4πc∇ ~ ∧ J~
= +c2 ∆B~ + 4πc∇
~ ∧ J,~

~ ·B
where in the last step we have used the already shown fact that ∇ ~ = 0. Thus we have a
~
wave equation for B, but this time with a source:

1 ∂2 ~ ~ = 4π ∇
~ ∧ J~
2 2
B − ∆B
c ∂t c
Thus, using the formula for the solutions to the wave equation with sources, (1.8), and
~ ~x) to express its time derivative in terms of E(t,
again the evolution equation for B(t, ~ ~x) at
t = 0, and so in terms of F~ (~x) we get,

~ ~x) = ∂ ~ x)) − tcMct (∇ ~ ∧ F~ (~x))

B(t, (tMct (G(~
∂t Z
t
+ 4πc t̃Mct̃ (∇~ ∧ J~(t − t̃, ~x)) dt̃
0

~ ~x)
Using this expression and integrating forward in time the evolution equation for E(t,
we obtain the values of the electric field for all times.

Exercise: Make this last calculation. Redo the calculation getting first a wave equation for
~ ~x), using its constraint equation, and see that one arrives to the same results.
E(t,

Exercise: Use Duhamel’s formula, 1.12, and the homogeneous solution to Maxwell’s equa-
~ and B)
tions (in terms of E, ~ to find the inhomogeneous solution directly.

Remark: We see now that the solution has two different type of terms: One type only
depends on the initial data, and are vacuum (homogeneous) solutions with the sought initial
data. The other is a solution to the equations with sources (inhomogeneous equation) but
with vanishing initial data. The validity of this splitting is a generic and very useful property
of linear systems.
1.2. INITIAL VALUE FORMULATION II: SOURCES 37

To summarize we have shown that given a current density everywhere in space and time,
~
J(t, ~x) and an initial charge density, ρ(0, ~x) = ρ0 (~x), (and obtaining ρ(t, ~x) by integration
of the continuity equation), we have constructed the set P(ρ0 ,J) of all solutions to Maxwell’s
equations, and display its explicit dependence on initial data sets, ID(ρ0 ). Note that the map
~ ~x), that is, Φ = Φ .
relating these sets now depends on J(t,
J~
Even taking into account the discussion at the beginning of this section warning the reader
that this type of sources are very special, it would seem we have made a big break through
into Maxwell’s theory and it would only remain just to explicitly display the set ID(ρ0 ), which
in fact can be done quite easily. This is not so for various reasons:

• As we have discussed in general the motion of the sources do depend on the electro-
magnetic fields they generate or are acted upon from the outside. This includes all
macroscopic media discussions.

• The presence of sources brings into the problem different time and space scales. In
general, it is necessary to use some extra physical arguments to say in which scale the
equations are to hold. We shall see this when we study continuous media.

• Even if we could find all solutions for the case where the interaction of the sources,
between each other and with the electromagnetic fields were taken into account, (some-
thing which probably could be done in a few years thanks to the ever growing computer
power) it is very difficult to extract physically relevant information from general so-
lutions, or “tables of solutions" as the computers would generate. As we shall see in
the following chapters, Maxwell’s equations have an incredible richness which hardly
could be extracted by these methods. Such richness is best discovered by studying the
equation (rather than the specific solutions), and by studying many simple solutions,
which in turn act as building blocks of general, very complicated ones, and at the same
time are suitable for displaying all the physical phenomena behind electromagnetism.
38 CHAPTER 1. MAXWELL’S EQUATIONS: AN OVERVIEW
Chapter 2

Energy and Momentum of the

Electromagnetic Field.

2.1 The Electromagnetic Field Energy

Energy and Momentum are very important concepts in physics, for their conservation 1 implies
a limitation on the regions of phase space the system can visit along its evolution. For instance
if we initially have a particle at ~xo = 0, with velocity ~v o , and we know the forces acting upon
it do not increase its kinetic energy, then we know that for all times we shall have | ~vt |≤| ~v o |
and so | ~x(t) |≤| ~vo | t. That is, without doing any calculation we have obtained very useful
information about the system. Arguments of this sort become more important the more
complicated a dynamical system gets, that is, the more difficult to solve it becomes, since
arguments like the one above do not make use of the explicit knowledge of solutions.
We already have used these arguments in electromagnetism when we show the uniqueness
of solutions as characterized by their initial data. In that proof we used an expression which,
we claim, is the energy stored in the electromagnetic configuration. To see this we now study
the increase in energy caused by work done from outside the system.
We assume that at some given time we have a smooth current-charge distribution (ρ, J), ~
and given smooth electric and magnetic fields (E, ~ B).
~ The charges and currents are assumed
to be fixed in space by some mechanical artifact. We suddenly remove such artifact and so
the charges are free to move under the action of the electromagnetic field. In doing so, a
force will act upon them and so certain amount of work will be done. Since the total energy
of the system is constant (no external forces are acting) the work on the charges should be
compensated by exactly the same amount as a decrease in the electromagnetic field energy.
Using the Lorentz force expression we can compute the total initial power transfer to the
charge distribution,

~v ~
Z Z
PTo charges = ~+
∧ B) · ~v d3~x =
ρ(E ~ · ~v d3~x,
ρE
V c V

which includes the well known fact that the magnetic field does not exert work upon charges.
1
Some times they are not conserved, but even in many those cases they are useful for their variation can
be computed without difficult calculations.

39
40CHAPTER 2. ENERGY AND MOMENTUM OF THE ELECTROMAGNETIC FIELD.

~ the power released by the field would be,

Using ρ~v = J,
Z
PFrom fields = − J~ · E
~ d3~x
V

This should be equal to the time derivative of the total energy of the electromagnetic field
configuration. Using now Maxwell’s evolution equation for E, ~

~
∂E ~ ∧B
= c∇ ~ − 4π J,
~
∂t
to replace the J~ in the expression above we get,

1 ~
∂E ~ d3~x = 1 ∂ ~ ~
Z Z
PFrom fields = ( ~ ∧ B)
− c∇ ~ ·E { ~ ∧ B)
(E · E) − 2c(∇ ~ · E}
~ d3~x,
4π V ∂t 8π V ∂t
~ ∧W
to handle the second term we use the identity, (∇ ~ )·V
~ − (∇
~ ∧V
~)·W
~ = −∇
~ · (V
~ ∧W
~ ),
and obtain,

1 Z ∂ ~ ~ ~ ∧ E)
~ ·B
~ −∇
~ · (E
~ ∧ B))}
~
PFrom fields = { (E · E) − 2c((∇ d3~x
8π v ∂t
~
We now use the evolution equation for B,

∂ ~ ~ ∧ E,
~
B = −c∇
∂t
and get,

1 ∂ ~ ~
Z
PFrom fields = { (E ·E+B ~ · B)
~ + 2c∇
~ · (E
~ ∧ B)}
~ d3~x.
8π V ∂t
Finally, we integrate by parts the last term using Gauss theorem and take out of the
integral sign the time derivative to obtain:

d d 1 Z ~ ~ ~ ~ 3
Z
~ · ~nd2~s.
E(t) = {E · E + B · B} d ~x = PFrom fields − S
dt dtZ8π V Z ∂V
~ ~ 3
= − J · Ed ~x − ~ · d2~s,
S
V ∂V

where, we have introduced,

1 Z ~ ~ ~ · B}
~ d3~x,
E(t) := {E · E + B
8π V
~=
the volume energy of the electromagnetic field, and, again, the Poynting Vector, S c ~
(E ∧
4π
~ ∂V denotes the boundary of the space region, V .
B).
There are several remarks about the above formula:
2.1. THE ELECTROMAGNETIC FIELD ENERGY 41

Remark: • Lorentz formula appeared only to assert that the work (density) done by unit
of time was,
W
Z
= − J~ · Ed~ 3~x.
δt V

In essence, it is used to “glue" the electromagnetic energy to other energies, like the mecha-
1
nical one, through the interaction of the systems. In particular this gives us the constant 8π
in front of the definition. Once we have this expression for the energy we can even apply to
the case where no sources are present, as we did in the previous chapter.
• The existence of surface term, the integral of the scalar product of the Poynting vector
~
S, with the unit normal to the boundary of the region, shows that in moving charges -or
fields- around we also produce electromagnetic radiation which leaves the region. That is,
after the redistribution of the fields in general there will be less or more energy stored in the
region than the one remaining after taking into account the work strictly needed for doing
the job.
• Care has to be exercised in interpreting Poynting vector as a radiation flux. This is not
a very precise interpretation, for instance the Maxwell solution,
E~ = (0, E, 0), E = cte, B ~ = (0, 0, B), B = cte., in Cartesian coordinates has a non-zero
Poynting vector, S ~ = ( c EB, 0, 0), but, since the solution is static there is no energy exchange
4π
and therefor no energy flux.
It is only theR integral of the Poynting vector the one which makes physical sense. In the
example above, ∂V S ~ · d2~s = 0.
• In the presence of a material obeying Ohm’s law, J~ = σ E, ~ we see that the presence of
any electric field in the region of that material would cause Joule dissipation,

dE
Z Z
= 3
−J~ · E
~ d3~x = −
3
~ ·E
σE ~ d3~x < 0,
dt lR lR

where we have integrated over the whole of lR3 , (although the only regions which contribute
to the integral are the ones where σ 6= 0), and assumed that the Poynting vector vanishes
near infinity.
The above argument tell us that in the presence of finite resistivity materials the stationary
solutions have zero electric field in the regions occupied by the material.
• We have only used the evolution equations. We shall see latter what is the role of the
constraint equations.
• If we integrate the above expression along a time interval [0, t] we get,

Z tZ Z tZ
EV (t) − EV (0) = −J~ · E
~ d3~xdt − ~ · d2~s.
S
0 V 0 ∂V

See figure XXX, thus the change in field energy in a fixed in time region V has two contri-
butions, a bulk one due to Joule work, and a boundary one, interpreted as the energy flux
leaving or entering the volume boundary as the time goes.
42CHAPTER 2. ENERGY AND MOMENTUM OF THE ELECTROMAGNETIC FIELD.

2.2 The Electromagnetic Field Momentum

To find out what is the expression for the total momentum carried by the electromagnetic
fields we shall use the same strategy as with the energy. We shall look for an analog to the
equation of motion of mechanics, ddtp~ = F~ .
On the right-hand side we shall write the total Lorentz force acting on the particles along
a direction ~k,
Z
~k · F~ V = − ~ + J/c
(ρE ~ · ~k d3~x.
~ ∧ B)
V

There is a minus sign because it is the force caused by the charge density upon the field.
In analogy with the case for the energy, we want to find an expression P~ V such that, when
contracted with any constant vector ~k, we get,

d(P~ V · ~k)
Z Z
~ ~ ~ ~ 3
= − (ρE + J/c ∧ B) · k d ~x − T~ ~ · n̂ d2 S,
dt V ∂V k
where T~ ~ is a vector which also depends linearly on ~k.
k
The last term in the above equation, the surface term, will be needed, for radiation will
not only take energy away, but also momentum. Let us see that P~ = c12 S ~ satisfies our desired
equation:

∂ P~ · ~k 1 ∂E~
~ + (E
~
~ ∧ ∂ B )} · ~k
= {( ∧ B)
∂t 4πc ∂t ∂t
1
= {[c∇~ ∧B~ − 4π J]
~ ∧B ~ + (Ê ∧ [−c∇ ~ ∧ E])}
~ · ~k
4πc
=
−1 ~
{ B ∧ (∇ ~ + 4π J~ ∧ B
~ ∧ B) ~ +E ~ ∧ (∇ ~ · ~k
~ ∧ E)}
4π c
−1 ~ ~ ~ ~ ~ · ∇)(
~ B)~ · ~k + (~k · ∇)(
~ E) ~ ·E ~ − (E~ · ∇)(
~ E) ~ · ~k} + (J/c
~ ∧ B)~ · ~k,
= {(k · ∇)(B) · B − (B
4π
−1 1 ~ ~ ~ ~ ~ · ∇)(
~ B ~ · ~k) + 1 (~k · ∇)(
~ E ~ · E)
~ − (E ~ · ∇)(
~ E ~ · ~k)} + (J/c
~ ∧ B)~ · ~k,
= { (k · ∇)(B · B) − (B
4π 2 2

where we have used the identity U~ ∧ (V

~ ∧W ~ )=V ~ (U
~ ·W
~ )−W ~ (U~ · V~ ), but keeping in mind
~ is a derivative and so must only on the terms where it is supposed to. Using now the
that ∇
constraint equations to add to the above expression terms with the divergences of E, ~ and B ~
we get,

∂ P~ · ~k −1 1 ~ ~ ~ ~ ~ · ∇)(
~ B ~ · ~k) − (B~ · ~k)(∇
~ · B)
~ +
= { (k · ∇)(B · B) − (B
∂t 4π 2
1 ~ ~ ~ ~ ~ · ∇)(
~ E ~ · ~k) − (E~ · ~k)(∇
~ · E)
~ + 4πρ(E~ · ~k)} + (J/c ~ · ~k).
~ ∧ B)
(k · ∇)(E · E) − (E
2
regrouping terms we get,
2.2. THE ELECTROMAGNETIC FIELD MOMENTUM 43

∂ P~ · ~k −1 ~ ~ ~ ~ ~ · B)]
~ − 2[E(
~ E~ · ~k) + B(
~ B~ · ~k)]} − (ρ + J~/c ∧ B)
~ E~ · ~k).
= ∇ · {k[(E · E) + (B
∂t 8π
Integrating over a volume V and using Gauss theorem, we get

d(P~ V · k) Z
~ · ( 1 {2[E(
~ E~ · ~k) + B( ~ · ~k)] − ~k(E
~ B ~ ·E
~ +B
~ · B)}
~
= {∇
dt V 8π
− (ρE~ + J/c
~ ∧ B)~ · ~k} d3~x
Z
= − (ρE ~ ∧ B) · ~k d3~x
~ + J/c
V
Z
1 ~ ~ ~ ~ B~ · ~k) − ~k(E
~ ·E
~ +B
~ · B)]
~ · ~n d2 S.
+ [2E(E · k) + 2B(
∂V 8π

Thus we see that P~ = S/c

~ 2 is the correct expression with

1 ~ ~ ~ ~ · ~k) − 2B( ~ · ~k)].

T~ ~ = ~ · B)
[k(E · E + B ~ − 2E(
~ E ~ B
k 8π
Using again indices we see that,
1 ij ~ ~ ~ iE
~ j − 2B
~ iB
~ j ]kj := T ij kj
(T~ ~ )i = ~ · B)
[δ (E · E + B ~ − 2E
k 8π
ij
The object T , whose contraction with a vector gives another vector, is called a 2-tensor, and
it is a genuinely geometrical, (i.e. physical), quantity.
In an equilibrium situation, i.e. when there is no change in the momentum, we have,
Z Z
F~ V · ~k = ~ + J/c
{ρE ~ · ~k d3~x = −
~ ∧ B} ~k · T~ · ~n d2 S.
V ∂V
That is, in this case, the net Lorentz force acting on a material inside a volume V can be
expressed as a surface integral which only involves the electromagnetic fields and not the
sources. That is T~ can be computed just from the knowledge of the fields at ∂V , it does not
matter what is inside! The 2-tensor T~ is called Maxwell’s Stress tensor.
Question: In a static situation, what is the value of the above expression when there are
no sources around? Imagine a static situation in which one could use the above expression
to compute forces among systems parts. Is there any momentum transfer along a surface in
between two parallel capacitor plates?
44CHAPTER 2. ENERGY AND MOMENTUM OF THE ELECTROMAGNETIC FIELD.
Chapter 3

The Symmetries of Maxwell’s

Equations

A very important tool for disentangling physics phenomena has been the study of their symme-
tries. Through Noether’s theorem symmetries and conserved quantities are intimately related.
Observable conserved charges on physical systems imply symmetries which put constraints in
the possible mathematical models describing them.
What is a symmetry? It is a property of the equations which allows to find more solutions
once one is given. Sometimes, as we saw in the case of the wave equation, all solutions. More
precisely, it is a map which sends solutions into solutions and preserves certain structure, the
precise description of which is beyond the scope of this book.

3.1 Time Translation

A simple example, which already appears in classical mechanics, is the symmetry under
time translation: Since Maxwell’s equations do not depend explicitly on the time variable,
that is there is no preferred origin of time, if (E(t, ~ ~x), B(t,
~ ~x)) is a solution with sources
~ ~ ~ ~ − T, ~x)) is also a solution with
(ρ(t, ~x), J(t, ~x)), then E T (t, ~x), B T (t, ~x)) := (E(t − T, ~x), B(t
sources (ρT (t, ~x), J~T (t, ~x)) = (ρ(t − T, ~x), J(t~ − T, ~x)).
To see this, we insert the potentially new solution into Maxwell’s equations, define u :=
t − T , and use the chain rule to obtain:

~T
∂E ∂E~ ~
∂u ∂ E
(t, ~x) = (u, ~x) = (u, ~x)
∂t ∂t ∂t ∂u
∂E~
= (u, ~x) = c∇ ~ ∧ B(u,
~ ~ ~x)
~x) − 4π J(u,
∂u
~ ∧B
= c∇ ~ T (t, ~x) − 4π J~T (t, ~x).

For the other equations the procedure is identical and is left as an exercise. Time translation
symmetry is associated to energy conservation.

45
46 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS

Exercise: To see that there are equations which do not have such a symmetry consider the
E~ ~ ∧ E.
~ Show that this one does not have the above time symmetry. Does it
equation ∂∂t = t∇
have another symmetry?

3.2 Space Translations

Time translation can easily explained by saying: Since in physics there is no privileged time
instants (not so in Cosmology), then any starting point is equivalent and so if we have a given
evolution -solution- we also could have an “advanced one", by any arbitrary time interval.
By the same reasoning, since no spatial (Cartesian) coordinates explicitly appear, (nor in
any of the fundamental equations of physics) there is no preferred origin and therefore if we
~ ~x), B(t,
have a solution, (E(t, ~ ~x)), corresponding to a charge distribution (ρ(t, ~x), J(t, ~ ~x)), we
~ (t, ~x), B
also have its translated one, (E ~ (t, ~x)) = (E(t,
~ ~x − L),~ B(t,
~ ~x − L))~ which correspond
~
L ~
L
to the translated charge distribution, (ρ ~ (t, ~x), J~~ (t, ~x)) = (ρ(t, ~x − L),
~ J(t,
~ ~x − L)),
~ where L ~
L L
is a constant vector which indicates the distance and direction of the translation. Again, we
check this for one of the Maxwell’s equation, leaving the checking on the others as an exercise.

~
∂E ~
~
L ∂E ~
(t, ~x) = (t, ~x − L)
∂t ∂t
∂E~
= (t, ~x′ )
∂t
~ ′ ∧ B(t,
= c∇ ~ ~x′ ) − 4π J(t,
~ ~x′ ),

~ Since
where we have defined ~x′ := ~x − L.
~
∂B ~
∂x′j ∂ B ~ ~
j ∂B ∂B
= = δi = ,
∂xi ∂xi ∂x′j ∂x′ j ∂x′i

~ ′ ∧ B(t,
∇ ~ ~x′ ) = ∇
~ ∧ B(t,
~ ~x′ (~x)) = ∇
~ ∧ B(t,
~ ~x − L)
~ =∇
~ ∧B
~ (t, ~x),
~
L
and the result follow immediately. We attribute this symmetry to the homogeneity of space.
It intimately related to momentum conservation.

3.3 Rotation
Space not only does not have any preferred origin, but neither have any preferred direction
or orientation, all of them are alike. Thus, if we rotate any given solution we should expect
to get another solution, and this should be so irrespectively of origin we choose to perform it.
3.3. ROTATION 47

Figure 3.1: A 90 degree rotation

48 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS

Since in Maxwell’s equations we are dealing with vectorial quantities, we must be a bit more
careful and note that the vectors also must rotate! See figure.
The rotated solution would be

~ R (t, ~x), B
(E ~ R (t, ~x)) = (R(E(t,
~ R−1 (~x))), R(B(t,
~ R−1 (~x))))

corresponding to the rotated source:

(ρR (t, ~x), J~R (t, ~x)) = (ρ(t, R−1 (~x)), R(J(t,
~ R−1 (~x))))

~ i = Ri j E j and [R−1 (~x)]i = [R−1 ]i j xj .

where [R(E)]
We now check this symmetry transformation for the evolution and constraint equations
~ The others are similar.
for E.
For the constraint equation for E ~ we have,

~ ·E
∇ ~ R (t, ~x) = ∇
~ · R(E(t,~ R−1 (~x)))
∂ l ~
= R k [E(t, ~x′ )]k
∂xl
∂x′i l ∂ ~
= R k ′ i [E(t, ~x′ )]k
∂xl ∂x
∂ ~
= [R−1 ]i l Rl k ′ i [E(t, ~x′ )]k
∂x
∂ ~
= [E(t, ~x′ )]k
∂x′k
~ ′ · E(t,
= ∇ ~ ~x′ )
= 4πρ(t, ~x′ )
= 4πρ(t, R−1 (~x))
= 4πρR (t, ~x).

~ we have,
For the evolution equation for E

~R
∂E ∂ ~
(t, ~x) = ~ R−1 (~x))) = R( ∂ E (t, R−1 (~x)))
R(E(t,
∂t ∂t ∂t
′
~ ~ ~
= cR(∇ ∧ B(t, ~x ) − 4πRJ(t, R−1~x′ )
′

= cR(∇~ ′ ∧ B(t,
~ R−1 (~x)) − 4πRJ(t,
~ R−1 (~x)),

where the curl is with respect to the variable ~x′ = R−1 (~x). We write now the curl in compo-
nents,
e
~ ~x′ )]i = E ijk ∂ Bk (t, ~x′ ) = E ijk ∂x′ ∂Bk (t, ~x),
~ ′ ∧ B(t,
[∇
∂x′j ∂x j ∂xe
′ ∂xe
but xe = Re n x n and so ∂x′j
= Re j .
3.4. DISCRETE SYMMETRIES 49

On the other hand, since B ~ appears in the curl with its “index down", and B ~ ·B
~ = B k Bk
must be a scalar, we must have that B ~ ·B~ =B ~R · B
~ R , so the index down version of B
~ must
change with the inverse transformation, [BR ]k = [R ] k [B]l . Therefore, [B]k = Rkl [BR ]l ,
−1 l

[Notice that this is the same change as the one for the differential ∇i .] so,

~ ′ ∧ B]
[R∇ ~ i = Rli E ljk Re j Rn k [∇]
~ e [B
~ R ]n .

But
E ljk Ri l Re j Rn k = det(R)E ien ,
and for a rotation, detR = 1. Another way to see this is to notice that the vector product is
an invariant operation, and so the vector product of two rotated vectors should be the rotated
of the vector product of the vectors prior to the rotation, that is R(v) ∧ R(w) = R(v ∧ w),
but distinguishing between vectors with the index up to the ones with the index down in the
formulae, this implies,
E ijk Re j Rn k = [R−1 ]i m E men .
With this formula, we see that,

~ ′ ∧ B(t,
cR(∇ ~ ~x′ ) − 4π/cJ(t, ~ ~x′ ))
~ ∧B
= c∇ ~ R (t, ~x) − 4π J~R (t, ~x)

and the result follows. We attribute this symmetry to the isotropy of space.

Exercise: Check that: E ijk Rl i Rm j Rn k = det(R)E lmn , for any 3 × 3 matrix Rl i ; and that
det(R) = 1 for rotations. Hint: for the last part use that any rotation can be expressed terms
of five known consecutive transformations, each one of determinant one. Or realize that the
final expression is independent of the coordinate system and choose the coordinates after
giving the rotation so as to leave invariant one of the axis. What is the general expression for
the determinant of a n × n matrix?

3.4 Discrete Symmetries

~
The symmetries we were considering until now were continuous, because the parameters, T, L
and R could be made to vary continuously from point-to-point, even more, they could be
deformed up to the identity transformation: (T → 0, L ~ → 0, R → I). We now look at
transformations which do not have these properties.

3.4.1 Time Inversion

Maxwell’s equations not only do not have a preferred origin of time but also do not have an
arrow of time. This property, which is also shared with all fundamental equations of physics,
says that if we have a solution, we also have its time reserved one, that is, where all happens
50 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS

as if we were watching a movie starting from the end. As in the case of rotations, in this
case the symmetry does not only involve the change t → −t, but it also must involve same
~ and B.
change on the fields, E ~ For simplicity in the presentation we shall assume each of one
these vectors changes just by a constant factor. Since the application of two consecutive time
inversion is the identity (t → −t, (−t) → −(−t) = t), those constant factors must have unit
square, and so they must be +1 or −1. Thus, we assume the symmetry has the form:
~ I (t, ~x), B
(E ~ I (t, ~x)) = (aE E(−t,
~ ~
~x), aB B(−t, ~x))

with a2E = a2B = 1, both real. Using now the equations we shall fix these values.
Since this symmetry concerns only with the time variable the constraint equations must
hold identically, so in particular we must have the obvious result, ρI (t, ~x) = aE ρ(−t, ~x).
Since
∂E~I ~
∂E
(t, ~x) = aE (−t, ~x)
∂t ∂t
∂(−t) ∂ E ~
= aE (−t, ~x)
∂t ∂(−t)
~
∂E
= −aE (u, ~x),
∂u
we have,

∂E~I
~ ∧ B(u,
(t, ~x) = −aE [c∇ ~ ~ ~x)]
~x) − 4π J(u,
∂t
~ ∧B
= −aE [aB c∇ ~ I (t, ~x) − 4π J(−t,
~ ~x)]

similarly,

∂B~I ∂B~
(t, ~x) = −aB (u, ~x)
∂t ∂u
= −aB [−c∇~ ∧ E(u,
~ ~x)]
~ ∧E
= +aB aE [c∇ ~ I (t, ~x)]

thus we must have, aB aE = −1 and J~I (t, ~x) = −aE J(−t,

~ ~x) ρI (t, ~x) = aE ρ(−t, ~x). It is then
~
natural to take aE = 1 and aB = −1, for since J is a current, (charge density times velocity)
it should change sign with time reversal, for velocities certainly do. Thus, we conclude that
~ and J.
time reversal involves a sign change in B ~

3.4.2 Space Inversion

Space inversion is the symmetry which you experiment daily when you comb the hair of that
ugly guy which look at you from inside the mirror. That guy is the space inverted symmetric
of yours, in that case the axis perpendicular to the mirror has been inverted. So we now,
instead of changing the time from t to −t, we change the axis ẑ, say, from z to −z. This is a
3.5. GALILEAN TRANSFORMATIONS. 51

linear transformation which changes -inverts- the ẑ axis and leave invariant the other two, so
it is like a rotation, but with determinant equal to −1.

Exercise: Check this last assertion writing down explicitly the inversion matrix. Namely,
~ Invz (t, ~x) = RInvz E(t,
(E ~ RInvz ~x), H
~ Invz (t, ~x) = −RInvz H(t,
~ RInvz ~x)).

Exercise: Check that inverting two axis is a transformation with determinant one. Check
that indeed this can be accomplished by a rotation.

Since the transformation is linear we do not have to redo the calculation we made for
rotations, it is similar to it except that since the determinant of the transformation is minus
one, in analogy with the time inversion symmetry, B ~ must have an extra −1 change!
As we shall see latter this comes about because B ~ itself is in fact made out of the curl
of a proper vector, i.e. one which changes like E ~ under a space inversion, and so, since the
Levi-Civita symbol changes sign under space inversion, so does B. ~

3.5 Galilean Transformations.

According to the Galilean principle of relativity, the laws of physics should be the same for
systems which are moving with respect to each other with constant relative velocities [one
in uniform motion with respect to the other.] Consider, for instance, the equation of motion
for a system of two particles which are subject to a force which only depends on the relative
position of that particles, that is,

d2~x1 (t)
m = F~ 1 (~x1 (t) − ~x2 (t))
dt2
d2~x2 (t)
m = F~ 2 (~x1 (t) − ~x2 (t)),
dt2
as it would be the case for motion under the influence of their gravitational attraction. The
Galilean principle of relativity, for this case would then imply that (~x′1 (t), ~x′2 (t)) = (~x1 (t) −
~v t, ~x2 (t) − ~vt) is also a solution, if (~x′1 (t), ~x′2 (t)) was one. Indeed,
d2~x′1 d2
m = m (~x1 (t) − ~v t)
dt2 dt2
d2
= m 2 ~x1 (t)
dt
~
= F (~x1 (t) − ~x2 (t))
= F~ (~x1 (t) − ~v t − (~x2 (t) − ~v t))
= F~ (~x′1 (t) − ~x′2 (t)).

Exercise: Check this for the case of two particles interacting through a spring of Hooke
52 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS

constant k. That is, find the solution corresponding to the particles (and the spring) traveling
with constant velocity ~v .

In the case of electromagnetism this principle would imply, for instance, that if one has
two equal capacitors with the same potential difference between its plates, and one is in the
surface of the earth and the other over a ship sailing nearby, then the forces between the
plates should be equal for both capacitors. In Galilean terms we are then saying that, if
~ ~x), B(t,
(E(t, ~ ~x)) is a solution, then,

~ ~x − ~v t) + G B(t,
(Eg (t, ~x) := F~v E(t, ~ ~x − ~v t), B
~ j (t, ~x) := F̃ E(t,
~ ~x − ~v t) + G̃ B(t,
~ ~x − ~v t)
~v ~v ~v

should be also a solution, for some set of linear transformations, (F~v , G~v , F̃~v , G̃~v ) where
~ i = F i [E]
[F~v E] ~ j , etc.
~v j
Since ~v is a constant velocity, not only in time but also in space, and because of the
linearity of Maxwell’s equations the linear transformations can neither depend on the fields
E~ nor B,~ they must be constant in both time and space and only depend on ~v. Furthermore,
since the transformation should be the identity for vanishing velocities,

F~o = Id. G~o = 0, F̃~o = 0, G̃~o = Id.

Notice that we have to allow for this type of transformation, which mixes both E~ and B,~
for in the above example the capacitor’s plates would be moving, and so generating currents,
this in turn, would generate magnetic fields, which in turn, would contribute to the total
force.
Let us see now whether these new fields we have defined also satisfy Maxwell’s equations.
To simplify the calculation we consider only vacuum solutions (or just look at the solution in
an empty region), and decouple them taking a time derivative to the equations to get,

∂2 ~ ~ =0
E − c2 ∆E
∂t2
∂2 ~ ~ = 0.
B − c2 ∆B
∂t2

We apply these two equations to E~ g and B

~ g respectively, if they are not solutions to these
equations, then they can neither be solutions to the complete set of Maxwell’s equations.
~ we get,
Substituting in the equation for E

2 2
(∂t2 E ~ g ) = F ( ∂ E(t,
~ g − c2 ∆E ~ ~x′ )) + G ( ∂ B(t,
~ ~x′ ) − c2 ∆′ E(t, ~ ~x′ ) − c2 ∆′ B(t,
~ ~x′ ))
~v ∂t2 ~v ∂t2

−2~v · ∇~ ′ ( ∂ (F E(t,~ ~x′ ) + G B(t,~ ~x′ )))

∂t ~v ~v
~ ′ )(~v · ∇
+(~v · ∇ ~ ′ )(F E(t,
~ ~x′ ) + G B(t,
~ ~x′ )),
~v ~v
3.5. GALILEAN TRANSFORMATIONS. 53

where we have used that ∂t ∂

|~x f (t, ~x − ~v t) = ∂f ~ ′ f (t, ~x′ ), and that ∇f
(t, ~x′ ) − ~v · ∇ ~ (t, ~x − ~v t) =
∂t
′
~ f (t, ~x′ ) with ~x′ = ~x − ~vt.
∇
Using that the original E ~ and B ~ satisfy the wave equation, and the evolution equations
we see that E ~ g satisfies the wave equation if and only if

~ ′ )(F (∇
−2c(~v · ∇ ~ ′ ∧ B(t,
~ ~x′ )) − G (∇~ ∧ E(t,
~ ~x′ )))
~v ~v
~ ′ )(~v · ∇
+(~v · ∇ ~ ′ )(F (E(t,
~ ~x′ ) + G (B(t, ~x′ ))) = 0
~v ~v
Consider now this relation at t = 0. We see that, since there are no time derivatives, this
is a relation among the initial data, and that this relation is different from the constraint
equations (for it involves curls). So this relation is enforcing further constraints to the ones
already imposed by Maxwell’s equations. So there are plenty of solutions to Maxwell’s equa-
tions which do not conform with Galilean relativity, namely all solutions coming form initial
data sets not satisfying the above relatio 1 .

axis (t,x=0) axis (t,x´=x − vt = 0)

Solution at
rest in syst.
(t,x) Solution at
rest in syst.
(t,x´)

axis (0,x) and (0,x´)

Figure 3.2: A wave seen by two observers

We are in trouble, either Galilean Relativity is wrong or Maxwell’s equations are wrong.
The two together are inconsistent. One has two logically consistent formalism, only exper-
iments can decide. The incredible amount of phenomena described by Maxwell’s equations
tell us that it is the Galilean Principle of Relativity the one which must be abandoned. We
shall come back to this point latter in the book.
One could have guessed the result of our calculation, for we already remarked that in
Maxwell’s equation there appears a constant with dimensions of velocity. At that moment
we should have asked: What is moving with that velocity? And: With respect to what are
1
To see this in more detail, assuming our transformations are smooth note that F~v = Id + O(| ~v |) and
G~v = O(| ~v |), since the relation has to vanish order by order we see that to first order we must have,
~ ′ )(∇
(~v · ∇ ~ ′ ∧ B(t,
~ ~x′ )) = 0, which is clearly an extra constraint.
54 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS

we computing that velocity? The answer to the first was already given, it is the velocity
of propagation of electromagnetic waves. The second will be answered latter in the book.
For the moment we shall content ourselves to seeing how the constant c appears on Maxwell
equation.

3.5.1 The origin of the constant c on Maxwell’s equations

To see how it appears we shall resort to three experimental facts:
1.-) Charge Conservation. Mathematically this translates into the continuity equation,
∂ρ ~ ~ L
+ ∇ · J = 0 from which [ρ] = [J]
∂t T
From this we can obtain the dimensions of the electric current across a surface, S, I =
R
~ · ~nds. Therefore, 2
J
s
L
[I] = [J]L2 = ([ρ] )L2 ,
T
where L is length and T time. But q = v ρdv and so,
R
3
[I] = ( L[q]3 ) LT = [q]
T
.
2.-) Coulomb’s Law. The electric force between two particles of charge q and q ′ respectively
is,
qq ′
Fc = k1 2
r
where r is the distance between them. Thus,
L2
[k1 ] = [F ] .
[q]2
3.-) Ampère’s Law. The force between two parallel wires of length l and separation d
carrying currents I and I ′ respectively is:
l
FA = k2 II ′
d
From here we see that,

[F ] [F ]T 2
[k2 ] = 2 = .
[I] [q]2
Therefore,
[k1] L2
= 2 = [c]2
[k2] T
Of course these laws not only allow to know the dimensions of c, put also its value! In
spite of the fact that they are basically stationary experiments, where nothing travels to the
speed c. In fact, when Maxwell unified electricity and magnetism, and found this quantity
noticed that its value was close to the speed of light! It took some time to relate both, for
the optical phenomena were really not thought, at that time, as having anything to do with
the electric or magnetic phenomena.
2
Square brackets here means “the dimension of" the quantity inside them.
3.6. OTHER SYMMETRIES 55

3.6 Other Symmetries

Exercise: Check that in vacuum Maxwell’s equations also have the symmetry:

~˜ = cos(α)E
E ~ + sin(α)B
~
~˜ = cos(α)B
B ~ − sin(α)E.
~ (3.1)

Generalize it further.

Exercise: Check that the energy does not change under the above symmetry.

Exercise: Use the vacuum Maxwell’s equations to find the time derivative and divergence of
~ := E
W ~ + iB
~
56 CHAPTER 3. THE SYMMETRIES OF MAXWELL’S EQUATIONS
Chapter 4

Stationary Solutions: Electrostatics

4.1 Stationary and Static Solutions

Definition: We call a solution Stationary if it is invariant under time translation, that is

~ T (t, ~x) := E(t

E ~ − T, ~x) = E(t,
~ ~x), and
~ T (t, ~x) := B(t
B ~ − T, ~x) = B(t,
~ ~x), ∀T,

that is, they are independent of time. Naturally for this to happens the sources must also be
time independent. For these solutions all time derivative vanish and so the equations becomes,

~ ∧ B(~
c∇ ~ x) = 4π J(~
~ x)
~ ∧ E(~
c∇ ~ x) = 0
~ · E(~
∇ ~ x) = 4πρ(~x)
~ · B(~
∇ ~ x) = 0.

Note that the equations are now decoupled, we have two equations for E(~ ~ x) and two
~ ∂ρ
equations for B(~x). Note also that since ∂t = 0, the continuity equation implies ∇~ · J(~
~ x) = 0,
a necessary condition for the first equation above to have a solution. If furthermore we require
the solutions to be invariant under time inversion, that is,

~ I (t, ~x) := E(−t,

E ~ ~ ~x),
~x) = E(t,
~ I (t, ~x) := −B(−t,
B ~ ~ ~x),
~x) = B(t,
ρI (t, ~x) := ρ(t, ~x) = ρ(t, ~x),
J~I (t, ~x) := −J(−t,
~ ~ ~x),
~x) = +J(t,

~ x) = J(~
we see that the independence of the solution on the time variable imply B(~ ~ x) = 0, and
~ remain. We shall call these solutions static solutions, and
so only the two equations for E
their study electrostatics. The study of the stationary solution is completed by studying
~ x)) and is called magnetostatics.
the remaining two equations (for B(~

57
58 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

4.2 Electrostatics
The equations of electrostatics are:

∇~ ∧ E(~
~ x) = 0 , ∇
~ · E(~
~ x) = 4πρ(~x).
To solve them we make the ansatz, E ~ = −∇φ.
~ Then it is ease to see that the first equation
~ ~ ~ ~
is trivially satisfied, ∇ ∧ E = −∇ ∧ ∇φ ≡ 0, while the second becomes Poison’s equation:
∆φ(~x) = −4πρ(~x).
We shall discuss this equation during the rest of the chapter.
How general is the ansatz we have used? Or in other words, are all solutions of the
electrostatic equations gradients of some functions?
The answer to this problem is completely known and there are precise conditions which
guarantee uniqueness. When these conditions are not meet, there are other solutions. In
most cases in electrostatics the ansatz suffices, but we shall see in magnetostatics an example
where it fails.

4.2.1 The General Solution for Isolates Systems

An isolated system in physics means a system on which no external force acts, that is a
system left on its own. In electrostatics means a system of charges that has finite extension
and which generates its own electric field. This last condition is imposed requiring the electric
~ |= 0( 12 )) at large distances 1 . This is supposed
field generated to decay sufficiently fast (| E r
to incorporate the common experience that the influence of a system into another decreases
when the distance between them increases.
We now look to the general solution with this asymptotic condition. To do that we observe
that Poison’s equation can be considered the limit when c → ∞ of the inhomogeneous wave
equation,
1 ∂2φ
− ∆φ = 4πρ.
c2 ∂t2
But we saw that the inhomogeneous wave equation had as general solution,
∂ ∂ϕ(t, ~x)
φ(t, ~x) = (tMct (ϕ(0, ~x))) + tMct ( |t=0 )
∂t Z ∂t
t
+ 4π c2 t̃Mct̃ (ρ(t − t̃, ~x)) dt̃.
0
2
Taking t 6= 0, replacing in the last integral ct̃ by r, and taking the limit c → ∞ we get,
tc Mr (ρ(t − r/c, ~x)) 2
Z
φ(t, ~x) = lim 4π r dr;
c→∞ 0 r
1 ~ = −∇φ
Since E ~ we see that φ(~x) → φ0 , as, r → ∞, and we choose without loss of generality this constant
to be zero, thus, φ(~x) → 0 as r → ∞.
2
We need t 6= 0 so that we can get rid of the initial data, which in the limit propagates with faster and
faster speed and so goes away.
4.2. ELECTROSTATICS 59

where we have used the condition ϕ(0, ~x) → 0,, and ∂ϕ

∂t
(t, ~x) as | ~x |→ ∞ to eliminate the first
two terms. Using the definition of Mr , a time independent ρ, and a new variable ~x′ = ~x − r~n,
we get,

ρ(t − r/c, ~x − r~n) 2

tc
Z
φ(t, ~x) = φ(~x) = lim 4π r dr dΩ
c→∞ 0 r
ρ(t − r/c, ~x − ~y ) 3
Z tc
= lim 4π d ~y
c→∞ 0 |~y |
Z
ρ(~x − ~y ) 3
= 3
4π d ~y (4.1)
lR |~y |
ρ(~x′ )
Z
= 3 |~ ′ d3~x′
lR x − ~x |

and this, we claim, is the general solution to Poison’s equation in lR3 . It is instructive to
check this directly, we do that next:

ρ(~x′ ) 1
Z Z
3 ′
∆φ(~x) = 3
∆( ′ ) d ~
x = 3
∆( x′ ) d3~x′
′ )ρ(~
lR | ~x − ~x | lR | ~
x − ~
x |

But
1 ~ ~x − ~x′ 3 (~x − ~x′ ) · (~x − ~x′ )
∆( ) = −∇ · ( )=− +3 =0
| ~x − ~x′ | | ~x − ~x′ |3 | ~x − ~x′ |3 | ~x − ~x′ |5
if ~x 6= ~x′ .
Therefore, if ∆ 1
′ were a function at ~x = ~x′ , as is for ~x 6= ~x′ , then the integral above
|~
x−~x |
3 1
would give zero, and we would have, ∆φ(~x) = 0! Thus, we conclude that ∆ ′ is not a
|~
x−~x |
1
function. The problem arises because we are trying to take derivatives of a function, ( ′ ),
|~
x−~x |
at a point where it is not differentiable, and then we further pretend to integrate is as if it
were.
Thus, we must proceed with certain care and treat the equation in the sense of distribu-
tions. That is, we think about it as a map, as applied to a smooth, compactly supported, test
function ψ(~x), into the reals. We define,
Z Z
ρ(~x′ )ψ(~x) 3 ′ 3
TKρ (ψ) := d ~x d ~x (4.2)
lR3 lR3 |~x − ~x′ |
We want to show now that,
Z
∆TKρ (ψ) = −4πTρ (ψ) := −4π ρ(~x)ψ(~x) d3~x. (4.3)
lR3
But by definition,
3
The integral of a function which zero everywhere except at a point vanishes.
60 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

∆TKρ (ψ) := TKρ (∆ψ)

ρ(~x′ )∆ψ(~x) 3 ′ 3
Z Z
= d ~x d ~x
lR3 lR3 |~x − ~x′ |
Z Z
~ 1 ~ x) d3~x′ d3~x
= − 3 ρ(~x′ )∇ · ∇ψ(~
lR lR 3 |~x − ~x′ |
Z Z
′ (~
~ x)
x − ~x′ ) · ∇ψ(~
= ρ(~x ) d3~x′ d3~x (4.4)
lR3 lR3 |~x − ~x′ |3
1
where in the third line the integration by parts is valid because the gradient of ′ is
|~
x−~x |
integrable, and the boundary term vanishes because ψ(~x) is of compact support. We now
perform a change of variables to ~y = ~x − ~x′ , d3~y = d3~x and then a change to spherical
coordinates in the new variable ~y , to we obtain:

Z Z ~ y + ~x′ )
~y · ∇ψ(~
∆TKρ (ψ) = ρ(~x′ ) d3~x′ d3~y
lR3 lR 3 |~y |3
Z Z ∞ Z
= 3
ρ(~x′ )∂r ψ(r~n + ~x′ ) drdΩ d3~x′
ZlR Z0 S2

= ρ(~x′ )[−ψ(~x′ )] dΩ d3~x′

lR3 ZS 2
= −4π 3 ρ(~x′ )ψ(~x′ ) d3~x′
lR
(4.5)
~ = ∂r ψ, and again that ψ is compactly supported to eliminate
where we have used that ~n · ∇ψ
one term in performing the radial integration. We finally have used that the solid angle
integral gives 4π.
Even when it is clear that ∆( 1 ′ ) is not a function, it is often treated like one to make
|~
x−~x |
formal calculations, and a multiple of it is called the Dirac’s Delta function, δ(~x − ~x′ ),
1
∆( ) := −4πδ(~x − ~x′ ). (4.6)
| ~x − ~x′ |
Which has to be understood as above, namely,

1
Z
3
x′ ) d3~x′ = −4πψ(~x) ∀ smooth and compactly supported φ(~x)
′ ∆ψ(~ (4.7)
lR |~x − ~x |
Notice that the calculation done would imply (if the Dirac’s Delta were a function),
Z
3
δ(~x − ~x′ )ρ(~x′ ) d3~x′ = ρ(~x),
lR
for ρ(~x) sufficiently smooth. This is a convenient abuse of notation for doing some calculations,
but many times leads to nonsense.
4.2. ELECTROSTATICS 61

Exercise: Show directly (4.6).

4.2.2 The Weak Form

Actually there is an intermediate treatment of the Poisson equation which does not need
to use Dirac’s distributions, but milder ones, it is called the weak form of the equation: A
solution is an element of the space H 1 (lR3 ) such that,
Z
~ · ∇φ
[∇ψ ~ + 4πρ ψ] d3~x = 0 ∀ ψ ∈ H 1 (lR3 ) (4.8)
lR3
where the space H 1 (lR3 ) is the Hilbert space of functions with the norm:
Z
φ2 ~ 2 ] d3~x.
||φ||2H 1 = [ + |∇φ|
1 + r2
This weak, or variational form, is very important to devise numerical methods to solve
the equations and allows a nice theory of existence and uniqueness of solutions. It is found
in many different contexts in physics and mathematics.

4.2.3 Uniqueness
So far we have found a solution corresponding to an isolated system of charges, how unique
is that solution? Are there more? The answer is that clearly there are lots of solutions, any
solution to Laplace’s equation (the homogeneous, i.e. vacuum, Poison’s equation) can be also
added to the one already found and the result will also be a solution. For instance, we could
add:

φ0 (~x) = φ0 = cnst,
φ1 (~x) = ax + by + cz,
φ2 (~x) = a1 yz + a2 xz + a3 xy + ã1 (y 2 − z 2 ) + ã2 (x2 − z 2 ) + ã3 (x2 − y 2 ),

where, φ0 , a, b, c, a1 , a2 , a3 , ã1 , ã2 and ã3 are arbitrary constants.

Exercise: Show that in the class φ2 (~x) there are only five linearly independent terms.

Exercise: Find all (linearly independent) solutions which are cubic in the combination
x, y, z, that is terms with xα , y β , z γ and α + β + γ = 3, with α, β, γ positive or null integers.

Thus, to single out a unique solution one must impose boundary conditions, in this
case, asymptotic conditions, which are given by the physics of the problem. As we already
saw, the condition in this case should be: φ(~x) → 0 as | ~x |→ ∞. If we do impose them, we
do get uniqueness.
62 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

Theorem 4.1 If φ(~x) decays sufficiently fast at infinity, and satisfies Poisons equation 4 ,

∆φ(~x) = 4πρ(~x).

Then φ(~x) is unique and therefore given by

ρ(~x′ )
Z
φ(~x) = d3~x′ .
| ~x − ~x′ |

Proof: Assume that there are two solutions with the required decay properties, φ1 (~x) and
φ2 (~x). Then their difference, ϕ(~x) = φ1 (~x) − φ2 (~x) also has the required decay properties and
satisfies Laplace’s equation,
∆ϕ(~x) = 0
Multiply this equation by −ϕ and integrate over all space,
Z
0 = 3
−ϕ∆ϕ d3~x
ZlR
= ~ · ∇ϕ
−ϕ∇ ~ d3~x
3
ZlR
= ~ · (−ϕ∇ϕ)
[∇ ~ + ∇ϕ
~ · ∇ϕ]
~ d3~x
3
ZlR Z
= ~ · ∇ϕ
∇ϕ ~ d3~x − lim ~ d2 S,
−ϕ ~n · ∇ϕ
3
lR r→∞ s(r)

where in the last steep we have used Gauss theorem.

~ x) |< c for sufficiently big | ~x |, then the surface integral tends
If | ϕ(~x) |< c and | ∇ϕ(~
|~
x| |~
x|2
to zero as r → ∞, and we get Z
~ · ∇ϕ
∇ϕ ~ d3~x = 0.
lR3
~ x) = 0 and since ϕ(~x) → 0 as r → ∞ we have ϕ(~x) ≡ 0, everywhere and
Thus, ∇ϕ(~
uniqueness is established.
Remark: The application of Gauss theorem above is called first Green Identity.

4.2.4 Existence of the Potential Function

We just finish showing that the solution found by Poison’s formula is unique. But, is the
electric field unique? If the electric field can only be expressed as minus the gradient of a
function then it is unique. But how do we know that this is so? Could there be static electric
fields which are not of that form?
~ x), any curve
To see that this is not the case consider any smooth electric field in lR3 , E(~
γ(s) = ~x(s) and define,
4
Some conditions are needed on ρ(~x) to ensure this, basically the source must be smooth enough and decay
at infinity fast enough so that the solution behaves in such a way that all integrals in the proof that follows
are well-defined. The decay condition, | ρ(~x) |< c 3 , is sufficient.
|~
x|
4.2. ELECTROSTATICS 63

~ x(s)) · d~x(s) ds.

Z t
φ(~x) = − E(~
0 ds
If we take any vector field for doing the above construction then in general it would not result
in a function of the point but actually a function of the curve. That is, the value of φ at ~x
would depend on the curve used it reaching the point. But in the case the vector field satisfies
that its curl vanishes the procedure results in a well-defined function. Indeed, let consider
two curves, γ1 (s), and γ2 (s), such that γ1 (0) = γ2 (0) and γ1 (t) = γ2 (t) = ~x, we want to see
what is the value of:

~ x(s)) · d~x(s)γ1 ds + ~ x(s)) · d~x(s)γ2 ds

Z t t Z
φγ1 (~x) − φγ2 (~x) = − E(~ E(~
0 ds 0 ds
d~
x (s)
~ x(s)) · d~x(s)γ1 + E(~
Z t −1
γ2
= − E(~ ~ x(s)) · ds
0 ds ds
I d~x(s)γ −1 ∗γ1
= − E(~~ x(s)) · 2
ds
Z ds
= − (∇ ~ ∧ E)
~ · n̂ dS
S
= 0

where in the second line we have defined the inverse of a curve, namely the curve which runs
in the opposite direction, from ~x to the initial point. The union of the two curves results in
a closed loop that we have called γ2−1 ∗ γ1 . It is the boundary of a surface, which we call S.
Stokes theorem is used in the fourth line to express that circulation integral in terms of a
~ Finally, the fact that E
surface integral of the curl of E. ~ is curl-free has been used to conclude
that the difference vanishes. Thus, we are constructing a well-defined function. If we take
different curves, so as to cover the whole of lR3 with them, all of them starting from the same
point, we would get a function on the whole of lR3 .
~ = −∇φ.
It remains to see that this function satisfies E ~ To see this connect any two nearby
points by a curve, γ(s), we then have,

~ x(s)) · d~x(s)γ ds ≈ E(~

~ x(s)) · d~x(s)γ (0) ds.
Z δt
φ(~x + δ~x) − φ(~x) = − E(~
0 ds ds
Therefore,

~ · d~x(s)γ (0) = −E(~

∇φ ~ x(s)) · d~x(s)γ (0).
ds ds
The fact that both, ~x and δ~x are arbitrary implies that the equality holds for any velocity
vector, and so we have that,

~ x) = −E(~
∇φ(~ ~ x),

and so that in lR3 we can always write a curl-free vector field as the gradient of a function.
64 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

4.2.5 Weak Solutions

There is an intermediate concept of solutions, where they are not considered as distributions
but neither as twice differentiable solutions. This is the case if we consider that the solution
has integrable derivatives, so we can differentiate once inside the integral. In that case we say
that φ(~x), with integrable first-derivative satisfies Poison’s equation if for all first-derivative
integrable functions ψ,
Z Z
~ x) · ∇ψ(~
∇φ(~ ~ x) d3~x = 4πρ(~x)ψ(~x) d3~x.
3
lR lR3
This concept is very useful both, for theoretically showing existence of solutions and for finding
numerically those solutions. Boundary conditions are imposed by tailoring the functions
defined to be in spaces satisfying the requested conditions so that the integration by parts
needed to be performed to obtain the classical expression does not contribute.

4.3 Conductors
So far we have shown that given a system or distribution of charges in an isolated region of
space we have a unique solution, and we have an explicit formula for it. One could think
that now it is a question of just doing the integral for the given source. Unfortunately this is
not so for in most cases we do not know where the charges are, for charges, if not completely
fixed, are going to move and accommodate until an equilibrium configuration is found, and
that equilibrium configuration depends on the electric field.
To see an example of the type of problem we envision consider any object, of copper say,
of which you know its shape and which you know it has a given amount of charge, q. One
does not know before finding the solution where the charges will be, for since the object is a
conductor, the charges would move in its interior until all electric field inside vanishes, that
is, until all forces upon them vanish, which is in this case the equilibrium configuration.
So this problem can not be solved using the formulas we gave. Although it will satisfy the
equations!
How do we solve it? Let us see what we know about the solution. First we know that
whatever the solution is, it should go to zero at infinity. Second that inside the conductor
~ = 0 or φ(~x) = φ0 inside V , where V is the volume
there should no be electric fields, that is ∇φ
occupied by it. But since the charges are in the conductor and since ∆φ = −4πρ, we see that
they must be at the surface of it. This implies, in the mathematical approximation we are
making of a conductor, that ∇φ ~ must be discontinuous across the surface of the conductor,
for ρ is discontinuous there, but it is only discontinuous along the normal to the conductor,
so one expects that at the conductor’s surface only the normal component of ∇φ ~ will be
discontinuous. To see this we consider the following line integral along γ, see figure.
Z
L= ~l · ∇φds,
~
γ

where γ = γ(s), l = dds~x , that is “the velocity at which we circulate along γ if the parameter s
were the time".
4.3. CONDUCTORS 65

CONDUCTOR VACUUM
γ2

γ1 γ3

γ4

Figure 4.1: A loop with zero circulation

The value of that integral is the difference of the value of φ at the end of the curve and
the value of φ at the beginning of the curve, so, since in this case the curve is closed we have,
L = 0. That is
Z Z Z Z Z
0= ~l · ∇φ
~ dS = ~l · ∇φ
~ dS + ~l · ∇φ
~ dS + ~l · ∇φ
~ dS + ~l · ∇φ
~ dS,
γ γ1 γ2 γ3 γ4

where we have split the integral into four ones, we now consider the limit when the curve
approaches more and more the boundary of the conductor, in that case, γ1 → γ3 and γ2 →
~ discontinuous but finite, we see that the integrals along
0, γ4 → 0. Since we are assuming ∇φ
γ2 and γ4 go to zero, for γ2 and γ4 go to zero and so,
Z
0= ~l · (∇φ
~ i − ∇φ
~ e ) dS,
γ3

where (∇φ~ i ), (∇φ

~ e ) are the limiting value of ∇φ
~ from the inside and outside respectively.
Since the loop we were considering was arbitrary we conclude that ~l · ∇φ ~ i (~x) = ~l · ∇φ
~ e (~x),
where ~l is any tangential vector to the boundary and so that the tangential components of
~ are continuous across the boundary of a conductor. Thus, since inside E
∇φ ~ i = −(∇φ)
~ i = 0,
we see that at the boundary (∇φ)~ e can not have any tangential component. This implies that
φe at the boundary of the conductor is constant. This implies φ|S = φ0 .
What about the normal component? To study its discontinuity –it clearly must have one–
we consider now the following “pill box" integral, see figure. That is, a flat box with one face
inside the conductor and one outside, just the thin sides go across the boundary.
Z
~ d2 s
~n · ∇φ
S
66 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

CONDUCTOR
VACUUM

normal

Figure 4.2: A pill box

where ~n is the unit normal to the surface of the box, denoted by S, pointing towards the
outside of the box.
Using Gauss theorem we see that its value is
Z Z
~ · (∇φ)
∇ ~ d3~x = −4π ρ(~x) d3~x
box box
Taking now the limit when the faces of the box go flat on the boundary we see that the rim
of the box in that limit does not contribute. Thus,
Z Z Z Z
~ i+
n̂i · ((∇φ) ~ e ) d2 s =
n̂e (∇φ) ~ i − (∇φ)
n̂e · ((∇φ) ~ e ) d2 S = −4π σ(~xs ) d2 S,
S S S S

5
where σ(~xs ) is now the surface charge density, that is ρ(~x) = δ(~x − ~xs ) · σ(~xs ).
~ i = 0 and so we have,
In the present case, (∇φ)
Z Z Z
~ e d2 S = −
n̂e · (∇φ) ~ d2 S = −4π
n̂ · E σ(~xs ) d2 S.
S S S

~ |S = 4πσ |S . Thus, the normal component

Since the box was arbitrary, we see that n̂ · E
~ and so of ∇φ,
of E, ~ at the boundary of the box is given by the charge density there, but
we do not a priori know that charge density, so we do not a priori know the value of that
component. Notice that we do know for this problem the total charge, which was assumed to
be given. So we only know Z
~ d2 S = 4πq,
n̂ · ∇φ
S
where S is the conductor surface.
To summarize then, we only know that:
~ T |s = 0 that is φ |s = φ0 ← unknown constant.
1. (∇φ)
5
R
That is, Tρ (φ) = S σ(x)φ|S dS
4.3. CONDUCTORS 67

2.
R
~ d2 S = −4πq.
n̂ · ∇φ
S

3. φ(~x) → 0 when | ~x |→ ∞.

Are these three conditions enough to determine φ(~x) everywhere? The answer is in the
following theorem:

Theorem 4.2 Given a surface S and knowing that

1. φ |S = cnst. (not given)

2.
R
~ d2 S = −4πq
n̂ · ∇φ
S

3. φ(~x) → 0 | ~x |→ ∞

Then there is a unique solution to the problem:

∆φ = 0 inside and outside S.

Proof: We shall only prove uniqueness, to show existence amounts to know the core of
the theory of elliptic equations, Let φ1 and φ2 two solutions to ∆φi = 0 in D = lR3 − V , where
V is the volume occupied by conductor.

D = lR3 − V

Conductor n̂

S = ∂V

Figure 4.3: Geometrical setting for theorem 4.2

Then φ1 |s = c1 and φ2 |s = c2 , with c1 and c2 two constants, and

Z
~ i d2 S = −4πq.
n̂ · ∇φ
S

Thus, ϕ = φ1 − φ2 satisfies ϕ |s = c = c1 − c2 ,
R
~ = 0 and ϕ(~x) → 0 as | ~x |→ ∞, and
n̂ · ∇ϕ
s
furthermore 6 ∆ϕ = 0 in D.
6
Note that the same proof of uniqueness holds if there are (fixed) charges in the region D, for the difference,
ϕ, also satisfies Laplace’s equation.
68 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

Multiplying ∆ϕ by −ϕ and integrating over D we get,

Z
0 = [−ϕ∆ϕ] d3~x
ZD
= ~ · (ϕ∇ϕ)
[−∇ ~ + ∇ϕ
~ · ∇ϕ]
~ d3~x
ZD Z Z
= ~ · ∇ϕ]
[∇ϕ ~ d3~x − ~ d2 S +
ϕ n̂ · ∇ϕ ~ d2 S.
ϕ n̂ · ∇ϕ
D S(∞) S

Where in the second step we have used Gauss theorem, and in the integral over S we use
as ~n the outer normal. From the decay assumptions we have that the first surface integral
(at infinity) vanishes, while the second gives,
Z Z
~ d2 S = c
ϕn̂ · ∇ϕ ~ d2 S = 0
n̂ · ∇ϕ
S s
Thus,
Z
~ · ∇ϕ]
[∇ϕ ~ d3~x = 0
D
~ = 0 in D, and ϕ(~x) = cnst in D. But ϕ(~x) → 0 as | ~x |→ ∞ and so ϕ(~x) = 0 in
and so ∇ϕ
D.

Example: Conducting ball with charge Q

We are given a ball of radius a, say, and charge Q and ask to find the potential outside it.
We know that at the surface’s ball the electric field must be normal to it. That is, taking
the coordinates origin at the center of the ball, it must be radial so, φ(~x)||~x|=a = φ0 , a
constant.
Thus the exterior problem is:

1. ∆φ = 0 on lR3 − B(r = a).

2. φ |S = cnst. (not given)

3.
R
~ d2 S = −4πQ
n̂ · ∇φ
S

4. φ(~x) → 0 | ~x |→ ∞,

and it has a unique solution.

The uniqueness of the solution and the symmetry of the problem imply the electric field
is radial everywhere. Indeed, assume for contradiction that at some point ~x0 it would not be
radially pointing. Then performing a rotation which keeps ~x0 fixed we would find another
~ x0 ) will then be different from E(~
solution [for RE(~ ~ x0 )]. But since the boundary conditions
are not changed by a rotation the solution should not change, and we reach a contradiction.
Thus, everywhere we must have φ(~x) = φ(r). But the only solution to Laplace’s equation
which only depend on the radial coordinate are any linear combination of φ(r) = φ0 , a constant
4.3. CONDUCTORS 69

solution, and φ(r) = 1r Since the solution must go to zero at infinity the constant contribution
is ruled out and so φ(r) = rc . To find the value for c we must impose the condition on the
total charge, namely,
Z
~ d2 S = −
Z
c
4πQ = − n̂ · ∇φ ∂r ( )|r=a r 2 d2 Ω = 4πc
S S r
Q
Thus the solution is φ(~x) = r , and the value of the potential at the boundary is φ0 = Qa .
Notice that the relation among the potential and the charge is Q = aφ0 so the sphere
capacity is C = a.

Exercise: Show uniqueness in a region surrounded by a conductor.

Exercise: Show that φ |s = f , a given function at S, ∆φ = 4πρ outside S, and φ(~x) → 0 as

| ~x |→ ∞, implies φ(~x) is uniquely determined outside S.

Exercise: Find an example that shows that, even if φ is known everywhere outside a surface
S, we do not know its value inside, nor can determine ρ inside. This is particularly strange
because if ρ were an analytic function, then φ would be analytic also, and so the knowledge of
φ in an arbitrary open neighborhood of any point would determine completely φ everywhere.

Exercise: Show that nevertheless some knowledge about ρ inside S we can obtain from the
knowledge of φ at S, although indirectly. Hint: Use Coulomb’s Law.

4.3.1 The principle of superposition

Since Maxwell’s equations are linear, to any solution to the system one can add a homogeneous
solution and obtain a new one (in general with different initial and boundary conditions). This
is very helpful! We illustrate this here:
Suppose we have the following configuration (see figure) of fixed charge density ρ, and
conductors {Si }, i = 1..N, each with charges {Qi }. Thus we are looking for a solution
satisfying:

1. φ|Si = φi0 , i = 1..N at the conductor surfaces.

2. ∆φ = −4πρ in the space outside de conductors.

This seems to be complicated, but we can reduce it to simpler problems: We look for
N + 1 solutions. The first one satisfying,

1. φ|Si = 0, i = 1..N
2. ∆φ = −4πρ
70 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

This problem has a unique solution which we call φI .

We then solve the following set of equations:

1. φj |Sj = 1

2. φj |Si = 0, j 6= i

3. ∆φj = 0

They also have unique solutions, φ(~x)i . We now scale them by the constants φi0 so that they
satisfy the required boundary conditions. Then,

φi0 φi (~x)
X
φ(~x) = φI (~x) +
i

satisfy the required boundary conditions for the first problem. Shortly we shall see how to do
it when the total charges at each conductor is given.
We could have also split the sources in simpler ones and solve a problem for each one of
them and then sum all of them together. This is also very handy in solving some problems.

Exercise: Show using these techniques, and Gauss theorem the solution to the following
situation: There are two infinite flat conducting slabs held parallel to each other at a distance
L. Each one of thickness l1 and l2 respectively and charges q1 and q2 . Show:

• The charge densities on each of the internal faces of each slab are equal and of opposite
sign.

• The charge densities on each of the two external faces are equal.

• The value of the fields and charge densities are independent of the length.

• What happens when the two charges are equal but opposite?

Hint: Think on the charge densities and forget about the conductors, except to impose
the conditions that the field is zero inside them.

4.3.2 Example: [Method of Images]

Let us try to compute the potential field of a charge q outside a conducting sphere at zero
potential. We choose coordinate axis with origin at the center of the conducting sphere,
which we assume has a radius a, and such that the point-like charge is located at coordinates
~xq = (r, 0, 0). We notice that the problem has azimuthal symmetry around de x-axis, that
is, symmetry under rotations along that axis, so the solution we are seeking should also have
that symmetry. Indeed, assume that this is not the case, then there would exist a solution
without that symmetry, which nevertheless vanishes at the surface of the sphere of radius a
and has a source at ~x = (r, 0, 0). If we rotate that solution along the symmetry axis, the
4.3. CONDUCTORS 71

resulting function would also be a solution to Poison’s equations with a charge at ~x = (r, 0, 0)
and would also vanish at the surface of radius a. But there could be at most one solution
with those characteristics, so the rotated solution must be the same as the original, thus it
must have the required symmetry.

~x
a n̂
q′ q
r′ r

Figure 4.4: Point like charge outside a conducting sphere

If we pretend to continue the solution inside the sphere, then at the symmetry axis the
equipotential surfaces can only become points or cross orthonormal to it. This suggests that
we look there for a solution with point-like sources along the symmetry axis. We try with
just one charge there, say at the point ~xq′ = (r ′ , 0, 0). Then, the potential due to these two
charges is:
q q′
φ(~x) = + ,
| ~x − rx̂ | | ~x − r ′ x̂ |
where x̂ = (1, 0, 0).
At the surface of the sphere we have ~x = an̂ and so,

q q′
φ(~x) = +
| an̂ − rx̂ | | an̂ − r ′ x̂ |
q q′
= √ + √
a2 + r 2 − 2a rn̂ · x̂ a2 + r ′2 − 2a r ′ n̂ · x̂
q q′
= q + q . (4.9)
a 1 + r 2 /a2 − 2r n̂·x̂
a
r ′ 1 + a2 /r ′2 − 2a n̂·x̂
r′

The potential would then vanish at |~x| = a if we choose:

q′ q
′
=− ,
r a
and
a r
′
= ,
r a
72 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

Figure 4.5: Contour plot on the z = 0 plane of the potential of two charges as in the problem.

′ 2
that is q ′ = − qra = − qa
r
, and r ′ = ar < a. See figure (4.5) with the contour plots in the z = 0
plane.
What have we done? We have constructed a solution to Poison’s equation which vanishes
at | ~x |= a and which outside the sphere of radius a has just a single point-like source at
~xq = (r, 0, 0). But this is what we were looking for! For the solution to the problem outside
the sphere is unique.
What we have done is to mimic the effect of the surface charge density at the surface of
the conducting sphere with a point-like source in its interior, by giving to it a special location
and strength. Indeed, we can now use the formula +4πσ = −n̂ · ∇φ ~ |S to compute such
distribution.
qa
4πσ d2 S = q ′ =
R
Exercise: Compute σ using the formula above and check that r=a r
.

Exercise: How would you find a solution for the same problem but assuming now that the
sphere is at potential φ 6= V ? Compute the total charge that would have to be at the sphere.

Note that it is easy to compute the total force the sphere exerts on the charge q, it is just
the force q ′ would exert on q would it exist, namely
F~ = (F, 0, 0),

−q 2 a −q 2 ra
F = =
r | r − a2 /r |2 (r 2 − a2 )2

Exercise: Use the principle of action-reaction to compute the total force exerted by the
4.4. CAPACITIES 73

charge q on the conducting sphere. Compare with the result obtained using Maxwell stress
tensor.

An interesting limiting situation is the limit a → ∞ but ∆x := r − a finite. In this case

−qa
q′ = → −q
a + ∆x
a2 a2 + a∆x − a2 a∆x
∆x′ := a − r ′ = a − = = → ∆x
a + ∆x a + ∆x a + ∆x
and
−q 2 a(a + ∆x) −q 2 a(a + ∆x) −q 2
F = = → ,
((a + ∆x)2 − a2 )2 (2a∆x + (∆x)2 )2 4(∆x)2
that is the force exerted by a charge of strength q at a distance of 2∆x of the charge q. This
is in fact the force exerted by an infinite conducting plane at distance ∆x of q.
The general strategy of this method is as follows. Let us assume we have a region V of
3
lR where we want to solve ∆φ = −4πρ, with ρ given and subject to a boundary condition
φ|∂V = f . We first extend ρ inside V − lR3 as vanishing there and use Poison’s formula to
obtain a φs solution to ∆φs = −4πρ in the whole of lR3 . Of course this solution in general
would not satisfy the boundary condition. So we call g = f − φs |∂V and solve in V ∆φe = 0
subject to the boundary condition, φe∂V = g, and decaying to zero at infinity. Thus, in V
we have now the solution we were looking for, φ = φs + φe satisfies both the equation and
the boundary condition. We now extend the solution inside V − lR3 . φs is already defined
there by Poison’s formula, while we can extend φe any way we want. Since there φe can be
shown to be analytic we can extend it in a region inside also having vanishing Laplacian. But
there would be a point where we can not extend it any longer as a smooth function without
keeping its Laplacian vanishing. Thus we will have some deviation and that would produce
some internal sources, ρ̃ = −14π
∆φe . Those are the mirror charges.
Note that ρ̃ has no physical meaning on what counts for the problem inside region V .
In fact it is by no means unique. Indeed suppose we have found a φ in lR3 corresponding
to a ρT = ρ + ρ̃. Then we can take a second smooth potential φ′ which is identical to the
first inside V and outside a region surrounding V , but different in the region in between, this
would satisfy everything we want, but would result (computing its Laplacian) in a different
ρ̃.

4.4 Capacities
Consider a configuration of conductors as shown in the figure (4.6) below, on an otherwise
empty space.
Imagine that to get to this distribution we fixed the potentials, V i , i = 1...n on each
conductor and solved,

∆φ = 0 in lR3 − {conductor’s volume} (4.10)

φ|∂Si = V i (4.11)
74 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS

D = lR3 − Vi
S
i
V1
V2
S1 = ∂V1

V3

Figure 4.6: A configuration of conductors

Correspondingly there would be fixed ammount of charges on the surface of each conductor,
given by,

1
Z
Qi = ~ dS 2
n̂ · ∇φ (4.12)
4π ∂Si

If we had fixed another set of potentials, Ṽ i , then we would have obtained another poten-
tial, φ̃ and correspondingly another set of charges Q̃i . Now, if we consider the configuration
where the potentials at the conductors are given by V̂ i = V i + cṼ i where c is any constant,
then the the potential field would be φ̂ = φ + cφ̃. Indeed, it follows from the linearity of the
equations and of the boundary conditions that φ̂(~x) satisfies the equation and the boundary
conditions, but uniqueness of the solutions implies this is the right one. But then, from the
linear dependence of the charges on the potential field, (4.12), the total charge would be,
Q̂i = Qi + cQ̃i . That is, there exists a linear relation between potentials at conductors and
their charges! Thus, there must exist a matrix, Cij such that,

Qi = Cij V i (4.13)
This matrix, called the capacities matrix, can only depend on the geometrical configuration
of the conductors distribution.

Exercise: Show that this matrix in invertible by setting the problem of fixing the charges
and finding the potentials.

Exercise: Show that the matrix is symmetric, that is, Cij = Cji. Hint: choose any pair
of conductors, say the i − th and the j − th, solve the potential problem for the boundary
conditions {V i = 1, v k = 0, k 6= i} call that solution φi . Do the same for the j − th conductor
and call the solution so obtained, φj . We then have, for the j − th solution,

1 Z ~ j dS 2 = 1
Z
~ j dS 2 ,
Cik V k = Cij V j = Cij = n̂ · ∇φ φi n̂ · ∇φ (4.14)
4π Si 4π Si
4.4. CAPACITIES 75

where in the last equality we have used that φi |Si = 1. Find a similar formula for Cji, subtract
the first from the second and use the divergence theorem to convert the two surface integrals
in a volume one. Finally show that the volume integral vanishes.
76 CHAPTER 4. STATIONARY SOLUTIONS: ELECTROSTATICS
Chapter 5

Static Problems – Constructing

Interesting Solutions

5.1 Method of Separation of Variables in Cartesian Co-

ordinates
We would like now to solve problems where the boundary is a rectangular surface, that is
the boundary of a box. Thus, the coordinates adapted to such a boundary are the Cartesian
coordinates aligned perpendicular to the rectangular faces of the box. Here we shall be solving
the problem of finding the solution inside such a box under the assumption that in one of the
faces the potential is a given function while in the others vanishes. Thus a general solution
(with a potential arbitrarily given on each of the faces) can be obtained by adding six solutions
as the one we shall find.
To fix ideas consider the following rectangle {(x, y, z)|0 ≤ x ≤ a, 0 ≤ y ≤ b, 0 ≤ z ≤ c}
and assume the boundary conditions to be:

φ(0, y, z) = φ(a, y, z) = φ(x, 0, z) = φ(x, b, z) = φ(x, y, 0) = 0 φ(x, y, c) = V (xy)

In Cartesian coordinates (x, y, z) the Laplacian takes the form:

∂2 ∂2 ∂2
∆φ = φ + φ + φ.
∂x2 ∂y 2 ∂z 2
Thus, if we assume a solution of the form,

φ(x, y, z) = X(x)Y (y)Z(z),

we find,

∂2 ∂2 ∂2
∆φ = Y (y)Z(z) X(x) + X(x)Z(z) Y (y) + X(x)Y (y) Z(z)
∂x2 ∂y 2 ∂z 2
or
1 ∂2 1 ∂2 1 ∂2
∆φ = [ X(x) + Y (y) + Z(z)]φ(x, y, z)
X(x) ∂x2 Y (y) ∂y 2 Z(z) ∂z 2

77
78CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

b y

Figure 5.1: A conducting box with only its upper face at non-zero potential.

Since each term inside the brackets is a function of a different variable, each term must be
constant and the sum of them must be zero. That is, we need,

∂2 ∂2 1 ∂2
X(x) = α2 X(x) Y (y) = β 2 Y (y) Z(z) = γ 2 Z(z)
∂x2 ∂y 2 Z(z) ∂z 2
with the constants satisfying,
α2 + β 2 + γ 2 = 0.
We have used squared quantities for later convenience, there is no loss of generality on that for
we shall consider them as complex numbers. We shall see later how to construct the general
solution as linear combination of these basic ones.
Consider any of the above equations, say,

∂2
2
X(x) = α2 X(x)
∂x
The general solution to it is given by, 1

X(x) = A+ eαx + A− e−αx .

Likewise, we will have,

Y (y) = B + eβy + B − e−βy ,

1
Every second order equation has two linearly independent solutions, since one can prescribe two indepen-
dent values (for the function and its first derivative) at any given point and integrate the equation from there
on.
5.1. METHOD OF SEPARATION OF VARIABLES IN CARTESIAN COORDINATES 79

and

Z(z) = C + eγz + C − e−γz .

We impose now the boundary conditions, φ(0, y, z) = φ(a, y, z) = 0. They are just condi-
tions on X(x) for they must hold whatever are Y (y) and Z(z) there. Evaluation of X(x) at
the boundaries gives,

A+ + A− = 0 A+ eαa + A− e−αa = 0.
That is,

A+ (eαa − e−αa ) = 0.
Thus, to have a nontrivial solution we need,

eαa = e−αa or eαa = e−αa or e2αa = 1

iπn
Thus, α = a
, and the solutions have the form,
πnx
X(x) = An sin( )
a
2
for any given n = 1, 2, .....
Similarly, for Y (y),
πmy
Y (y) = Bn sin( )
b
for any given m = 0, 1, 2, .....
But then we must have, γnm 2
= ( πn
a
)2 + ( πm
b
)2 , and so γ must be real. Thus, the solution is

Z(z) = C + eγnm z + C − e−γnm z

and the boundary condition φ(x, y, 0) = 0 implies C − = −C + .
Thus, we have found solutions of the form,
πnx πmy
φnm (x, y, z) = Cnm sin(
) sin( ) sinh(γnm z).
a b
We claim now that we can construct the general solution we are seeking from these ones. For
that we consider a sum of solutions as above,
∞ X
∞
X πnx πmy
φ(x, y, z) = Cnm sin( ) sin( ) sinh(γnm z)
n=1 m=1 a b
and the boundary condition is now,
∞ X
∞
X πnx πmy
φ(x, y, c) = Cnm sin( ) sin( ) sinh(γnm c) = V (x, y).
n=1 m=1 a b
2
The solutions for negative n are the same, and for n = 0 vanishes.
80CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

3
Thus, the coefficients Cnm just proportional to the Fourier coefficients of V (x, y),
4 Z aZ b
πnx πmy
Cnm = V (x, y) sin( ) sin( ) dx dy.
ab sinh(γnm c) 0 0 a b

Exercise: Check the above formula by multiplying φ(x, y, c) by sin( πpx

a
) sin( πqy
b
) and inte-
R1 1
grating with respect to x and y using that 0 sin(πnx) sin(πpx)dx = 2 δnp .

Example: Box with all faces at zero potential except one at V0 . We take the
coordinate system with the axis perpendicular to the faces and centered at one of boxes
corner, so that the non-zero potential face corresponds to z = c. The Fourier coefficients are
obtained from the above formula or from a table if one considers a potential function defined
in the rectangle [−a, a] × [−b, b] in such a way that is odd under x → −x or y → −y. In any
case, the coefficients are:
16V0
Cnm = n, m odd Cnm = 0, all other cases
π 2 nm

Figure 5.2: The value of the potential at z=c with 10 Fourier modes

3
To see this consider the Fourier transform on the rectangle [−a, a]×[−b, b] and describe only odd functions
in both variables, thus you need only the sin function and can dispense all the cos ones. The factor 4 that
appears in the formula is due to the fact that we perform the integrals just in one quarter of the whole
rectangle.
5.1. METHOD OF SEPARATION OF VARIABLES IN CARTESIAN COORDINATES 81

Figure 5.3: The value of the potential at y=b/2 (0.5)

Figure 5.4: The value of the potential at y=b/2 (5) for a wide box

Figure 5.5: The value of the potential at y=5 with only 3 Fourier modes
82CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

5.2 Method of Separation of Variables in Spherical Co-

ordinates
In spherical coordinates (r, θ, ϕ) the Laplacian takes the form:

1 ∂2 1 ∂ ∂φ 1 ∂2φ
∆φ = (rφ) + (sin θ ) +
r ∂r 2 r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂ϕ2

Exercise: Show this using the chain rule.

We shall look for solutions to Laplace’s equation using the method of separation of vari-
ables, with the hope to find enough of them to account, using linear combination, for all
possible solution.
To this end we propose solutions of the form φ(r, θ, ϕ) = U r(r) P (θ)Q(ϕ). Substituting in
2 sin2 θ
the above formula for the Laplacian and multiplying the result by rφ(r,θ,ϕ) we obtain:

1 d2 U 1 d dP 1 d2 Q
r 2 sin2 θ[ + (sin θ )] + = 0.
U dr 2 P r 2 sin θ dθ dθ Q dϕ2
Since the first term can only be a function of r and θ, while the second only a function of ϕ
and r, we conclude that in order to cancel each one of them must be constant, thus:
d2 Q
2
= α2 Q, for some constant α,
dϕ
that is Q(ϕ) = Qα eαϕ . If we look for solutions defined for all angles ϕε[0, 2π], then we must
have Q(ϕ) = Q(ϕ + 2π) = Q(ϕ)e2πα and therefore α = ±im, that is α2 = −m2 , and the
general solution is:
Q(ϕ) = Q+ me
imϕ
+ Q−me
−imϕ
,
where Q+ −
m and Qm are two arbitrary constants to be determined latter when imposing bound-
ary conditions.
The rest of the equation becomes then,

r 2 sin2 θ d2 U sin θ d dP
2
+ (sin θ )] + α2 = 0.
U dr P dθ dθ
Thus, since the first term is the only one that depends on r, we conclude that
d2 U βU
2
= 2, for some constant β.
dr r
The solutions to this equation are:

Uν (r) = Ar ν , for β = ν(ν − 1).

√
1± 1+4β
Note that for given β there are two ν ′ s, ν+ and ν− , ν± = 2
, and ν+ + ν− = 1, thus,
for given β we have,
5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 83

Uβ (r) = Uβ+ r ν+ + Uβ− r ν−

= Uβ+ r ν+ + Uβ− r −(ν+ −1) .

In terms of the potential we need, U(r)/r, so it is customary to express these equations

in terms of the parameter l = ν+ − 1, which we shall use from now on.
The equation for P (θ) is a bit more complicated and is:
d dP
(l(l + 1) sin2 θ + α2 )P + sin θ (sin θ )=0
dθ dθ

√ It becomes d simpler if we change variables to x = cos θ, then dx = − sin θdθ, sin θ =

1 − x2 , and dθ = dx d
dθ dx
d
= − sin θ dx . Therefore, the above equation becomes,

d dP
(1 − x2 ) ((1 − x2 ) ) + (α2 + l(l + 1)(1 − x2 ))P = 0,
dx dx
or
d dP α2
((1 − x2 ) ) + ( + l(l + 1)P = 0,
dx dx 1 − x2
which is known as Legendre’s equation.
We shall first treat the case α = 0, that is we shall consider only solutions with azimuthal
symmetry. Latter we shall generalize to α2 6= 0. In this case the equation becomes,
d dP
((1 − x2 ) ) + l(l + 1)P = 0.
dx dx
We shall look for solutions valid for the whole interval of θ, namely [0, π], that is for
solutions valid in the x-interval [−1, 1]. Assuming that in that interval the solutions admit a
convergent power series representation,
∞
aj xj ,
X
P (x) =
j=0

and substituting into the equation we find,

∞
d
((1 − x2 )jaj xj−1 ) + l(l + 1)aj xj }
X
0 = {
j=0 dx
∞
{(1 − x2 )j(j − 1)aj xj−2 − 2jaj xj + l(l + 1)aj xj }
X
=
j=0
∞
{[−j(j − 1) − 2j + l(l + 1)]aj + (j + 2)(j + 2 − 1)aj+2 }xj = 0.
X
=
j=0

Therefore, we must have,

j(j + 1) − l(l + 1)
aj+2 = aj
(j + 1)(j + 2)
84CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

Thus, we can split the solution into two series, one with a0 6= 0, a1 = 0 and the other with
a0 = 0, a1 6= 0. One is even in x, the other odd, so they are linearly independent solutions.
Since every second order ordinary differential equation has two linearly independent solutions
at every point, we know we have found all possible solutions.
j
For j > l, |aj+2 | < j+2 |aj | < |aj |, so |aj | < c for some c ≥ 0. Thus, in the interval of
interest, x ∈ [−1, 1], we have,
∞ ∞ ∞
|aj xj | ≤ C + |aj xj | ≤ C + c |x|j
X X X

j=0 j>l j>l

where the series in the last term converges for all |x| < 1. Thus, the series converges for all
|x| < 1. But in general it diverges at the boundaries, (see exercise below) unless it terminates
for finite j. So if we want to consider solutions which are everywhere smooth, including the
poles of the sphere, then the series must terminate. This happens when l is an integer, and
the series becomes a polynomial of order l in x. They are called the Legendre Polynomials,
and the first ones are:

P0 (x) = 1,
P1 (x) = x,
P2 (x) = 1/2(3x2 − 1),

Notice that the first corresponds to the choice a0 = 1, a1 = 0,, l = 0, the second to
a0 = 0, a1 = 1, l = 1, and the third to a0 = −1/2, a1 = 0, l = 3. They are normalized in
such a way that,
Pl (1) = 1

Exercise: Compute P3 and P4 using the recursion relation found and the normalization
condition

Exercise: Convince yourself that indeed the above series diverge at |x| = 1 unless they
terminate. Hint: First show that for any ε > 0 there exists integer J such that |aj+2 | >
(1 − ε)|aj | ∀ j > J. Second use the previous point to establish that |aj+J | > (1−εj ) J |aJ |.
j
j
j+2
2j 2j+1
Compare this result with the series ∞ s
j=0 2j , and
∞ s
j=0 2j+1 . Show, using that 1−s =
1 ∞
j=0 s
j
P P P

that the above series represent the functions 12 ln(1 + s)(1 − s) and 12 ln( 1−s
1+s
) respectively and
so that they diverge for |s| = 1. Conclude, using s = x(1 − ε), and bounding one series with
the other, that the series diverges for all |x| ≥ 1.

Since for each l we obtain a polynomial of degree l, with linear combinations of then
we can obtain all powers of x. But since every smooth function on a compact set can be
approximated to arbitrary precision with a polynomial with appropriated coefficients. We
can approximate any function by linear combination of the Legendre polynomials, thus they
5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 85

form a complete basis to expand smooth functions in the interval [−1, 1]. If we consider the
scalar product in the space of smooth functions in [−1, 1] given by,
Z 1
< f, g >= f¯(x)g(x)dx,
−1

then one can see that the Legendre polynomials form a orthogonal basis with respect to this
scalar product, indeed,

Z 1
− [l(l + 1) − l′ (l′ + 1)] Pl (x)Pl′ (x)dx
−1
d1 dPl d dPl′
Z
= [(1 − x2 )
[ ] Pl′ − [(1 − x2 ) ] Pl ] dx
−1 dx dx dx dx
dPl dPl′ dPl′ dPl
Z 1
= [−(1 − x2 ) + (1 − x2 ) ] dx
−1 dx dx dx dx
dPl dPl′
+ (1 − x2 ) Pl′ |1−1 − (1 − x2 ) Pl |1−1 = 0
dx dx
thus, Z 1
Pl (x)Pl′ (x)dx = 0 if l 6= l′ .
−1
On the other hand it can be seen using special properties of these polynomials that
1 2
Z
Pl2 (x)dx = .
−1 2l + 1

Exercise: Use the explicit formula of the first four Legendre polynomials and the orthogo-
nality relations to compute Z 1
x4 Pl (x)dx.
−1

Example: Two spherical cups at opposite potentials.

For problems with azimuthal symmetry, that is, where the boundary conditions do not
depend on the azimuthal angle ϕ we can already use the solutions found. In this case the
general solution has the form:
∞
(Ul+ r l + Ul− r −(l+1) )Pl (cos(θ))
X
φ(r, ϕ) = (5.1)
l=0

For constants Ul+ , Ul− to be determined from the boundary conditions.

Imagine you have two spherical conducting cups of radios a, separated by an insulator
sheet. The north one at potential φ0 , the south one at −φ0 . We want to find the field at the
external region outside the conductors.
86CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

Thus, we are solving the following boundary problem:

∆φ = 0 at V = lR3 − Ba
φ|∂V + = φ0
φ|∂V − = −φ0
φ(~x) → 0 as |~x| → ∞ (5.2)

Where we have divided the sphere into two parts, V + , and V − according to the the sign of
the coordinate z. The asymptotic condition implies Ul+ = 0 ∀l so we have only to find the
coefficients Ul± from the other two boundary conditions. In terms of the coordinates used
these two boundary conditions result in the following one:
∞
Ul− a−(l+1) Pl (cos(θ)) = φ0 (cos(θ))
X
φ(a, θ) = (5.3)
l=0

where we have defined, (in terms of x := cos(θ)):

(
−φ0 x ∈ [−1, 0]
φ0 (x) := (5.4)
φ0 x ∈ [0, 1]
Using the orthogonality of the Legendre polynomials we find that [?]:

2l + 1
Z 1  0 l even
Ul− a−(l+1) = Pl (x)φ0 (x) dx =  (− 1 )(l−1)/2 (2l+1)(l−2)!! l odd (5.5)
2 −1 2 2( l+1 )! 2

And therefore the potential takes the form:

∞
1 (4m + 3)(2m − 1)!! r −(2m+2)
(− )m
X
φ(a, θ) = ( ) P2m+1 (cos(θ)) (5.6)
m=0 2 2(m + 1)! a
Evaluation at r = a with up to m = 9 is shown in the plot below.
The potential with twenty coefficients is shown below.

Exercise: Find the solution inside the cups (assuming them hollow). Then apply Gauss law
and compute the surface charge distribution.

Exercise: Integrate the surface charge distribution and find the configuration’s capacity.

Example: Charged spherical crust with a circular hole.

We consider the case of a spherical crust of constant surface density σ with a circular hole.
This configuration has azimuthal symmetry around an axis which connects the center of the
circular hole with the center of the sphere. Thus, it is convenient to use spherical coordinates
5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 87

Figure 5.6: Potential at cusps with 9 coefficients.

Figure 5.7: Potential with 20 coefficients.

88CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

with the z axis along the symmetry axis. The parameters describing the problem are then,
the charge density, σ, the radius of the crust, a, and the angular aperture of the hole, α.
The procedure to solve this problem is as follows, we know that inside the crust the
potential satisfies Laplace’s equation, and also outside. Thus, we can propose an internal and
an external solution and match them at the interface r = a. They are given by,

∞
Ulint r l Pl (cos(θ))
X
φint (r, θ) := r < a
l=0
∞
Ulext r −(l+1) Pl (cos(θ))
X
φext (r, θ) := r > a
l=0

The coefficients will be determined by imposing the two interface conditions:

φint (a, θ) = φext(a, θ) Continuity

∂ ∂
{ φint (r, θ) − φext (r, θ)}|r=a := 4πσ(θ) Gauss relation,
∂r ∂r
where (
0 0≤θ≤α
σ(θ) :=
σ α<θ≤π
The first says,
∞
[Ulint al − Ulext a−(l+1) ]Pl (cos(θ)) = 0.
X
φint (a, θ) − φext (a, θ) =
l=0

But the Legendre polynomials are linearly independent, so we conclude:

Ulint al − Ulext a−(l+1) = 0 ∀l = 0 . . . ∞.

To obtain the other relation we need to express σ(θ) as a series in Legendre polynomials,
∞
X
σ(θ) := σn Pn (cos(θ)). (5.7)
n=0

In that case, Gauss relation gives,

lUlint al−1 + (l + 1)Ulext a−(l+2) = 4πσl

Using the first relation we get,

(2l + 1)Ulint al−1 = 4πσl

Multiplying the expression on both sides of (5.7) by Pl (cos(θ)), integrating, and using the
orthogonality relation among Legendre polynomials we find,
2l + 1 π 2l + 1 π
Z Z
σl = σ(θ)Pl (cos(θ)) sin(θ)dθ = σPl (cos(θ)) sin(θ)dθ.
2 0 2 α
5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 89

To perform the integral it is useful to use the following relation among Legendre polynomials,

∂ ∂
Pl+1 (x) − Pl−1 (x) − (2l + 1)Pl (x) = 0.
∂x ∂x
Using it we get,

σ cos(α) ∂ ∂
Z
σl = [ Pl+1 (x) − Pl−1 (x)] dx
2 −1 ∂x ∂x
σ
= [Pl+1 (cos(α)) − Pl+1 (−1) − Pl−1 (cos(α)) + Pl−1 (−1)]
2
σ
= [Pl+1 (cos(α)) − (−1)l+2 − Pl−1 (cos(α)) + (−1)l ]
2
σ
= [Pl+1 (cos(α)) − Pl−1 (cos(α))]
2
Thus, the second interface conditions shields,

4π 2π
Ulint = σl = [Pl+1 (cos(α)) − Pl−1 (cos(α))],
2l + 1 2l + 1
and we have solved the problem.

Exercise: Express both potential in all details.

Exercise: Using the above result find the first two multipole moments in terms of the total
charge of the crust, Q, the radius, a, and α.

Exercise: Find the solutions when there are two holes, one in the north pole and one on the
south, of aperture α+ and α− respectively. Look at the limit, when the angles go to π/2 and
σ → ∞ but keeping the total charge constant.

5.2.1 Associated Legendre Functions and Spherical Harmonics

We return now to the task of solving Legendre’s equation when azimuthal dependence is
present, that is when m 6= 0. Recall that in that case one is seeking a solution to:

d dP m2
[(1 − x2 ) ] − P = −l(l + 1)P,
dx dx 1 − x2
where we have already assumed l integer for this is needed to obtain smooth solutions in the
whole sphere.
It can be seen that the following are solutions to this equation:
90CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

dm m 2 m/2
Plm (x) := (−1) (1 − x ) Pl (x) m ≥ 0.
dxm
The coefficient, (−1)m , appears only for convention and does not play any important role.

Exercise: Show that these are solutions by first proving that

dm+2 dm+1 dm dm
(1 − x2 ) P l (x) − 2(m + 1) P l (x) − m(m + 1) P l (x) = l(l + 1) Pl (x)
dxm+2 dxm+1 dxm dxm

Remark: Note that the solutions are no longer polynomials for m odd.
Remark: Note also that Plm = 0 for m > l.
For fixed m these functions, called generalized Legendre functions, satisfy the following
orthogonality condition (same m):
1 2 (l + m)!
Z
Pl′ m Plm dx = δll′ .
−1 2l + 1 l − m)!
In practice, it is convenient to work with the whole angular part of the solutions to the
Laplace’s equation, that is to multiply Plm by Qm (ϕ) = eimϕ .
Renormalizing them for convenience we define the spherical harmonics of type (l, m):
v
u 2l + 1 (l − m)!
u
Ylm (θ, ϕ) := t Plm (cos θ)eimϕ , m ≥ 0.
4π (l + m)!
But we also need solutions with m negative, those we define as:
v
u 2l + 1 (l−| m |)!
u
Ylm (θϕ) := t Pl |m| (cos θ)(−1)m eimϕ , m < 0,
4π (l+ | m |)!
or in other words:
Yl−m (θ, ϕ) = (−1)m Ylm
⋆
(θ, ϕ),
where ⋆ means complex conjugation.

Exercise: Compute all the spherical harmonic functions for l = 0, 1, 2. Express them also in
Cartesian coordinates.

They have the following properties:

• Orthonormality
Z 2π Z π
dϕ sin θdθ Yl⋆′ m′ (θ, ϕ)Ylm(θ, ϕ) = δll′ δmm′
0 0

• Completeness They form a complete set of functions to expand arbitrary (smooth)

functions on the sphere. Note that this means, that given any smooth function f (θ, ϕ) on the
sphere, there exist constants Clm such that,
5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 91

Figure 5.8: Absolute value of the real part of Y20 (θ, φ)

Figure 5.9: Absolute value of the real part of Y21 (θ, φ)

92CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

Figure 5.10: Absolute value of the real part of Y30 (θ, φ)

Figure 5.11: Absolute value of the real part of Y31 (θ, φ)

5.2. METHOD OF SEPARATION OF VARIABLES IN SPHERICAL COORDINATES 93

∞ X
X m
f (θ, ϕ) = Clm Ylm (θ, ϕ),
l=0 l=−m

with Z
Clm = f (θ′ , ϕ′ )Ylm
⋆
(θ′ , ϕ′ )dΩ.
S2

But then,
∞ X
m Z
f (θ′ , ϕ′ )Ylm
⋆
(θ′ , ϕ′ )dΩYlm (θ, ϕ)
X
f (θ, ϕ) =
2
l=0 l=−m S
Z ∞ X
m
f (θ′ , ϕ′ ) ⋆
(θ′ , ϕ′ )Ylm (θ, ϕ)dΩ,
X
= Ylm
S2 l=0 l=−m

and so,
∞ X
l
⋆
(θ′ , ϕ′ )Ylm (θ, ϕ) = δ(ϕ − ϕ′ )δ(cos θ − cos θ′ ).
X
Ylm
l=0 m=−l

This is a convenient way of writing Dirac’s delta on the surface of a sphere. It shows also
that the commutation of the integral with the series is not an allowed step, and that the last
equality is just formal, neither the double series converges nor the integral makes sense, since
the integrand is not an integrable function.

Exercise: Write down Dirac’s delta on lR3 but in spherical coordinates.

What are these spherical harmonic functions on the sphere? Can we characterize them in
some invariant way? If we take the Laplacian in lR3 in spherical coordinates, set r = 1 and
dismiss all “derivatives" along the radial direction, we obtain the Laplacian of the unit sphere,

1 ∂ ∂φ 1 ∂2φ
∆S 2 φ := (sin θ ) + .
sin θ ∂θ ∂θ sin2 θ ∂ϕ2
The spherical harmonics we have defined are a complete, linearly independent set of solu-
tions to the following eigenfunction - eigenvalue problem:

∆S 2 Ui = λi Ui ,

that is,
∆S 2 Ylm (θ, ϕ) = −l(l + 1)Ylm (θ, ϕ).
Note that there are several, 2l + 1, solutions for each eigenvalue, as m ranges from −l to l.

Exercise: Check this explicitly.

So the Laplacian in the sphere splits naturally the space of smooth functions into invariant
subspaces of dimension 2l + 1.
94CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

Exercise: Show that,

Z Z
ψ∆S 2 φ dΩ = φ∆S 2 ψ dΩ,
S2 S2

for ψ and φ arbitrary smooth functions.

Exercise: Show the orthogonality property of the spherical harmonics with different l using
the previous exercise. Exercise: Use that for each l the spherical harmonic functions satisfy

∂2
Y
∂ϕ2 lm
= −m2 Ylm to show their orthogonality.

5.3 Application: Multipole moments of a static config-

uration
Let us assume that we have a fixed charge distribution, ρ, which is of compact support. That
is, it is only different from zero in a bounded region D ⊂ lR3 .
As we already know, the potential field corresponding to this situation is given by

ρ(~x′ ) 3 ′
Z
φ(~x) = d ~x . (5.8)
lR3 |~x − ~x′ |
This solution is unique and smooth outside D.
Outside D the potential φ satisfies the homogeneous solution, ∆φ(~x) = 0, so it must be
of the form:

∞ X
l
4π
Clm r −(l+1) Ylm (θ, ϕ),
X
φ(~x) = ~x ∈
/D (5.9)
l=0 m=−l 2l + 1
4π
for some constants Clm (the factor 2l+1 is for compatibility with standard definitions). These
constants are called the multipole moments of the electrostatic field. They characterize
uniquely the field outside the support of the static sources region. The bigger the l the faster
they decay at large distances and each one of them has a characteristic angular dependence.

Exercise: Compute the first moments (l = 0, 1, 2) as expressed in Cartesian coordinates.

Compute the corresponding electric field.

We shall see now how to compute these constants as functions of the source distribution ρ.
The first one, C00 is immediate by applying Gauss theorem. Indeed, if we perform a surface
integral of the normal component of the electric field at a surface r = R surrounding the
sources, that is, outside D, we have:
5.3. APPLICATION: MULTIPOLE MOMENTS OF A STATIC CONFIGURATION 95

Z Z
3
4πQ = 4πρ(~x) d ~x = − ∆φ(~x) d3~x
lRI3 I lR
3

= − ~ · d~s = −
∇φ ∂r φ|r=R R2 dΩ
S(R) S2
∞ X
l
4π
I
Clm (l + 1)R−(l+2) Ylm (θ, ϕ)R2 dΩ
X
=
S 2 l=0 m=−l 2l + 1
3
∞ X
l
(4π) 2
I
Clm (l + 1)R−l Ylm (θ, ϕ)Y ∗00 (θ, ϕ) dΩ
X
=
S 2 l=0 m=−l 2l + 1
3
= (4π) 2 C00 (5.10)

Where in the last step we have included Y00∗ (θ, ϕ) = √14π and use the orthogonality relations
among spherical harmonics. So we see that the first one, the one that decays as 1/r at long
distances, carries the information about the total mass of the charge system. We use now the
same strategy to find the others, recalling that ψ = r l Ylm
∗
is a solution to Laplace’s equation
we get,

Z Z
l ∗
4πqlm := 3
4πρ(r, θ, ϕ)r Ylm (θ, ϕ) dV = − 3
∆φ(r, θ, ϕ)r l Ylm
∗
(θ, ϕ) dV
ZlR IlR
= ~
∇φ(r, ~ l Y ∗ (θ, ϕ)) dV −
θ, ϕ)∇(r ∂r φ|r=R Rl Ylm
∗
(θ, ϕ) R2 dΩ
3 lm
lR S(R)
Z I
= − 3
∆(r l Y ∗lm (θ, ϕ))φ(r, θ, ϕ) dV + φ|r=R ∂r (r l Ylm
∗
(θ, ϕ))|r=R R2 dΩ
lR S(R)
I
− ∂r φ|r=R Rl Ylm
∗
(θ, ϕ) R2 dΩ
S(R)
∞ l′
I
4π ′ ∗
Cl′ m′ l R(l +1)−(l−1) Yl′ m′ (θ, ϕ)Ylm
X X
= (θ, ϕ)) dΩ
S 2 l′ =0 m′ =−l′ 2l′+1
∞ l′
4π
I
′ ∗
Cl′ m′ (l′ + 1) R(l +2)−(l+2) Yl′ m′ (θ, ϕ)Ylm
X X
+ (θ, ϕ)) dΩ
S2 l′ =0 m′ =−l′
2l′+1
∞ l′
4π
I
′ ∗
Cl′ m′ (l + l′ + 1) Rl −l Yl′ m′ (θ, ϕ)Ylm
X X
= (θ, ϕ)) dΩ
S 2 l′ =0 m′ =−l′ 2l′+1
= 4π Clm (5.11)

where in the fourth line we have used that ψ = r l Ylm ∗

is a solution to Laplace’s equation
and in the last one the orthogonality relation among spherical harmonics. Notice that the
integration by parts used above are a special case of the second Green’s identity.
Z I
[ψ∆φ − φ∆ψ] dV = [ψ∂n φ − φ∂n ψ] dS 2 (5.12)
V ∂V

Exercise: Show the above identity.

96CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

So we see that we have an explicit and simple relation between the multipole moments
of the fields and certain integrals over the sources, which are called the multipole moments
of the sources. Do this multipole moments of the sources characterize completely them?,
namely, can we recuperate ρ from our knowledge of the {qlm }’s? Notice that this would imply
that by knowing the field outside the sources we could know the sources themselves! This is
not so, because it is easy to find different source distributions for which the corresponding
fields outside them are identical. The simplest example is to take two different spherically
symmetric charge distributions with the same total charge, but it is easy to build many
examples. Nevertheless, the multipole moments are a very important tool to describe the
qualitative behavior of the fields knowing the source distributions and also the other way
around, knowing the field distribution obtain information about the source distribution. This
is mostly used to perform geological studies, where instead of the electric field potential people
look at the gravitational potential, which satisfies the same equation but the matter density
as the source.

Exercise: Find two different charge distributions with no symmetry at all but giving the
same external field (which can have symmetries). Hint: do not construct them explicitly, but
just start from their potentials and work backwards.

Exercise: Show that for real charge distributions we have qlm = (−1)m ql,−m
∗
.

Exercise: Compute the expressions in Cartesian coordinates of the first three source mo-
ments.

5.3.1 The Addition Theorem for Spherical Harmonics

In the above section we have shown that outside the sources
∞ X
l
4π
qlm r −(l+1) Ylm (θ, ϕ)
X
φ(~x) = (5.13)
l=0 m=−l 2l + 1
using now the expression for qlm obtained we get,

l
∞ X
4π
qlm r −(l+1) Ylm (θ, ϕ)
X
φ(~x) =
l=0 m=−l 2l + 1
∞ X
l
4π Z
[ ρ(~x′ ) (r ′ )l Ylm
∗
(θ′ , ϕ′ ) d3~x′ ] r −(l+1) Ylm (θ, ϕ)
X
=
l=0 m=−l 2l + 1 V

∞ X
l
4π (r ′ )l ∗ ′ ′
Z
ρ(~x′ ) Y (θ , ϕ ) Ylm (θ, ϕ) d3~x′
X
= (l+1) lm
V l=0 m=−l 2l + 1 r
(5.14)
5.3. APPLICATION: MULTIPOLE MOMENTS OF A STATIC CONFIGURATION 97

Since we are in the region far from the sources, r ′ < r and one can show the expression
′
converges (the terms, taking a factor 4π
r
out, are bounded by ( rr )l ).
But since (5.8) holds, and these equations are for arbitrary ρ(~x′ ), it follows that,
∞ l
1 1 (r ′ )l X ⋆
(θ′ , ϕ′ )Ylm (θ, ϕ),
X
′ = 4π Ylm
l=0 2l + 1 (r)
| ~x − ~x | l+1
m=−l

whenever r ′ < r. But this expression clearly diverges for r ′ > r, so can not be true in this case
and our claimed identity fails, the reason being that in the previous calculations we commute
some integrals with series and in the present case this is not allowed. From the symmetry of
the expression on the left, it is clear that the same expression holds when we interchange ~x
and ~x′ , that is r and r ′ , and (θ, ϕ) and (θ′ , ϕ′ ). Since only matters the angle difference we
must have,
∞ l
1 1 (r < )l X ⋆
(θ′ , ϕ′)Ylm (θ, ϕ),
X
= 4π Ylm
| ~x − ~x′ | l=0 2l + 1 (r > )l+1
m=−l

where r < = min{r, r ′ }, r > = max{r, r ′ }. This expression being always convergent whenever
~x 6= ~x′ . This useful relation is known (or rather equivalent to) as the addition theorem for
spherical harmonics. A direct proof follows:

Direct proof: (complicated but illustrate some tricks)

Any smooth function, f (θ, ϕ), can be expanded in spherical harmonics:
∞ X
X l
f (θ, ϕ) = Alm Ylm (θ, ϕ),
l=0 m=l

with Z
⋆
Alm = f (θ, ϕ)Ylm (θ, ϕ)dΩ
S2

Note that at θ = 0, ( q
2l+1
m=0
Ylm (θ = 0, ϕ) = 4π
0 m 6= 0,
and so, s
∞
X 2l + 1
f (θ, ϕ) |θ=0 = Al0 ,
l=0 4π
with s
2l + 1
Z
Al0 = f (θ, ϕ)Pl (cos θ)dΩ.
4π S 2
We shall use this property in what follows. We now want to see that if we define the angle
γ by cos γ = n̂′ · n̂, then
l
4π X
Pl (cos γ) = Y ⋆ (θ′ , ϕ′ )Ylm (θ, ϕ),
2l + 1 m=−l lm
98CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

n′
~
θ′

~
n
γ
θ

φ′

Figure 5.12: The relation among angles

where (θ, ϕ) and (θ′ , ϕ′ ) are the angular coordinates defining n̂ and n̂′ respectively, see figure.
Since

n̂ = (sin θ cos ϕ, sin θ sin ϕ, cos θ)

n̂′ = (sin θ′ cos ϕ′ , sin θ′ sin ϕ′ , cos θ′ ),

we have,

cos γ = sin θ′ cos ϕ cos ϕ′ + sin θ sin θ′ sin ϕ sin ϕ′ + cos θ cos θ′
= sin θ sin θ′ [cos ϕ cos ϕ′ + sin ϕ sin ϕ′ ] + cos θ cos θ′
= sin θ sin θ′ [n̂xy · n̂′xy ] + cos θ cos θ′
= sin θ sin θ′ cos(ϕ − ϕ′ ) + cos θ cos θ′ ,

where n̂xy is the normalized projection of n̂ to the xy plane.

Note that in particular the above identity implies
l
4π X
1 = Pl (1) = Pl (cos 0) = | Ylm (θ, ϕ) |2
2l + 1 m=−l

To get the result we are seeking, we first write the general expression,
∞ X
l ′

Al′ m (θ′ , ϕ′ )Yl′ m (θ, ϕ),

X
Pl (cos γ) =
l′ =0 m=−l′
5.3. APPLICATION: MULTIPOLE MOMENTS OF A STATIC CONFIGURATION 99

where we think of Pl (cos γ) as a function of (θ, ϕ), where (θ′ , ϕ′ ) are some fixed parameters,
the right-hand side is just the (true) assumption that any smooth function an the unit sphere
can be expressed as a linear combination of spherical harmonics.
But Pl (cos γ), being just a rotated Legendre polynomial, is a solution to

∆S 2 U = −l(l + 1)U
It must be a linear combination of its eigenfunction, thus,
l
X
Pl (cos γ) = Am Ylm (θ, ϕ),
m=−l

we see that only the l′ = l terms can be different from zero on the above expression, that is,
Pl (cos γ) can be written by just a linear combination of spherical harmonics with the same l,
l
Alm (θ′ , ϕ′ )Ylm (θ, ϕ),
X
Pl (cos γ) =
m=−l

with Z
′ ′ ⋆
Alm (θ , ϕ ) = Pl (cos γ)Ylm (θ, ϕ) dΩ.
S2

We now, for fixed (θ′ , ϕ′ ), make a coordinate change in the integration variables and integrate
with respect to (γ, β) where β is the new angular variable around the axis defined by n̂′ .
⋆
Thus, we interpret Ylm (θ, ϕ) as a function of (γ, β), i.e. θ = θ(γ, β), ϕ = ϕ(γ, β).
It can be shown that this change of variables leaves the surface element unchanged.

Exercise: Give an argument for which this should be true.

Thus, Z
Alm (θ′ , φ′ ) = ⋆
Pl (cos γ)Ylm (θ(γ, β), ϕ(γ, β))dΩ(γ, β).
S2
⋆
But then we can think of this as the computation of the l coefficient of Ylm (θ(γ, β), ϕ(γ, β)),
for
s
2l + 1
Z
⋆
Al0 = Pl (cos γ)Ylm (θ(γ, β), ϕ(γ, β))dΩ(γ, β)
4π S2
s
2l + 1
= Alm
4π
⋆
Since there is only one l in the expansion of Ylm (θ(γ, β), ϕ(γ, β)) [for the same reason that
there was only one l for the expansion of Pl (cos γ).] We have
s s
∞
⋆
X 2l′ + 1 2l + 1 2l + 1
Ylm (θ(γ, β), ϕ(γ, β)) |γ=0 = Al′ 0 = Al0 = Alm
l′ =0 4π 4π 4π
100CHAPTER 5. STATIC PROBLEMS – CONSTRUCTING INTERESTING SOLUTIONS

But when γ = 0 θ = θ′ , ϕ = ϕ′ , so
4π
Alm = Y ⋆ (θ′ , ϕ′ ),
2l + 1 lm
and the result follows.
1
The above identity allows to expand the function ′ with respect to arbitrary axes:
|~
x−~x |
∞ ∞
1 1 (r < )l X ⋆
(θ′ , ϕ′ )Ylm (θ, ϕ),
X
′ = 4π Ylm
l=0 2l + 1 (r )
| ~x − ~x | > l+1
m=−l

where (θ, ϕ) and (θ′ , ϕ′ ) are respectively the angular coordinates of ~x and ~x′ with respect to
an arbitrary coordinate system.
To show this result, first choose axes so that ~x′ points in the ẑ direction. Thus, the above
expression satisfies Laplace’s equation away from ~x′ , and has azimuthal symmetry (think of
it as the potential of a point charge which is on the z axis (at ~x′ = (0, 0, r ′)). And so it can
be written as,
∞
1
Cl r −(l+1) Pl (cos(γ))
X
′ =
| ~x − ~x | l=0
where γ is the angle between both directions. We shall assume that r := |~x| > r ′ |. To find
the coefficients we take the case where ~x is also on the z axis, namely, ~x = (0, 0, r). In that
case we should have,
∞ ∞
1 −(l+1)
Cl r −(l+1)
X X
′
= Cl r Pl (1) =
| r − r | l=0 l=0

Comparison with the geometric series gives Cl = (r ′ )l and the result is established The
symmetry of the expression under interchange of ~x with ~x′ implies that if r ′ > r the result is
obtained by interchange of r with r ′ in the final expression, that is,
∞
1 X (r < )l
= Pl (cos γ),
| ~x − ~x′ | l=0 (r > )l+1
where r < = min{r, r ′ }, r > = max{r, r ′ } and γ was the angle between ~x and ~x′ .
Chapter 6

Eigenfunction Expansions

6.1 The General Picture

We have seen several instances of the following phenomena:
• In the task of solving particular problems we were led to solve an eigenvalue - eigenfunc-
tion problem:
Given a linear (differential) operator A, find numbers {λi }, called eigenvalues, and asso-
ciated functions {Ui }, called eigenfunctions, such that

AUi = λi Ui .
d 2 2
2 d d
In our examples A was dx 2 , (1 − x ) dx2 − 2x dx , or the Laplacian on the sphere.

• We found that solutions to the specific problems we had satisfied an orthogonality

relation:
< Ui , Uj >= 0 if λi 6= λj ,
where <, > indicates a certain scalar product which in general is some integral.
In our examples,
a 2πj
Z
< Ui , Uj > d2 = Ūi Uj dx, Uj = sin( )
dx2 0 a
Z 1
< Ui , Uj >(1−x2 ) d2 d
−2x dx
= Ūi Uj dx, Uj = Pl (x)
dx2
Z−1
< Ui , Uj >S 2 = Ūi Uj dΩ, Uj = Ylm (θ, φ)
S2

• We found a large enough number of solution such that we could expand any (smooth)
function in terms of them. In the first of our examples we found the trigonometric functions,
Un (x) = sin( πnx
a
), and the theory of Fourier Series granted us that using them we could
expand an arbitrary continuous function in the interval [0, a], provided it vanished on the
extremes. In the second example the fact that we got polynomials of arbitrary order and
Weierstrass approximation theorem granted us similar expansions properties.
We shall see now that these three phenomena are deeply related, and are consequences of
certain property of the operators A which appear in physics.

101
102 CHAPTER 6. EIGENFUNCTION EXPANSIONS

We shall treat in detail the finite dimensional case, that in when A is just a n × n matrix
and so Ui are n-vectors. Then we discuss the infinite dimensional case remarking the analogies
and differences.

6.2 The Finite Dimensional Case

We consider the following problem: find U, a n-vector and λ a number such that,

AU = λU,

or
Dλ U = 0,
with Dλ := A − λI where I is the identity matrix. Clearly eigenvectors are only determined
up to a factor, if U is one, so is cU, with the same eigenvalue.
Using the fundamental theorem of linear algebra we have:
The above equation has a solution if and only if detDλ = det(A − λI) = 0.
Now this determinant is a polynomial of degree n in λ, and so we know it has at least one
solution, λ0 . Thus we conclude:
There is at least one pair (λ0 , U0 ) solution to the above eigenvalue – eigenvector
problem.

Let us assume now we have two solutions, (λi , Ui ), (λj , Uj ). If λi 6= λj , then

(λi − λj ) < Ui , Uj >=< AUi , Uj > − < Ui , AUj >

and so for them to be orthogonal it is necessary that the right-hand side vanishes. This is
automatically satisfied if < AU, V >=< U, AV >, in that case we say that A is self-adjoint.
Notice that this property depends on what scalar product we are using, with the conventional
one the self-adjoint matrices are the symmetric ones. Most of the operators which appear in
physics are self-adjoint, so we restrict the discussion to them form now on.
What happens if there are several eigenvectors with identical eigenvalues? Let the set
{Ui } i = 1, . . . m, be eigenvectors with the same λ as eigenvalue. Then any linear combina-
tion of then is also an eigenvector with the same eigenvalue, indeed,
m m m m
ci Ui ) = ci AUi = ci λUi = λ ci Ui .
X X X X
A(
i=1 i=1 i=1 i=1

Thus, we can use the Gram - Schmidt procedure to get a orthonormal base for the subspace
expanded by the original ones. Thus, we conclude: The eigenvectors of an arbitrary self-
adjoint matrix A, can be chosen to be orthonormal.
We turn now to the question of completeness, namely whether the above set of eigenvectors
suffices to expand any vector in lRn . In finite dimensions this is simple, we only must check
that we have n linearly independent eigenvectors. Again this is true if A is self adjoint with
respect to some scalar product, <, >.
6.2. THE FINITE DIMENSIONAL CASE 103

To see this consider the function f : lRn → lR given by

< U, AU >
f (U) = .
< U, U >
Notice that, f (σU) = f (U), for any σǫlR, so we can think of this function as defined in
the unit sphere in lRn , S n . Since S n is compact, and since f is continuous there, we know it
must reach its minimum at a point of S n , say U0 . We call that minimum value λ0 , that is,
λ0 := f (U0 ).
Since f (U0 ) is a minimum and f is differentiable we must have that the derivative of f at
the point U0 , along any direction, must vanish. Thus, defining Uλ = U0 + λδU, with δU an
arbitrary vector we must have,
d < U0 , AδU > + < δU, AU0 >
0= f (Uλ ) |λ=0 =
dλ < U0 , U0 >
< U0 , AU0 > (< U0 , δU > + < δU, U0 >)
−
< U0 , U0 >2
2
= [< δU, AU0 > −λ0 < δU, U0 >]
< U0 , U0 >
2
= < δU, AU0 − λ0 U0 > .
< U0 , U0 >
Since δU is arbitrary we can take it to be δU = AU0 − λ0 U0 and then the positive definiteness
of a scalar product implies AU0 −λ0 U0 = 0. Thus, we have found an eigenvector corresponding
to the lower eigenvalue. But now we consider the subspace of lRn perpendicular (w.r.t. <, >)
to U0 , H1 = {U0 }⊥ = {U | < U, U0 >= 0 } and restrict f to that space. Again f takes value
in the unit sphere on H1 , S n−1, and reaches there its minimum value, λ1 , at some point U1 .
Repeating the calculation as before we see that this vector U1 is an eigenvector with eigenvalue
λ1 = f (U1 ). Repeating the above procedure with H2 = {U0 , U1 }⊥ , H3 = {U0 , U1 , U2 }⊥ and
so on we obtain a set of n orthonormal eigenvectors. Its linear independence implies that any
vector U can be written as linear combination of them,
n
ai Ui ,
X
U=
i=1
with
ai =< Ui , U > .

Exercise: Find a 2 × 2 matrix where this is not true, namely a matrix which has only a
single l.i. eigenvector.

Alternative to the proof given above we have the following, which admits an easy gener-
alization to infinite dimensional vector spaces:
Let (H, <, >) a vector space with scalar product, and A : H → H a linear map which
is self-adjoint, < V, AU >=< AV, U >. Then their eigenvectors form an orthonormal base
which expands the space.
104 CHAPTER 6. EIGENFUNCTION EXPANSIONS

We have already seen that there is at least one eigenvector,

AU0 = λ0 U0
Consider the space H0 = {V ∈ H, < V, U0 >= 0} that is the space of all vectors perpendicular
to U0 . This is a subspace of H and A : H0 → H0 , indeed, < AV, U0 >=< V, AU0 >= λ0 <
V, U0 >= 0 ∀V ∈ H0 . Thus, AV ∈ H0 . Therefore, by the same argument as above, there
exist an eigenvector-eigenvalue pair in this subspace,

AU1 = λ1 U1 .
Continuing that way we get a complete set of eigenvectors and eigenvalues pairs.

6.3 The Infinite Dimensional Case

We turn now to the infinite dimensional case.
Here we include the functions as a possible case of vectors in an infinite dimensional space,
for we can think of a function as an infinite list of components, f (~x), one for each point ~x
of lRn . Functions form vector spaces, for we can add them and multiply them for any real
number. In this case it is natural to take as a scalar product an integral, the most common
being the L2 scalar product, Z ∞
< f, g >L2 = f ⋆ g dx,
−∞

where we take the complex conjugate of the first member to allow for the case of complex
valued solutions, and we are considering functions defined over lR, the case for lRn or other
spaces follows by trivial generalizations.
The first difference with the finite dimensional case is that here there exist vectors whose
norm is infinite! So the first step is to exclude them from consideration: From now on
we consider only classes of functions whose norms are finite w.r.t. the scalar product under
consideration.
There are other very important differences with the finite dimensional case, but they are
subtle, and we shall not discuss them here. In the infinite dimensional case, linear differential
operators can be taken as our A operator in the eigenvalue - eigenfunction problem. A straight
forward generalization to self adjoint operators follows. Repeating the proof of orthonormality
we made for eigenvectors, we also obtain in this case the orthogonality of eigenfunctions as a
consequence of the selfadjointness of the operator under consideration.
We now turn to the problem of completeness of eigenfunctions. The first difference here
is that, since the expression
∞ N
al Ul (x) := lim al Ul (x),
X X
f (x) =
N →∞
l=0 l=0

involves a limiting procedure, it is not well-defined until we give the norm with respect to
which the limiting procedure holds.
There are at least two important norms which one should consider:
6.3. THE INFINITE DIMENSIONAL CASE 105

6.3.1 The L2 norm.

In this case the distance between two functions f (x) and g(x) is given by
sZ
q
k f (x) − g(x) kL2 = < f − g, f − g >L2 = | f − g |2 dx
lR
For this norm we say that the set {Ui } expands f if given ǫ > 0 there exists N, and N
coefficients {cN
i } i = 1...N such that
N
cN
X
kf− i Ui kL2 < ǫ
i=1

It is interesting to see what is the best choice of coefficients {cN

i }. Since the best choice
would minimize the error, if we define cN i (λ) = c N
i + λδ i and take the derivative of (for
simplicity) the square of that error, it should vanish.

N
d
cN 2
X
0 = kf− i (λ)Ui k |λ=0
dλ i=1
N N
d
cN cN
X X
= <f− i (λ)Ui , f − j (λ)Uj >|λ=0
dλ i=1 j=1
N N N N
δi⋆ < Ui , f − cN cN
X X X X
= − j Uj > − δj < f − i Ui , Uj >
i=1 j=1 j=1 i=1
N
δi⋆ (< Ui , f > − cN
X
= − i ) + complex conjugate.
i=1

since the δi ’s were arbitrary we conclude that cN

i = ci =< Ui , f > is the best choice.
Note that in this case the best choice of coefficient does not depend on N, thus, if we had
done the computation for a given N and then needed to do it for a larger integer N ′ we would
have needed just to compute the extra coefficients, cN +1 . . . cN ′ .
Note also that for this choice,
N N
2
k2L2 | ci | 2 ,
X X
kf− ci Ui k =k f −
i=1 i=1
and one can compute just the right-hand side to estimate the error made.
With this norm, and assuming that:
• A is self adjoint w.r.t. this norm.
• A is bounded by below (or above), that is, there exists a constant C > 0 such that
< U, AU >≥ −C < U, U >,
for all U smooth.
One can go along the steps of the finite dimensional proof and establish the completeness
of the eigenfunction of A in the norm L2 , that is in the sense defined above the eigenfunctions
of A expand any function whose L2 norm is finite.
106 CHAPTER 6. EIGENFUNCTION EXPANSIONS

6.3.2 Point-wise Norm

In this case the distance between two functions f (x) and g(x) is given by,

k f (x) − g(x) k0 := supxǫlR | f (x) − g(x) |,

which is the distance which one is more used to.

Here we shall say that f (x) can be approximated by the set of functions {Ui },if given any
ǫ > 0 there exists N > 0 and constants cNi such that

N
cN
X
kf− i Ui k0 < ǫ.
i=1

In this norm is not so clear what the best choice of coefficients is, but if we take the ones
of the case before, which computationally is very handy, in many cases of interest one can see
that continuous functions are well approximated by eigenfunctions of self adjoint operators.
Chapter 7

The Theory of Green functions

One of our first results in electrostatics was the general solution to Poisson’s equation for
isolated systems,

ρ(~x′ ) 3 ′
Z
φ(~x) = d ~x .
lR3 |~x − ~x′ |
We shall call ψ(~x, ~x′ ) = 1
′ the Green function of the problem and notice that,
|~
x−~x |
∆ψ(~x, ~x′ ) = −4πδ(~x − ~x′ ).
So, the function which one integrates to obtain de general solution is just the solution
corresponding to a point source at point ~x′ . Thus, if we know those solutions for all points
where the sources can be, we can integrate and obtain the generic solution. Can one find
general solutions for other boundaries conditions, for instance and to fix ideas, for arbitrary
distributions of charges in the presence of a conducting body? The answer is yes, and the
machinery to find them is based in the following identity, called the second Green Identity:
Z I
3
(ψ∆φ − φ∆ψ) d ~x = [ψn̂ · ∇φ − φn̂∇ψ] dS,
V ∂V

which follows easily by integration by parts and application of Gauss theorem.

Fixing ~x′ , and taking ψ(~x) = ψ(~x, ~x′ ) such that ∆~x ψ(~x, ~x′ ) = −4πδ(~x − ~x′ ) [which is
satisfied by any function of the form ψ(~x, ~x′ ) = 1 ′
x, ~x′ ) such that
′ + F (x, x ) with F (~
|~
x−~x |
∆~x F (~x, ~x′ ) = 0, and φ(~x) a solution to Poisson’s equation, ∆φ(~x) = −4πρ(~x), the second
Green identity gives,

[ψ(~x, ~x′ )(−4πρ(~x)) + 4πφ(~x)δ(~x − ~x′ )d3~x =

R
V Z
= ~ x) − φ(~x)n̂ · ∇ψ(~x, ~x′ )] dS.
[ψ(~x, ~x′ )n̂ · ∇φ(~
∂V

That is,

1
Z I
φ(~x′ ) = ψ(~x, ~x′ )ρ(~x) d3~x + ~ x) − φ(~x)n̂ · ∇ψ(~x, ~x′ )] dS.
[ψ(~x, ~x′ )n̂ · ∇φ(~
V 4π ∂V

107
108 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

Notice that this is not yet a formula for a solution, for it is still an equation, since φ(~x′ )
appears in both sides of the expression. But we can transform it into a formula for a solution
if we choose conveniently ψ(~x, ~x′ ), that is F (~x, ~x′ ) as to cancel, with the help of the boundary
conditions we want to impose on φ(~x), the boundary integrals.
For instance, if we would want to give as a boundary condition the value of φ(~x) at
∂V, φ(~x)|∂V = f (~x), we would choose F (~x, ~x′ ) such that ψ(~x, ~x′ )|∂V = 0 and so we would then
have,
1
Z I
φ(~x′ ) = ψ(~x, ~x′ )ρ(~x) d3~x − ~ x, ~x′ ) dS.
f (~x)n̂ · ∇ψ(~
V 4π ∂V
We call this function, ψ(~x, ~x′ ) the Green function of the Dirichlet problem for V , and no-
tice that it depends on V and its boundary.
Notice that this function satisfies:

∆~x ψ(~x, ~x′ ) = −4πδ(~x − ~x′ )

ψ(~x, ~x′ ) |~xǫ∂V = 0,

that is satisfies the Dirichlet problem for a point-like source of strength one at ~x′ with homo-
geneous boundary conditions.

Exercise: Show that ψ(~x, ~x′ ) is unique.

Similarly, we can find other Green functions adapted to other boundary problems.

7.1 The Dirichlet Green function when ∂V is the union

of two concentric spheres.
We have already found two Green functions, namely the one corresponding to a unit charge
in an otherwise empty space, ψ(~x, ~x′ ) = 1 ′ , and the one corresponding to a unit charge in
|~
x−~x |
the presence of a conducting sphere of radius a at zero potential, namely equation (4.9),
1 a
ψ(~x, ~x′ ) = ′ − .
|~x − ~x | |~x||~x′ − a2 ~
x|
|~
x|2
We shall deduce again this expression using the machinery we have learned in the previous
chapter.
We want to find the Green Function when ∂V is the union of two concentric spheres. That
is a solution to:

∆~x ψ(~x, ~x′ ) = −4πδ(~x − ~x′ )

ψ(~x, ~x′ ) ||~x|=a = 0
ψ(~x, ~x′ ) ||~x|=b = 0
7.1. THE DIRICHLET GREEN FUNCTION WHEN ∂V IS THE UNION OF TWO CONCENTRIC SP

~
x
a

x′
~
q

Figure 7.1: Two concentric spheres and a point charge in between

First we notice a general property of Dirichlet Green’s functions:

ψ(~x, ~x′ ) = ψ(~x′ , ~x). (7.1)

Indeed,

−4π[ψ(~x′ , ~x) − ψ(~x, ~x′ )]

Z
= −4π[ψ(~y , ~x)δ(~y − ~x′ ) − ψ(~y , ~x′ )δ(~y − ~x)]d3 ~y
ZV
= [ψ(~y , ~x)∆~y ψ(~y , ~x′ ) − ψ(~y , ~x′ )∆~y ψ(~y , ~x)]d3~y
ZV
= ~ ψ(~y , ~x′ ) − ψ(~y , ~x′ )n̂ · ∇
[ψ(~y , ~x)n̂ · ∇ ~ ψ(~y , ~x)]d2 S
∂V
~y ~y

For the Dirichlet Problem ψ(~y , ~x) |~yǫ∂V = 0, therefore the surface term vanishes and so the
result follows. It is clear that this is also true for the homogeneous Neumann problem. Had we
had considered complex green functions, then the result would have been, ψ(~x′ , ~x) = ψ̄(~x, ~x′ ).
We now return to the problem at hand. It is clear that it is convenient to write all
expressions in spherical coordinates. We already know the expression for the right-hand side,

1
δ(~x − ~x′ ) = δ(r − r ′ ) δ(ϕ − ϕ′ )δ(cos θ − cos θ′ )
r2
∞ X l
1 X
= δ(r − r ′ ) 2 ⋆
Ylm (θ, ϕ)Ylm (θ′ , ϕ′ )
r l=0 m=−l
:= δ(r − r ′ )σ(r, θ, θ′ , ϕ, ϕ′ ), (7.2)
110 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

where we can think of σ(r, θ, θ′ , ϕ, ϕ′) as a surface distribution. We also know the general
expression for a solution to Laplace’s equation, which will be valid at all points where ~x 6= ~x′ ,

∞ X
l
ψ(~x, ~x′ ) = [Alm (r ′ , θ′ , ϕ′ )r l + Blm (r ′ , θ′ ϕ′ )r −(l+1) ]Ylm (θ, ϕ).
X

l=0 m=−l

It will be discontinuous at ~x = ~x′ , so we expect the solution to have a given form for r < r ′
and a different one for r > r ′ Since ψ(~x, ~x′ )|r=a = ψ(~x, ~x′ )|r=b = 0 these expressions must be,

∞ X
l
a2l+1
ψ(~x, ~x′ ) = [A< ′ ′ ′ l
r < r′
X
lm (r , θ , ϕ )(r − )]Ylm (θ, ϕ)
l=0 m=−l r l+1
∞ X
l
b2l+1
ψ(~x, ~x′ ) = [A> ′ ′ ′ l
r > r′
X
lm (r , θ , ϕ )(r − )]Ylm (θ, ϕ)
l=0 m=−l r l+1

Continuity of the potential at r = r ′ (at points other than where the point charge is)
implies,

∞ X
l ∞ X l
a2l+1 b2l+1
[A< ′ ′ l > ′ ′ l
X X
lm (r, θ , ϕ )(r − )]Y lm (θ, ϕ) = [Alm (r, θ , ϕ )(r − )]Ylm (θ, ϕ)
l=0 m=−l
r l+1 l=0 m=−l
r l+1

Using the orthogonality of the spherical harmonics we see that for each pair l, m we must
have,

a2l+1 b2l+1
A< ′ ′ l
lm (r, θ , ϕ )(r − ) = A>
lm (r, θ ′
, ϕ ′
)(r l
− ),
r l+1 r l+1
and so,

l b2l+1
A< ′ ′ ′ ′ ′ ′ ′
lm (r , θ , ϕ ) = Elm (r , θ , ϕ )(r − )
r ′ (l+1)
′l a2l+1
A>
lm (r ′ ′
, θ , ϕ ′
) = Elm (r ′ ′
, θ , ϕ ′
)(r − )
r ′ (l+1)
For some functions Elm (r ′ , θ′ , ϕ′ ). To find them we look now at the matching conditions at
r = r ′ , we can consider the delta function there as a surface distribution (which in itself, at
(θ, ϕ) = (θ′ , ϕ′ ), is a distribution). So we must have,

∂r ψ(r, r ′ , θ, θ′ , ϕ, ϕ′ )− − ∂r ψ(r, r ′ , θ, θ′ , ϕ, ϕ′)+ = 4πσ(r, θ, θ′ , ϕ, ϕ′ )

∞ X l
4π X ⋆
= Ylm (θ, ϕ)Ylm (θ′ , ϕ′ )
r 2 l=0 m=−l
7.1. THE DIRICHLET GREEN FUNCTION WHEN ∂V IS THE UNION OF TWO CONCENTRIC SP

where ∂r ψ(r, r ′, θ, θ′ , ϕ, ϕ′ )−(+) = limrրr′ ∂r ψ(r, r ′ , θ, θ′ , ϕ, ϕ′)(limrցr′ ∂r ψ(r, r ′ , θ, θ′ , ϕ, ϕ′ )). Af-

ter canceling all the (θ, ϕ) angular part –using orthogonality of the spherical harmonics– we
get,

b2l+1 a2l+1 b2l+1 ′l a2l+1

limr′ →r Elm (r ′ , θ′ , ϕ′ )[(r ′l − )(lr l−1
+ (l + 1) ) − (lr l−1
+ (l + 1) )(r − ′l+1 )]
r ′l+1 r l+2 r l+2 r
2l+1 2l+1 2l+1 2l+1
b a b a
= Elm (r, θ′ , ϕ′ )[(r l − l+1 )(lr l−1 + (l + 1) l+2 ) − (lr l−1 + (l + 1) l+2 )(r l − l+1 )]
r r r r
2l + 1
= Elm (r, θ′ , ϕ′ ) 2 (a2l+1 − b2l+1 )
r
4π ⋆ ′ ′
= 2 Ylm (θ , ϕ )
r
and so,
4π 1
Elm (r, θ′ , ϕ′ ) = Y ⋆ (θ′ , ϕ′ )
2l + 1 a2l+1 − b2l+1 lm
Thus,
∞ X
l
1 b2l+1 l a2l+1
ψ(~x, ~x′ ) = −4π l ⋆ ′ ′
X
(r> − l+1 )(r < − l+1 )Ylm (θ, ϕ)Ylm (θ , ϕ ).
l=0 m=−l (2l + 1)(b2l+1 − a2l+1 ) r> r<

Exercise: Use the symmetry under interchange of variables of the Green function to get the
solution up to constants depending only on l and m.

There are several limiting cases of interest:

Case 1: a → 0, b → ∞
∞ X
l l
r< 1
ψ(~x, ~x′ ) = 4π ⋆
(θ′ , ϕ′ )
X
l+1 Ylm (θ, ϕ)Ylm
l=0 m=−l >r 2l + 1
1
= ,
|~x − ~x′ |
and we recuperate the Green function for isolated systems.
Case 2: a → 0, b fixed, or interior problem.

∞ X
l
′
X 1 l b2l+1 l ⋆ ′ ′
ψ(~x, ~x ) = −4π (r > − l+1 )r< Ylm (θ, ϕ)Ylm (θ , ϕ )
l=0 m=−l
(2l + 1)b2l+1
r>
∞ X
l
1 (rr ′ )l r<l
⋆
(θ′ , ϕ′ )
X
= −4π (b 2(l+1) − l+1 )Ylm (θ, ϕ)Ylm
l=0 m=−l 2l + 1 b r >
−b 1
= + ′ ,
b2 ~
x
|~x||~x − 2 | |~x − ~x |
′
|~
x|
112 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

which was the result already obtained by the method of images.

1
Exercise: Use the expansion of ′ and the substitution r, r ′ → r, b2 → r ′ and r < r ′ to
|~
x−~x |
check the last line in the above deduction.

Case 3: a fixed, b → ∞, or exterior problem

∞ X
l
1 a2l+1 1
ψ(~x, ~x′ ) = 4π l ⋆
(θ′ , ϕ′ )
X
(r< − l+1 ) l+1 Ylm (θ, ϕ)Ylm
l=0 m=l 2l + 1 r< r>
∞ X
l l
1 r< a a2 l ⋆
(θ′ , ϕ′ )
X
= 4π ( l+1 − ′ ( ′ ) )Ylm (θ, ϕ)Ylm
l=0 m=−l 2l + 1 r> rr rr
1 a
= ′ − .
|~x − ~x | |~x||~x − a2~x |
′
|~
x|2

It is interesting to recall what we have done in solving the equation for ψ(~x, ~x′ ). That
equation is second order so in general one should specify two boundary conditions, for in
general there would be two linearly independent solutions. If we depart from r = b inward
and towards r = r ′ , we must spend one boundary condition in giving ψ(~x, ~x′ )|r=b = 0. If we
depart from r = a outwards and towards r = r ′ we also waste one in setting ψ(~x, ~x′ )|r=a = 0.
For fix r ′ we must use another to set equal both solutions at r = r ′ . It remains one boundary
condition to be settled. One could imagine using it to set both first derivatives equal at r = r ′,
but that can not be done, unless all boundary conditions are taken to be zero, for the only
solution to the homogeneous equation with zero boundary conditions at r = a and r = b is
the zero solution. The last parameter has to be chosen then as to regulate the strength of the
source of the equation, it sets the jump on first derivatives. See figure.

7.2 Some examples using the spherical Green functions

7.2.1 Empty sphere at potential V (θ, ϕ)
Recall that if we have Green’s function for Dirichlet boundary conditions, then

1
Z I
φ(~x) = ′ ′
ρ(~x ) ψ(~x, ~x ) d ~x − 3 ′ ~ ′ ψ(x, ~x′ ) d2 S,
V (~x′ )n̂ · ∇
V 4π ∂V ~x
where is the volume inside the sphere of radius b, satisfies

∆φ(~x) = −4πρ(~x)
φ(~x)|∂V = V (~x).

~ ′ ψ(~x, ~x′ ).
Thus, we must compute n̂ · ∇~x
7.2. SOME EXAMPLES USING THE SPHERICAL GREEN FUNCTIONS 113

~
x

q
x′
~

r′

Figure 7.2: The interface problem

For the Green function corresponding to Case 2, which is well suited to study interior (to
the sphere) problems we have,

~ ′ (~x, ~x′ ) = ∂ψ (~x, ~x′ )|r′ =b

n̂ · ∇~x ∂r ′
∞ X l
∂ X 1 1 rr ′ l r<l
⋆ ′ ′
= [−4π ( ( ) − l+1 )Ylm (θ, ϕ)Ylm (θ , ϕ )]|r =b
′
∂r ′ l=0 m=−l 2l + 1 b b2
r >

Since in this case r ′ = b, and therefore r < r ′ , it suffices to compute

d 1 r′r l rl l r′r l l + 1 rl 2l + 1 l
( ( ) − )| r ′ =b = ( ( ) + )| r ′ =b = r,
dr ′ b b2 (r ′ )(l+1) br ′ b2 r ′ (r ′ )(l+1) bl+2
so
∞ X l
1 ~ x, ~x′ )| ′ = rl ⋆
(θ′ , ϕ′ )
X
− n̂ · ∇ψ(~ |~
x |=b ( )Ylm (θ, ϕ)Ylm
4π l=0 m=−l bl+2

Thus, for instance, the solution to Laplace’s equation (ρ ≡ 0) inside a sphere at potential
V (θ, ϕ) is
∞ X
l
r
Z
( )l Ylm (θ, ϕ)[ V (θ′ , ϕ′ )Ylm
⋆
(θ′ , ϕ′ )dΩ′ ]
X
φ(r, θ, ϕ) =
l=0 m=−l b S 2 (r=b)

7.2.2 Homogeneously charged ring inside a sphere at zero potential

In this case the charge density is given by
Q
ρ(~x′ ) = δ(r ′ − a)δ(cos θ′ ),
2πa2
114 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

Figure 7.3: Homogeneously charged ring inside a sphere at zero potential

where a is the ring radius and b the one of the sphere. See figure.
∞ l
Q 1 ra r<
Z
ρ(~x′ )ψ(~x, ~x′ )d3~x′ = (−4π) ( ( 2 )l − l+1
X
φ(~x) = )Pl (cos θ)Pl (0),
V 2π l=0 b b r>
with r> = max{a, r}, r< = min{a, r}.
But 
 0 l odd
Pl (0) = (−1)l/2 (l−1)!!

2l/2 (l/2)!
l even
so,
∞
X (−1)l (2l − 1)!! 1 ra 2l 2l
r<
φ(r, θ, ϕ) = −2Q ( ( ) − 2l+1 )P2l (cos θ).
l=0 2l l! b b2 r>

7.2.3 Uniformly charge rod inside a grounded sphere

Q
a

Figure 7.4: Uniformly charge rod inside a grounded sphere

7.3. CONSTRUCTION OF GREEN FUNCTIONS USING EIGENFUNCTIONS. 115

In this case,
Q 1
ρ(~x′ ) = [δ(cos θ′ − 1) + δ(cos θ′ + 1)]
2b 2πr ′ 2
Using the Green function of the problem we obtain,
∞
−Q X Z b
1 rr ′ l
r<
φ(~x) = [Pl (1) + Pl (−1)]Pl (cos θ) [ ( 2 )l − l+1 ]dr ′ .
b l=0 0 b b r>

Let
b 1 rr ′ l
r< b 1 rr ′ r (r ′ )l b rl
Z Z Z Z
Il = [ ( 2 )l − l+1 ] dr ′ = [ ( 2 )l dr ′ − ] dr ′ − ] dr ′ ,
0 b b r> 0 b b 0 r l+1 r (r ′ )l+1
then for l 6= 0,

rl bl+1 1 r l+1 1 r l b
Il = − + ′l |r
b2l+1 l + 1 l + 1 r l+1 lr
r l 1 1 r l1 1 r 2l + 1
= ( ) − + ( ) − = [( )l − 1] ,
b l+1 l+1 b l l b l(l + 1)

while,
b r
I0 = − ln = ln
r b
Since Pl (1) = 1 and Pl (−1) = (−1)l we have,
∞
Q r r 4l + 1
φ(~x) = [− ln + 2 [1 − ( )2l ]
X
P2l (cos θ)]
b b l=1 b 2l(2l + 1)

Exercise: Compute the surface charge density induced on the sphere.

7.3 Construction of Green Functions Using Eigenfunc-

tions.
Let us assume we have a linear differential operator, Ax , the Laplacian, say. we are interested
in finding its Green Function, for certain fixed boundary condition.
That is, we want to find a function ψ(~x, ~x′ ) such that:

Ax (ψ(~x, ~x′ )) = −4πδ(~x − ~x′ ) in V

Bx ψ(~x, ~x′ ) = 0, in ∂V
116 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

where Bx is some linear, possible differential operator in ∂V , which gives the boundary con-
ditions.
For instance Bx = 1 gives Dirichlet’s boundary condition, while Bx = n̂ · ∇ ˆ x gives Neu-
mann’s.
To find ψ(~x, ~x′ ) one can look at the associated problem:

Ax Ui (~x) = −λi Ui (~x) in V

Bx Ui (~x) = 0 in ∂V.

That is, the eigenvalue - eigenfunction problem for Ax with the same boundary conditions
as we want for the Green function.
If Ax is self-adjoint for these boundary conditions, then the set of eigenfunctions, {Ui }
i = 0, . . . form a complete set and therefore the Green function, as a function of ~x, can be
expanded as:
∞
′
ai (~x′ )Ui (~x),
X
ψ(~x, ~x ) =
i=0

for some coefficients ai (~x), but then

∞
Ax ψ(~x, ~x′ ) = − λi ai (~x′ )Ui (~x) = −4πδ(~x − ~x′ ).
X

i=0

~ j (~x), integrating over V and using the orthonormality relations for the Uj′ s
Multiplying by U
we obtain:
aj (~x′ )λj = −4πUj∗ (~x′ )
and so ∞
4π ∗ ′
ψ(~x, ~x′ ) =
X
(Uj (~x )Uj (~x)).
i=0 λj

Remark: For this formula to be valid the operator Ax does not have to have an eigenfunction
with zero eigenvalue for the required boundary conditions. If there is an eigenfunction U0 with
zero eigenvalue, then we can consider

′ Ui∗ (~x′ )Ui (~x)

∞
X
ψ(~x, ~x ) = 4π λi 6= 0
i=1 λi

but now only be able to find solutions for sources which do not have components along U0 ,
that is, sources such that: Z
U0 (~x)ρ(~x) d3~x = 0
V
If this condition, called an obstruction, is not satisfied, then there is no solution to the
problem. Note also that solutions are now not unique, if φ is a solution to our problem, then
φ + U0 is also a solution.
Remark: If the operator Ax is self-adjoint for the boundary conditions Bx , then if Ui is an
eigenfunction, so is Ūi , and both for the same real eigenvalue. This implies that ψ(~x, ~x′ ) is
real and so that ψ(~x, ~x′ ) = ψ(~x′ , ~x).
7.3. CONSTRUCTION OF GREEN FUNCTIONS USING EIGENFUNCTIONS. 117

Remark: For most self-adjoint operators in compact sets the set of eigenvalues do not
accumulate around any finite value and so the number of obstructions are finite.

~ =g
Exercise: Analyze the obstruction for the Neumann problem ∆φ = −4πρ in V , n̂ · ∇φ
in ∂V for finite V . Give a physical interpretation to it.

Answer: according to Gauss theorem,

Z
~ d2 S = −4πQ
n̂ · ∇φ
∂V

where Q is the total charge,

Z
Q := ρ(~x) d3~x.
V

Thus, there can only be a solution when,

Z
g d2 S = −4πQ. (7.3)
∂V

How do we see this form our construction? The problem,

∆Ui = λi Ui ∈ V, ~ i |∂V = 0,
n̂ · ∇U
q
has a eigenfunction with zero eigenvalue, U0 = 1/ vol(V ). This is the unique one, indeed
assume U is an eigenfunction with zero eigenvector, then ∆U = 0, so, integrating over the
~ =0
volume UδU, integrating by parts, and using the boundary condition we find that ∇U
from which we conclude that U is constant.
Thus, the obstruction is that

Q
Z
ρ U0 = q =0
V vol(V )

for the homogeneous boundary condition.

We shall see now how to proceed when the boundary condition is not homogeneous. For
that we first define δφ(~x) = φ − φ̃, where φ̃ is any smooth function defined in V such that
n̂ · φ̃|∂V = g. Thus we have now the homogeneous problem,

∆δφ = −4πρ − ∆φ̃ := −4πδ ρ̃

~
n̂ · ∇δφ| ∂V = 0

So we can apply our Green’s function construction. Indeed, if (7.3) holds, then
Z
ρ̃ d3~x = 0.
V

For those pairs (ρ, g) the solution will be given by,

118 CHAPTER 7. THE THEORY OF GREEN FUNCTIONS

Z
δφ(~x) = ψ(~x, ~x′ )ρ̃(~x′ ) d3~x′
V
1
Z
= ψ(~x, ~x′ )(ρ(~x′ ) +
∆φ̃) d3~x′
V 4π
1 1
Z I
= ′ ′
ψ(~x, ~x )ρ(~x ) + ′ 3 ′
∆ψ(~x, ~x )φ̃ d ~x + ~ φ̃ − φ̃ n̂ · ∇ψ]
[ψ n̂ · ∇ ~ d2 S
V 4π 4π ∂V
Z
′ ′ 1 I ~ φ̃ d2 S
= ψ(~x, ~x )ρ(~x ) − φ̃(~x) + ψn̂ · ∇ (7.4)
V 4π ∂V
where in the second and the last line we have used the properties of Green’s functions, and on
the third Green’s identity. Thus, we reach to the known expression for this class of problems,
but knowing that it is only valid when the integrability condition (7.3) is satisfied.
If the condition is not satisfied, the above expression will give some function, but it will
not satisfy the equations for the ρ and g given, but rather for a pair projected into a subspace
where the condition holds.
Chapter 8

Dielectrics

8.1 The Nature of the Problem

So far we have solved two types of electrostatic problems (and their linear combinations):
1.) Fixed charges problems.
2.) Completely free charges inside conductors.
For the first case, we give some charge distribution ρ(~x) and look for solutions to Poisson’s
equation for that distribution, which can be done, for instance using the appropriate Green
function. For the second, the property of charges to move freely translate into conditions
the potential must satisfy at the boundaries of the conductors. In some simple cases the
Green functions can be computed and then used to solve the problems, or one can use series
expansions in simple solutions to construct the complete solution.
We want to discuss now a case in between these two, which shall allows to treat many
situations of practical interest. It is the case where the sources are almost fixed and only
react weakly to the presence of an external field. We want to find the extra electric field that
this small reaction produces.
To fix ideas we consider the following simple case:
Example: Uniformly charged sphere hanging from a spring in the presence of an external
electric field.
We consider a uniformly charged sphere of radius a hanging from a spring in an external
field which is also directed along the vertical direction.
Before applying the external field, and choosing the equilibrium position as the coordinate
origin the initial electric field is:
Q
(
~i = r2
n̂, r>a
E Qr
a3
n̂, r<a
where Q is the total charge of the sphere.
When we apply an external field E ~ 0 we exert a net force on the sphere of strength F~ 1 =
QE~ 0 , and correspondingly there will be a displacement of the sphere a distance d = QE0 ,
k
where k is the spring constant. That displacement means that if the initial charge density
~ We want to know the field configuration after this
was ρ0 (~x) it is now ρ(~x) = ρ0 (~x − d).
displacement. To avoid having to carry along the fields of the original charge configuration

119
120 CHAPTER 8. DIELECTRICS

a ~
E

Figure 8.1: Charged sphere hanging form a spring

we shall modify the problem and assume the we have now two spheres superimposed with
charges Q and −Q respectively, so that initially the field they produce cancel each other and
so the total field due to the charges vanishes.
Of course the answer is simple, the new configuration is the sum of the external field
and of the fields corresponding to the two sphere of uniform charge in their new positions,
~
d~ = ± QE 0 . To first order the new field satisfies:
k

~ · δ E(~
∇ ~ − ρ0 (~x + d)]
~ x) = 4π[ρ0 (~x − d) ~ (8.1)
≃ −8π d~ · ∇ρ
~ 0 (~x) (8.2)
≃ −8π ∇ ~ 0 (~x)),
~ · (dρ (8.3)

or

∆δφ(~x) = 8π d~ · ∇ρ
~ 0 (~x)
−3Q
= δ(r − a)d~ · n̂
a3
= −4πσ(~x)δ(r − a),
+3Q ~
with σ(~x) := d · n̂.
4πa3
Thus, our problem is to solve ∆δφ = 0 inside and outside the sphere, with the boundary
conditions:
~
n̂ · ∇(δφ out − δφin )|r=a = −4πσ
(δφout − δφin )|r=a = 0.
8.1. THE NATURE OF THE PROBLEM 121

a ~
E

Figure 8.2: Charged sphere hanging form a spring

~ the only possible combinations of solutions to

Since the solution must depend linearly on d,
the Laplace equation which are allowed are:

d~ · ~x
δφout (~x) = B ,
r3
δφin (~x) = C d~ · ~x.

The matching conditions then imply:

∂ B~
~
n̂ · ∇(δφ out − δφin )|r=a = ( d · n̂ − Cr d~ · n̂)|r=a
∂r r 2
2B 3Q
= d~ · n̂(C + 3 ) = 3 d~ · n̂
a a
B ~
and (δφout − δφin )|r=a = a(−C + 3 )d · n̂ + 0
a

Thus, B = Q, C= Q
, ~ 0,
or in terms of E
a3

~ 0 ·~x

Q2 E
r>a

δφ = k
Q2 ~
r3

E
ka3 0
· ~x r<a

or, adding to it the external field contribution:

~ 0 ·~x


−E~ 0 · ~x + Q2 E r>a
φ= k r3
Q2 ~

(−1 + ka3 )E 0 · ~x r < a
122 CHAPTER 8. DIELECTRICS

We see that outside the sphere the electric field difference, besides the constant field, is of
2~
a dipolar field configuration corresponding to a dipole ~p = −QkE 0 , while inside the constant
2
field has diminished its strength due to the presence of an extra constant field −Q ~ 0.
E
ka3
The total field is (again dismissing the field of the initial charge configuration):

~ 0 ·~n ~n−E
~0

~ =
 ~ 0 + 3 Q2 E
E r>a
E k r3
Q2 ~

(1 − ka3 )E 0 r<a
Remark:
1.) If we define

~ 0 (~x)
P~ (~x) = 2dρ
Q ~ 0,
= ρ0 (~x)E
k
3Q2 ~
= E 0 θ(a − r)
4πka3
~ = E(~
then it follows from 8.1 that, D ~ x) + 4π P
~ (~x) satisfies,

~ ·D
∇ ~ = 0.
This field, called the electric displacement vector has a continuous radial component, but
the angular components have jumps across r = a. To see this, contract the electric field with
the normal to the sphere (r = a) both inside and outside it and compare.
2.) Note that E ~ and P~ are linear in E
~ 0 , to the approximation we are working with, and
~ ~
so P is linear in E. We define the electric susceptibility χe by the relation:

P~ = χe E
~
3Q2
For this case: χe = 4π(ka3 −Q2 )
.
~ = εE,
Defining, ε, the electric permittivity by the relation, D ~ in this case we have,

ka3 + 2Q2
ε = 1 + 4πχe =
ka3 − Q2

8.2 A Microscopic Model

With the above example in mind we want to discuss now how real materials behave when
electric fields act upon them. Matter is made out of groups of molecules, each one of these
molecules can be modeled as a set of spheres hold in some equilibrium positions due to springs
connecting them. Of course for this case we can not compute all displacements due to some
external field and compute the resulting fields. Instead, we content ourselves by looking to
8.2. A MICROSCOPIC MODEL 123

some averaged fields and, after finding equations for them, impose certain phenomenological
relations between them.
To fix ideas we consider a solid with a charge density given by

XX
ρ(~x) = ρf ree (~x) + qj(m) δ(~x − ~xm − ~xj(m) ),
m j(m)

where, ρf ree (~x), is the charge density of freely moving charges, ~xm is the center of mass
coordinate of the present, fixed molecules, and ~xj(m) is the, relative to the center of mass,
displacement of the “j - sphere" in the molecule m. These last displacements are supposed to
change when external fields are applied.

~
xm,1 ~
xm,3

~
xm,2

~
xm,4

~
xm

Figure 8.3: Individual molecule

The free charge contribution to the solution is dealt with as in conductors and so will not
be discussed further. Notice that if that free charge is present in any reasonable amount, then
the material would behave as a conductor and so no polarization effect would be present in
the bulk of the material. It basically amounts to have an infinite value for ε.
We shall consider now space averages over distances which are big compared with the size
of individual molecules, but still very small in comparison with the smallest regions on which
we measure macroscopic electric fields. A typical distance in solids is L = 10−6 cm, in the
corresponding volume are housed typically about 106 molecules.
Given smooth, compactly supported function in lR+ , ϕ(s), such that ϕ(0) = 1,
R∞
4π 0 ϕ(s)s2 ds = 1, we define for each function f : lR3 → lR
Z
fϕ (~x) = 3
ϕ(|~x − ~y |)f (~y)d3~y
lR

The most important property of this averages is:

124 CHAPTER 8. DIELECTRICS

Averaging region

Figure 8.4: Molecules in a solid

Z
~ ϕ (~x) =
∇f ~ x ϕ(|~x − ~y |)f (~y )d3~y
∇
lRZ3
= − ∇~ y ϕ(|~x − ~y |)f (~y)d3~y
3
Z lR
= ϕ(|~x − ~y |)∇ ~ y f (~y )d3~y
3
lR
= (∇~ x f (~x))ϕ .

Since the surface integral at infinity that results from the application of Gauss theorem
in the third step vanishes, for ϕ is compactly supported. If we apply these averages to the
static Maxwell’s equations we obtain,

~ ∧E
∇ ~ϕ = 0
~ ·E
∇ ~ ϕ = 4πρϕ .

These are now equations for averaged quantities over much larger scales than those charac-
teristics of the matter scales. 1 This field is not the field which is felt by individual molecules,
but rather the averaged field produced by other molecules and external fields, the own molec-
ular field can be much bigger, but mostly participate in creating the initial molecular equi-
librium configuration.
The task now is to accurately enough describe ρϕ in terms of known quantities. Taking
the molecular charge distribution we had defined above we find:

Z
(ρm )ϕ (~x) = ϕ(|~x − ~y |)ρm (~y )d3~y
lR3
X
= qj(m) ϕ(|~x − ~xm − ~xj(m) |).
j(m)

1
The fact that we get the same averaged equations for any average is due to the linear character of the
equations, which we shall also assume in the way the electrostatic field affects the matter.
8.2. A MICROSCOPIC MODEL 125

Expanding ϕ(|~x − ~xm − ~xj(m) |) in Taylor series around ~x − ~xm we obtain

~
X
(ρm )ϕ (~x) = qj(m) [ϕ(|~x − ~xm |) − ~xj(m) · ∇ϕ(|~
x − ~xm |) +
j(m)

1 i ∂2
xj(m) xkj(m) i k ϕ(|~x − ~xm |) + . . .
2 ∂x ∂x
~
= qm ϕ(|~x − ~xm |) − p~m · ∇ϕ(|~ x − ~xm |)
2
1 ik ∂
+ Sm ϕ(|~x − ~xm |) + · · · ,
2 ∂xi ∂k k
with
X
qm := qj(m)
j(m)
X
p~m := qj(m)~xj(m)
j(m)
ik
qj(m) xij(m) xkj(m) .
X
Sm :=
j(m)

This in turn can be written as,

~ · p~mϕ (~x) + 1 ∂ 2 Smϕ

ik
(ρm )ϕ (~x) = ρmϕ − ∇ (~x) + · · ·
2 ∂xi ∂xk
where,
Z
ρmϕ (~x) = qm ϕ(|~x − ~xm |) = 3
ϕ(|~x − ~y |)qm δ(~y − ~xm )d3~y ,
ZlR
~pmϕ (~x) = p~m ϕ(|~x − ~xm |) = 3
ϕ(|~x − ~y |)~pm δ(~y − ~xm )d3~y ,
lR
etc. are smooth out (with ϕ) molecular averages, and where we have used,

Z
~ x − ~xm |) =
−~pm · ∇ϕ(|~ ~
−~p · ∇ϕ(|~ x − ~y |)δ(~xm − ~y ) d3~y
lR3 Z m
= −∇~ · p~ ϕ(|~x − ~y |)δ(~xm − ~y ) d3~y
3 m
lR
~
= −∇ · ~pmϕ .

Summing now over all molecules we get,

~ · P~ ϕ + 1 ∇
ρϕ = ρ̄ϕ − ∇ ~ · (∇
~ ·S
~ ϕ) + · · · ,
2

where, P~ ϕ := m ~pmφ , etc. Since all except the first term are divergences we could write:
P

~ ·D
∇ ~ = 4π(ρ̄ + ρf ree )
126 CHAPTER 8. DIELECTRICS

with D~ϕ = E~ ϕ + 4π P~ ϕ − 2π ∇
~ ·S
~ϕ + · · ·
The above, plus the equation ∇ ~ ∧E~ ϕ = 0 are the two equations we shall now be interested
in. But they can not be solved until we relate D ~ ϕ and E~ ϕ.
The fundamental assumption we shall be doing is that P~ , S,
~ and all other similar averaged
quantities are caused by the applied external electric field, and that these relations are all
local. That is, we shall assume there exists a smooth relation fo the form:

~ x) = D(
D(~ ~ E(~
~ x))
Expanding this relation in Taylor series,
~ x)]i = [D
[D(~ ~ 0 (~x)]i + εi (~x)[E(~
~ x)]j + εi (~x)[E(~
~ x)]j [E(~
~ x)]k + · · · ,
1j 2jk

we note that usually the averaged interior fields vanish, and so D ~ 0 (~x) = 0. This is for sure
the case for isotropic media, for there is no privileged directions and so necessarily D~ 0 (~x) = 0.
In general the first non-trivial term in the series suffices to treat most materials with weak
external fields, and so
~ x)]i = εi (~x)[E(~
[D(~ ~ x)]j .
1j

On isotropic media εi1j , as a linear map, can not have any preferred subspaces and so
εi1j = εδ i j, thus,
~ x) = ε(~x)E(~
D(~ ~ x).
For the isotropic case then the equations to solve now are:

~ ∧E
∇ ~ = 0
~ · (εE)
∇ ~ = 4π(ρf ree + ρ̄),

~ = −∇φ,
or with the ansatz E ~

~ · (ε∇φ)
∇ ~ = ε∆φ + ∇ε
~ · ∇φ
~ = −4π(ρf ree + ρ̄).

Example: A point like charge q in a medium with permittivity ε > 1, constant.

In this case ρf ree = qδ(~x) and ρ̄ = 0 2 . We have chosen the coordinate origin at the
~ and since the problem has
position of the point like particle. Gauss law is now valid for D
spherical symmetry D ~ can only have radial component, D~ = D(r)n̂. Therefore,
Z
4πq = D(r)r 2dΩ = 4πD(r)r 2,
S 2 (r)

that is,
q
D(r) = ,
r2
2
Actually this is not a free charge, in the sense that we have fixed it to stay in a point, but we call it free
to distinguish it from the averaged charges, for here it is used just to create an external field
8.2. A MICROSCOPIC MODEL 127

and
E~ = 1D ~ = q n̂ .
ε ε r2
Since ε > 1 we see that the electric field the charge generates is in this case smaller than
if the charge had been in vacuum. These phenomena are called screening and is due to the
microscopic dipole alignment along the electric field direction and so any sphere centered at
the point-like source has a real charge given by qε .

Example: Point like charge at the center of dielectric sphere of radius a.

ε ε=1

Figure 8.5: Charge at the center of a dielectric sphere

Inside the sphere the result must be identical to the one of the prior example, that is:
D ~ = qn̂2 . Outside D
~ = qn̂2 , E ~ = E(ε
~ = 1) and E ~ = qn̂2 . We see that D
~ is continuous across
r εr r
the surface delimiting the dielectric, that is, where ε jumps, while E ~ is not. The jump in E
~
corresponds to a surface charge density given by:

~ in )|r=a = q (1 − 1 ).
~ out − E
4πσ = n̂ · (E
a2 ε

8.2.1 Matching conditions for dielectrics

Most of the problems one has to deal with when dielectrics are present consists of bodies with
constant permittivity in their interiors, thus we depart from Poisson or Laplace’s equation
only at their boundaries. That is, in most cases the presence of dielectrics manifests itself
only through boundary conditions.
128 CHAPTER 8. DIELECTRICS

In general, given a surface S where ε has a discontinuity we have:

~ 1 · n̂ = D
D ~ 2 · n̂,

or
~ 1 = ε2 n̂ · ∇φ
ε1 n̂ · ∇φ ~ 2,

and consequently, a surface charge density,

~1 − E
~ 2 ) · n̂ = ( 1 1 ~
4πσ = (E − )D · n̂.
ε1 ε2
This can be seen applying Gauss theorem to an infinitely thin pillbox surface enclosing a
section of the discontinuity surface.

DIELECTRIC
VACUUM

normal

Figure 8.6: Pill box in a dielectric material

~ ∧E
Since ∇ ~ = 0, the same argument as the one used in the vacuum case can be applied
and integration along a very thin loop at the surface implies:
~1 − E
(E ~ 2 ) ∧ n̂|S = 0

or
~ 1 − ∇φ
(∇φ ~ 2 ) ∧ n̂|S = 0.

Thus, all derivatives tangential to the interface S of φ1 − φ2 vanish and so (φ1 − φ2 )|S = c for
some constant c. Therefore, if we can arrange for φ1 − φ2 = 0 at some point of S,

(φ1 − φ2 )|S = 0

This in general can be done, for it is just the requirement that there will be no discontinuous
increase of energy when bringing a test charge from infinity through the interface. Thus, we
find that the potential must be continuous along the boundary.
8.2. A MICROSCOPIC MODEL 129

DIELECTRIC VACUUM
γ2

γ1 γ3

γ4

Figure 8.7: Loop in a dielectric material

~0
E
a

Figure 8.8: Dielectric sphere on a constant external field

130 CHAPTER 8. DIELECTRICS

Example: Dielectric sphere in a homogeneous external field.

~ 0 , both, D
Since there are no free charges present in absence of a external field, E ~ and E
~
should vanish. Thus, the solution should depend linearly on E ~ 0.
But then,
φin = αE ~ 0 · ~x
~
φout = β E ~ 0 · ~x + γ E 0 · ~x
r3
for these are the only regular solutions to Laplace’s equations that can be build out of just a
radial coordinate and a constant vector, and are linear on that constant vector 3
Since φ(~x) → −E ~ 0 · ~x when |~x| → ∞, to recuperate the constant field away from the
influence of the sphere, we have, β = −1. The boundary conditions at the surface of the
sphere should give us the two other constants. They are
φin (r = a) = φout (r = a)
∂φin ∂φout
εin |r=a = εout |r=a .
∂r ∂r
Since all terms have the same angular dependence, (n̂ · E~ 0 ) we can factor it out. The first
condition give us,
γ
α = −1 + ,
a3
while the second,
2γ
εin α = εout (−1 − )
a3
Solving this system we obtain,
−3εout γ εin − εout
α= ; 3 = ,
2εout + εin a εin + 2εout
therefore,

~ in = 3εout ~
E E0
2εout + εin
3 3
~ out = E
E ~ 0 (1 + a εin − εout ) − 3a εin − εout (E ~ 0 · n̂)n̂ (8.4)
r 3 εin + 2εout r 3 εin + 2εout
When εout = εin we just get the external field everywhere. When εout = 1 and εin → ∞
(uncharged conductor), E ~ in → 0 and E ~ out → E
~ 0 + a33 (E~ 0 − 3(E
~ 0 · n̂)n̂).
r
Notice also that if εout < εin , as is the case if outside the sphere we are in vacuum, them
3εout
2εout +εin
< 1 and again we have screening.

3
Recall that Laplace’s equation solutions which are regular at the origin are of the form S ij...n xi xj . . . xn ,
~ 0 , and linearity
with S ij...k symmetric and without trace. Thus, in our case S ij...k can only be build out of E
implies that only the dipolar field survives. Without imposing linearity we would have also, for instance
S ij = E0i E0j − 13 δ ij E
~0 · E
~ 0.
8.3. THE ELECTROSTATIC ENERGY OF DIELECTRICS 131

8.3 The Electrostatic Energy of Dielectrics

We want to study now the energetics of a system as the one on the figure bellow, that is a
dielectrics body between an array of conductors.

Figure 8.9: Dielectric material in an external field

If we bring from infinity to the surface of, say, a conductor C1 a charge δq, then we would
be giving to the system an energy given by

−V 1
I
1
δE = V δq = ~ · n̂ dS,
δD
4π ∂C1

where we have used that, since C1 is a conductor, δq would distribute in its surface creating
an increment in the surface charge distribution, δσ, which in turn can be expressed as an
~ δσ = − 1 (δ D
increment in D, ~ · n̂)|∂C1 , where we have taken the normal towards the inside of
4π
the conductor.
Since the potential φ|∂C 1 = V 1 we have,

1 I ~ · n̂ dS
δE = − φδ D
4π ∂C1
1 Z ~ ~ d3~x
= − ∇ · (φδ D)
4πZ V
−1 ~ · ∇φ
~ + φ∇ ~ · δ D]
~ d3~x,
= [δ D
4π V
where V is the space outside the conductors.
Using now that E ~ = −∇φ~ and ∇ ~ ·D
~ = 0 – we are assuming there are no free charges
outside the conductors – we have,

1
Z
δE = ~ · δD
E ~ d3~x.
4π V
Thus, we have an expression for the infinitesimal change in energy in the above configu-
~ and D.
ration of dielectric and conductors in terms of E ~
~ ~ ~ ~
If we assume a linear between E and D, D = εE, then,
132 CHAPTER 8. DIELECTRICS

1 Z ~ ~ 3
δE = εE · δ E d ~x.
4π V
This variation corresponds to an energy given by,
1 Z ~ ~ 3 1 Z ~ ~ 3
E= εE · E d ~x = E · D d ~x.
8π V 8π V
To see this in some detail, consider reaching the final configuration of conductors and
dielectrics by slowly increasing the potential of all conductors simultaneously and at the same
rate, thus the change in the electric potential at the conductors surface will be of the form,
~
V i (λ) = λV i , thus, E(λ) =λE ~ and correspondingly we will have δ E ~ = Edλ.
~ That is, the
field will increase from zero to a final value without changing its direction nor their respective
magnitudes from point-to-point. We then have,

1 1
Z Z
E = [ λεE ~ ·E~ dλ] d3~x
4π V 0
1
Z 1 Z
= [ λ dλ] εE ~ ·E
~ d3~x
4π Z 0 V
1 ~ ·E~ d3~x
= εE
8π ZV
1 ~ ·D~ d3~x.
= E (8.5)
8π V
Chapter 9

Stationary Solutions: Magnetostatics

9.1 The General Problem

Recall that the equations satisfied by the stationary solutions are:

~ ∧E
∇ ~ = 0
~ ·E
∇ ~ = 4πρ
~ ∧B
∇ ~ = 4π J~
c
~ ~
∇ · B = 0.
~ ~x) = J~(~x) , with ∇
We shall assume that ρ(t, ~x) = ρ(~x), J(t, ~ · J(~
~ x) = 0, are given.
We already know how to solve the first pair of equations, so we now concentrate in the
second pair, called the magnetostatics equations:

~ ∧B
∇ ~ = 4π J~ (9.1)
c
~ ~
∇ · B = 0, (9.2)
~ ~x) = J(~
where J(t, ~ x), with ∇~ · J(~
~ x) = 0, is given.
~ x) is needed, for the identity ∇
Notice that this last condition in J(~ ~ · (∇
~ ∧ V~ ) = 0 applied
to the first equation above implies it.
Notice that in contrast with the static equations the sources in this case have vectorial
character, and so they appear in the vectorial equation and not in the scalar one as in the
static case.
As in the electrostatic case our strategy shall be to first find a way to solve trivially the
source-less equation, in this case the scalar one, and then concentrate in the other one. To
do that we introduce the vector potential, A, ~ that is, assume the magnetic field is of the form
~ ~
B = ∇ ∧ A. ~
With this ansatz, then the second equation, the source-less one, is identically satisfied, as
follows from the vector calculus identity mentioned above. The curl equation becomes now,
~ ∇
∇( ~ · A)
~ − ∆A ~ = 4π J,
~
c

133
134 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

where we have used the vector calculus identity

~ ∧ (∇
∇ ~ ∧V
~ ) = ∇(
~ ∇~ ·V
~ ) − ∆V
~.

The above equation can not be considered a system of three equations, one for each
component of J, ~ for three unknown, each component of A, ~ for they are not independent
equations. Indeed, if we take the divergence of the left-hand side we get identically zero. In
fact, it is easy to see that the above equation does not determine uniquely A,~ for if A~ is a
′
~ then A
solution for a given J, ~ =A ~ + ∇λ
~ is also a solution, for any given smooth function λ,
~ ∇
for ∇( ~ · ∇λ)
~ − ∆∇λ ~ = ∇(∆λ)
~ ~ = 0.
− ∆∇λ
~
The lack of uniqueness of A does not affect the uniqueness of B ~ as solutions to the
magnetostatics equations, for
~′ = ∇
B ~′ = ∇
~ ∧A ~ ∧A
~+∇
~ ∧ (∇λ)
~ =∇~ ∧A
~ = B.
~

We can use this freedom in the choice of vector potential to get a simpler equation for it.
We do that imposing to the potential an extra condition which makes it unique. That extra
condition is only for mathematical convenience and has no physical meaning. The most
~ ·A
convenient one is to require that ∇ ~ = 0, which is called the Coulomb Gauge. In this case
~ becomes,
the equation for A

~ x) = − 4π ~
∆A(~ J(~x), (9.3)
c
which, if we express the vectors in Cartesian components, is a system of three decoupled
Poisson equations. In cases where this equation has a unique solution we obtain a unique
~ x).
vector potential A(~
Can one always find a gauge where ∇ ~ ·A~ = 0? The answer in most cases is affirmative,
suppose you have a solution in some other gauge, that is a A ~ ′ such that its divergence is not
zero, then one can solve ∆λ = −∇ ~ ′ for some field λ (provided ∇
~ ·A ~ ′ decays sufficiently fast
~ ·A
asymptotically), and so the new potential, A ~=A ~ ′ + ∇λ
~ will have, ∇~ ·A ~=∇ ~ ·A~ ′ + ∆λ = 0
and so will be divergence-less

9.1.1 Isolated systems of currents

In particular for isolated systems we already know the solution to the above equation (9.3),

1
Z ~ y )d3~y
J(~
~
A(~x) = ,
c lR3 |~x − ~y |
~ and J~ are expressed in Cartesian coordinates.
provided that A

~ as defined by the above integral, satisfies ∇

Exercise: Show that A, ~ ·A
~ = 0.

But, do we get by this procedure all solution to the original magnetostatics equations?
For isolated systems this is so, as the following theorem asserts:
9.1. THE GENERAL PROBLEM 135

Theorem 9.1 Systems (9.1 - 9.2) and (??) are equivalent, in the sense that given J(~ ~ x) in
~ · J(~
lR3 , of compact support and with ∇ ~ x) = 0, for each solution B(~
~ x) of (9.1- 9.2) there exists
~ x) of (??) and vice-versa, provided both decay asymptotically sufficiently
a unique solution A(~
fast.

Proof: Let J~(~x) be given and let B(~ ~ x) and A(~

~ x) be the corresponding unique solution of
(9.1 - 9.2) and (??) respectively, decaying sufficiently fast at infinity.
~ satisfies equations (9.1 - 9.2) then it is unique. To see this we take
We first show that if B
the curl of the first equation to get,

~ ∧ (∇
0=∇ ~ ∧ δ B)
~ = ∇(
~ ∇~ · δ B)
~ − ∆δ B,
~

~ is the difference between two solutions with the same given J.

where δ B ~ We use now the
second equation to eliminate the first term on the right and get,

~ = 0.
∆δ B

Expressed in Cartesian coordinates this is a decoupled system of three Poisson equations and
from their uniqueness, for the case that the magnetic field decays asymptotically, it follows
~ = 0. Thus, if we show that B
that δ B ~ ′ := ∇
~ ∧A ~ also satisfies the (9.1 - 9.2 ) equations, we
′
would conclude that B~ =B ~ =∇ ~ ∧ A.
~ The curl of B ~ ′ is

∇ ~′ = ∇
~ ∧B ~ ∧ (∇~ ∧ A)
~ = ∇(~ ∇
~ · A)
~ − ∆A
~
~ ∇
= ∇( ~ + 4π J.
~ · A) ~
c

~ · J~ = 0 we get,
But, taking the divergence of equation (??), and using that ∇

~ · A)
∆(∇ ~ = 0,

~ ·A
and the uniqueness of Poisson’s equations then implies ∇ ~ = 0, so we conclude B
~ satisfies
system (9.1 9.2).

Example: Circular Current Loop

In this case,

J~ = Jϕ êϕ
= Jϕ (− sin ϕ êx + cos ϕ êy )
′
with Jϕ = I sin θ′ δ(cos θ′ ) δ(r a−a) .

Exercise: Show that across the plane ϕ = ϕ0 the flux of J~ is I.

136 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

z
θ

ϕ
x

Figure 9.1: Circular Current Loop

1
Expanding ′ is spherical harmonics we get,
|~
x−~x |
l l
~ θ, ϕ) = I sin θ′ δ(cos θ′ )δ(r ′ − a) X ⋆
Ylm (θ, ϕ)Ylm (θ′ , ϕ′)
Z X
A(r, 4π
c lR3 a l=0 m=−l
2l + 1
l
r< ′ ′ ′2 ′ ′ ′ ′
l+1 (− sin ϕ êx + cos ϕ êy )r dr sin θ dθ dϕ
r>
∞ X l l Z 2π
4πIa X Ylm (θ, ϕ) r< ⋆ π
= l+1 Ylm ( , ϕ′ )(sin ϕ′ êx + cos ϕ′ êy ) dϕ′
c l=0 m=−l 2l + 1 r> 0 2

But,

⋆ π
Z 2π
Ylm ( , ϕ′ )(sin ϕ′ êx + cos ϕ′ êy ) dϕ′
0 2
iϕ′ ′ ′ ′
Z 2π
⋆ π ′ e − e−iϕ eiϕ + e−iϕ
= Ylm ( , ϕ )( êx + êy ) dϕ′
0 2 2i 2
and
⋆ π
Z 2π ′
Ylm ( , ϕ′ )eiϕ dϕ′
0 2 v
s
2l + 1 u (l − 1)!
u
= πδm,1 t Pl1 (0)
4π (l + 1)!

and so,
∞ l
~ θ, ϕ) = πIa
X r< 1 1 (l − 1)! iϕ
A(r, l+1 [Pl (0)Pl (cos θ) e (iêx + êy )
c l=0 r> (l + 1)!
(l + 1)! −iϕ
+Pl−1(0)Pl−1 (cos θ) e (−iêx + êy )].
(l − 1)!
9.2. BOUNDARY CONDITIONS - SUPER CONDUCTORS 137

Using now, Pl−1(x) = (−1) (l−1)! P 1(x), the above expression becomes,
(l+1)! l

∞ l
~ θ, ϕ) = πIa r< 1
P 1 (0)Pl1(cos θ)[−2 sin ϕêx + 2 cos ϕêy ]
X
A(r,
c l=0 r> l(l + 1) l
l+1

∞ l
2πIa X r< 1
= êϕ [ l+1 Pl1 (0)Pl1 (cos θ)],
c l=0 r> l(l + 1)

where, 
 0 l even
Pl1 (0) =  (−1)n+1 Γ(n+ 23 )
Γ(n+1)Γ( 32 )
l = 2n + 1
Recall that the Gamma function takes the following values,
√
1 π
Γ(n) = n!, Γ(n + ) = n (2n − 1)!!
2 2
Far away from the source, (r> = r, r< = a), the leading contribution will come from the
first non-null term, that is,
2 2
~ θ, ϕ) ≡ − πIa P 1 (cos θ)êϕ = πIa sin θêϕ
A(r,
cr 2 1 cr 2
2
πIa
≡ (−yêx + xêy )
cr 3
m
~ ∧ ~x
≡ ,
r3
πa2 I
~ :=
with m c
k̂ = (Area of circular loop) Ic k̂. To this approximation the magnetic field is,

~ =∇
B ~ ∧A ~ ∧ (m
~ ≡ ∇ ~ ∧ ~x
) = ~ 1 ) ∧ (m
∇( ~ ∧ ~
x ) +
1
(m(
~ ∇~ · ~x) − (m ~ x)
~ · ∇)~
r3 r3 r3
−3~x 3m~ m
~
≡ ( 5 ) ∧ (m ~ ∧ ~x) + 3 − 3
r r r
−3 2m~
≡ 5
(m(~
~ x · ~x) − ~x(~x · m))
~ + 3
r r
3(n̂ · m~ )n̂ − m~
≡ ,
r3
which in analogy with the electrostatic expression is called a magnetic dipole.

9.2 Boundary Conditions - Super conductors

In magnetostatics, superconductors play a similar role to the one ordinary conductors play in
electrostatics. For our purposes a superconductor can be defined as a body inside of which
no magnetic field can be present.
138 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

The same integral argument used for conductor boundaries in the electrostatic case (8.2.1)
says here that:

1. From ∇~ ·B~ = 0, the normal component of the magnetic field must be continuous and
~ · n̂|∂V = 0.
therefore B

~ B
2. From ∇∧ ~ = 4πJ~ , the jump on the tangential component is due to a superficial current
c
density, 4πc k̂ = −B~ ∧ n̂|∂V .

Since the second condition now includes a vector source we deduce it again. Following the
notation in figure 8.7 we integrate the magnetic field along the curve γ, as in that case we
call n̂ the unit normal, ˆl the unit vector along the outer part of the curve γ. It is understood
that we are in the limit of a small curve so that their pieces along both sides of the interface
are almost straight lines and the pieces entering the interfaces are very small. Since these two
vectors are perpendicular among each other, the vector ŝ = n̂ ∧ ˆl is unitary, perpendicular to
both, and normal to the surface enclosed by γ. In our figure it would be a vector entering the
page.
Using Stokes theorem we get:

Z l1 Z
~ · ˆl dl =
B ~ · d~l
B
l0 γ
Z
= ~ ∧ B)
(∇ ~ · dS
~
S(γ)
4π
Z
= J~ · dS
~
c S(γ)
4π l1 re ~
Z Z
= J · ŝ dl dr
c l0 ri

where r is a coordinate going perpendicular to the interfaces so that ∂r ∂ ~

= n̂ · ∇.
In the limit ri r0 → 0, r0 − re → 0, where r0 is the coordinate value at the interface, the
integral vanishes unless there is a distributional surface current, J~ = δ(r − r0 )~k with ~k · ~n = 0.
In the distributional case the result becomes,

~ · ˆl dl = 4π
Z l1 l1 Z
B ~k · ŝ dl
l0 c l0
Since the integrals have arbitrary length we conclude that,

4π ~ ~ · ˆl
k · ŝ = B
c
=~ · (ŝ ∧ n̂)
B
= −B~ ∧ n̂ · ŝ,

from this, and the arbitrariness of ˆl, hence, ŝ, a the above interface matching condition follows.
9.2. BOUNDARY CONDITIONS - SUPER CONDUCTORS 139

This type of problems can be solved using the vector potential equation with the following
boundary conditions:

A~ ∧ n̂|∂V = 0
~ · [n̂(n̂ · A)]|
∇ ~ ∂V = 0.

The first condition guarantees B ~ · n̂|∂V = 0, for B~ · n̂ contains only tangential derivatives
of the tangential components of A. ~ Depending on the physical situation this condition is too
restrictive, we shall see this latter in an example. The second implies, together with the first,
that ∇~ · A|
~ ∂V = 0, a boundary condition sufficient to ensure that ∇ ~ ·A
~ = 0 everywhere outside
the superconductor, and so that B ~ =∇ ~ ∧A ~ is the solution sought.

Example: Super conducting sphere of radius a in the presence of a constant

magnetic field.

~k
~0
B
a

Figure 9.2: Super conducting sphere

Choosing the z axis along the constant external magnetic field we have,

~ 0 = B0 k̂.
B
This field has a vector potential given by

~ 0 (~x) = B0 (k̂ ∧ ~x),

A
2
indeed,

~ 0 (~x) = B0 ∇
~ ∧A
∇ ~ ∧ (k̂ ∧ ~x) = B0 (k̂ ∇
~ · ~x − (~k · ∇)~
~ x)
2 2
B0
= (3k̂ − k̂) = B0 k̂.
2
140 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

~ 0 does not single out any point in space, and so any coordinate origin,
Notice that while B
~ 0 (~x) does. That means that the difference between two vector potentials for B
A ~ 0 with
different origins must be just the gradient of a function, indeed,
~ ~ ~
~ −A
A ~ 0 = B 0 (k̂ ∧ ((~x + ~r)) − B 0 (k̂ ∧ ~x) = B 0 k̂ ∧ ~r
~r 2 2 2
~
= ∇(~ B 0 (k̂ ∧ ~r) · ~x).
2
For this problem it is convenient to choose as coordinate origin the center of the supercon-
ducting sphere and a external vector potential centered on it.
The external field would induce currents on the surface of the superconducting sphere
which would rotate along circular loops, all of them perpendicular to the k̂ direction, these
in turn would generate a vector potential, which as in the case of the single circular current
loop, would have only component along the eˆϕ direction. Thus, the induced vector potential
would have the form:
A~ I (~x) = AI (r, θ)(k̂ ∧ ~x).
This vector is tangent to all spheres centered at the origin, and so the boundary condition,
~ · [n̂(A
∇ ~ · n̂)]|r=a = 0

is satisfied trivially.
The other boundary condition then implies,
~ θ, ϕ)|r=a = ( B0 + AI (r, θ))(k̂ ∧ ~x)|r=a = 0,
A(r,
2
but then AI (a, θ) = − B20 , independent of θ. But we have already found a solution with these
characteristics, namely,

~ I (~x) = m
A
~ ∧ ~x
,
r3
3
~ = − B02a k̂.
with m
The total solution is then,
3
~ x) = (k̂ ∧ ~x)( B0 − B0 a ),
A(~
2 2r 3
and
~ B0 a3
B(~x) = B0 k̂ − (3(n̂ · k̂)n̂ − k̂).
2r 3
Notice that A~ has only component along φ̂, vanishes at r = a, and growth as r → ∞.
This last property is not relevant, for it is only a property of the gauge.
The induced surface current is:
~k = c (B~ ∧ n̂)|r=a = c B0 (k̂ ∧ n̂ − 3 (n̂ · k̂)(n̂ ∧ n̂) + 1 k̂ ∧ n̂) = 3cB0 k̂ ∧ n̂.
4π 4π 2 2 8π
9.3. THE MAGNETIC SCALAR POTENTIAL 141

Figure 9.3: Magnetic field: dipole

9.3 The magnetic scalar potential

The problem on the last example can be also solved in the following alternative way. Outside
the sphere we are in vacuum and therefore we have ∇ ~ ∧B
~ = 0. Therefore, in that region there
~ = −∇ϕ
will also be an scalar potential, ϕm (~x), called the magnetic potential, such that, B ~ m.
So now we must solve for
~ ·B
∇ ~ = −∇
~ · (∇ϕ
~ m ) = −∆ϕm = 0

outside the sphere, with the boundary condition,

~ · n̂|∂V = n̂ · ∇ϕ
B ~ m |∂V = 0,

That is, a Neumann boundary value problem. In this case, ϕ0 (~x) = −B0 k̂·~x and ϕI (~x) = α k̂·r~3x ,
the only other solution to Poisson’s equation with this angular dependence. The boundary
condition implies,
3α ~
~ 0 + ϕI )|r=0 = −B0 k̂ · ~n − α~
n̂ · ∇(ϕ k · ~
n + k · ~n
a3 a3
2α
= (−B0 − 3 )~k · ~n = 0
a
3
Thus, α = − Bo2a , and the problem is solved. We see that for boundary value problems in
vacuum this method is very useful, for it reduces the problem to a single Poisson equation
which we can handle very easily with the techniques already learn for electrostatics. But one
has to be very careful, for there are situations where the magnetic potential can not be defined
everywhere outside the superconducting bodies.
To see how this problem arises we shall first give an argument showing why the definition
of a potential does always works in the electrostatic case. Given a curve γ in lR3 with starting
142 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

point ~x0 and ending point ~x we can define

~ x(s)) · d~x ds,

Z Z 1
φγ (~x) := − ~ · d~l =
E E(~
γ 0 ds
where the curve γ is given by the map ~x(s) : [0, 1] → lR3 , with ~x(0) = ~x0 , ~x(1) = ~x.

~x

~
E

~
E
γ

~
E
~x0

Figure 9.4: Integrating the electric potential

If we take any another curve γ̃, also starting at ~x0 and ending at ~x, we can also define
φγ̃ (~x), we claim φγ (~x) = φγ̃ (~x) and so this procedure really define a function in lR3 , φ(~x),
independent of any particular curve chosen to compute it.

~x
γ̃
~
E

~
E
γ

~
E
~x0

Figure 9.5: Integrating the electric potential along another path

Indeed,
Z Z
φγ (~x) − φγ̃ (~x) = ~ · d~l −
E ~ · d~l
E
γ γ̃
I
= ~ · d~l,
E
γ̃ −1 γ
9.3. THE MAGNETIC SCALAR POTENTIAL 143

where γ̃ −1 is the curve “going backwards" of γ̃, i.e. if γ̃ is given by ~x˜(s), then γ̃ −1 is given by,
(~x˜−1 (s) = ~x˜(1 − s). But then the curve γ̃ −1 γ is a closed curve, starting at ~x0 , going (with γ)
up to ~x and returning (with γ̃ −1 ) to ~x0 . Using Stokes theorem we then have,
Z
φγ (~x) − φγ̃ (~x) = ~ ∧ E)
(∇ ~ · n̂ dS = 0,
S

where S is any surface having as boundary the curve γ̃ −1 γ, [see figure], and we have used the
~ ∧E
electrostatic equation, ∇ ~ = 0.

~x
γ̃ −1
~
E

~
E
γ

~
E
~x0

Figure 9.6: Electric potential: path independence

Exercise: Choose families of curves which go along the three coordinate axis to show that
~ = −E.
∇φ ~

Thus, we conclude that the electrostatic potential is always well-defined on lR3 . On the
contrary this is not the case for the magnetic potential, for, if we similarly define
Z
ϕmγ (~x) = − ~ · d~l,
B
γ

then it is clear from the above argument that if currents are somewhere present, then the
potential does depends on the loop. To see this consider the following current distribution,
(see figure) where we have a closed current loop.
Thus, if we go to the point ~x along γ we get some value for φm (~x), φmγ (~x), if we go along
γ̃ we get some other value, its difference is
I Z
ϕmγ (~x) − ϕmγ̃ (~x) = ~ · d~l =
B ~ ∧ B)
(∇ ~ · n̂ dS
γ̃ −1 γ S
4π 4π
Z
= J~ · n̂ dS = I,
c S c
that is, proportional to the total current along the loop.
144 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

~
B ~x
γ̃

~
B
γ

~
B
~x0

Figure 9.7: Magnetic potential: path dependence

9.3.1 Wires
We consider now the idealization of an infinitely thin current loop, that is a current line o
wire. If the current circulates along a closed loop γ : [0, 1], given, say, by ~x(s) : [0, 1] → lR
with ~x(1) = ~x(0). Then a current density is a distribution given by:
1 d~x(s)
Z
~ x) =
J(~ I δ(~x − ~x(s)) ds,
0 ds
where I is a constant denoting the total current flowing along the loop, and d~xds(s) is the
circulation velocity of the chosen parametrization of the curve. On the above expression only
the tangency of that velocity is relevant, for one can see that the integral does not depend on
the parametrization chosen to describe the curve.
wire

d~
x(s)
ds

~x(s)

Figure 9.8: A current wire

To see how to deal with such distributions, we shall check now that it satisfies charge
conservation, namely,
9.3. THE MAGNETIC SCALAR POTENTIAL 145

~ · J~(~x) = 0. In the sense of distributions, this means

∇
Z
3
J~ · ∇ϕ
~ d3~x = 0
lR

for all smooth functions ϕ of compact support, but

d~x 1
Z Z Z
~ x) · ∇ϕ(~
J(~ ~ x) d3~x = ~ x) ·
I ∇ϕ(~ dsδ(~x − ~x(s)) d3~x
3 3
lR lR 0 ds
Z 1
d~x ~ Z 1
dϕ(~x(s))
= I · ∇ϕ(~x(s))ds = I ds
0 ds 0 ds
= I[ϕ(~x(0)) − ϕ(~x(1))] = 0,

since ~x(0) = ~x(1).

The vector potential for this current, assuming it is an isolated system, is

~ x) = 1
~ y )d3~y
J(~ I d~l
Z I
A(~ = ,
c |~x − ~y | c γ |~x − ~x(s)|

where we have defined d~l = d~xds(s) ds.

~ x)
dA(~

~x − ~x(s) d~
x(s)
ds

~x

~x(s)

Figure 9.9: Vector potential of a wire

Similarly,
~ x) = −I (~x − ~y ) ∧ d~l
I
~ x) = ∇
B(~ ~ ∧ A(~ ,
c γ |~x − ~y |3
which is Biot - Savart’s law.
Contracting the vector potential with a constant vector field, ~k, using Stokes theorem we
146 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

have,
I ~
~ x) = I
~k · A(~ k · d~l IZ ~ ~k
= ∇~y ∧ ( ) · n̂ dS
c γ |~x − ~y | c S |~x − ~y |
IZ ~ 1
= (∇~y ( ) ∧ ~k) · n̂ dS
c S |~x − ~y |
I ~ ~ ( 1 )) dS,
Z
= k · (n̂ ∧ ∇ ~y |~x − ~y |
c S

where S is any surface whose boundary is γ, and n̂ its unit normal. Thus,

~ x) = − I ~ ( 1 ) ∧ n̂ dS,
Z
A(~ ∇~y |~x − ~y |
c S

and so,

~ x) = −I ∇ ~ ( 1 ) ∧ n̂) dS
Z
B(~ ~ ∧ (∇
c S ~
x ~y |~x − ~y |
−I ~ ( 1 )) + (n̂ · ∇ 1
Z
= ~ ·∇
[−n̂(∇ ~ )(∇ ~
c S ~x ~y |~x − ~y | ~x ~y |~x − ~y | )] dS
I ~ ( 1 )) + (n̂ · ∇ 1
Z
= [−n̂(∇ ~ ·∇ ~ )(∇ ~ )] dS
c S ~
y ~
y |~x − ~y | ~
y ~
y |~x − ~y |
I ~ ( 1 )] dS
Z
= [n̂ 4πδ(~x − ~y ) − (n̂ · ∇ ~ )∇
c S ~
y ~x |~x − ~y |
−I ~ ~ ( 1 ) dS]
Z
= ∇~x [ n̂ · ∇ ~y |~x − ~y |
c S

1
where in the second step we have used that, ∆( ) = −4πδ(~x − ~y ), and in the third that
|~
x−~y |
~x is not a point along the line current. So we see that

I 1 I n̂ · (~x − ~y )
Z Z
ϕm (~x) = ~ (
n̂ · ∇ ) dS = dS.
c S
~y |~x − ~y | c S |~x − ~y |3

From the figure, and changing the coordinate origin to the point ~x, that is, using a new
integration variable ~y˜ = ~y − ~x, the above integral becomes,
I n̂ · m̂
Z
ϕm (~x) = dS,
c S |~y˜|2
˜
with m̂ = ~y˜ = (~x−~y ) . Note that n̂ · m̂ dS is just the area as seen from ~x, that is, the solid
|~
y| |~
x−~y |
angle that differential spans times the square of the distance to ~x. So, n̂·˜m̂2 dS is just the
|~
y|
differential of solid angle spun by dS as seeing from ~x. Correspondingly the integral is just
the solid angle spun by the whole surface S. If we approach that surface from bellow n̂· m̂ > 0
the value of that solid angle in the integration tends to 2π, while if we approach the surface
9.3. THE MAGNETIC SCALAR POTENTIAL 147

n̂ S
~x − ~y

γ
~x
~y

Figure 9.10: Magnetic potential

n̂ S
m̂

Figure 9.11: Magnetic potential

148 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

from above, n̂ · m̂ < 0, the solid angle covered tends to −2π, and so we have a jump of 4πI
c
on
ϕm . The surface S is arbitrary, as long as its boundary is γ, and so we can choose the jump
whatever we please, but it has to be somewhere.

Example: The surface integral of J~

We want to see now that if

1 d~x(s)
Z
~ x) = I
J(~ δ(x − x(s)) ds,
0 ds
then,
Z
J~ · n̂ dS = I.
S

where S is any surface punctured just once by ~x(s) for some s.

The above expression is just a short-hand for,

1 d~x(s)
Z
~
J(φ) =I φ(x(s)) ds,
0 ds
and this distribution is not so wild that one can even apply to other distributions, in particular
to surface distributions. Among surface distributions is n̂S which has support just on a smooth
surface S and there is its unit normal. When applied to a smooth compactly supported test
vector li it gives the flux of such a vector across S. We claim that J i (n̂i ) = I and so represents
the integral flux of J~ across S. To see this consider a thickening of the surface S and the
following distributional normal,

1 ~
(
∇r 0≤r<ε
n̂εi = 2ε (9.4)
0 r≥ε
where r is the distance from S into a neighborhood of it, and we are taking ε small enough
so that the distance is well-defined and smooth. In the limit ε → 0 this gives a good integral
representation of n̂S .
But,
i ε I Z s(ε)+ d~x(s) ~
J (n̂i ) = · ∇rds,
2ε s(ε)− ds
where s(ε)± is the value of s for which r(~x(s)) = ε before and after the loop goes into the
region where r < ε. If ε is small enough the relation r = r(~x(s)) can be inverted, and we can
define s(r), so after using r as variable in the above integral we get,

I
Z +ε
d~x(r) ~
J (n̂εi ) =
i
· ∇rdr = I.
2ε −ε dr
Thus we can take the limit ε → 0 and get the correct result.
9.4. NON-SIMPLY CONNECTED SUPER-CONDUCTORS 149

9.4 Non-Simply Connected Super-Conductors

The use of a magnetic potential allows to quickly conclude that, in the absence of an external
~ ≡ 0. Indeed, in this case
field, for the superconducting sphere the only possible solution is B
the potential is well-defined everywhere outside the sphere, and so we have the problem,

∆φm = 0 outside the sphere (9.5)

~ m = 0
n̂ · ∇φ at the sphere. (9.6)

Multiplying by φm the first equation, integrating on the whole space outside the sphere, and
assuming the fields decay at infinity we find that φm = cons, and so that B ~ ≡ 0. The same
argument follows for any superconductor body whose topology implies that φm is well-defined
everywhere outside it. But this is true provided any closed loop outside the body can be
continuously deformed to zero. Bodies with this property are called simply connected.
What happens in the case of bodies which are not simply connected? As is the case of a
superconducting ring?
Physically it is reasonable that we can have configurations where a current of arbitrary
total intensity flows along the ring. In fact, we shall show latter how to build such a configu-
ration. Thus, we expect in this case to have many non-trivial solutions.
Mathematically we can look for these solutions in the following way: Since the magnetic
potential will necessarily have discontinuities, –but we can choose where they are going to
be– we set the following boundary value problem (see figure):

SR

D+
D−

Figure 9.12: Superconducting ring.

∆φm = 0 in V = lR3 − {Ring} − {Disk closing the Ring} (9.7)

~ m = 0 in
n̂ · ∇φ SR (9.8)
4πI
(φ+ −
m − φm ) = At Disk (9.9)
c
φm → 0 as |~x| → ∞. (9.10)

where I is the total current we have flowing along the ring.

Lemma 9.1 The above problem has a unique solution.

150 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

Proof: We only prove uniqueness and not existence. First notice that at the disk inside the
~ + = n̂ · ∇φ
ring, D, we must have n̂ · ∇φ ~ − . Indeed, applying Gauss theorem to ∇ ~ ·B~ = 0 in
m m
a pill-box containing a piece of the disk and taking the limit on which the pill-box flattens
over the disk the assertion follows. Assume now we have two solutions, φ1 and φ2 satisfying
the above problem, then δφ ≡ φ1 − φ2 satisfies,

∆δφ = 0 in V = lR3 − {Ring} − {Disk closing the Ring} (9.11)

~
n̂ · ∇δφ = 0 in SR (9.12)
+ −
(δφ − δφ ) = 0 At Disk (9.13)
~ + − n̂ · ∇δφ
(n̂ · ∇δφ ~ −) = 0 At Disk (9.14)
δφ → 0 as |~x| → ∞. (9.15)

But then,

Z
0 = − δφ∆δφ d3~x (9.16)
V
Z Z
= ~
∇δφ ~
· ∇δφ d3~x − ~
δφn̂ · ∇δφ ~
d2 S (9.17)
V SR
Z Z
− ~ + d2 S
δφ+ n̂ · ∇δφ ~− ~ − d2 S
δφ− n̂ · ∇δφ ~ (9.18)
D+ D−
Z
= ~
|∇δφ| 2 3
d ~x (9.19)
ZV
− ~ + − n̂ · ∇δφ
δφ+ (n̂ · ∇δφ ~
~ − ) d2 S, (9.20)
ZD
= ~
|∇δφ| 2 3
d ~x, (9.21)
V

where in the third equality it was used that at the disk, δφ+ = δφ− , and in the fourth that
at the disk both gradients were also the same, and we have taken the normal to be the one
incoming into the upper disk. This shows that ∇δφ~ = 0, and since δφ → 0 as |~x| → ∞, we
conclude δφ = 0 everywhere. Thus, the total current flow suffices to determine uniquely the
solution. 1
In electrostatics, we saw that we could give either the potential V or the total charge Q
on a conductor (or the corresponding arrays of potential or charges in the case of an array of
conductors) and that –together with the geometry of the bodies– would determine a unique
solution. Does there exist in the case of superconductors another quantity that one could
specify and so determine a unique solution? The answer is affirmative and the other quantity
is the total magnetic flux, Z
Φ≡ B ~ · n̂ d2 S, (9.22)
S

1
In fact one could have avoided this calculation since δφ is smooth at the disk surface, and so we can
extend the Gauss surface pass that surface and just wrap up the superconductor, but at that surface we just
~
use n̂ · ∇δφ|S = 0 to finish the argument.
9.4. NON-SIMPLY CONNECTED SUPER-CONDUCTORS 151

where S is any surface whose boundary, γ, meets the body. In the case of the ring one could
take, for instance, the disk D.
Using that B ~ =∇ ~ ∧ A,
~ and Stokes theorem we get,
I
Φ= ~ · d~l.
A (9.23)
γ

Exercise: Check that this alternative definition is gauge independent.

~
B ~
B
~
S B
n̂
~
B

~
A
γ

Figure 9.13: Magnetic Flux

Notice that this can be taken as an integral condition on the magnetic field, and so it is, in
some sense, like the condition that the integral of the normal to the electric field on the surface
of a conductor is the total charge it contains. In fact, it is the total flux which is the analog to
the total charge and not the total current, for if we have an array of conductors with potentials
and charges, (V i , Qi ), and move them around changing their geometrical configurations, then
their potentials V i will change values, but not their total charges Qi . On the other hand, if
we have an array of superconducting bodies and move them around, then their currents I i
will change, but not their fluxes Φi , as follows from the following calculation,

dΦ ∂ ~
Z
= B · n̂ d2 S (9.24)
dt S ∂t
Z
= −c (∇ ~ ∧ E)~ · n̂ d2 S (9.25)
IS
= −c ~ · d~l
E (9.26)
γ
= 0. (9.27)

Since the last integral is along the border of the superconductor and there E~ can only have
normal component. Since this calculation is valid for any one of the bodies it shows that the
flux on each one of them is constant in time, and so they will not change if we change the
superconductor configuration.
152 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

This constancy is used to induce currents in superconductors, in particular in supercon-

ducting electromagnets: One takes a ring in its normal (non-superconducting) phase and
place it in an external magnetic field. This field generates on it the desired flux. One then
cools it down to the superconducting phase and then takes it away from the external field.
Since the flux remains constant it has to be now due to an internal current.
The uniqueness proven above implies that

Φi = L̃ij I j , (9.28)

for some matrix L̃ij , that is, the flux is a function of the currents, and it is a linear function. We
claim, without proving it, that this relation is also invertible. The coefficients Lij ≡ cL̃ij are
called inductances and, as the capacities, they only depend on the geometrical configuration
of the system.

Φ1 I1 Φ2

I2

Φ3
I3

Figure 9.14: Inductances

9.5 Multipolar expansion of the magnetostatics field

Recall that for isolated systems of currents we have the formula,

1
Z ~ y)
J(~
~
A(~x) = d3~y ,
c lR3 |~x − ~y |
valid in Cartesian coordinates, or contracting with an arbitrary constant vector ~k,

~k · J(~
~ y)
~ x) = 1
Z
~k · A(~ d3~y ,
c lR3 |~x − ~y |
in an arbitrary coordinate system.
9.5. MULTIPOLAR EXPANSION OF THE MAGNETOSTATICS FIELD 153

But this expression is identical to the expression for the scalar potential en electrostatics,
so we can proceed as in electrostatics and make a Taylor series expansion of the function
1
,
|~
x−~y |
1 1 ~ 1 ) + ...
= − ~y · ∇(
|~x − ~y | |~x| |~x|
and so obtain the leading behavior of the magnetostatic field at large distances away from
the current sources,

~ x) = 1 ~ 1 ) + . . .] d3~y ,
~k · J~(~y )[ 1 − ~y · ∇(
Z
~k · A(~
c lR3 |~x| |~x|
1 1 ~ 1 )·
Z Z
= [ ~k · J(~
~ y ) d3~y − ∇( ~y ~k · J(~
~ y ) d3~y + . . .]. (9.29)
c |~x| lR3 |~x| lR3
~ ~k · ~x)) = (~k · ~x)∇
~ · (J(
But ∇ ~ · J~ + J~ · ~k = J~ · ~k, for ∇
~ · J~ = 0, so
Z Z Z
~k · J(~
~ y ) d3~y = ~ ~k · ~y )) d3~y =
~ · (J(
∇ J~(~k · ~y ) · n̂ dS 2 = 0, (9.30)
3 3
lR lR S 2 (∞)

since the sources are assumed to have compact support. Thus, since −∇( ~ 1 ) = n̂ , we have
|~
x| |~
x|2
the very important fact that the magnetostatic field does not have any monopole contribution,
that is, it decays one order (in 1r ) faster than the electric field, as we go away from the sources.
The divergence free property of J~ can also be used to write in a more transparent way the
subsequent terms in the series. To do that, notice that in general,

~ · (Jf
∇ ~ ) = f∇
~ · J~ + J~ · ∇f
~ = J~ · ∇f.
~
And since J~ has compact support,
Z Z
3
J~ · ∇f
~ d3~y =
3
~ · (Jf
∇ ~ ) d3~y = 0.
lR lR
To handle the first term we used f = ~x · ~k. For the second it is convenient to use a vector,
namely f = ~x(~x · ~k), then,

J~ · ∇f ~ x · ~k) + ~x(J~ · ~k).

~ = J(~
Therefore,
Z Z
3
~x(~k · J)
~ d3~x = − ~ ~k · ~x) d3~x
J(
lR lR3
1
Z
= [~x(~k · J~) − J(
~ ~k · ~x)] d3~x
2 lR3
1Z ~ ~ d3~x
= k ∧ (~x ∧ J)
2 lR3Z
1
= ~k ∧ ~ d3~x
(~x ∧ J)
2 lR3
:= c~k ∧ m
~, (9.31)
154 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

where we have defined the magnetic momentum ,

1 Z
~ := −
m (J~ ∧ ~x) d3~x,
2c lR3
and its density,
~ (~x) := − 1 J~ ∧ ~x
M
2c
usually called magnetization .
The first non identically null term in the series is then,
1 ~ 1 1 −1
Z
− (∇( )) · 3
~y (J~ · ~k) d3 ~y = 3 ~x · (~k ∧ m) ~ ) · ~k,
~ = 3 (~x ∧ m
c |~x| lR |~x| |~x|
that is,

~ x) ≡ m
A(~
~ ∧ ~x
,
|~x|3
and correspondingly,

~ x) = 3n̂(n̂ · m)
B(~
~ −m~
.
|~x| 3

For a line current we have,

−1 I
Z I
~ =
m 3
J~ ∧ ~x d3~x = ~x ∧ d~l.
2c lR 2c
~l|
If the current loop is contained in a plane, then |~x∧d2
is the area of the infinitesimal triangle
2
of the figure below , and so

I
× Circuit area
~ =
|m|
c
while the direction of m
~ is perpendicular to the plane where the circuit lies.
If the current is due to point like charges in motion, then

J~ =
X
qi~vi δ(~x − ~xi ),
i

and so,
1 X
~ =
m qi (~xi ∧ ~v i ).
2c i
If the charges have mass mi and therefore angular momentum,
2
To see this, first notice that Area = h(b12+b2 ) . Choosing the ê1 vector along the vector ~x + d~l, we have,
h = −d~l · ê2 = ~x · ê2 , b1 = d~l · ê1 , b2 = ~x · ê1 . Thus, h b1 = ~x · ê2 d~l · ê1 , h b2 = −d~l · ê2 ~x · ê1 . And so,
h(b1 + b2 ) = |d~l ∧ ~x|.
9.5. MULTIPOLAR EXPANSION OF THE MAGNETOSTATICS FIELD 155

d~l
~x + d~l

~x

Figure 9.15: Circuit and area differential

ê1

ê2 b1 d~l
~x + d~l h

b2
~x

Figure 9.16: The area differential

156 CHAPTER 9. STATIONARY SOLUTIONS: MAGNETOSTATICS

~ i = mi (~xi ∧ ~v i ),
L
then,
1 X qi ~
~ =
m Li .
2c i mi
In the case of identical particles, or in the more general case of particles with the same charge
to mass ratio, we have,
q ~
~ =
m L,
2cm
where L~ is the total angular momentum of the system. This relation is called the giro-magnetic
momenta. It works well for even orbital electrons, if we do not take into account their intrinsic
angular momentum or spin. For their intrinsic angular momentum this relation fails in the
sense that the numerical factor (in the case above 21 ) takes another value.
Chapter 10

The Energy of the Magnetostatic Field

As we have seen, the energy stored in a magnetic field in a volume V is given by,
1 Z ~ ~ 3
E= B · B d ~x. (10.1)
8π V
In the case of a stationary field this quantity can be expressed in terms of the currents which
~ B
generate the fields. To do that we first use that since ∇· ~ = 0, there exists a vector potential,
~ ~ ~ ~
A such that B = ∇ ∧ A. Second we substitute it for one of the B’s ~ above and use the identity,
~ · (∇
V ~ ∧W ~ )−W ~ · (∇
~ ∧V~)=∇ ~ · (W
~ ∧V ~ ). Integrating by parts we obtain,

1 Z ~ ~ ~ d3~x + 1
I
~ ∧ B)
~ d2 S.
E= (A · ∇ ∧ B) n̂ · (A (10.2)
8π V 8π ∂V
Finally we use now one of Maxwell’s stationary equations, ∇ ~ ∧B~ = 4π J~, to obtain,
c

1
Z I
E= ~ · J~ d3~x +
A ~ ∧ B)
n̂ · (A ~ d2 S. (10.3)
2c V ∂V

Exercise: Check that the above expression, as the original one, does not depend upon any

~ That is, check that the expression remains the same if we use
particular gauge chosen for A.
′
~ =A
A ~ + ∇λ,
~ for any arbitrary smooth function λ. a.-) Consider only the case V = lR3 and
assume the fields decay sufficiently fast at infinity so that no surface integral contributes. b.-)
Consider the case of arbitrary region V .

Example: Energy change due to the introduction of a superconducting sphere

into a constant magnetic field. There are two contributions to this energy difference.
One is the contribution due to the lost of the constant magnetic field inside the volume. The
other is the one due to the magnetic field produced by the induced surface current ~k.
The first is given by,
3 3
−1 ~ 0 |2 d3~x = −1 4πa B 2 = −a B 2
Z
∆E1 = |B 0 (10.4)
8π r≤a 8π 3 6 0
The second term is the energy difference in the region V outside the sphere. That is,

157
158 CHAPTER 10. THE ENERGY OF THE MAGNETOSTATIC FIELD

1 Z ~ ~ I )2 d3~x − 1
Z
~ 2 d3~x = 1
Z
~ I + 2B~ 0) · B
~ I d3~x
∆E2 = (B 0 + B B (B
8π V 8π V 0 8π V
where B~ 0 is the constant field and B ~ I the induced one due to the presence of the sphere. We
now proceed as in the general case and use that B ~ I + 2B
~0 = ∇~ ∧ (A~ I + 2A ~ 0 ) to get, after
integration by parts,

1 Z ~ ~ I + 2A ~ 0) · B~ I d3~x
∆E2 = ∇ ∧ (A
8π ZV
1 ~ · ((A~ I + 2A ~ 0) ∧ B ~ I ) − (∇~ ∧B
~ I ) · (A
~ I + 2A
~ 0 )) d3~x
= ∇
8π ZV
−1 ~ I + 2A ~ 0) ∧ B ~ I ) · n̂) d2 S
= ((A
8π ZS
−1 ~0 ∧ B~ I ) · n̂) d2 S
= (A (10.5)
8π S
where we are using the normal outgoing from the sphere surface, that is incoming into V . In
the third equality we used that the current is zero in V and in the fourth that at the surface
~I + A
of the sphere the total potential, A ~ 0 vanishes. From the previous chapter computation
we know that

~ 0 (~x) = B0 (k̂ ∧ ~x)

A
2
and
3
~ I (~x) = −B0 a (3(n̂ · k̂)n̂ − k̂).
B
2r 3
So,
−1
Z
∆E2 = ~0 ∧ B
(A ~ I ) · n̂) d2 S
8π S
B02 a
Z
= ((k̂ ∧ n̂) ∧ (3(n̂ · k̂)n̂ − k̂)) · n̂) d2 S
32π S
B02 a
Z
= (k̂ ∧ n̂) · ((3(n̂ · k̂)n̂ − k̂) ∧ n̂) d2 S
32π S
B02 a
Z
= (k̂ ∧ n̂) · ((−k̂) ∧ n̂) d2 S
32π S
B02 a
Z
= (k̂ ∧ n̂)2 d2 S
32π S
B02 a3
Z
= sin(θ)3 d2 Ω
32π S 2
B02 a3 8π
=
32π 3
B02 a3
=
12
(10.6)
10.1. THE ENERGY OF CURRENT LINE DISTRIBUTIONS. 159

k̂

$region 2$

n̂
$region 1$

~k

Figure 10.1: Piece of surface with superficial current distribution.

a3 B02
Thus, ∆E = ∆E1 + ∆E2 = − 12
.

10.1 The Energy of Current Line Distributions.

In the case of line currents the formula simplifies considerably if we use the expression for the
potential vector in terms of the currents,

1 ~ x) · J(~
J(~
Z Z ~ y)
E = d3~x d3~y , (10.7)
2c2
V V |~x − ~y |
1 X i j d~li · d~lj
I I
= I I , (10.8)
2c2 i,j γi γj |~ xi − ~xj |

where γi and I i are respectively the path and current of the i-th current line.
This quantity is ill-defined, for each of the integrals with i = j diverges. As in the case
of point charges in electrostatics we redefine the energy by dropping all self-energies, and get
the interaction energy ,
X
EI = Eij , (10.9)
i<j

where,

I iI j d~li · d~lj
I I
Eij = 2 . (10.10)
c γi γj |~xi − ~xj |
160 CHAPTER 10. THE ENERGY OF THE MAGNETOSTATIC FIELD

J~1
J~2

J~3

Figure 10.2: A wire distribution

It is instructive to obtain another expression for these interaction energies. Since,

1
Z
Eij = J~i · A
~ j d3~x, (10.11)
c V

where J~i is the i-th current line distribution and A

~ j is the potential due to the j-th current,
we get,

Ii I ~
Eij = Aj · d~li (10.12)
c γi
Ii
Z
= ~ j · n̂i d2 Si
B (10.13)
c Si
Ii
= Φij (10.14)
c
where in the first step we have used the explicit formula for the i-th current distribution, in
the second Stokes theorem to transform a line integral into a surface integral, where Si is
any surface such that its boundary, ∂Si = γi , and in the last step we have defined Φij as the
magnetic flux due to the j-th current across the i-th current loop.

10.2 Inductances and Magnetic Fluxes.

The formula found in the preceding section for the interaction energy between line currents
shows that this energy depends on the product of the currents present times a purely geo-
metrical factor,
10.2. INDUCTANCES AND MAGNETIC FLUXES. 161

d~li · d~lj
I I
Lij = , (10.15)
γi γj |~xi − ~xj |
called the inductance of the i-th circuit with respect to the j-th circuit. It is clear that
Lij = Lji and that,
1 X
EI = 2
Lij I i I j . (10.16)
c i<j
Comparing with the expression for Eij in terms of the magnetic flux we see that the total flux
across the i-th loop due to the magnetic fields generated by the other loops is,
1X
Φi ≡ Lij I j . (10.17)
c j6=i
Notice that there is a close analogy between the triplets (I i , Φi , Lij ) and (Qi , V i , Cij ) of
electrostatics of perfect conductors. Can this analogy be extended to more realistic circuits?
The answer is yes, but one has to include the more general case the self inductances of circuits,
for now they are no longer divergent and so play a role in the energetics and fluxes. We treat
now superconducting circuits, for there the idealization allows for a strong argument. We
start with an example:

Example: The Self-Induction of a Superconducting Ring.

We consider a circular superconducting ring. We have,
1 Z ~ 2 3
E= |B| d ~x, (10.18)
8π V
~ ∧B
where V is the volume outside the ring. Since outside ∇ ~ = 0, and B
~ = −∇φ
~ m , with the
remark that φm has a jump somewhere, we get,

−1
Z
E = B~ · ∇φ
~ m d3~x (10.19)
8π V
1 1
Z I
= ~ ~ 3
φm ∇ · B d ~x − ~ · n̂ d2 S,
φm B (10.20)
8π V 8π ∂V
where we have used Gauss theorem and taken as the boundary for V the surface shown in
the figure, for φm is taken to be discontinuous on the inner ring plane. The contribution from
the part of ∂V at the surface of the ring vanishes, for there B ~ · n̂ = 0, but the part in the
plane gives,

1
Z
E = (φ+ − φ− ~ 2
m )B · n̂ d S (10.21)
8π ∂V m
φ+m − φm
− Z
~ · n̂ d2 S
= B (10.22)
8π ∂V
I
= Φ, (10.23)
2c
162 CHAPTER 10. THE ENERGY OF THE MAGNETOSTATIC FIELD

SR

D+
D−

Figure 10.3: Boundary for V

4π
where we have used that φ+ −
m − φm = c
I, and defined,
Z
Φ= ~ · n̂ d2 S,
B (10.24)
∂V
that is, the self-flux across the ring. But on the other hand, by definition,
LI
Φ= . (10.25)
c
so we see that
1
E= LI 2 , (10.26)
2c2
For a superconducting ring of radius a and section radius b one can compute L and find,
8a
L = 4πa[ln( ) − 2]. (10.27)
b
In the line current limit, (a → 0), L → ∞, and correspondingly Φ, but in such a way that
their ratio, which is Ic , stays constant.
Notice that,

Z
Φ = ~ · n̂ d2 S
B (10.28)
ZS
= ~ ∧ A)
(∇ ~ · n̂ d2 S (10.29)
IS
= ~ · d~l,
A (10.30)
γ

where S is any surface with boundary a loop γ at the surface of the ring, and we have used
in the last equality Stokes theorem. Thus, since L is a geometrical factor, a non-zero current
~ must have a non-zero component along
results then in a non-zero flux and so, in this case, A
the boundary of the ring.
Chapter 11

Magnetic Materials

In this chapter we treat macroscopic fields resulting from averages upon materials which
interact with magnetic fields producing changes which can be accounted by defining new
averaged fields and a constitutive relation among them.
As in the treatment of macroscopic fields on electrically susceptible materials we start
with the microscopic equations,

~ ∧B
∇ ~ = 4π J~ (11.1)
c
~ ~
∇ · B = 0, (11.2)

and take averages, since them commute with derivatives the only effect on the equations is to
change the sources, in this case the currents. We won’t go into the details of that calculation
for two reasons, first because the one that can be done it is very similar to the one performed
for dielectrics, second because in magnetic materials most of the effect is quantum in nature,
basically due to the presence of spins on electrons and orbitals which get only discrete values,
and all of them non-vanishing, thus, having an influence even in the absence of external fields.
On making the calculations one obtains to first order on the field strength,

~ ∧B
∇ ~ ϕ = 4π J~ϕ = 4π J~f ree + 4π ∇
~ ∧M
~ (11.3)
c c
~ ·B
∇ ~ϕ = 0 (11.4)

where

~ x) =
X
M(~ Ni (x) < m
~i> (11.5)
i

with Ni the average molecular number density, and m

~ i the average molecular magnetic mo-
ment, given by:

I I
I
~ · ~k =
m (~x ∧ d~l) · ~k = Areak , (11.6)
2c c

163
164 CHAPTER 11. MAGNETIC MATERIALS

where Areak = projected area of current loop on the plane normal to ~k

It is convenient to define a new field,

~ ≡B
H ~ − 4π M,
~ (11.7)
and write the equations as:

∇ ~ = 4π J~f ree
~ ∧H (11.8)
c
~ ~
∇·B = 0 (11.9)

where we have dropped the average indicator.

To solve the system we need a constitutive relation between H ~ and B.
~ In practice there
are two types of relations which are good at describing many interesting situations:
~ = µH
1. Linear relation: B ~

~ = B(
2. Hysteresis curve: B ~ H)
~ (see figure)

The first relation applies to normal materials and is only valid for small values of the
fields, the second has a saturation range, which is expected of many materials but also has
two more special features, one is that gives a nonzero value for B ~ even when there is no H,~
that is, it describes magnets. The other is that it is not a single valued function, the value of
B~ depends on the history of the material.

Figure 11.1: Hysteresis curve

11.0.1 Matching conditions on material boundaries

We shall assume now we are at an interface between two materials of different magnetic
properties. From similar arguments as the ones used for dielectrics, using Gauss theorem
in a pill box on the interface and Stokes theorem on a loop also at the interface, using the
equations above (11.8) we get,
11.1. CONSTANT MAGNETIZATION 165

~2 − B
(B ~ 1 ) · n̂|S = 0 (11.10)
~2−H
(H ~ 1 ) ∧ n̂|S = 4π ~k l (11.11)
c

where ~k l is a possible free current on the interface. We see that the normal component of B ~
~
is continuous while the tangential components of H have a jump proportional to the possible
free current distributions there.
~ = µH,
If the relation is linear, B ~ and there is no superficial free currents, then we have,

~2 − B
(B ~ 1 ) · n̂|S = 0 (11.12)
~ 2 − µ2 B
(B ~ 1 ) ∧ n̂|S = 0 (11.13)
µ1
or

~2− µ1 ~
(H H 1 ) · n̂|S = 0 (11.14)
µ2
~2−H
(H ~ 1 ) ∧ n̂|S = 0 (11.15)
~ field.
if one prefers to work with the H

11.1 Constant Magnetization

~ and H,
Within all possible relations between B, ~ or between B
~ and M
~ , we have one in which
~ ~
M does not depend upon B, that is, when the magnetization is constant. We shall look now
at problems of this sort.
In this case we have, if no free current is present,

~ ∧H
∇ ~ = 0, (11.16)
~ ·B
∇ ~ = ∇
~ · (H
~ + 4π M)
~ = 0, (11.17)
~ = −∇φ
and –for simply connected bodies– we can use a magnetic potential φm such that H ~ m.
Thus we obtain,

~ ·M
∆φm = 4π ∇ ~ ≡ −4πρM , (11.18)
~ ·M
with ρM ≡ −∇ ~ . Thus, if M
~ is smooth,

~ ·M
∇ ~ (~y )
~y
Z
φm (~x) = − 3
d3~y , (11.19)
lR |~x − ~y |
which, integrating by part can be transformed into,
166 CHAPTER 11. MAGNETIC MATERIALS

1
Z
φm (~x) = ~ (~y ) · ∇
M ~ ( )d3~y (11.20)
lR 3 ~y
|~x − ~y |
~ ( 1 )d3~y
Z
= − 3M ~ (~y ) · ∇ (11.21)
lR ~x |~x − ~y |
Z
M~ (~y )
= −∇~ ·( d3~y ). (11.22)
3 |~x − ~
y |
lR
In particular, far away from the source,

1
Z
~
φm (~x) ≈ −∇( ) ~ y )d3~y ,
M(~ (11.23)
|~x| lR3
m
~ · ~x
≈ , (11.24)
r3
where m
R
~ ≡ lR3 M ~ (~y )d3~y is the total magnetic moment of the magnetic source.
If M~ is discontinuous, as is the case if the material ends abruptly at a surface S, some of
the formulas above are incorrect, and we have to proceed with care, for in this case ∇ ~ ·M~ is
only a distribution. In this case we interpret equation (11.18) as a distributional equation in
the following sense: If φm and M ~ were smooth, then equation (11.18) would be equivalent to
the following infinite set of relations:

Z
~ · M(~
u(~x)[∆φm (~x) − 4π ∇ ~ x)]d3~x = 0, for all smooth u(~x) of compact support. (11.25)
lR3
But this set of equations in turn are equivalent to:

Z
~ x)·M(~
[∆u(~x)φm (~x)+4π ∇u(~ ~ x)]d3~x = 0, for all smooth u(~x) of compact support, (11.26)
3
lR
where we have just integrated by parts. But these expressions make sense even when M ~ , and
~
φm are discontinuous, so we interpret the above equation for discontinuous M ’s as these set
of relations. That is, as distributions.
In the case that we have a continuous medium which ends abruptly in a 2-surface S, then
~
M is only piece-wise continuous and it can be seen that in that case, φm is continuous but
not differentiable at S. So we should use the equation in the distributional sense.
Recalling that the equation
1
∆~x ( ) = −4πδ(~x − ~y ), (11.27)
|~x − ~y |
should in fact be interpreted as,

1
Z
3
(∆u(~x)) d3~x = −4πu(~y ) for all smooth u(~x) of compact support. (11.28)
lR |~x − ~y |
11.1. CONSTANT MAGNETIZATION 167

From (11.26), and using (11.28), we get, for all smooth u(~x) of compact support,

Z Z
3
∆u(~x)φm (~x) d ~x = −4π ~ u(~y )) · M(~
(∇ ~ y ) d3~y
lR3 lR3 ~y
1
Z Z
= (∆∇ ~ (~y )(
~ u(~x)) · M ) d3~xd3~y
3
lR lR 3 ~x |~x − ~y |
~ (~y )( 1 ) d3~xd3~y
Z Z
= ~ (∆u(~x)) · M
∇
3
lR lR 3 ~
x |~x − ~y |
~ ( 1 ) d3~xd3~y
Z Z
= − 3 ~ y) · ∇
∆u(~x)M(~
lR lR 3 ~x |~x − ~y |
~ ( 1 ) d3~xd3~y
Z Z
= ~ y) · ∇
∆u(~x)M(~
3
lR lR 3 ~y |~x − ~y |
~ ( 1 ) d3 ~y ] d3~x
Z Z
= ∆u(~x)[ 3 M ~ (~y ) · ∇ (11.29)
lR 3
lR ~y |~x − ~y |

which is equation equivalent to (11.20). Thus, we see that the correct, or more general,
expression for the potential is,
1
Z
φm (~x) = − ~ (~y ) · ∇
M ~ d3~y (11.30)
lR 3 ~x |~x − ~y |

which is a valid expression, even when M ~ is only piece-wise continuous. Using Gauss theorem
in both sides of a discontinuity surface S we get,

~
~ (~y ) 1 d3~y + n̂ · [M]
Z Z
φm (~x) = ~ ·M
∇ d2 S, (11.31)
3
lR −S ~y |~x − ~y | S |~x − ~y |
where [M ~]≡M ~ in − M
~ out measures the jump of the magnetization across the discontinuity
and n̂ is the outward normal to S (that is from in to out).
Had we chosen to solve the above problem using the vector potential, A,~ such that B ~ =
~ ∧ A,
∇ ~ then,

~ ∧H
∇ ~ =∇
~ ∧ (B
~ − 4π M)
~ = 0, (11.32)
and so,

~=− 4π ~
∆A JM, (11.33)
c
with J~M = c∇
~ ∧M
~ . Thus, if M
~ is smooth,

~ ∧M
∇ ~ (~y )
~y
Z
~ x) =
A(~ d3~y . (11.34)
3 |~x − ~y |
lR
If the magnetization is not smooth, then an argument parallel to the one given above
shows that the correct expression for the vector potential is,
168 CHAPTER 11. MAGNETIC MATERIALS

~ x) =
Z
~ (~y ) ∧ ∇
~ ( 1
A(~ 3
M ~x )d3~y , (11.35)
lR |~x − ~y |
and, in the presence of a surface discontinuity,
Z ~ ∧M
∇ ~ (~y ) Z ~ ] ∧ n̂
[M
~ x) =
A(~ 3
d ~y + d2 S. (11.36)
lR3 −S |~x − ~y | S |~
x − ~y |

Example: Uniformly Magnetized Sphere Let be a uniformly magnetized sphere of radius

a, and let choose the z axis in the direction of that magnetization. In this case the contribution
to the magnetic potential (or vector potential) just comes from the surface S give by r = a.
Thus, we are merely solving Laplace equation in and outside the sphere subject to some
boundary conditions, namely,

~ out − B
(B ~ in )|r=a · n̂ = (H
~ out − H
~ in )|r=a · n̂ + 4π(M
~ out − M
~ in ) · n̂ = 0, (11.37)
that is,

~ out − φin )|r=a = 4π M

−n̂ · ∇(φ ~ · n̂, (11.38)
m m

and the continuity of the tangential component,

(φout in
m − φm )|r=a = 0. (11.39)
~ = k̂M), and only
Since the solution can only depend on k̂, the magnetization direction, (M
linearly, we must have,

φin in
m = A r k̂ · n̂, (11.40)
k̂ · n̂
φout
m = Aout . (11.41)
r2
Therefore, continuity implies,

inAout
A = 3 , (11.42)
a
and the jump on the normal derivatives,

Aout in
A + 2 3 = 4πM. (11.43)
a
Therefore Ain = 4π
3
M, and Aout = 4π
3
a3 M.
~ = H,
Outside the sphere, B ~ is a magnetic dipole with magnetic moment given by, m
~ =
4π 3 ~
3
a M.
4π 3 ~
~ =
m a M. (11.44)
3
Inside the sphere,
11.2. PERMANENT MAGNETS 169

~ = − 4π M
H ~, (11.45)
3
~ = H
B ~ + 4π M~ = 8π M.
~ (11.46)
3

~ to solve this problem by direct

Exercise: Use the previously found formulae for φm , and A
integration.

11.2 Permanent Magnets

If we place a sphere in a constant magnetic field we know that the material out of which it is
made would react to it creating some magnetization vector field, M ~ , which, like the external
field is constant inside the sphere. 1 Thus, from the calculation done in the previous section
we have,

8π
Bin = B0 + M, (11.47)
3
4π
Hin = B0 − M, (11.48)
3
where we have just used the norms of the vectors for all of them point along the same direction.
To solve this system we need to specify some relation between the three unknowns. In practice
the following to cases are of interest:
Case 1: If B ~ = µH,~ then, since Bin + 2Hin = 3B0 , we have(µ + 2)Hin = 3B0 , and so,

3B0
Hin = , (11.49)
µ+2
3µB0
Bin = , (11.50)
µ+2
3(µ − 1)B0
M = . (11.51)
4π(µ + 2)

We see that for linear relations between the fields both fields, H~ in , and B
~ in vanish when the
external field is not present.
Case 2: If we have a nonlinear relation like the one in the figure bellow –called hysteresis
curve–, Bin = Bin (Hin ), then the solution would be the intersection of that graph and the
above linear relation, Bin + 2Hin = 3B0 .
1
This is a property of the sphere and does not hold for bodies of other shapes.
170 CHAPTER 11. MAGNETIC MATERIALS

B B
B0 = 00

B = B(H) B = µH
H
B = B0 − 2H

Figure 11.2: Intersection of hysteresis curve and linear relation

We see that there could be one solution, or two solutions, depending on the value of B0 .
In particular for B0 = 0 there are two nontrivial solutions, one the reverse of the other,
corresponding to permanent magnets. In particular, we see that if we start at B0 = 0 in a
unmagnetized material and slowly increase its value we shall be moving along the point which
crosses the hysteresis curve Bin (Hin ) and the linear relation Bin + 2Hin = 3B0 . We see that
when we return to B0 = 0 the material would remain magnetized.

11.3 Generalized Forces on Charged Conductors and

Circuits
In this section we want to find expressions for the forces needed to keep a set of charged
conductors or a set of circuits in place, that is the force needed to compensate the electric or
magnetic force acting between different components to these sets.
To obtain such expressions we imagine we make an infinitesimal displacement, ∆~x of one
of the elements. This would cause a change on the total energy of the configuration which
would be equal to the work done when making the displacement, thus,

F~ · ∆~x = −∆ET , (11.52)

or

∂ET
F~ = − , (11.53)
∂~x
where ET is the total energy of the system, and not just the electromagnetic energy. Let us
see an example.

Example: Forces between the parallel plates of a capacitor.

11.3. GENERALIZED FORCES ON CHARGED CONDUCTORS AND CIRCUITS 171

Let us assume we have a pair of infinite parallel plates separated a distance L apart, with
potential difference V . We want to compute the force by unit area between them, that is, the
pressure on one of the plates
To do this we assume the displacement is done keeping the plates isolated, that is, their
charges by unit area σ, is constant, and compute the change in its total energy, which in this
case is just the change in its electromagnetic energy

ET E 1 ZL~ ~
= = E · E dx. (11.54)
Area Area 8π 0
~ = k̂E, E =
Now, E V
= 4πσ, therefore,
L

E
= 2πσ 2 L, (11.55)
Area
and

E 1 V2
( )∆L = 2πσ 2 ∆L = ∆L. (11.56)
Area 8π L2
Thus, the pressure, namely the force per unit area is given by

−1 ~ ~ V2
P = E ·E = − 2. (11.57)
8π L

Notice that to compute the change in the total energy we first have to decide how to do
the displacement. We choose to make it keeping the charges constant. Since in that case the
only energy change is in the electrostatic energy, we just did,

∂ET ∂E
= ( )Q . (11.58)
∂~x ∂~x
We could have done the displacement in another way. We could have connected a battery
with a potential difference V to the plates and then make the displacement. In that case
the change in the total energy would have to include the change in the energy stored in the
battery, because in this case there would certainly have been a current across the circuit. So
even if we could compute

∂E
( )V , (11.59)
∂~x
this would not had help us much, for it is different from ET . 2
∂~
x
Nevertheless, notice that, since
2
Notice that since we are computing the force at the given configuration, the result is independent on the
way we choose to make the displacement. It is just a virtual displacement.
172 CHAPTER 11. MAGNETIC MATERIALS

1
E = Qi V i (11.60)
2
1
= Cij V i V j (11.61)
2
1 −1 ij
= (C ) Qi Qj , (11.62)
2
where (C −1 )ij Cjk = δ i j , and summation over repeated indices is assumed. Thus,

∂E 1 ∂Cij i j
( )V = ( )V V (11.63)
∂~x 2 ∂~x
−1 ∂(C −1 )kl
= Cik Clj V i V j (11.64)
2 ∂~x
−1 ∂(C −1 )kl
= Qk Ql (11.65)
2 ∂~x
∂E
= −( )Q , (11.66)
∂~x
where we have used that 0 = ∆δ i k = ∆(C −1 )ij Cjk + (C −1 )ij ∆Cjk , and Qk = Ckj V j .
Thus, we see that if in some circumstances it is simpler to compute ( E )V than to compute
∂~
x
( E )Q , we can go ahead and compute it, and then use the relation above to compute ∂ET using
∂~
x ∂~
x
( ∂E )Q . This is usually formalized by introducing the free energy, F = E − Qi V i = −1 C i j
ij V V .
∂~
x 2
For then ∂F | = E∂xT .
∂x V
We consider now the case of circuits. Here we have a similar situation with currents and
vector potentials (or fluxes) instead of charges and potentials, and inductances instead of
capacities. Thus, the following relation must also hold,
∂E ∂E
(
) ~ = −( ) ~ . (11.67)
∂~x A ∂~x J
But in contrast with electrostatics here the adiabatic displacement, that is the displacement
where no external energy sources are needed is at constant vector potential.
There are basically two ways to reach this conclusion: One is to do the calculation at
constant current, keeping truck of all external electromotive forces needed for this to happens.
If this is done one finds,
∂ET ∂E
) ~ = −( ) ~,
( (11.68)
∂~x J ∂~x J
that is the external energy sources must do twice –and with the opposite sign– the work of
the circuit. Thus,
∂ET ∂ET ∂E ∂E
F =− = −( ) ~ = ( ) ~ = −( ) ~ . (11.69)
∂~x ∂~x J ∂~x J ∂~x A
The other way to reach this conclusion is to replace the original circuit by a superconduct-
ing one. In order for this not to change significantly the configuration we imagine replacing
11.4. THE ENERGY OF MAGNETIC MATERIALS 173

the original circuit for a bunch of very thin superconducting wires, thus modeling more and
more precisely, in the limit when the section of the wires goes to zero, the original current
distribution. But for superconducting circuits the magnetic flux is constant under displace-
ments, and so is the vector potential at the surface. Therefore, this is the natural quantity to
keep constant in the adiabatic variation.

11.4 The Energy of Magnetic Materials

We have seen that the magnetic energy is given by,
1 Z ~ · J~ d3~x.
E= A (11.70)
2c lR3
The expression con be thought of a function of A(~ ~ or as a function of
~ x), with J~ = −c ∆A,
4π
~ x), with A(~
J(~ ~ x) a solution of ∆A
~ = −4π J~. Usually one is not interested in this expression, but
c
~
rather in its derivative –keeping A fixed at some boundary– with respect to some parameter
present in the problem, which we call s. That is one is interested in,
∂E
) ~,
( (11.71)
∂s A
~ fixed, but rather some boundary value for it.
where we stress that we are not holding A
Thinking for the moment that E is a function of both arguments, and using,
δE ~ 1
Z
(δ A) = J~ · δ A
~ d3~x, (11.72)
~
δA c lR3
we can define the free energy,

1Z
F ≡ E− J~ · A
~ d3~x (11.73)
c lR3
= −E, (11.74)
and then,

∂F δE ∂ A ~ ∂E 1 ~
~ · ∂ A d3~x
Z
( ) = ( ) ( ) + ( ) − J (11.75)
∂s J~ ~
s
~
δA ∂s ∂s A c lR3 ∂s
∂E
= ( ) ~. (11.76)
∂s A
Thus, for calculations it is equivalent to know F as a function of J~, than to know E as a
~ But,
function of A.

~ ~
δF ~ δE δ A ~ −1 ~ · δ A · δ J~ d3~x − 1
Z Z
(δ J) = ( · δ J) J ~ · δ J~ d3~x
A (11.77)
~
δJ ~ ~
δ A Zδ J c lR 3
δJ~ c lR 3

1 ~ · δ J~ d ~x.
3
= − A (11.78)
c lR3
174 CHAPTER 11. MAGNETIC MATERIALS

~ adding infinitesimal currents,

We can use now this expression to integrate and obtain F (J)
even in the case that the relation between J~, and A
~ is not linear, as often occurs in the
presence of magnetic materials.
~ and J~ as averaged quantities and using ∇
For this we consider now A, ~ ∧H ~ = −4π J,
~ we
c
get

δF ~ 1
Z
(δ J) = − ~ ·∇
A ~ ∧ δH
~ d3~x. (11.79)
δ J~ 4π lR3
1
Z
= − ~ · (∇
{δ H ~ ∧ A)
~ +∇
~ · (δ H
~ ∧ A)}
~ d3~x (11.80)
4π 3
lR
1
Z
= − ~ · δH
B ~ d3~x. (11.81)
4π 3
lR
~ = µH,
When B ~ then

1
Z
F =− ~ ·H
B ~ d3~x. (11.82)
8π lR3
~ = B(
In the case that B ~ H),
~

1 Z I ~ ~ ~ d3~x,
∆F = − H(B) · δ B (11.83)
4π lR3
where the integral is along a hysteresis cycle, that is, it is just the area inside the hysteresis
curve.

Example: The difference between the free energies of a set of circuits in vacuum
and the same set in the presence of a magnetizable medium.

−1
Z
∆δF = [B ~ − ~h · δ~h] d3~x,
~ · δH (11.84)
4π lR3
where ~h is the field generated by the circuit, given by certain J(~ ~ x), as if it were in vacuum,
i.e. µ = 1.
The above expression can be rewritten as,
−1 Z ~ − ~h) · δ~h + B ~ − δ~h) + (B
~ · (δ H ~ · δ~h] d3~x.
~ − H)
∆δF = [(H (11.85)
4π lR3
The first term is

Z Z
~ − ~h) · δ~h d3~x =
(H ~ − ~h) · (∇
(H ~ ∧ δ A)
~ d3~x (11.86)
3 3
lR ZlR
= ~ · (δ A
[∇ ~ − ~h)) + δ A
~ ∧ (H ~·∇ ~ − ~h)] d3~x (11.87)
~ ∧ (H
lR3
= 0, (11.88)
11.4. THE ENERGY OF MAGNETIC MATERIALS 175

~ H
for the first term vanishes upon application of Gauss theorem and the second because ∇∧ ~ =
~ ∧ ~h = 4π J.
∇ ~
c
The second term also vanishes, as can be seen using similar arguments upon substitution
~
of B by ∇ ~ ∧ A.~ Thus,

1
Z
∆δF = ~ − H)
(B ~ · δ~h d3~x (11.89)
4πZ lR 3

= − 3M ~ · δ~h d3~x (11.90)

lR
If the magnetization grows linearly with the applied field, then

−1
Z
∆F = M~ · ~h d3~x (11.91)
2 lR 3
−1
≈ ~ · ~h,
m (11.92)
2
where
Z
m
~ ≡ ~ d3~x,
M (11.93)
lR3
and we have assumed that near the medium ~h ≈ ~h0 , a constant field.
If the magnetization does not depend on the external field, then

~ · ~h.
∆F ≈ −m (11.94)
176 CHAPTER 11. MAGNETIC MATERIALS
Chapter 12

Examination Questions 1

Problem 1 State the Cauchy problem (1.1) for electromagnetism and explain it.

Problem 2 Prove uniqueness of solutions in the Cauchy problem for electromagnetism (1.1).

Problem 3 Given any smooth function g(~x) : R3 → R, then

φ(t, ~x) := t Mt (g(~x))

with
1
Z
Mt (f )(~x) := f (~x + t~n) dΩ
4π S2
satisfies the wave equation,

∂t2 φ(t, ~x) = ∆φ(t, ~x).

Prove that

a)
∂
φ(t, ~x) = tMt (φ1 (~x)) + (tMt (φ0 (~x))),
∂t
satisfies also the wave equation.
∂φ
b) Furthermore, it has as initial conditions: φ1 (~x) = ∂t
(t, ~x) |t=0 and φ0 (~x) = φ(0, ~x).

Problem 4 Assume E ~ = E(t,

~ ~x) is a solution to the vacuum Maxwell’s equations. Find an
~ = B(t,
expression for the corresponding B ~ ~x) if its value at t = 0, B
~ 0 (~x), is given.

Problem 5 Assume now B ~ = B(t,

~ ~x) is a solution to the vacuum Maxwell’s equation. Find
~ = E(t,
an expression for the corresponding E ~ ~x) if its value at t = 0, E
~ 0 (~x), and J~ = J(t,
~ ~x)
are given.

Problem 6 Show that if the constraint equations are satisfied at t = 0 then they are satisfied
for all times provided the fields satisfy the evolution equations.

177
178 CHAPTER 12. EXAMINATION QUESTIONS 1

Problem 7 Given two smooth vector fields in lR3 , ~l(~x) y m(~x) produce everywhere smooth
~ and B.
solutions to the vacuum constraint equations for E ~

Problem 8 Find the expression for the energy of the electromagnetic field starting from the
expression of the power in terms of the work done by the Lorentz force.

Problem 9 Find an example of a nonzero Poynting vector in a situation where there is no

radiation.

Problem 10 You are left alone in a region without any house or other reference, except a
power line. To reach a city you decide to follow the line in the assumption that it is feeding
power to it. How you determine the direction you should go without cutting the power line?

Problem 11 Show that Maxwell’s equations are invariant under time and space translations.

Problem 12 Show that Maxwell’s equations are not invariant under Galilean transforma-
tions.

Problem 13 Prove, using the concept of distributions that,

1
∆( ) = −4πδ(~x − ~x′ )
|~x − ~x′ |
d
Problem 14 Show, using the concept of distributions that, dx
Θ(x) = δ(x) where Θ(x) is the
step function, Θ(x) = 0 x < 0, Θ(x) = 1 x ≥ 0.

Problem 15 Prove Theorem 4.1

Problem 16 Deduce the matching conditions at the surface of a conductor.

Problem 17 Prove Theorem 4.2

Problem 18 Deduce the existence of the Capacities matrix, eqn. 4.13. Given three conduc-
tors, how would you measure the component C23 of the capacity matrix? In the experiment
you can set up any given potential distribution among the conductors (only once) and can
measure the induced charge at only one of the conductors of your choice.

Problem 19 Prove that the Capacity matrix is symmetric.

Problem 20 Compute P3 and P4 using the recursion relation found and the normalization
condition

Problem 21 Deduce the expression for the multipole constants, eqn. 5.11

Problem 22 Find two different charge distributions with no symmetry at all but giving the
same external field (which can have symmetries). Hint: do not construct them explicitly but
just start from their potentials and work backwards.
 179

Problem 23 Deduce the expression for the Dirichlet Green’s function corresponding to two
concentric conducting spheres.

Problem 24 Deduce the symmetry of the Dirichlet Green function, equation 7.1

Problem 25 It is required to solve the Poisson equation inside a volume V where there is
defined a smooth source function ρ(~x), and a Neumann boundary condition ∂n φ(~x)|∂V = g.
What condition must g satisfy in order for a solution to exist?

Problem 26 Deduce the matching conditions at a boundary of a dielectric material.

Problem 27 Deduce the electrostatic energy of a dielectric material.

Problem 28 Show theorem 9.1

Problem 29 Deduce the boundary conditions at the surface of a superconductor for the mag-
netic field.

Problem 30 Find the vector potential corresponding to a constant magnetic field, show that
the dependence on a given origin is pure gauge.

Problem 31 Deduce the boundary conditions at the surface of a superconductor for the scalar
magnetic potential. Explain the multi-valuate nature of it, and give the expression for the value
of the needed discontinuity when a non simply connected superconductor is present.

Problem 32 Deduce that ∇ · J~ = 0 (in the sense of distributions) for a wire current distri-
bution.

Problem 33 Prove lemma 9.1

Problem 34 Prove that the definition of magnetic flux using the vector potential eqn. (9.23)
is gauge independent.

Problem 35 Show that the flux across a surface bounded by a superconductor is constant in
time.

Problem 36 Why there exists a inductance matrix?

Problem 37 Deduce that the first term in the multipolar expansion of the magnetic potential
vanishes, and find the following two terms.

Problem 38 Find the expression for the energy of the magnetostatic field in terms of the
inductance matrix.

Problem 39 Deduce the matching conditions for a surface in a magnetic material.

Problem 40 Deduce, using distributions, the true formula for the magnetic potential when
the magnetization is not differentiable.
180 CHAPTER 12. EXAMINATION QUESTIONS 1
 Part II

Second Course

181
Chapter 13

The Symmetries of Maxwell’s

Equations Continued

13.1 Introduction
In chapter 3 we saw Maxwell’s equations have a number of symmetries, that is, transformations
that take one solution into another. It was clear that most of them arise because of the
underlying symmetries of space and time. Indeed, time translation, or the homogeneity of
time implied that if (E(~~ x, t), B(~
~ x, t)) was a solution, then (E
~ T (~x, t), B
~ T (~x, t)) = (E(~
~ x, t +
~
T ), B(~x, t + T )) was also a solution (with the corresponding time translated sources, if any
present). The same for space translations or rotations.
But we also saw that the assumed Galilean symmetry of space-time was not a symmetry
of these equations. Galilean symmetry can be stated in different ways, the ampler is the
statement that by means of local experiments one can not determine the state motion of our
system with respect to others objects outside it, namely there is no notion of absolute speed.
This conception, very obvious today, was sharpened by Galileo as a justification why we do not
feel earth motion in its orbit around the sun. The previous conception of space and time was
due to Aristotle, which though there was a natural motion state for earthly bodies, namely
to be at rest, meaning with respect to earth. 1 Since then this conception has been central
for our understanding of physics. But here we are presented with equations which describe
with incredible precision the electromagnetic phenomena and yet have on it a parameter with
dimension of velocity and its solutions propagate which such a speed. So the natural way
of interpreting this was to throw away Galileo’s conception and say that after all there was
an ether, namely, something with respect to which these solutions were moving. From a
mechanistic conception, still strong at that time, electromagnetic waves where mechanical
waves of the ether, that is local modifications of the state of such a material which would
propagate along it with the "sound speed" characteristic of it. A very stiff material indeed! 2
1
Galileo’s reasoning against this beliefs, thinking on situations where friction forces where smaller and
smaller imagining a limit where they were absent and so that in this idealized situation the bodies would
continue to move forever, marks the beginning of one of the pillars
q of the scientific thinking.
2 Y
Recall that the sound speed of a material is given by Cs = ρ where Y is Young’s module and measures
the stiffness of a material. For steel Cs ≈ 6 106 m
s .

183
184 CHAPTER 13. THE SYMMETRIES OF MAXWELL’S EQUATIONS CONTINUED

Here, instead, we shall investigate whether there is a symmetry after all reflecting Galileo’s
principle of relativity of motion more general than the one already tried in chapter 3.
To that end, and to facilitate the calculations, we shall consider the symmetries of the
wave equation,

∂2 ∂2 ∂2
✷φ := ( ∂t2 − c2∆)φ := ( ∂t2 − c2δij ∂xi ∂xj )φ = 0
where the last expression is valid in Cartesian coordinates. We know that in these coordinates
each component E ~ or B~ satisfies the wave equation, so we know that a symmetry of Maxwell’s
equation necessarily must be a symmetry of the wave equation. On the other hand in this
simpler equation we already have the fact that solutions propagate at a given speed.
So we want to find a transformation among solutions which represent a given solution
but as seen in constant motion from the original coordinate system. So we want to look for
a coordinate system transformation which depend only on a constant vector ~v, namely the
relative speed between the solutions.
The most general linear transformation we can write depending on just a vector is:

˜v ∧ ~x
~x′ = a~x + b~v t + d(~v · ~x)~v + d~ (13.1)
t′ = et + f (~v · ~x) (13.2)

Redefining a the last term in the first transformation can be seen to be a rotation, and we
have already seen that rotations are symmetries by themselves, so we can safely remove that
term. We also know that changing the value of a amounts to a redefinition of the space scale,
and that would give rise to a trivial symmetry (assuming c or t are also scaled, and so also
the equation sources, if there were). So we can set a = 1 without lost of generality. Choosing
to keep the value of c as fixed, we see we can not re-scale t. Furthermore, if we want this
transformation to really represent a motion with speed ~v , then it should be the case that a
trajectory of the form ~x(t) = ~xo −~v t should transform into a stationary trajectory, namely we
should have dtd ~x′ (t) = 0. Plugging these two conditions on the above transformation (where
we have already set a = 1, d˜ = 0) we find,

d ′ d d
~x = ~x + b~v + d(~v · ~x)~v (13.3)
dt dt dt
2
= −(1 + dv )~v + b~v (13.4)

so the stationarity condition implies that,

b = 1 + dv 2 . (13.5)
So now we assume we have a solution to the wave equation, φ(~x, t) and want to see
whether φ~v (~x, t) := φ(~x′ (~x, t), t′ (~x, t)) is also a solution. This will impose conditions on the
remaining free coefficients and determine them completely. Notice that these coefficients √ can
only depend on the two scalar parameters of the problem, namely c, and v := ~v · ~v .
13.1. INTRODUCTION 185

Using the chain rule we have,

∂ ∂t′ ∂ ∂x′i ∂
φ~v = φ + φ
∂t ∂t ∂t′ ∂t ∂x′i
∂ ∂t′ ∂ ∂x′l ∂
φ = φ + φ
∂xi ~v ∂xi ∂t′ ∂xi ∂x′l
and for the second derivatives, using the fact that the Jacobian of the transformation is
constant,

∂2 ∂t′ ∂t′ ∂ 2 ∂t′ ∂x′i ∂ 2 ∂x′i ∂x′j ∂2

φ = φ+2 φ+ φ
∂t2 ~v ∂t ∂t ∂t′2 ∂t ∂t ∂t′ ∂x′i ∂t ∂t ∂x′i ∂x′j
∂2 ∂t′ ∂t′ ∂ 2 ∂x′l ∂t′ ∂2 ∂x′l ∂t′ ∂2 ∂x′l ∂x′m ∂2
φ = φ + φ + φ + φ
∂xi ∂xj ~v ∂xi ∂xj ∂t′2 ∂xi ∂xj ∂x′l ∂t′ ∂xj ∂xi ∂x′l ∂t′ ∂xi ∂xj ∂x′l ∂x′m
Using now that

∂x′l
= δ l i + dv l vi
∂xi
∂x′i
= bv i
∂t
∂t′
= e
∂t
∂t′
= f vj
∂xj
where vi := δij v j , being δij the inverse of δ ij , δij δ jk = δi k , ~x · ~v = δij xi v j , etc.
Therefore, we have,

∂2 2 2 2 2 ∂2
φ
✷ ~v = ′2
φ[e − f c v ] + ′ ′i
φ[2ebv i − 2f c2 (1 + dv 2 )v i]
∂t ∂t ∂x
∂2
+ φ[−c2 δ lm − v l v m (2adc2 + d2 v 2 c2 − b2 )] (13.6)
∂x′l ∂x′m
2 2
Using that φ satisfies the wave equation and so, −c2 δ lm ∂x′l∂∂x′m φ = − ∂t∂ ′2 φ we get,

∂2 2 2 2 2 ∂2
φ
✷ ~v = φ[e − f c v − 1] + φ[2eb − 2f c2 (1 + dv 2)]v i
∂t′2 ∂t′ ∂x′i
∂2
+ φ[v l v m (−(2 + dv 2 )dc2 + b2 )] (13.7)
∂x′l ∂x′m
So, if we want that φ~v satisfies the wave equation we need that every term on the right-hand
side must vanish. At any given point in space and any time we can find solutions to the
wave equation for which any of the second partial derivatives are zero except for one of them.
Thus, every term in brackets must vanish.
186 CHAPTER 13. THE SYMMETRIES OF MAXWELL’S EQUATIONS CONTINUED

Exercise: Check this! Hint: choose Cartesian coordinates (x, y, z) such that ~v = (1, 0, 0)
and try with the following solutions to the wave equation: tx, c2 t2 − y 2 , c2 t2 − x2 .

Using that b = (1 + dv 2) in the third term we get a second order polynomial in d. From
which we get that,
s
±γ − 1 v2
d= γ := 1−
v2 c2
and correspondingly,

b = ±γ.
We take the plus sign, for this is the value that gives the Galilean transformation in the limit
of small ~v . Since b ≥ 1 we can factorize it in the second term and get,
e
f= .
c2
Substitution of this relation on the first term gives,

1
e2 = 2 or e = ±γ.
1 − vc2
We shall keep the plus sign, the other corresponds to a time inversion.
So, we have found all coefficients and the final transformation is:

~x′ = ~x + γ~vt + d(~v · ~x)~v = γ(n̂ · ~x)n̂ + γ~v t + (~x − (n̂ · ~x)n̂) (13.8)
(~v · ~x)
t′ = γ(t + ) (13.9)
c2

where n̂ := ~v .
|~
v|
Finally, then we have found a symmetry transformation for the wave equation that re-
sembles a Galilean transformation as close as possible. Together with the transformations
already found en chapter 3 they are all symmetry transformations of the wave equation (of
course one can compose many of them one after the other and obtain more transformations,
but they are all generated, in this way, from the ones already found).
Let us analyze it in detail. Choose coordinate axis so that the velocity is in the ~e1 direction,
~v = (−v, 0, 0). Then the symmetry transformation becomes,

x′ = γ(x − vt)
y′ = y
z′ = z
vx
t′ = γ(t − 2 )
c
13.1. INTRODUCTION 187

ct

ct′

x′

Figure 13.1: The Lorentz map.

ct ct′

x′

Figure 13.2: How a square is transformed.

188 CHAPTER 13. THE SYMMETRIES OF MAXWELL’S EQUATIONS CONTINUED

v
In the following figure we see what such a transformation does for a value of c
= 54 , that
is, γ = √ 1 16 = 35
1− 25

Exercise: The above plot uses the value c = 1, or alternatively it is plotted using as variable
ct instead of t. Make a similar plot but now with take c = 0.1.

Exercise: Consider the solution to the wave equation given by f (x − ct) = {1 ∀ x − ct ∈

[0, 1], 0 x ∈
/ [0, 1]}

We can see from the example that points along the line x = ct are transformed among
each other. Let us see this in more detail.

vx 2
−c2 (t′ )2 + (x′ )2 + (y ′ )2 + (dz ′ )2 = −c2 (γ(t + 2
)) + (γ(x + vt))2 + y 2 + z 2
c
= t2 (γ 2 (−c2 + v 2 )) + x2 (γ 2 (−v 2 c−2 + 1) + y 2 + z 2
= −c2 t2 + x2 + y 2 + z 2 (13.10)

Thus we see that not only the lines x = ct are transformed among each other, but also the
lines x = −ct and all the hyperbolas of the form −c2 t2 + x2 + y 2 + z 2 = const..

Exercise: Draw in a diagram these hyperbolas for positive and negative constant values.
Chapter 14

Special Relativity

14.1 Introduction
In the previous chapter we found a new symmetry for the wave equation, which can be
extended also to Maxwell’s equations. This symmetry is a transformation on space and time
which leaves invariant the lines which move at the speed of light. To understand what all this
means we have to change our concepts of space and time.
The building block of our description are happenings in phenomena, they will be called
"events", that is an idealization of something happening in a limit where the duration of it
goes to zero, as well as the size of the space where this is taken place. We imagine we have
a big bag with all events which have occurred in history, not only the important events, but
also all events where nothing important happened, or just where nothing happened at all! We
can also imagine having all events that will happens in the future. That will be the sand-box
where we want to describe phenomena, we shall refer to it as space-time. If we think on
our usual description of events, we see that we can describe them using four numbers, for
instance if I have an appointment at the doctor, I need the time, the street name (which I
can code with some number), the building number and the floor. In general, besides the time,
one needs to define a position relative to some other objects, and for that it suffices with
three other numbers. This way we can mark events with four numbers, and so we have some
notion that they form a continuum in the sense that given two events with their respective
numbers we can find some other whose numbers are all equidistant to those of both events.
That way we can describe, say the events of a given molecule, by a continuous line, centered
at the events occupied by its center of mass. That assumption might ultimately be wrong as
we probe further and further smaller distances, but it suffices for the present description of
the world. This is all what we meant when we say that we live in four dimensions.
We can put some order on these events and draw them on a four dimensional space.
Representing the label time going in an upright direction, so that events at the same time
are represented by horizontal hyper-surfaces, which in our drawings have one dimension sup-
pressed.
We will now describe some phenomena with these tools, so as to get acquainted with the
concepts.
In figure 14.1 we describe the encounter of two persons in space-time by marking the events

189
190 CHAPTER 14. SPECIAL RELATIVITY

hart-beat of each one of them. They are not proper events for hart-beats are not localizable
in time or space in arbitrarily small regions, but they suffice to describe the encounter pretty
well. Of course, we can imagine that we look at more and more localized events, like the
firing of a nerve cell in the hart, some given molecule passing a cell membrane, and so on
(disregarding the quantum world for this discussion) and so as to describe a continuum of
events. In that limit each of our persons is geometrically characterized in our space-time as
a line. Each line is called the world–line of the person, the encounter is the point where
these lines intersect. The important thing to realize here is that each person, or molecule
inside the person is a one-dimensional object! The event “encounter” is a point, that is zero-
dimensional, but each person, in our rough description, is a line. In figure 14.2 we show this
idealized situation.

Figure 14.1: Describing the event date.

Figure 14.2: Idealized model.

14.1. INTRODUCTION 191

Consider now a hair belonging to one of the persons in the previous encounter. If we
imagine events taking place along the hair, say in its cells, then they would describe in our
space-time a two-surface! We see then that in this way of describing phenomena a hair is no
longer though as a “line”, that is a one dimensional object, but rather as a two-dimensional
one. If one develops the intuition to picture this situation, one is half-way in understanding
relativity. Here is a figure 14.3 showing our hair in space-time.

t
a hair

Figure 14.3: Events history of a hair.

Having introduced the concepts of event and of space-time we can now interpret on it our
usual concepts of space and time.

14.1.1 The Aristotelian viewpoint

The usual, every day, concept of space and time we use is called the Aristotelian view point.
For Aristotle the “earthly” bodies, in contrast with the “heavenly” ones which he though
underwent permanent circular motions, have a natural state of “rest”. So for him the surface
of the earth was at rest and all things tend to lay in this state. As one can see when one
throws a stone and it rolls for a while until stopping. Thus, on space-time there are world lines
which are preferred in the sense that they represent objects which are at rest, for instance
if one of the persons in our past description is sitting somewhere, see figure 14.4. On top of
this, this view believes in a sense of simultaneity of events, that is a sense in which we can say
when two non–coinciding events occur “at the same time”. Thus, we can order the events in
three-dimensional hyper-surfaces containing events occurring simultaneously. Thus, we have
192 CHAPTER 14. SPECIAL RELATIVITY

an absolute time, namely a label for these hyper-surfaces, namely a function from space-time
into the reals. The real value this function has on each of these hyper-surfaces is not very
important for now, the important thing is that it distinguishes each one of these hyper-surfaces
among them. Thus, we have the following picture of this view, see figure 14.5. Notice that
since the time changes for the objects at rest, their world-lines transverse the simultaneity
hyper-surfaces and through these world-lines we can identify points at different hyper-surfaces
as occupying the “same space along time”. The identification of all these hyper-surfaces with
each other along the preferred world-lines is what we call “space”.

t
t=3

t=2

t=1

t=0

Figure 14.4: Aristotelian view

We now describe in mathematical terms the structure this view assumes. We have al-
ready introduced a time function and its level surfaces, the simultaneity hyper-surfaces. The
differential of this function, dt, allows to say which vectors are tangent to the simultaneity
hyper-surfaces. If we take coordinates describing points of space-time, xµ , µ = 0, .., 3. Then
∂t
a vector a of components aµ is tangent to the simultaneity surfaces if a(dt) := aµ ∂x µ = 0.

This follows from the fact that the previous expression is the derivative of the function t on
the direction of a. And this derivative vanishes if and only if this vector is tangent to the
level surfaces of t. Then we have the preferred world-lines of objects at rest, their tangent
vectors, (four-dimensional vectors) provide a vector field over the whole space-time. To obtain
them we use as parametrization of these lines the values of the time function as they cross the
simultaneity surfaces. Then the preferred world lines are given by xµ (t) where t is the value of
µ ∂xµ (t) ∂t
the function t. Thus, we have a vector tµ = ∂xdt(t) , and tµ (dt)µ := tµ ∂x
∂t
µ = dt ∂xµ
dt
= dt = 1.
There is an extra mathematical structure present in this view, namely a notion of distance
between events. This notion is very specific, it says that the distance between two events (in
the same simultaneity hyper-surface) is given by the square root of the sum of the square
of the components of the vector connecting the events when expressed in some preferred co-
ordinate systems. These are the so-called Cartesian coordinate systems of space. One way
of encoding this information without referring to particular coordinate systems is to use a
metric tensor, namely an object which takes two vectors, and provides a number, called
14.1. INTRODUCTION 193

t
t=3

t=2

t=1

t=0

space

Figure 14.5: Space in the Aristotelian view

the square of its norm. This is a bi-function on a vector space V , (an entry for each vector),
h(·, ·) which is linear on each entry, namely,

h(v, aw + u) = ah(v, w) + h(v, u), ∀u, v ∈ V, a ∈ lR,

and symmetric, h(v, w) = h(w, v) ∀u, v ∈ V . In Cartesian coordinates this tensor has
components, hij = δij , the Kronecker delta. 1 The object just described lives in our “space”,
now how do we extend it to the space-time? One procedure is the following, we define the
Cartesian coordinates in one of the simultaneity hyper-surfaces, that is, at a given time. And
then transport it in time requiring that the points along the preferred world-lines have the
same coordinate values for all times. We then extend the metric to all space-time by requiring
that its component in the Cartesian coordinates remain constant for all times. This is not the
whole prescription for symmetric two-tensors in space-time have more components, 10 instead
of 6. We define the other components completing the four dimensional coordinate system with
the time function as the zero-index coordinate, x0 = t, and defining h00 = 0, h0i = 0, i =
1 . . . 3. Notice that in this way, the distance between two events which are occurring at the
same space point but at different time can be measured, and it has the value zero. While the
distance between two events at different space points have a given, positive distance among
them. The “time” between two events is just the difference between the times of each of their
simultaneity hyper-surfaces, so for Aristotle it makes perfect sense to measure time intervals
and distances between arbitrary intervals. Alternatively given a vector in space-time we can
decompose it in a unique way into a vector along t and another perpendicular to it,
1
The existence of Cartesian coordinates, in particular global ones, is a very nontrivial topic, and implies a
certain structure for the space-time. Since this topic is too complex to be dealt with here, we just postulate
its existence.
194 CHAPTER 14. SPECIAL RELATIVITY

X = [X(dt)] t + X̃
since

X̃(dt) = X(dt) − [X(dt)]t(dt) = 0.

Thus, the space norm of the vector X is just the norm of the vector X̃.

t t̃
t=3

t=2

t=1

t=0

Figure 14.6: Galilean view

This completes the description of the mathematical structure of the Aristotelian view of
space-time. We have the set of all events which we call space-time. In space-time, we have
a set of preferred world-lines, representing the events taking part at objects at rest. We also
have a time function t from this set to the reals, and we call simultaneity hyper-surfaces at the
level set of it. With this function we can define the tangent vector to each preferred world-line
and so have a vector field t. We can identify these hyper-surfaces with each other through
the preferred world-lines thus having a notion of “space”. In “space” we have a notion of
Cartesian coordinates and distance among objects. Which we uplift to space-time using the
present structure, giving a (degenerate) metric tensor h in space-time. Time is parametrized
by a clock carried by any of the observer at rest. Note that time and space translations, and
space rotations leave invariant all this structure.

14.1.2 The Galilean viewpoint

While trying to argue that the earth revolved around the sun, Galileo had to dispose of the
concept of “natural rest state” of things. He argued that we can imagine a situation where a
ball is rolling on a flat horizontal surface and that this surface is made of different materials
in such a way that the friction force acting on the ball by the different surface materials is
smaller and smaller, one can also consider families of balls made out of different materials so
that their size are equal but their weight is growing so that the influence of air friction is each
time less and less important, all the balls thrown with the same initial velocity. In this limit
14.1. INTRODUCTION 195

the ball will roll indefinitely, showing that the concept of “natural rest” is not fundamental, it
exists in our mind due to our inability to experience the limit of frictionless motion. Once he
realized this, he proposed the relativity principle which with our present concepts we can re-
state as saying that there are no preferred world-lines and so no natural identification among
simultaneity hyper-surfaces, and so no “space” in the way we have introduced it before.
If no preferred world-lines exist what does remains from the Aristotelian view? We still
have simultaneity hyper-surfaces and a time function, but no preferred vector t. What about
the notion of distance? Assume we have a given Aristotelian frame, that is a vector t, then
we are in the Aristotelian framework and can assume there is a space with its 3-metric h
defined. We also have a coordinate system xµ obtained by pulling up from one simultaneity
hyper-surface a Cartesian system and declaring the other coordinate to be the label of the
hyper-surfaces. Then in this coordinate system we can also lift the metric tensor hij = δij ,
other components vanishing. Now imagine another family of world-lines moving at a velocity
v i = (v, 0, 0) with respect to the previous one, that is, we now have coordinates which relate
to the previous one as,

t′ = t
x′ = x + vt
y′ = y
z′ = z

Then the components of the metric tensor above introduced in this coordinates would
∂xσ ∂xρ 2
be, (using that components of a tensor transform like hµ′ ν ′ = ∂x µ′ ∂xν ′ hσρ ), h0 0 = v hxx ,
′ ′

h0′ i′ = −2vhxi , hi′ j ′ = hij . Thus, we see that the possible lifts will generate different 4-
tensors, so there is no preferred one. The only intrinsic tensor is the one acting on tangent
vectors to the simultaneity hyper-surfaces. The reason is simple: given any two points at
different simultaneity hyper-surfaces and an Aristotelian frame we can project one of them
into the simultaneity hyper-surface of the other one by following the rest lines of that frame,
and then measure the distance among them at that hyper-surface. But this prescription for
measuring distances depends on the Aristotelian frame chosen, in fact, taking v = ∆x AB
∆tAB
this
number can be made to vanish! Thus, we see that in the Galilean framework we can only
measure distances among events when they are at the same simultaneity hyper-surface, still
we can measure the time between events as before. This does not mean that we are restricted
on what we can describe in nature, on the contrary this framework tells us what we should
expect from nature, namely that natural phenomena can not depend on “distances” among
events at different times, but only on distances between events at the same simultaneity
hyper-surface. Thus, for instance, the force between two bodies in the Newtonian theory of
gravitation can not depend on their distances other that when measured at the same time.
This framework order our ideas about how nature behaves.

Exercise: Check that given any two non-simultaneous events there is a world-line for which
the distance between events can be made to vanish. Check that nevertheless it can not be
made to be negative.
196 CHAPTER 14. SPECIAL RELATIVITY

The Galilean view is democratic, in the sense that any two sets of world-lines, and so
any two sets of Aristotelian frameworks are equivalent. Physics can not depend on them,
and one way to make sure that one is considering physically relevant concepts is to check
such equivalence. That is why in classical mechanics one requires the theories to not only to
be invariant under space and time translations and space rotations, but also under arbitrary
Galilean transformations.

14.1.3 There comes the light!

The first successful measurement of the speed of light was made by Ole Römer in 1676.
Looking at the time series of the intervals for which Saturn’s moon, Io, was hidden behind
it. Since the earth is moving with respect to Io in its orbit around the Sun that time interval
changes due to the fact that the light has to cover a longer distance when the earth is going
away from Io than when is coming towards it. This travel difference, when added across
several of Io’s orbits was measurable with the technology existing at that time. In analogy
with bullets, one could also ask what happens with the velocity of light when it leaves Io
going away from the earth or when it leaves Io when it is coming in the direction of the
earth. If we think of the light as bullets then in one case the light would come faster than in
the other and the velocity difference would be twice the orbital speed of Io. Unfortunately
that speed is not big enough, given our distance to Io, and so can not very easily measured.
But, nowadays we observe binary systems (pulsars) very far away and with orbital periods of
milliseconds so if this analogy with bullets were true we would see quite amazing things, like
seeing the pulsar when coming to us much younger than when going away! We don’t observe
this, because light does not behave like particles o bullets. Light travels at its own velocity,
c, independent of the emitter velocity! This is an undisputed observation nowadays, which
follows, for instance, form Michelson’s experiment, and it should surprise us! Light has its
own routes on space-time, its own paths. Space-time has some estrange structure that light
detects and so it is guided along certain paths and not along others! But this is against our
Galilean construction of space-time! Indeed, simultaneity, together with a constant speed of
light is equivalent to an Aristotelian framework, for given a simultaneity surface then we can
construct a time direction using light rays and mirrors. Indeed, take an event where a mirror
sphere is uncovered. Previously at different times send light rays from different events with
a code saying from where and when were they sent collect them into the future at different
events. If they reach an event at the same time into the future with respect to the event
where the mirror sphere was uncovered as the time into the past when it was emitted, then
we say that the emission event and the receiving event are at the same rest point. Doing this
with many events we get a whole set of rest observers, and so an Aristotelian framework.
When people realized this the first attempt of an explanation was to say, after all Aristotle
was right and there must be a medium against which light moves, like a sound or elastic waves
in matter. But all attempts to measure the velocity difference on earth when it was moving
in different directions with respect to the presumed ether gave negative results.
So we have to revise our space-time concepts to account for the observations.
14.1. INTRODUCTION 197

14.1.4 The Einstein viewpoint

We come back now to the symmetries of Maxwell’s equations, namely the Lorentz transfor-
mations. That symmetry tell us that when we apply it to points of space-time it changes
our simultaneity hyper-surfaces, what were two simultaneous events are no longer so. This
is a strong hint that to save Galileo we must abandon simultaneity. Simultaneity, actually
is a very unjustified assumption, aside from the fact that worked very well in our models for
describing physical phenomena aside from electromagnetism. So we just abandon it. So what
are we left with? It seems that with very little, we don’t have any longer a way of determining
the time interval between two events nor can we determine the space distance between any
event, for we can not even say when two events are occurring at the same time. Nevertheless,
it remains a quantity that we can determine, namely the quantity which we saw was invariant
under a Lorentz transformation. Given any Aristotelian framework and a vector connecting
any two events, xAB we can compute,
d2AB := gµν xµAB xνAB := (x0AB )2 − xiAB xjAB δij ,
and this quantity is invariant under any of the symmetry transformations of the wave equation
(here we take x0 = ct so that all coordinates, and therefore tensor components, have the same
dimensions). This is the only invariant quantity one can built out of relative position vectors,
and so the only thing that makes physical sense. If this quantity is negative it means that the
events are separated in a space-like way, that is, we can not reach one event from the other
traveling at speeds smaller or equal to the speed of light. That is, with light or any other
material traveling at smaller speeds, we can not influence from one of these events what is
going on on the other. Furthermore, via a Lorentz transformation, we can always find another
Aristotelian framework
q for which these two events are simultaneous, and the distance between
2
them is just −dAB . This new Aristotelian system has a relative speed to the previous one
which is smaller than c. If the above quantity is null, then we can see that one event can be
reached from another one by via a light ray, but not by any material thing flying slower than
light. We shall say these events are light-related. These events are along the special paths
light travels on space-time. If the above distance is positive, then we can find, via a Lorentz
transformation, another Aristotelian framework in which both events are at the same q space
point, that is are along a given preferred world-line of the framework. In this case d2AB is
the time interval between these two events, we say that they are time-like related. Again
the relative velocity of this new Aristotelian framework with respect to the former one is less
than the speed of light. This is what remains of our simultaneity surfaces, prior to this we
could say whether two events were one into the future of the other or vice-versa or whether
they were simultaneous (occurring simultaneously). We could order them according to their
time of existence. Now that is not longer the case, but still we can say whether one event is
into the future of another, or whether they are light related or space related. This is also an
ordering system, but only partial. This is all what is needed for doing physics.

We are ready to define now our subject, special relativity, which is just the following
assumptions.
198 CHAPTER 14. SPECIAL RELATIVITY

Figure 14.7: Different causal relations among events: events A and B are space-related,
events A and B are light-related, events A and D future-time-related, and event A and E
past-time-related.

1. The set of events is four dimensional, that is it suffices four numbers to label all
events and describe their interactions. More specifically at each neighborhood Up of
an event p there exist four functions (x0 , x1 , x2 , x3 ) form Up into the reals (a coor-
dinate system), such that any function, f from Up into the reals can be written as
˜ 0 (q), x1 (q), x2 (q), x3 (q)) ∀q ∈ Up .
f (q) = f((x

2. At each point of this space of events there is a metric tensor, namely an invertible,
symmetric, by-tensor of signature (+, −, −, −).

3. There are global coordinate systems in which the metric tensor is constant and is diag-
onal with components, (1, −1, −1, −1). We call them Cartesian systems.

This is all what we need to have the arena where physics is played. This space-time is called
the Minkowski space-time . The first two assumptions are though to be very basic. But
they are just assumptions and so are constantly contested in modern physics, in particular the
notion of a four dimensional space-time is abandoned in string theory, although these theories
have not yet succeeded in describing more physics than before. The third assumption is clearly
false because implies the space-time geometry does not have any curvature, and we know that
matter curves space-time, for instance it is known form Eddington expedition to the island of
PrÃncipe near Africa to watch the solar eclipse of 29 May 1919 that sun’s gravity pull bends
light rays. Thus, they do not travel along the straight lines a global coordinate system as the
one above would imply. Nevertheless, this assumption is a good first step in the sense that
with it we can model many physical situations (where gravity is not important) with very
high accuracy and with simple mathematical tools. In particular for this book this is enough.
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 199

In the next sections we shall develop the geometry corresponding to this new concepts, in
particular the kinematics and dynamics of particles and other fields.

14.2 The geometry of Minkowski space-time.

14.2.1 4-vectors
The simplest non-trivial structure in this space are the tangent vectors, they are, obviously,
four-vectors.
At every point/event of the Minkowski space-time we have the pseudo-metric ηµν , thus
the vector space at each of these points is divided in three regions,

• ηµν v µ v ν > 0, we call this vectors time-like .

• ηµν v µ v ν < 0, we call this vectors space-like .

• ηµν v µ v ν = 0, we call this vectors null .

Clearly if we multiply any vector by any real number the vector remains in the same set,
its nature does not change. The zero vector is the only vector in the intersection of these sets.

Exercise: Show that the time-like vectors form a double cone. That is, if v 1 and v 2 are
time-like then either v 1 + αv 2 is time-like ∀α ∈ [0, ∞) or v 1 − αv 2 is time-like ∀α ∈ [0, ∞).
Conclude that the non-zero time-like vectors are in either of two cones. One of them is called
the future cone, the other the past one. Reach the same conclusion for the non-space-like
vectors.

Exercise: Find two time-like vectors whose sum is space-like.

Exercise: Show that the sum of two null vectors is never a null vector unless they are
proportional to each other.

Exercise: Show that the sum of two future directed null vectors is a future-directed time-like
vector.

Exercise: Find two space-like vectors whose sum is time-like.

200 CHAPTER 14. SPECIAL RELATIVITY

Figure 14.8: Sum of future directed time-like vectors

14.2.2 Reference frame descriptions.

Reference frames are useful to bring the description of physical phenomena to the more familiar
Galilean or Aristotelian views. They are also important, for there are many physically relevant
quantities, measurable quantities, which depend on the observer who is measuring them, like
for instance, frequencies.
Given a unit time-like vector t, η(t, t) = 1, at point p in space-time, we can define the
simultaneity hyperplane at that point as the set of space-time vectors such that the tips of
them can be reached by a light ray coming from a point in the past given by −at, a ∈ lR,
positive, and from which one can send light which will reach the point at into the future, that
is the points from which light takes the same time, according to an observer moving along t,
to go than to come back. So these are the points which can be written as,

e = −at + k
(14.1)
e = at − l
(14.2)

for a > 0, and l, and k two future directed vectors. We immediately see that,

1
e = (k − l)
2
(14.3)
1
t = (k + l).
2a
(14.4)
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 201

Therefore,

−1
η(e, e) = η(k, l) ≤ 0
2
(14.5)
1
η(t, e) = η(k + l, k − l) = 0.
4a
(14.6)

Thus, e is space-like (the scalar product of two future directed null vectors is positive or zero,
being zero only when the vectors are zero or proportional to each other, this last condition
can not hold since t was taken to be time-like), and it is perpendicular to t. Thus, the
simultaneity hyper-surfaces are space-like and perpendicular to the observer world line.
Therefore the space of vectors perpendicular to a time-like vector t is a three-dimensional
space with a norm, indeed −η(x, x) > 0 if x 6= 0 for all the above vectors. When appropriate,
for vectors at a simultaneity space we shall write,

−η(x, y) = ~x · ~y ,

and so make contact with the Galilean view.

It is useful to consider the following coordinate system: Given the vector t at a point p
we can propagate this vector to the whole space-time by applying translations, either space-
like or time-like and propagating it as a constant vector. Thus we have a constant vector
field on the whole space-time. The simultaneity spaces of this vector field are integrable,
that is they form global hyper-surfaces, global simultaneity spaces. The Minkowski metric
induced in these simultaneity spaces is constant an positive definite, so we can choose there
a Cartesian coordinate system, namely any coordinate system for which the induced metric
becomes, hij = δij . We then extend this system to the whole space-time by declaring this
coordinate system to be constant along the integral lines of the vector t. Furthermore, we
complete the coordinate system by choosing an extra coordinate, the time, by labeling the
simultaneity hyper-surfaces using the parameter of an integral curve of t which passes through
p. In this coordinate system the vector t has components tµ = (1, 0, 0, 0). This way we get
all elements of a global Aristotelian viewpoint. Notice that if on each hyper-surface we look
at the curves obtained by fixing all coordinate vales, except one, we can define at each point
a set of orthonormal vectors, {ei }, i = 1..3, namely the normalized vectors tangent to those
curves. Together with t they form a global vector frame, that is a global base. Its co-base is
given by θ i (·) = −η(ei , ·), i = 1..3, θ 0 (·) = η(t, ·). With the help of this preferred vectors
it is easy to get the components of any vector in the Cartesian coordinate system, namely,

ui = θ i (u) = −η(u, ei ) u0 = θ 0 (u) = η(u, e0 ) = η(u, t)

Given the vector t we can decompose any tensor in parts along it and parts perpendicular
to it. For instance, given a vector u we can write it in a unique way as,

u = at + ũ with η(t, ũ) = 0

202 CHAPTER 14. SPECIAL RELATIVITY

contracting with t we find that a = η(t, u) (so ũ = u − η(t, u)t). In particular, if we

have a curve in space-time, xµ (τ ) then the vector connecting two nearby points is given by
µ
dxµ := dx
dτ
dτ = udτ , thus, in the adapted coordinate system we have,

dx0 = η(t, u)dτ dxi = ũidτ

Thus, the velocity of this curve is, for the t observer,

dxi i ũi
:= β :=
dx0 η(t, u)

Note that we use β i to express this velocity, the reason is that we are using as time function
x0 = ct so the velocity is dimensionless, in terms of t the velocity is v i = cβ i . Note also that
this 3-velocity does not depend on the parametrization of the curve, and therefore not in the
magnitude of the 4-velocity vector.
Thus, it is better to express the above splitting in the following way,

u = η(t, u)(t + β)
Contracting u with itself we get,

η(u, u) = η(t, u)2 η(t + β, t + β)

= η(t, u)2 [1 + η(β, β)]
= η(t, u)2 [1 − β~ · β]
~ (14.7)

Therefore,
q
u= η(u, u)γ(t + β)
where,

η(t, u) 1
γ=q =q ≥ 1.
η(u, u) ~
1−β·β~

The view from two frames

Consider now two frames, obtained by two future directed unit time like vectors, t and t̂. We
have seen that we can write one as a function of the other, namely,

t̂ = γ(t + β),
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 203

Figure 14.9: Different unit time-like vectors in a given coordinate system.

t̃ l̃ := t̃ − ẽ1

l := t − e1
t

ẽ1

e1

k̃ := t̃ + ẽ1
k := t + e1

Figure 14.10: Different frames and simultaneity hyper-surfaces

204 CHAPTER 14. SPECIAL RELATIVITY

where β is the relative velocity of t̂ with respect to t, and so is a vector in the space of the
simultaneity surface perpendicular to t. Likewise, we can write, t in terms of t̂,

t = γ̂(t̂ + β̂).
We want to find now the relation among these quantities. Since both vectors are unit, we
have,

γ = η(t, t̂) = η(t̂, t) = γ̂,

so both normalization constants are the same. It should be the case that β̂ represents the
negative of the relative velocity β, but there is an important point here, these two vectors are
not in the same hyper-surface, one is in the simultaneity surface of t the other in the one of
t̂. Indeed, we get,

β̂ = t/γ − t̂
= t/γ − γt − γβ
1
= −γ([1 − 2 ]t + β)
γ
2
= −γ(β t + β),

so β̂ is a vector with a part sticking out of the hyper-surface perpendicular to t, and indeed
it is perpendicular to t̂. To see the components of this vector in that hyper-surface we take a
frame (and its associated coordinate system) {êi }, i = 0..3, with ê0 = t̂, and ê1 in the plane
spanned by t̂ and t. The orthonormality condition determines it to be,

ê1 = γ(βt + e1 ),
where e1 is the frame component perpendicular to t and in the plane spanned by t̂ and t,
(e1 = β/β, or β = βe1 ). Thus, we see that

β̂ = −β ê1 ,
so indeed, in components the relative velocities are identical except that the directions are
opposite. This could have been guessed from the fact that,

η(β̂, β̂) = η(β, β),

and that it is perpendicular to t̂ and in the plane spanned by t̂ and t.

Relative Velocities
Frequently one is given the world lines of two bodies in some arbitrary reference frame, t, and
would like to find their relative velocities, namely the velocity of one as seen from the other.
That is, we are given their velocities,
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 205

u = γu (t + β), and u′ = γu′ (t + β ′ ),

with η(β, t) = 0, and η(β ′ , t) = 0, and we want to find,

u′ = γuu′ (u + β̂),
with η(β̂, u) = 0. The normalization factor is easy,

γuu′ = η(u, u′ ) = γu γu′ η(t + β, t + β ′ ) = γu γu′ (1 + η(β, β ′ )) = γu γu′ (1 − β~ · β~′ ).

Exercise: Check that γuu′ ≥ 1, and so that β̂ < 1.

We also have,

β̂ = u′ /γuu′ − u,
form which it follows,

β̂ 2 := −η(β̂, β̂)
1 2η(u′ , u)
= −[ 2 + 1 − ]
γuu′ γuu′
1
= 1− 2
γuu′
(1 − β 2 )(1 − β ′2 )
= 1−
1 + η(β, β ′ )
−η(β − β ′ , β − β ′ ) + β 2 β 2 sin(θ)2
=
(1 − ββ ′ cos(θ))2

In the limit of small velocities, β << 1, β ′ << 1 we get,

β̂ 2 ≈ −η(β − β ′ , β − β ′ ),
which corresponds to the usual addition of velocities en Galilean kinematics.
We would like now to find the components of β̂ in the frame associated to u. If {ei },
i = 0..3 is the frame associated to t, and {êi } is the frame associated to u, (e0 = t, and
ê0 = u), such that both e1 and ê1 are in the plane spanned by these two vectors, and
ei = êi , i = 2, 3, we have,

e0 = γu (ê0 − β ê1 ), e1 = γu (ê1 − β ê0 ).

Therefore,

u′ = γu′ (e0 + β ′i ei ) = γu′ [γu (ê0 − β ê1 ) + γu β ′1 (ê1 − β ê0 ) + β ′2 ê2 + β ′2 ê2 ],
206 CHAPTER 14. SPECIAL RELATIVITY

and,
−β + β ′1
β̂ 1 := −η(β̂, ê1 ) = −η(u′ /γuu′ − u, ê1 ) = ,
1 + η(β, β ′ )
(14.8)
′i
β
β̂ i := −η(β̂, êi ) = i = 2, 3.
γu (1 + η(β, β ′ ))
(14.9)

14.2.3 The proper time of a trajectory.

The parametrization of the above curve, or equivalently, the magnitude of the velocity vector,
does not play a role in the definition of the 3-velocity this curve has with respect to some other
curve. This is so because physically the relevant information about a trajectory in space-time
is given by the set of points along the curve and not the curve itself. That set is a whole
history, a complete succession of events. The way we parametrize it is irrelevant.
It is customary to use a parametrization which uses the only scalar intrinsic property a
trajectory has, namely the length of the curve from some given point. To see this, consider two
events, A, and B that are time-like related, with B into the future of A and curves connecting
them. Restricting consideration to smooth curves whose tangent vector is everywhere along
the curve time-like or null, that is its norm is positive or null. We can define the distance or
interval along each curve as,
s
1 dxµ dxν
Z
TAB := ηµν dτ. (14.10)
0 dτ dτ
Notice that this quantity does not depend on the parametrization chosen for expressing the
µ µ ds
curve, indeed choosing another parametrization, s = s(τ ), and the chain rule, dx dτ
= dx
ds dτ
,
shows that the result is the same. Thus, this quantity depends only on the curve, as a set,
and on the metric tensor, it is the only intrinsic property we have. It is thought as the time
interval along the curve between these two events as measured by a local clock. If we choose
this length to parametrize the curve, then the tangent vector has unit norm. This is the
customary choice.
We therefore have,

dx0 = η(t, u)dτ = γds ≥ ds

when s is the proper time. Thus, the coordinate interval any observer assigns between two
infinitesimal intervals is always larger than the proper time between them.

14.2.4 The trajectories of free particles and a variational principle.

Consider different curves connecting the events A and B. See figure 14.11. Notice that
different curves would give different times between the events, so that different processes along
different curves if synchronized at A would reach B clicking different times. Nevertheless, we
do have a sense in which there is a time interval between events A and B.
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 207

xµ
1
γ0

γ1

Figure 14.11: Curve variations.

Lemma 14.1 The maximal interval corresponds to the straight line connecting both events.
The value of this interval corresponds to the proper time of an observer moving along such
curve.
Proof: Assume xµ0 (s) is the curve which maximizes the above quantity, and consider nearby
curves of the type xµλ (s) = xµ0 (s) + λxµ1 (s). Since they all must leave from A and reach B we
assume xµ1 (0) = xµ1 (1) = 0. Using them we construct a function,
s
dxµλ dxνλ
1
Z
TAB (λ) := dτ.
ηµν
0 dτ dτ
taking the derivative with respect to λ and setting it to zero when λ = 0, (the condition for
dAB (λ = 0) to be an extreme. We get,

dxµ
λ d dxνλ
dTAB (λ)
Z 1 ηµν ( )
|λ=0 = r ds dλ ds ds|λ=0
dλ 0 dxµ dxν
ηµν dsλ dsλ
dxµ ν
0 dx1
Z 1 ηµνds ds
= ds.
dxµ dxν
q
0 ηµν ds0 ds0
choosing, for simplicity, the parametrization so that the denominator is constant, and inte-
grating by parts we get,

dTAB (λ) 1 1 d2 xµ0 ν dxµ0 ν 1

Z
|λ=0 = q [ −ηµν 2 x1 ds + ηµν x | ]
dλ dxµ dxν
ηµν ds0 ds0 0 ds ds 1 0
1 d2 xµ0
Z 1
= q −ηµν (s)xν1 (s)ds
dxµ dx ν
ds 2
ηµν ds0 ds0 0
208 CHAPTER 14. SPECIAL RELATIVITY

where in the last step we have used the initial and final conditions for both curves to coincide
at A and B. Since xν1 (s) is arbitrary, and ηµν invertible we reach the conclusion that along
the maximum the second derivatives of the curve, that is the acceleration, must be zero. So
the curve has constant velocity vector, and therefore it is a straight line.

Exercise: Show with an example that if ηµν were not invertible we would not be able to
reach the above conclusion.

We shall say that in absence of forces particles travel along straight lines of this coordinate
systems. Not only that but with the parametrization used those lines can be described as
linear functions of the parameters, in particular their internal clocks mark the time interval
between the events they occur. Thus, imagine two persons which meet at event A at which
they synchronized their watches, then each one of them went on different paths until meeting
again at event B. In general their watches will mark different times, the shorter time would
correspond to the person which accelerated the most during its path. So if you move around,
you stay younger!

Exercise: Show that the time difference between person going through the longest past and
a person going along a nearby path is given by
Z s
∆TAB = η(δu, δu) ds,
0

where δu := λ dx
ds
1
is the variation of the velocity in the direction perpendicular to it. Notice
that if we parametrize the curve with the proper time, then, since η(u, u) = 1, η(u, δu) = 0
and so, η(δu, δu) is negative definite unless δu = 0. Hint, perform a second variation of the
above formula, and use Taylor’s formula to second order to approximate TAB in terms of λ.

14.2.5 The size of objects

Consider a piece of chalk moving freely in space all its points with constant velocity, (so there
is no rotation). We can model it in space-time as a two surface object in a way similar as
what we do to model a the history of a hair, that is suppressing its transversal directions, see
figure 14.3.
It should be clear by now that this is the object chalk, there is not a one dimensional object
we can extract in a natural way from it. At most, we can have a one dimensional representation
if we fix an observer and a given event along its world-line and consider all those events at
the two-surface chalk from which light emitted at them reaches our observer at the specified
event. But that one dimensional representation depends mostly on the observer, so it is not
natural by any means. Since the chalk is assumed to move at constant speed, or otherwise
taking the center of mass instantaneous velocity as a rest frame we have, nevertheless, one
preferred simultaneity hyper-surface and so a one dimensional object. So it is interesting to
see how this object compares with others other observers would see when moving with respect
to it. For instance, we can ask what would it be the sizes these representations would have for
14.2. THE GEOMETRY OF MINKOWSKI SPACE-TIME. 209

each one of them. In the next figure 14.12 we show the situation, we have our chalk moving
with velocity u from another frame, and show the two different representations by the vectors
x, and x̃.

t
u
chalk

x̃

Figure 14.12: Chalk description and sizes.

From the figure we see that there must exist α such that

x̃ = x + αu.
Since u = γ(t + v), η(t, x) = 0, and η(u, x̃) = 0, contracting the above expression with u
we obtain α, and so,

x = η(u, x) u + x̃ = γη(β, x) u + x̃
therefore,

η(x, x) = η(x̃, x̃) + γ 2 η(β, x)2 η(u, u) = η(x̃, x̃) + γ 2 η(β, x)2
and so,
x̃ · x̃ = −η(x̃, x̃) = −η(x, x) + γ 2 η(β, x)2 = x · x + γ 2 (β · x)2 .
we see that the largest distance the object has is precisely when viewed from the frame at
which it is at rest.
If we choose coordinates so that the relative velocity is along the x axis, v = (v, 0, 0), then
in the other frame we have,

x2 v 2
x̃2 + ỹ 2 + z̃ 2 = x2 + y 2 + z 2 + = γ 2 x2 + y 2 + z 2
c2 (1 − v 2 /c2 )
210 CHAPTER 14. SPECIAL RELATIVITY

14.2.6 Volume contraction and densities

Consider a cube of sides (∆x, ∆y, ∆z) on its rest frame, if we see it from a frame moving with
respect to it with speed v in the x direction, we would see a shorter cube ( ∆x γ
, ∆y, ∆z) and
V0
its volume will be, γ , where V0 is the volume at the frame at which is at rest. Consider now
an arbitrary body, and view it from some arbitrary frame moving with respect to it, since
we can approximate its volume by summing over small cubes, and in particular we can take
these cubes to have one of their sides aligned with the velocity direction we see (see figure
14.13) we have the same relation among volumes as for the individual cube, namely it will be
measured as smaller by a γ factor.

Figure 14.13: Volume contraction.

Consider now a set of N particles in a given volume, then the particle number density,
ρ := N
V
will depend on the observer that is viewing it, indeed, in some other reference frame
we will see a density given by ρ = VN0 = NV0γ = γρ0 . Nevertheless, there is an object which is
γ
invariant, the current density, defined by,

j := ρ0 u
where u is the four-velocity of the center of mass of the particles and ρ0 the density as
measured on that frame. If t is another observer, then, u = γ(t + v) with η(v, t) = 0 and
therefore,

η(j, t) = ρ0 η(u, t) = γρ0

so it transforms in the expected way from frame to frame. Furthermore, in the coordinate
system of this new frame we have, j µ = (ρ, ρβ), where ρ := ρ0 γ, and,

∂j µ 1 ∂ρ ∂(ρv i )
= [ + ]
∂xµ c ∂t ∂xi
14.3. RELATIVISTIC OPTICS 211

So we see we get the expression for particle number conservation when we set this four-
divergence to zero. In general, we can interpret it as the rate of particle creation. It also
applies to other currents, as the charge density and electric currents, which we see they get
together into a single four-vector.

14.3 Relativistic Optics

Light rays can be approximated by plane solutions of Maxwell’s equations. For simple propa-
gation it is enough to restrict attention to just rays of the wave equation, for any component
of the electric or magnetic field satisfies it. We shall see later the full plane wave solution
of Maxwell’s equations, they are needed to study reflection and refraction, but for simple
propagation in vacuum what follows ray propagation is just enough.
A plane wave of the wave equation is a solution of the form,

φ(xµ ) = f (xµ kµ )
for some smooth function f : lR → lR. Clearly the surface of constant f are hyperplanes in
space-time. Given a reference frame with Cartesian coordinates xµ = (ct, xi ) we have,

φ(t, ~x) = f (ωt + ~k · ~x)

where kµ = ( ωc , ~k) . We see that expression represents a function which is constant along the
lines perpendicular to k, namely the position vectors of the form,

−~k · ~x
t=
ω
These hyperplanes are the level sets of f .
The intersection of a simultaneity hyper-surface of this coordinate system with a level sets
of f consists the sum of a fixed (space) position vector plus an arbitrary vector perpendicular
to ~k. That is on each simultaneity hyper-surface f is constant along planes perpendicular to
~k. That is, the intersection of two hyper-planes is a 2-plane. We can think of those space 2-
planes as moving along the lines (t, ~x = −ω
k2
t). In the case that f (s) := f0 eis , ω is interpreted
as the wave frequency that observer at rest in such a frame will see the wave has.
If we apply the wave equation to this function we get,

∂2φ ω 2 ~ ~ ′′
0 = ✷φ(t, ~x) = − ∆φ = ( − k · k)f
c2 ∂t2 c2
thus we see that in order that this be a solution we need that the four-co-vector k be a null
vector, indeed the above expression is just

η µν kµ kν = 0

where η µν is the inverse of the metric,

212 CHAPTER 14. SPECIAL RELATIVITY

1 0 0 0
 
 0 −1 0 0 
η µν =  (14.11)
 
0 0 −1 0

 
0 0 0 −1
Therefore, we must consider ray co-vectors of the form, kµ = (± ωc , ωc k̂) with k̂ any unit
vector. We shall take the plus sign for the time like part, which corresponds to a wave
moving in the direction opposite to k̂. This is so because the vector form of kµ has the form,
k µ := η µν kν = ωc (1, −k̂). This will be used when discussing aberration §14.3.2. Note that
given any time like vector, t, any null vector can be decomposed as,

k = a(t + k̂).
Contracting with t we obtain, as for the case of time-like vectors, a = η(t, k). Contrary to
the case for time-like vectors, the value of this quantity can not be expressed in terms of the
norm of the space part of k. It is just the frequency the observer t sees the wave has in its
reference frame. A property of the whole vector k. Indeed, the null character of the vector
only implies that η(k̂, k̂) = −1, or k̂ · k̂ = 1.
The geometrical description of this is simple, if one marks, null planes at which the function
f reaches, say, its maximum, and several unit vectors, then the number of those planes pierced
through by that those vectors will be proportional to the frequency an observer moving along
the line defined by each one of those vectors will measure that frequency. See figure 14.14.

Figure 14.14: Null planes and unit vectors. The number of planes the unit vector pierced
through by the unit vector is proportional to the frequency the observer moving along it will
measure.

14.3.1 Doppler shift

For the observer at rest with our coordinate system t = (1, 0, 0, 0) the frequency of the wave
is just,
14.3. RELATIVISTIC OPTICS 213

ω = ct(k) = ctµ kµ ,
the speed of light here appears because x0 = ct. Like wise, for any other observer, t̃ = γ(t+β)
the frequency will be,

ω̃ = cη(t̃, k) = ωγη(t + β, t + k̂) = ωγ(1 + η(β, k̂))

and so,

ω̃ 1 + ~vc·k̂
=q ,
ω 1 − vc2
2

as expected from the usual derivation for sound waves, the frequency is bigger when the ob-
server moves at a faster speed towards the light ray than the previous observer. In particular,
we can consider a light source which emits at a given frequency determined by its internal
state. So on the rest frame of the source we will observe the real frequency as determined by
its internal state, ω0 . Any other observer will see this ray with a different frequency given by,

1 − ~vc·k̂
ω = ω0 q 2
1 − vc2

where ~v is the velocity of the observer respect to the light source, and k̂ is the wave direction
as pointing in the direction to the other observer (the negative of the previously used direction
vector). Notice that the Doppler effect has two different contributions, one is the usual we
know from sound waves, and is related to the relative velocities of sources and receivers, and
depends on the sign of that velocity. The other is a purely relativistic effect, a γ factor, which
is present even in the case when the source moves perpendicular to the observer direction.
For relativistic situations this factor can be dominant.

t t̃

light ray

k̂

Figure 14.15: Doppler Effect

214 CHAPTER 14. SPECIAL RELATIVITY

tr

light ray source path

ts

k̂

receiver path

Figure 14.16: Source-receiver Doppler Effect

Exercise: At rest a interstellar hydrogen cloud absorbs light at some given spectral line
frequencies, can we infer its relative motion with respect to a further away background star?

Exercise: Millisecond pulsars binaries. Using Newtonian gravity estimate the Doppler shift
of a millisecond pulsar rotating around a normal star with a period of 1 hour.

Exercise: What happens when the observer velocity is perpendicular to the line of sight of
the light source object?

Exercise: Deduce the standard Doppler effect for sound.

14.3.2 Aberration
Light aberration is also a classical (pre-relativistic) phenomena which has also its relativistic
counterpart. In fact was used for the second measurement of the light speed by Bradley in
1729.
The observation of this effect consists in realizing that the angle at which a star is seen in
the sky depends on the velocity the observer has with respect to the star. Indeed, when the
star is sufficiently far away all light coming from it can be assumed to reach the telescope at
parallel rays. If the telescope is at rest with respect to the star one would measure the correct
14.3. RELATIVISTIC OPTICS 215

angles, in particular the declination angle, θ. If the earth, and so the telescope is moving, for
simplicity in the same direction as the star is, then the light entering the telescope aperture
will arrive at the ocular when this has already traveled a distance from the original position,
thus resulting in a measurement of the declination angle smaller that when at rest with respect
to the faraway star.

L ~v
θ′
θ

B
R cos θ

L cos θ ′

Figure 14.17: Aberration

From the figure 14.17 we see that A = L sin θ′ = R sin θ, and R = c∆t, on the other hand,
B = L cos θ′ − R cos θ = v∆t. Therefore,
v
cos θ′ sin θ − cos θ sin θ′ =sin θ′
c
For small effects, defining dθ = θ − θ and to first order in vc we get,
′

v
dθ = − sin θ.
c
Using measurements of this effect when the earth was in opposite sides of its orbit around
the sun, Bradley, who discovered it, was able to measure the speed of light. Notice that,
as Doppler’s, it is an effect of order vc , so not so difficult to measure. It is not due to the
2
Minkowski metric, which only can give effects of order vc2 , and so only relativistic in the sense
that the speed that appears is a relative speed.
We now consider the full relativistic version, for this it is better to consider two stars at
slightly different positions and relate all angles to the angles between them. This is actually
216 CHAPTER 14. SPECIAL RELATIVITY

much closer to the real measurement, a plate is taken of a sector of the sky, and then a second
one 6 month latter. One then looks for the difference on the plate distance among the stars,
which is just proportional to the angles they have on the sky on both situations. Following the
situation in figure 14.18, we have two stars, each one sending light according to null vectors
k1 and k2 . At event A they are registered by two observers moving with four-velocities t and
t′ respectively. In term of these two vectors we have,

t
t′ stars

light rays

k̂2

k̂1

Figure 14.18: Relativistic Aberration

ω1 ω1′ ′ ′
k1 = (t + k̂1 ) = (t + k̂1 )
c c
ω2 ω2′ ′ ′
k2 = (t + k̂2 ) = (t + k̂2 )
c c
′
where k̂i ( k̂i ) are unit vectors perpendicular to t (respectively t′ ). Therefore,

ω1 ω2 ω1′ ω2′
η(k1 , k2 ) = (1 − k̂ 1 · k̂ 2 ) = (1 − k̂1′ · k̂2′ )
c2 c2
from where we get,

ω1′ ω2′ 1 − k̂1 · k̂2

= .
ω1 ω2 1 − k̂1′ · k̂2′
But from Doppler’s effect we have,

ω1′ ω2′ (1 + ~v·ck̂1 )(1 + ~

v ·k̂2
c
)
= 2 .
ω1 ω2 1 − vc2
Defining k̂1 · k̂2 = cos θ, ~v · k̂1 = v cos θ1 , ~v · k̂2 = v cos θ2 , so θ is the angle between k̂1 and
k̂2 , and so on, we have, 1 − k̂1 · k̂2 = 1 − cos θ = 2(sin 2θ )2 and,
14.4. ENERGY-MOMENTUM 4-VECTOR 217

(sin 2θ )2 (1 + v cos θ1
c
)(1 + v cosc θ2 )
′ = 2 ,
(sin θ2 )2 1 − vc2
This is the relativistic version of the aberration effect.

14.4 Energy-Momentum 4-vector

We want now to define the four dimensional dynamics, that is the dynamics in space-time. We
will assume that free particles travel at speeds lower that light and along straight lines, that
is, lines which minimize the proper time along the trajectories. Using this parametrization
along the straight trajectories they can be expressed as linear relations among coordinates.
Forces, that is interaction with external fields will cause a modification on these trajectories,
that is, they will modify them producing some acceleration, as we know from Newtonian
theory. So we need to generate an equation of motion, on one side should have something like
an acceleration and on the other something like a force. Since we are in four dimensions this
should be a relation among 4-vectors.
We consider a particle along a trajectory γ, for reasons we shall explain below we parametrize
it with its proper time, and so we get a unique tangent vector u with unit constant norm.
If we take an instantaneous reference system which is at rest with the particle at a given
event, then the time in this reference coordinate system coincides with the proper time of the
particle at that particular event. Since, with this parametrization, along the curve u · u = 1
we have,

dη(u, u)
= 0,
ds
u is perpendicular to the 4-velocity,
so the 4-acceleration a := dds

dη(u, u) du
0= = 2η( , u) = 2η(a, u).
ds ds
Thus, we expect to have only three equations out of a vectorial equation determining a. The
4-acceleration is a spatial vector with respect to the frame instantaneously co-moving with u,
that is, a = (0, c~a2 ). On it we impose Newtons equations,

m0~a = F~ , (14.12)
Here m0 is the mass of the particle as measured in the instantaneous rest frame of the particle,
an additive property of the matter. If no forces are present, then we see that the velocity
vector remains constant and so, in this coordinate system we have a linear relation among
the coordinates along the trajectory. That is why we choose the proper time parametrization,
any other parametrization not proportional to proper time would make the above relation
undetermined, unless the force would also have information on the parametrization chosen.
So we can remove now our frames and write an invariant four dimensional expression for the
above law.
218 CHAPTER 14. SPECIAL RELATIVITY

du
m0 = f ⊥ := f − η(u, f )u (14.13)
ds
where f is any given 4-vector in space-time representing forces, and we have taken in the
right-hand side of the equation the part of it perpendicular to u, since we have, η(a, u) = 0.
We have written explicitly the acceleration to remark that the derivative we are taking is with
respect to proper time and not coordinate time, which would be wrong here, for the above
expression is valid for general coordinate systems (so no reference to a particular coordinate
system can occur).
How this expression results in an arbitrary system? Consider then a system whose constant
4-velocity is t, then with respect to these systems u = γ(t + β), with γ = η(t, u) = q 1 2 .
1− v2
c
For the force we write,
F
f ⊥ := γ(λt + ),
c2
where, from the perpendicularity condition,
η(F , u) c2 η(F , u) 1
λ=− 2
= = 3 F~ · ~v,
c η(t, u) γ c

is the work generated by F~ by unit coordinate time. There result two equations, one when
contracting 14.13 with t, and the other when taking the perpendicular part, (recalling that t
is a constant vector),

dγ
m0 c2 = F ·v (14.14)
dt
dγv
m0 = F (14.15)
dt
Where we have replaced the proper time interval ds by cdt γ
. As already mentioned, there
are only three independent equations, and in fact it is easy to see that the first one is a
consequence of the others.
To see this consider the limit of v << c and perform a Taylor expansion of both equations
v2
above, since γ ≈ 1 + 2c 2,

2 2
2 dγ d( v ) d( m02v )
m0 c ≈ m0 2 ≈ ≈F ·v (14.16)
dt dt dt
dγv dv dv
m0 ≈ m0 ≈ m0 ≈F (14.17)
dt dt dt
therefore the first equation follows, up to an integration constant, from integrating the second
when contracted with v. It is the well known equation giving the variation of the kinetic
energy due to work done on the system.

Exercise: Show that the first equation follows from the second one. To see this, get an
2
equation relating γ dγ
ds
, v 2 and dv
ds
. Integrate it to get back the expression of γ as a function of
14.4. ENERGY-MOMENTUM 4-VECTOR 219

v 2 . Compare with the Newtonian version, obtained by Taylor expanding the above expression
and retaining the first non-zero terms.

We can take a further step and define p := m0 u, the 4-momentum of the particle. As in
Newtonian mechanics, in cases where the masses can change we generalize the above system
to,
dp
= f, (14.18)
ds
in terms of this 4-vector,
q
m0 := η(p, p) (14.19)
can be taken to be the definition of rest mass. We shall come back to this formula once we
formulate Maxwell’s fields as objects in four dimensions and introduce Lorentz force.
If we have several particles, then the total 4-momentum, is given by,
N
X
pT := pi (14.20)
i=1

Now, this vector is the sum of vectors of the form mi ui , with mi positive and ui , that is
vectors pointing into the future of the light cone, namely the cone whose boundary consists of
all vectors of norm zero, see figure 14.8. But the sum of any two vectors in this cone produces
another vector inside the cone, thus any sum of vectors inside the cone gives a vector also
inside it and so the total momentum vector is inside the light cone and future directed. So
we can use its direction to define a rest frame, the center of mass rest frame of the particle
system.
We shall assume it is conserved in absence of external forces. This statement does not
follow from the similar one in Newtonian mechanics, for it assumes contact forces (so as to
be taking place at the same event) which in general is not the case. The general proof of this
statement needs a different formalism, the introduction of interaction fields in a variational
setting. This conservation law unifies both energy and momentum conservation in just one
law. As it stands, without the existence of extra fields carrying the interactions, it can not
possibly be true. For instance consider the two situations in the figure 14.19, clearly on each
one of these two simultaneity surfaces the total momentum is different, for they are considered
at different times. The only unambiguous case is when all interactions happen at the same
event, and then all momenta are at the same point, at which, of course the calculation does
not make any sense. The situation in which does makes sense is when the interaction among
particles is limited to a region of space-time, and we look at the total 4-momentum much
before and much later the interaction region, in that case the trajectories are all constants
and it makes sense to add vectors at different points in space-time, this is what is called a
scattering situation, see figure 14.20.
220 CHAPTER 14. SPECIAL RELATIVITY

Figure 14.19: Summing vectors at different times.

interaction region

Figure 14.20: Scattering situation, the sums are done before and after all interactions have
finished.
Chapter 15

Relativistic Electrodynamics

15.1 Maxwell’s tensor

Following our description of physics in space-time we turn now to electromagnetism.
So far we have considered Maxwell’s equations as a set of coupled equations for two vector
quantities, E~ and B~ describing the electromagnetic fields and a scalar and a vector field, ρ
~
and J, representing charge and current distributions respectively. But we know that this can
not be the whole story, for we know that Lorentz transformations mix time and space and
what were space for one observer no longer is for some other, so spatial vectors, (3-vectors) do
not make much sense. We must search them for truly four dimensional objects. For matter
fields this was easy, we saw that densities must be part of a 4-vector, so we have,

j µ := ρ0 uµ

the charge density measured by an observer for which the charges are momentarily at rest,
times its 4-velocity. An observer with 4-velocity tµ will see a charge density given by ρ =
tµ jµ = ρ0 γ, and a charge current given by J~ = c(j − tρ) = ρ~v .
The objects we are seeking for describing the electromagnetic fields can not be a couple
of 4-vectors, this is because a Lorentz transformation mixes both components. This can be
seen from the following simple observation:
Consider an infinite conducting flat plate with constant surface charge density σ, say at x =
0. Then we know it will generate a constant electric field perpendicular to it, E ~ = (4πσ, 0, 0).
But an observer moving in, say the y direction, ~v = (0, v, 0), not only will see a bigger field,
E~ ′ = (4πσ ′ , 0, 0) = (4πσγ, 0, 0), but also a magnetic field, for he will see the charge density
′
moving and so generating a current in the opposite direction, namely, J~ = (0, −σγv, 0), which
in turn will generate a constant magnetic field given by B ~ ′ = (0, 0, −4πσγ v ) = γ ~v ∧ E.
~ Thus,
c c
both vectors must be part of a single four dimensional geometrical object.
Since at every point of space-time we have six independent components, we must ask
ourselves which kind of tensor have that many independent
   components.
  Since a single 4-
1 0 2
vector would not do we try next with tensors of type 1 , or 2 or 0 . Since we can transform
any of this kind of tensors in any other of them by contracting with the metric or its inverse,
to consider just one kind of them is enough. But they have 4 × 4 = 16 components! So

221
222 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

for this to work we must impose some conditions among their components so as to get only
6 independent ones. These conditions must be valid for the components in any coordinate
system, (in particular Lorentz transformations), that is they must be geometrically invariant.
There are only three type of conditions which satisfy this, namely to ask for the tensor to be
symmetric: tµν = tνµ , antisymmetric: tµν = −tνµ , and/or trace free: η µν tµν = 0. Symmetric
four-dimensional bi-tensors, like the metric, have 10 independent components, while the trace
free condition only imposes 1 condition, so even if we require both conditions we are left
with 9 components. They are no good. But anti-symmetric four-dimensional bi-tensors
have precisely 6 components. So it seems that Maxwell’s fields must have a four-dimensional
realization as an antisymmetric bi-tensor, Fµν = −Fνµ . How do we relate it to our old friends,
~ and B?
E, ~ Given an observer t we can make,

Eµ := Fµν tν
notice that this is a vector in the space perpendicular to t, indeed from the anti-symmetry
of F we see that Eµ tµ := Fµν tν tµ = 0. So we assert that this will be our electric field for
this observer. How do we obtain B ~ now? Out of t, and F we need to obtain another space
vector, this can be done with the help of the completely antisymmetric tensor εµνσρ with
ε0123 = −1 in one of our Minkowskian Cartesian systems. 1 This is the Levi-Civita tensor in
four dimensions. Once we have fixed the value of this component all other components are
fixed because the antisymmetry of it. With the help of this tensor we define,
1⋆
F µν := εµνσρ Fσρ ,
2
This new tensor is also antisymmetric and so,

Bµ := −⋆ Fµν tν
~
is also a spatial vector perpendicular to t, this one is B!

Exercise: Check that in a coordinate system where tµ = (1, 0, 0, 0),

F00 F01 F02 F03 0 −E1 −E2 −E3

   
 F10 F11 F12 F13   E1 0 −B 3 B 2 
Fµν =  = (15.1)
   
3
F20 F21 F22 F23 E2 B 0 −B 1 
 
  
F30 F31 F32 F33 E3 −B 2 +B 1 0
−1 ijk
That is, Ei = Fi0 , B i = 2
ε Fjk , i, j, k = 1 . . . 3.

Exercise: Check that Fij = εijk B k

Exercise: Find the components of ⋆ Fi0 . To which vector corresponds 12 εijk⋆ Fjk ?
1
One could choose the opposite sign, this corresponds to a change in the direction of one of the coordinate
axis, that is to a parity transformation. With this sign tρ ερµνσ is the conventional volume 3-form.
15.2. MAXWELL’S EQUATIONS 223

Exercise: Check that,

⋆
F 00 ⋆
F 01 ⋆
F 02 ⋆
F 03 0 −F23 F13 −F12
   
⋆ 10 ⋆ 11 ⋆ 12 ⋆ 13
⋆
 F F F F   F23 0 F30 −F02 
F µν =  =
   
⋆ 20 ⋆ 21 ⋆ 22 ⋆ 23
F F F F −F13 −F30 0 F10
 
   
⋆ 30 ⋆ 31 ⋆ 32 ⋆ 33
F F F F F12 F02 −F10 0
0 B1 B2 B3
 
 −B 1 0 E3 −E2 
=  (15.2)
 
2
−B −E3 0 E1 


−B 3 E2 −E1 0

15.2 Maxwell’s equations

We now look at the covariant four-dimensional version of Maxwell’s equations. The fields we
have at our disposal are, the geometrical ones, the 4-metric, ηµν , and the Levi-Civita tensor,
εµνσρ , and Maxwell’s tensor. We need to find 4+4 = 8 equations, 3+3 = 6 evolution equations
and 1 + 1 constraints equations, that last description in terms of evolution and constraints
depends on the choice of time direction, it is not covariant. Since Maxwell’s equations are
linear in the fields and first order (only one derivative), we expect they would only include
one derivative operator and be also linear en F . As mention, we also expect a total of 8
equations, that is, two vector equations.
There are just two candidates,

η µν ∂µ Fνσ , (15.3)
and
−1 µνσρ
ε ∂ν Fσρ = ∂ν ⋆ F νµ . (15.4)
2
The first expression is a vector, while the second is a pseudo-vector, since it is built out of
Fµν , and εµνσρ , the Levi-Civita tensor, a pseudo-tensor since it changes sign when we perform
time or space inversion. Thus, since matter is vectorial in nature, we must have,

∂ µ Fµν = 4πjν (15.5)

and

∂ν ⋆ F νµ = 0. (15.6)

Exercise: Show that ∂ν ⋆ F νµ = 0 is equivalent to

1
∂[µ Fνσ] := [∂µ Fνσ + ∂σ Fµν + ∂ν Fσµ ] = 0. (15.7)
3
224 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Exercise: Show that if

1
Fνσ = ∂[ν Aσ] := [∂ν Aσ − ∂σ Aν ],
2
then
∂[µ Fνσ] = 0. (15.8)

Exercise: Use 24.1 and 15.5 to show that

✷Fµν = 8π∂[µ jν] .

To see that they are indeed Maxwell’s equations we write them in Cartesian coordinates,
x = (ct, x, y, z) (so that the metric is now ηµν = diag(1, −1, −1, −1)). Then tµ = (1, 0, 0, 0), is
µ

unitary, and we can write a 4-velocity vector as uµ = γ(tµ + vcµ ) (recalling that tµ = (1, 0, 0, 0))
The zero component, ν = 0, of 15.5 is,

∂ 0 F00 + ∂ i Fi0 = 0 − ∂i E i = −4πρ ⇒ ∂i E i = 4πρ,

So we regain the constraint equation corresponding to Gauss law for this frame. For ν = 1
we get,
4π 1
∂0 F01 − ∂1 F11 − ∂2 F21 − ∂3 F31 = −∂0 E 1 + 0 + ∂2 B3 − ∂3 B2 = ρv
c
so it is the 1 component of the equation
1 −4π i
∂t E i − εijk ∂j Bk = J.
c c
Equation 15.6 is the vacuum version of the previous analyzed case, but with E ~ → B ~
~ → −E
and B ~ see exercise in the previous section. So we reach the other set of Maxwell’s
equations,

∂t B i + cεijk ∂j Ek = 0 ∂i B i = 0
and again we see that the separations between constraint and evolution equations is observer
dependent.
Otherwise, for simplicity, we can take the alternative version of 15.6, 15.8 and compute
some of its components to see that they correspond to the above expression. Indeed, com-
puting

0 = ∂0 Fij + ∂j F0i + ∂i Fj0 = ∂0 Fij − ∂j Ei + ∂i Ej ,

15.3. INVARIANTS OF MAXWELL TENSOR 225

and choosing i = 1, and j = 2 we see that we get the third component of the first of the
above equations.

Exercise: Compute ∂[i Fjk] , how many equations are these? To what do they correspond?

So we see that the tensor Fµν has all the information of Maxwell fields and also can encode
in a very simple manner Maxwell’s equations. It is called the Maxwell tensor, and it is the
real four dimensional object behind the electromagnetic phenomena. It unifies in one entity
all magnetic and electric phenomena in a natural way.

15.3 Invariants of Maxwell Tensor

Given, Fµν , ηµν , and εµνσρ , how many independent scalars quantities can we make? They
are important for, since they are coordinate independent can tell us quickly whether we
are in presence of some type of solution or another. The quick answer is that there are
just two basic invariants, all others are functions of them, Fµν F µν = η µρ η νσ Fµν Fσρ , and
Fµν ⋆ F µν = 21 εµνσρ Fµν Fσρ . It is instructive, and we shall use this latter, to see their expression
in terms of electric and magnetic fields a given observer sees.
To get the expressions we first show that given a future-directed, unit time-like vector t
there exist two vectors perpendicular to it such that the Maxwell tensor can be decomposed
as follows,

F µν = 2E [µ tν] − εµν σρ B σ tρ (15.9)

Indeed, defining Eµ = Fµν tν we see that F µν − 2E [µ tν] is antisymmetric and perpendicular to
t, so it can be though to be an antisymmetric tensor in a three-dimensional space. But then
we know that it is equivalent to a vector, namely that in this three-dimensional space there
exists a vector B µ such that,

F µν + 2E [µ tν] = ε̃µν σ B σ = tρ ερµν σ B σ

from which the above expression follows. Notice that here we have used the fact that ε̃µνσ =
tρ ερµνσ is the Levi-Civita tensor in the three-dimensional space perpendicular to t.

Exercise: Use the above expression of F to show that B µ = −⋆ F µν tν . For that prove first
that
εµνσρ εκχσρ = −2(δµ κ δν χ − δν κ δµ χ ),
(Hint: the above expression is coordinate independent, so choose any easy coordinate basis
and compute it.)

Exercise: Show that

⋆
F µν = 2B [µ tν] + εµν σρ E σ tρ (15.10)
With the above expression, 15.3, we compute now,
226 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Fµν F µν = 2Eµ E µ tν tν + εµνσρ εµν κχ B σ tρ B κ tχ

= 2Eµ E µ − 2Bµ B µ . (15.11)

Using now 24.2 we get,

Fµν ⋆ F µν = 2E µ Bµ .
How do we know there are no more invariants? What about, for instance, Fµν F µσ Fσκ F κν ?
Notice that since, given any t, unit time-like, F , is equivalent to two vectors perpendicular to
it, then it follows, since vectors are not invariant under rotations on the space perpendicular
to t that all invariants must be also invariant under rotations and so can only depend on
E µ Eµ , B µ Bµ , or E µ Bµ . We already saw that the combinations Eµ E µ − Bµ B µ , and E µ Bµ are
invariants. We shall see later that the remaining combination, namely Eµ E µ + Bµ B µ , is not
an invariant (it changes values for different observers), but rather the component of a tensor.
So there are just two invariants.

Exercise: Show that if at some point of space-time Fµν F µν > 0 and Fµν ⋆ F µν = 0, that is,
if for any observer at that point the electric and magnetic fields are perpendicular and the
magnetic field is bigger in norm than the electric field, then there is an observer for which the
electric field vanishes.

Proof: For some observer t we have,

Eµ = Fµν tν Bµ = −⋆ Fµν tν
while for another, t̃, t̃µ = γ(tµ + β µ ) with tµ βµ = 0, tµ tµ = −1,

Ẽµ = Fµν t̃ν

= Fµν γ(tν + β ν )
= (−2E[µ tν] − εµνσρ B σ tρ )γ(tν + β ν )
= γEµ + (−2E[µ tν] − εµνσρ B σ tρ )γβ ν
= γ(Eµ + tµ Eν β ν ) + γtρ ερµνσ β ν B σ . (15.12)

Since E µ Bµ = 0, we can choose the space coordinate axis so that E µ = (0, E, 0, 0), B µ =
(0, 0, B, 0), we further choose a velocity in the perpendicular direction, β µ = (0, 0, 0, β), we
E
get, Ẽ0 = Ẽ2 = Ẽ3 = 0 and Ẽ1 = γ(E − βB). So, since |B| > |E|, taking β = B < 1 we
obtain Ẽ1 = 0.
In this frame there is no electric force, this is the case when studying electromagnetic fields
in a plasma. There the high mobility of electrons implies that there are no electrical forces
present, that is, the plasma acts like a perfect conductor for time scales larger that those
for which the electrons re-accommodate. Nevertheless, in this situation, magnetic fields are
important and the resulting theory is magneto-hydrodynamics.
15.4. THE ENERGY-MOMENTUM TENSOR 227

Exercise: Show that if for some time-like vector uµ , Fµν uν = 0 then ⋆ Fµν = 2b[µ uν] for some
space-like vector bµ . Show that it also follows that F µν ⋆ Fµν = 0 and F µν Fµν ≤ 0

15.4 The Energy-Momentum tensor

Besides the invariants discussed in the previous section there is another important object, a
tensor, that can be built out of Maxwell’s tensor,

−1 µσ νρ 1
T µν := (F F ησρ − η µν F σρ Fσρ ) (15.13)
4π 4
Note that it is symmetric and its trace vanishes, T := T µν ηµν = 0.
If we contract it twice with a unit, time-like vector tµ we get,

1 1 1
e := T µν tµ tν = (E 2 + (2(−E 2 + B 2 ))) = (E 2 + B 2 ),
4π 4 8π
which we recognize as the energy density of the electromagnetic field that the observer with
4-velocity t sees. Thus, the vector

pµ := T µν tν (15.14)
must be the 4-momentum of the electromagnetic field as seen by that observer. Thus, we see
that for the electromagnetic field the 4-momentum is not a basic object, but it depends on
the observer. Aside for particles all other physical objects have energy momentum tensors
from which observer dependent 4-momenta can be built.

~ and B
Exercise: Check that pµ in terms of E ~ is given by,

P~ 1
pµ := etµ + = ((E 2 + B 2 )tµ − 2tρ ερµσν Eσ Bν ),
c 8π

that is, the 3-momentum is P~ = c ~

4π
(E ~ = S,
∧ B) ~ where S
~ is Poynting’s vector.

Exercise: Check that pµ is time-like or null and future directed.

Proof: We have seen that its time component is positive, so it remains to see whether
~ ~
the vector is time-like or null. That is pµ pµ ≥ 0. So we compute pµ pµ = e2 − Pc·2P =
1 ~ ∧ B|
((E 2 + B 2 )2 − 4|E ~ 2 ). But |E
~ ∧ B|
~ ≤ |E||
~ B|,
~ so (E 2 + B 2 ) ≥ 2|E||
~ B|~ ≥ 2|E
~ ∧ B|
~ and
(8π)2
µ
p pµ ≥ 0.

Exercise: Check that if pµ tν = 0 for any future directed vector t, then at that point, Fµν = 0.
Hint, split the Maxwell tensor in terms of its electric and magnetic parts with respect to tµ .
228 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Exercise: Show that if a given time-like or null future directed vector when contracted with
another time-like future directed vector vanishes, then the given vector itself vanishes.

~ and B.
Exercise: Write down all components of T µν in terms of E ~

Exercise: Using that,

′ ′ ′
εµνσρ εµν σ ρ = −6δ ν [ν ′ δ σ σ′ δ ρ ρ′ ] .
a) Show that,

⋆ 1
F µσ ⋆ Fσν = δ µ ν F σρ Fσρ + F µσ Fσν . (15.15)
2
b) Check that,

⋆
F µν ⋆ Fµν = −F µν Fµν .
c) Write the energy momentum in terms of ⋆ F .

15.4.1 Energy conservation

The energy momentum tensor we have defined have the following important property:

∂µ T µ ν = j µ Fµν (15.16)
by virtue of Maxwell’s equations. Indeed,

−1 1
∂µ T µ ν = ((∂µ F µσ )Fνσ + F µσ ∂µ Fνσ − ∂ν (F σρ Fσρ ))
4π 4
−1 σ µσ 1
= (−4πj Fνσ + F (∂µ Fνσ − ∂ν Fµσ ))
4π 2
σ 1 µσ
= −j Fσν − F (∂µ Fνσ − ∂µ Fσν − ∂ν Fµσ )
8π
1 µσ
= −j σ Fσν − F (∂µ Fνσ + ∂σ Fµν + ∂ν Fσν )
8π
3 µσ
= −j σ Fσν − F (∂[µ Fνσ] )
8π
= −j σ Fσν (15.17)

where in the second line we have used one of Maxwell’s equations, in the third and fourth the
antisymmetry of F and in the fifth the other Maxwell equation.
To see the relevance of the above formula take a constant unit time-like vector t, (so in
some Cartesian coordinate system it has components tµ = (1, 0, 0, 0), when writing coordinate
15.4. THE ENERGY-MOMENTUM TENSOR 229

components we shall refer to them), and define pµ := T µν tν , the 4-momentum that this
observer assigns to the electromagnetic field. We have,

∂µ pµ = ∂µ (T µν tν )
= (∂µ T µν )tν T µν ∂µ tν
= −j σ Fσ ν tν
J~ ~
= · E, (15.18)
c
where we have used the constancy of t, namely ∂µ tν = 0. So we see that locally the energy
~ in the times the velocity of the
momentum loss is just the work done by the electric force ρE
particles, ~v .
We shall consider what happens with this result when considering a finite region, as the
one in the figure 15.1

Σ1 −t
t=T

n
T

t
Σ0
t=0

Figure 15.1: Energy conservation

Using the divergence theorem in four dimensions we have,

Z Z Z Z
∂µ pµ d4 x = pµ n0µ d3 x + pµ nTµ d3 x + pµ Nµ d3 x (15.19)
V Σ0 ΣT T

where the integral on the left is over a volume in space-time, and the others are over three-
dimensional hypersurfaces as shown in the figure. The normal vectors point outwards to
the hypersurfaces at t = 0 and t = T are respectively, n0µ = (−1, 0, 0, 0) = −tµ , and nTµ =
(1, 0, 0, 0) = tµ . 2 Using the above result we then have,
Z Z Z Z
pµ tµ d3 x = pµ tµ d3 x + pµ Nµ d3 x − J~ · E
~ d4 x. (15.20)
ΣT Σ0 T V

That is,
2
To see that these are the correct signs for the integrals over the t = 0 and t = T hypersurfaces assume pµ
has only time component and perform the integral of ∂0 p0 along x0 .
230 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Z Z Z Z
p0 d 3 x = p0 d 3 x + pµ Nµ d3 x − J~ · E
~ d4 x, (15.21)
ΣT Σ0 T V

the energy at ΣT is the energy at Σ0 minus the integral over the side-boundary and the work
done by the field to the charges, if any is present.
So we now examine the side-boundary term, taking any vector N perpendicular to t, that
is spatial, one sees that,

1 µσ 1
pµ Nµ = [F Fνσ tν − tµ F σρ Fσρ ]Nµ
4π 4
−1
= Eσ F µσ Nµ
4π
−1
= Eν Nµ (−2E [µ tν] − εµν σρ B σ tρ )
4π
1
= Eν Nµ εµν σρ B σ tρ
4π
−1 ρ
= t ερµνσ N µ E ν B σ
4π
1 ~ ~ ~
= N · (E ∧ B),
4π
1~ ~
= N ·S (15.22)
c

is Poynting’s vector, S ~ = cE ~ ∧B~ 3 Thus, we conclude that the change in energy form
4π
hypersurface Σ0 to hypersurface ΣT due to volumetric work done upon the charges integrated
over time and energy flux losses, the (time×2-surface) integral along the boundary of the
region. In the case the boundary is tangent to the time direction (so its normal is perpendicular
to t) that is the space integration region is the same at all times for this observer, the energy
flux is just Poynting’s vector.

Exercise: Assume the region under consideration is moving with four velocity uµ = γ(tµ +βµ )
with respect to the observer t with respect to which we measure energy. See figure 15.2. What
is the energy flux in this case?

Exercise: Show that pµ uµ = T µν tν uµ > 0 if both t, and u are time-like and future directed.

Proof: This follows from the fact that p is time-like or null and future directed, but it is
instructive to do the following calculation, taking, for simplicity, both vectors to have norm
one we and so uµ = γ(tµ + β µ ), with η(t, β) = 0, we have,

3 1
The factor c in the expression above is due to the fact that on the right-hand side we integrate power in
units of ct.
15.4. THE ENERGY-MOMENTUM TENSOR 231

Σ1
−t t=T

u
n T

t
Σ0
t=0

Figure 15.2: Moving region

pµ uµ = γ(pµ tµ + pµ βµ )
γ
= (E 2 + B 2 − 2β~ · (E
~ ∧ B))
~
8π
γ
≥ (E 2 + B 2 − 2βEB)
8π
γ
≥ (E 2 + B 2 )(1 − β)
8π
≥ 0,
~ ∧ B)|
since β < 1, |(E ~ ≤ |E||
~ B|,
~ and E 2 + B 2 ≥ EB. Notice that this quantity is zero only
when both electric and magnetic fields vanishes.

Exercise: Use the above result to show, that if two solutions to Maxwell equations coincide
at Σ0 then they must coincide inside a region like the one shown in figure 15.3 as long as the
normal to ΣT is time-like.

t
ΣT u

Σ0
t=0

Figure 15.3: Unicity of solutions

15.4.2 Momentum Conservation

Let us consider now a constant, unit, space-like vector, k. And, for simplicity we take a
constant time-like vector perpendicular to it, t, t · k = 0. From 15.17, we see the vector
232 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Tk := T µν kν satisfies,

∂µ Tkµ = ∂µ (T µν kν ) = j σ Fσ ν kν
Performing the same integral as before over a surface perpendicular to t, we get,

Z Z Z Z
− ∂µ Tkµ d4 x = T µν kν n0µ d3 x + T µν kν nTµ d3 x + T µν kν Nµ d3 x
V Σ0 ΣT T
Z Z Z
µν 3 µν 3
= T kν tµ d x − T kν tµ d x + T µν kν Nµ d3 x
Σ0 ΣT T
Z Z Z
ν 3 ν 3
= −p kν d x − −p kν d x + T µν kν Nµ d3 x
Σ0 ΣT T
(15.23)

So we see that from this equation we can get the change in the space components of the
energy-momentum 4-vector. Re-arranging this expression we get,
Z Z Z Z
pν k ν d 3 x = pν k ν d 3 x + T µν kν Nµ d3 x − j σ Fσ ν kν d4 x (15.24)
ΣT Σ0 T V

Again we see that the change in momentum has a volumetric term and a flux term. Notice
~ · ~k = 1 (E
that pµ kµ = −T µν tν kµ = 1c S ~ · ~k, so that the 3-momentum density that the
~ ∧ B)
4π
observer with 4-velocity t sees is the Poynting vector.
We now compute the momentum flux along k, −T µν Nµ kν , and the volumetric source,
−jµ F µν kν .
We first compute,

~ · ~k − tρ ερµσν Bσ kν
F µν kν = −(E µ tν − E ν tµ )kν − εµνσρ Bσ tρ kν = tµ E
So the momentum source becomes,

~ ~
~ · ~k − J · (B
−jµ F µν kν = jµ tµ E ~ · ~k − ( J ∧ B)
~ ∧ ~k) = −ρE ~ + 1 (J~ ∧ B))
~ · ~k = −(ρE ~ · ~k
c c c
So we see that the source of 3-momentum change is the Lorentz force. We turn now to
momentum flux,

1 µσ ν 1
T µν Nµ kν = [F F σ Nµ kν − ~k · N ~ F µν Fµν ]
4π 4
1 ~ ·N
~ − tρ ερσµν Bµ Nν )(−tσ E ~ · ~k − tρ ερσ µν Bµ kν ) + 1 (~k · N
~ )F µν Fµν ]
= [(−tσ E
4π 4
1 ~ · N)(
~ E ~ · ~k) + (B
~ ∧N ~ ∧ ~k) + (E
~ ) · (B 1 ~ ·E~ −B ~ ~k · N)]
~ · B)( ~
= [−(E
4π 2
1 ~ · N)(
~ E ~ · ~k) + B ~ ~k · N
~ · B( ~ ) − (B ~ · N)(
~ B ~ · ~k) + 1 (E~ ·E ~ −B ~ ~k · N)]
~ · B)( ~
= [−(E
4π 2
1 ~ ~ ~ ~k · N)
~ · B)( ~ − 2(E ~ ·N ~ )(E~ · ~k) − 2(B~ ·N ~ · ~k)]
~ )(B
= [(E · E + B (15.25)
8π
15.4. THE ENERGY-MOMENTUM TENSOR 233

which we recognize as Maxwell Stress tensor, 2.2.

Thus, we see that:

1. Momentum density is proportional to the energy flux, i.e. the Poynting vector.

2. The volumetric source of momenta is the densitized Lorentz force.

3. The momentum flux is Maxwell’s stress tensor, so in a static situation, when no mo-
mentum change occur, the total electromagnetic force over an object is proportional to
the surface integral of Maxwell stress tensor. T ij = −1
4π
[E i E j + B i B j − 21 eij (E 2 + B 2 )]

15.4.3 Angular Momentum conservation

The conservation theorems we have presented above used constant vectors to define, when
contracted with the energy-momentum tensor, other vectors whose four divergence was known
in term of sources and F . But constancy was a sufficient condition but it is not necessary
to define 4-vectors with known divergences. Indeed, contracting T µν with kµ and taking the
divergence in the index left we get,

∂µ (T µν kν ) = (∂µ T µν )kν ) + T µν ∂µ kν .
Thus, to have a conservation law we only need a vector kµ such that T µν ∂µ kν = 0, in particular,
since T µν is symmetric, T µν = T νµ , it follows that there are conservation laws for all vectors
such that ∂(µ kν) := 12 (∂µ kν + ∂ν kµ ) = 0. They are called Killing vectors. It can be shown
that in Minkowski space-time there are 10 linearly independent vectors with this property.
We have already seen four of them, one time translation and three space translations. The
rest can be constructed taking arbitrary constant antisymmetric tensors Aµν and defining,

k µ = εµνρσ xν Aρσ ,
where xµ = ηµν xν , are Cartesian coordinate functions.

Exercise: Check that ∂(µ kν) = 0 for the above vectors.

Notice, in particular, that

k µ = εµνρσ xν t̂ρ ẑσ ,

with t̂ρ = (−1, 0, 0, 0), and ẑσ = (0, 0, 0, 1) corresponds to

k µ = (0, y, −x, 0),

and the conserved quantity is angular momentum along the ẑ axis.

Exercise: Using a constant vectors base construct the remaining five quantities. Interpret
them as motions in space-time.
234 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

Exercise: The above definitions depend on the coordinate origin, show that if one change
origin then the above vectors change by a translation.

The formulas used to express the conserved quantities, fluxes and sources for space and
time translations can be used also for angular momentum, making the appropriate substitu-
tions, of course. They shall be used latter in an example.

~ and B
Exercise: Express in terms of E ~ the angular momentum along the ẑ axis.

Exercise: Find the angular momentum flux. Find the total torque an electromagnetic field
exerts over a static source in terms of a surface integral.

15.5 Killing Vector Fields in Minkowski space-time.

A killing vector field in Minkowski space-time is a vector field, k, which satisfies,

∂(µ kν) = 0.
In the presiding section we have seen that there is at least ten (10) of them which are
linearly independent, four (4), are just constant vectors, taking a constant time-like vector
t, we can split the remaining six into three (3) infinitesimal rotations, given by Aµν = tµ eνi
where the vectors ei represent the rotation axis, which for simplicity we take to be three
orthonormal space-like vectors perpendicular to t, and three (3) boosts , which we take to
be, those vectors obtained from Aµν = eµi eνj , i 6= j. They are just proper infinitesimal Lorentz
transformations along a 2-plane perpendicular to both frame vectors.
These vectors represent symmetries of the space-time and so are related to conserved
quantities. So it is important to know how many of them there are, for we will have as many
conserved quantities as Killing vectors.
To see this, we shall the following argument [2]: note that if we define

Lµν := ∂[µ kν] = ∂µ kν ,

where in the last equality we have used Killing’s equations.
We have, as for the vector potential integrability condition,

∂[σ Lµν] = 0.
Thus,

∂σ Lµν = −2∂[µ Lν]σ = −2∂[µ ∂ν] kσ = 0,

since partial derivatives commute
Thus, we see that the anti-symmetric derivatives of k must be constants. Taken these
constant to zero we obtain the four constant Killing vector fields. Taking different constant
15.5. KILLING VECTOR FIELDS IN MINKOWSKI SPACE-TIME. 235

anti-symmetric matrices Lµν (there are six linearly-independent) we obtain the other six
Killing vector fields. So we see that these are all possible (up to linear constant combinations)
infinitesimal symmetries of space-time.

Exercise: Prove that if a Killing vector field vanishes at a point together with its first
derivatives then it is the trivial vector field.

15.5.1 Conformal Killing Vector Fields.

Electrodynamics has further symmetries that come from the fact that the energy momentum
tensor is trace free (T µν ηµν = 0). Indeed, computing

∂µ (T µν kν ) = (∂µ T µν )kν + T µν ∂µ kν ,
we see that not only the last term vanishes when ∂(µ kν) vanishes, but also if it is proportional
to ηµν , that is

∂(µ kν) = φηµν .

This is called a conformal Killing vector field . Obviously φ = ∂µ k µ /4. How many of
these vector fields there are in Minkowski space-time? We already have the previous ten,
φ = 0. Are there more?
To see this we use the same argument as before, but now taking into account the fact that
the trace does not vanish.
For this case, we define,

Lµν := ∂[µ kν] = ∂µ kν − φηµν ,

and obtain,

∂σ Lµν = −2∂[µ Lν]σ = −2∂[µ (∂ν] kσ − φηνσ ) = 2ησ[ν Lµ] ,

where we have defined Lµ := ∂µ φ.
Taking another derivative and anti-symmetrizing we get,
1
0 = ∂[ρ ∂σ] Lµν = [ησν ∂ρ Lµ − ησµ ∂ρ Lν − ηρν ∂σ Lµ + ηρµ ∂σ Lν ].
2
σρ
Contracting with η we get,
1
0 = ∂ρ Lµ + ηρµ ∂ ν Lν ,
2
from which it follows that

∂ρ Lµ = 0.
Thus, we have another five possibilities, namely we can give the value of φ at some point and
the constants Lµ and obtain 5 extra conformal killing vector fields.
236 CHAPTER 15. RELATIVISTIC ELECTRODYNAMICS

If we take φ = 4, and Lµ = 0 we get, ∂µ kν = ηµν which can be integrated to give,

kµ = ηµν xν or k µ = xµ ,
this corresponds to an infinitesimal dilation or scale symmetry, namely a transformation
which changes the units of time and space in the same way.
If we choose at a point φ = 0, (and set it as the coordinate origin), and some value for
Lµ , then we have,

φ = Lµ xµ ,
and,

∂σ Lµν = ησν Lµ − ησµ Lν

which can be integrated to,

Lµν = 2L[µ xν] .

Thus,

∂µ kν = Lµν + φηµν = 2L[µ xν] + ηµν Lσ xσ ,

which in turn can be integrated to,
1
kν = Lσ xσ xν − xσ xσ Lν .
2
This completes the set of space-time symmetries of Maxwell’s equations.
Chapter 16

Vector Potential - Connection

16.1 Introduction
In the 3 + 1 dimensional treatment of Maxwell’s equations it is very useful to introduce
potentials so that certain set of equations are solved automatically and so, one can concentrate
in the rest. This idea can be further carried on in the four dimensional formalism but here it
acquires a different and more profound meaning. The potentials are not merely a simple way
of obtaining solutions or solving some equations, but have a geometrical meaning as curvature
carriers for a whole set of theories, called en general Gauge theories or Yang Mills theories.
In their more extended version they represent all force fields (bosonic fields) in the Standard
Model of particle physics, which, of course, includes among them the electromagnetic forces.
In the four dimensional setting we have just two set of Maxwell’s equations,

∂ µ Fµν = 4πJν (16.1)

and

∂ν ⋆ F νµ = 0. (16.2)
While there has been numerous experiments trying to detect sources for the second equations,
notably magnetic monopoles, none of them have give any positive result. The tendency of
theoretical physicists is to think about the second equation as intrinsically source-less and in
that case as acting as a integrability condition or compatibility condition to the existence of
another field of a more fundamental character than the tensor F , namely the existence of a
4-vector, A, such that
Fµν = 2∂[µ Aν] . (16.3)
Indeed, it easily follows, from the fact that partial derivatives commute, that

∂[µ Fνρ] = 2∂[µ ∂ν Aρ] = 0

and we have seen that this equation is equivalent to 16.2, so that this way we have already
solved one of Maxwell’s equations, and we can concentrate on the other.
It can be further shown that if a field Fµν satisfies,

237
238 CHAPTER 16. VECTOR POTENTIAL - CONNECTION

∂[µ Fνσ] = 0,
namely 16.2, then locally there always exists a vector potential, Aµ , in fact infinitely many,
such that 16.3 holds. Thus, we see that this equation is equivalent to the existence of vector
potentials.
Suppose now we have an Fµν and we have a potential for it, Aν , that is Fµν = 2∂[µ Aν] ,
then given any smooth function λ, õ := Aµ + ∂µ λ will also give the same Maxwell field,
indeed,

2∂[ν õ] = 2[∂[ν Aµ] + ∂[ν ∂µ] λ] = 2∂[ν Aµ] = Fνµ ,

since partial derivatives commute. Thus, we see that given F there is no unique A. How big
is this freedom of choosing different A’s? It can be shown that it is just the one shown above,
namely the addition of the differential of a function. This freedom is called the gauge freedom
and fields with this indeterminacy are called gauge fields. So the physical field is not the 4-
vector A but an equivalence class of fields, where we say that field A is equivalent to field Ã,
if their difference is the gradient of some function, Aµ − õ = ∂µ λ. So we arrive at something
new in physics, a physically relevant entity is not a tensor field, but an equivalent class of
them. To some extent we have a similar situation for the scalar potential in electrostatics,
since it was defined at the energy difference needed to carry a charge between two points it
has an indeterminacy of an overall constant. Here the indeterminacy is different at each point
of space-time, it is not the 4-vector A which has a meaning, but only the equivalent class to
which it corresponds. All physically relevant quantities we measure have to be independent
of this freedom, that is, they have to be functions on the equivalent class.

Exercise: Given a constant unit time-like vector t define Aµ = −φtµ + õ that is, φ = A0 =
A · t = A0 and à · t = 0. Check that,

B µ = ε̃µνσ ∂ν Aσ := tρ ερµνσ ∂ν Aσ ,

and
Eµ = ∂µ φ − tµ ∂0 φ − ∂0 õ = ∂˜µ φ − ∂0 õ ,
where ˜ means projection to the simultaneity surface. Check explicitly that both vectors
~ = −∇φ,
are gauge invariant. Notice that with this definition E ~ consistent with the usual
definition.

16.2 Gauge equation

We now concentrate in the remaining Maxwell equation, 16.1, which in terms of A becomes,

∂ µ Fµν = ∂ µ (∂µ Aν − ∂ν Aµ ) = ✷Aν − ∂ν (∂ µ Aµ ) = 4πJν (16.4)

This equation is not an evolution equation in the sense discussed for Maxwell’s equations
in the three-dimensional formulation. This follows easily from the fact that A has a gauge
16.2. GAUGE EQUATION 239

freedom, or indeterminacy, indeed if we substitute Aν by Aν + ∂ν λ where λ is any arbitrary

function in space-time, we obtain the same tensor F and so if Aν satisfied the above equation
so does Aν + ∂ν λ. Since λ is arbitrary, it can not be determined from any initial data given
at any t = t0 hypersurface. Indeed, choosing a λ function which is identically zero in a whole
neighborhood of such hypersurface but is non-vanishing latter in time we obtain two potential
vectors which are identical in a whole neighborhood of the hypersurface t = t0 , but differ at
late times. Thus, to obtain a deterministic evolution equation, an evolution equation which
will give a unique solution for given initial data, we must find a way to eliminate the gauge
freedom, that is fix a gauge. There are at least three interesting ways of doing this, and we
shall see them in the sections that follow.

16.2.1 Lorentz Gauge

We choose among all equivalent 4-potentials those which satisfy

∂µ Aµ = 0. (16.5)
To see that this is possible consider an arbitrary potential, Aµ , then make a gauge transfor-
mation, õ = Aµ + ∂µ λ, then we have,

∂µ õ = ∂µ Aµ + ∂ µ ∂µ λ
So we can solve for λ the equation,

✷λ = −∂µ Aµ
and the new potential would satisfy Lorentz gauge.
Notice that this does not fix completely the gauge freedom, the arbitrariness is the same
as the set of solutions of the wave equation, namely we can prescribe the value of λ and of
∂0 λ at some t = t0 hypersurface.
For this choice of gauge Maxwell’s equations reduces to just the wave equation,

✷Aµ = 4πj µ. (16.6)

So we know that giving j µ in space-time, and the values of Aµ and ∂0 Aµ at t = t0 there will
be a unique solution to the above equation, namely a unique A in the whole space-time. Does
the tensor F built out of this vector potential, A satisfies Maxwell’s equations? For that we
need to see that our A so constructed satisfies the gauge we imposed on the equation, namely
we have to check whether 16.5 holds.
To see that take the 4-divergence of the above equation to obtain,

∂µ ✷Aµ = ✷∂µ Aµ = 4π∂µ j µ = 0. (16.7)

So if we can prescribe initial data so that ∂µ Aµ = 0 and ∂0 ∂µ Aµ = 0 at t = t0 , then uniqueness
of the solutions to the wave equation would imply that ∂µ Aµ = 0 everywhere in space-time,
and so we would have checked consistency of the gauge choice. Thus, consistency of the gauge
fixing implies that not all initial data can be prescribed arbitrarily, in particular we must have,
240 CHAPTER 16. VECTOR POTENTIAL - CONNECTION

∂0 A0 |t=t0 = −(∂i Ai )|t=t0 = −∂i (Ai |t=t0 )

∂02 A0 |t=t0 = (✷A0 + ∆A0 )|t=t0 = (4πρ + ∆A0 )|t=t0 = 4πρ|t=t0 + ∆(A0 |t=t0 ). (16.8)

On the other hand,

∂0 ∂µ Aµ = 0 ⇒ ∂02 A0 = −∂i ∂0 Ai ,
so we have,

∂0 A0 |t=t0 = −(∂i Ai )|t=t0 = −∂i (Ai |t=t0 )

∆(A0 |t=t0 ) = −4πρ|t=t0 − ∂i (∂0 Ai )|t=t0 . (16.9)

Thus, we see that given Ai and ∂0 Ai at t = t0 , then values for A0 and ∂0 A0 are determined
by the above equations. So in this gauge the free initial data for Maxwell’s are the pairs
(Ai , ∂0 Ai ).

Exercise: Check that the equation for A0 (last equation) is gauge invariant. Confirm that
~ ·E
equation is just ∇ ~ = 4πρ

Exercise: Use the remaining gauge freedom, (in choosing λ at t = t0 ), to set,

∂0 A0 |t=t0 = 0
∆(A0 |t=t0 ) = −4πρ|t=t0 . (16.10)

16.2.2 Coulomb Gauge

In this case the gauge fixing is done by requiring,

∂i Ai = 0 (16.11)
This gauge is not space-time invariant, but it is very convenient in some calculations for it
resembles what happens in electrostatics and magnetostatics.
In this case the equations become,

∂ µ Fµν = ∂ µ (∂µ Aν − ∂ν Aµ ) = ✷Aν − ∂ν (∂0 A0 ) = 4πJν (16.12)

and so, in components we have,
16.2. GAUGE EQUATION 241

∂ 0 ∂0 A0 − ∂ i ∂i A0 − ∂ 0 ∂0 A0 = −∂ i ∂i A0 = 4πρ
✷Ai − ∂ i (∂0 A0 ) = 4πj i (16.13)

In this gauge one solves for A0 at each time slice, t = cons., using the first equation above
and then plugs the result into the second equations as a source to solve for Ai .
The consistency of the gauge fixing condition is granted by requiring for the initial condi-
tions for the second equation (Ai , ∂0 Ai )|t=t0 that they satisfy,

∂i Ai |t=t0 = ∂i (∂0 Ai )|t=t0 = 0. (16.14)

Indeed, we have,

✷∂iAi = ∂i ✷Ai
= ∂i ∂ i (∂0 A0 ) + 4π∂i j i
= ∂0 ∂i ∂ i A0 + 4π∂i j i
= 4π∂0 ρ + 4π∂i j i
= 4π∂µ j µ
= 0. (16.15)

So the conditions on the initial data ensure that the unique solution to this equation is
∂i Ai = 0 and the gauge fixing condition is granted.
It can be shown that given Fµν |t=t0 there is a unique initial data pair, (Ai , ∂0 Ai )|t=t0 such
that 16.14 holds. Notice that in this gauge,

~ = −∇φ
E ~ + ∂0 A,
~
with,

∆φ = −4πρ, ~ · (∂0 A)
and ∇ ~ = 0,
at all times.

Exercise: Check that given Fµν |t=t0 satisfying initially Maxwell’s equations, with ∂ i F0i and
∂ i Fij of compact support. There is a unique pair (Aµ |t=t0 , (∂0 Aµ )|t=t0 ) satisfying, Fµν =
2∂[µ Aν] , ∂µ Aµ = 0, (∆A0 + 4πρ)|t=t0 = 0, ∂0 A0 |t=t0 = 0, and decaying sufficiently fast at
infinity.

Proof: The Lorentz gauge plus the condition, ∂0 A0 |t=t0 = 0 implies, that ∂ i Ai |t=t0 = 0,
so we assume this. Consider now the relation, Fij = 2∂[i Aj] . Taking the divergence of this
expression we get,
∂ i Fij = ∆Aj − ∂j ∂ i Ai = ∆Aj
This is Poisson’s equation for each Aj , so if ∂ i Fij is of compact support, as assumed, there is
a unique Ai satisfying it and decaying at infinity.
242 CHAPTER 16. VECTOR POTENTIAL - CONNECTION

1 ~ ~ 1 i
The equation (∆A0 +4πρ)|t=t0 = 0 fixes uniquely A0 if ρ = 4π ∇· E = 4π ∂ Fi0 is of compact
0
support and A is assumed to decay sufficiently fast. Thus, Fi0 = ∂i A0 − ∂0 Ai determines
∂0 Ai uniquely. Notice, that
0 = ∂ i Fi0 − ∆A0 = ∂ i ∂0 Ai .
So this vector potential initially satisfies Coulomb gauge.

16.2.3 Temporal Gauge

In this case the gauge fixing is done requiring that A0 := 0.
In this case the equations become,

∂0 (∂i Ai ) = 4πρ
✷Ai − ∂ i (∂j Aj ) = 4πj i, (16.16)

so the first equation can be integrated in time to find ∂i Ai , this value can then be inserted into
the second equation and then this one is solved for Ai . Again, charge conservation ensures
consistency.

Exercise: Check the last assertion.

Chapter 17

Variational Formalisms

17.1 Introduction
Variational formalisms, that is the treatment of systems of equations as coming from a varia-
tional principle, are central in physics for they allow a different view point to understand these
systems, and, in many cases, to apply powerful mathematical techniques to solve them or to
find out how they behave. We start with a mechanical, finite dimensional system, namely
the variational principle ruling the motion of a charged particle in the presence of a given,
fixed, electromagnetic field. We then present the variational principle leading to Maxwell’s
equations interacting with charged matter. What we show here is just a glimpse of the beauty
of these theories and, in order not to lose the scope and purpose of this book we shall not dis-
cuss the many interesting properties and ramifications these approach renders. But we hope
the reader will get interested and will return to these topics at future instances of his/her
formation.

17.2 Variational Dynamics of a Charged Particle

We have already seen that the time interval along a trajectory, 14.10, namely the integral,
s
1 dxµ dxν
Z
TAB := ηµν dτ. (17.1)
0 dτ dτ
is maximal for a straight one, namely the trajectories of free particles. So we now ask the
question whether one can describe the interaction of a charged particle as coming from some
variational principle, so as being the consequence of the maximization (or more generally
and extreme) of some quantity. From our experience in non-relativistic mechanics we have
some reasonable expectation this can be accomplished, for after all most classical dynamical
systems allow for such principles. On the other hand, the integral we used above was a bit
strange, in the sense that was not just the square of the velocity expression we are used to
in the non-relativistic setting, and recall, we had the constraint in the possible equations of
motion that the 4-velocity of particles was determined only up to a local scale. A scale which
we took care by normalizing the velocities to 1 and requiring the forces to be perpendicular
to the velocities.

243
244 CHAPTER 17. VARIATIONAL FORMALISMS

The simplest expression which we can build out of the four velocity and the electromagnetic
field which is re-parametrization invariant is,

s
1 dxµ dxν e dxµ
Z
IAB := [m0 c ηµν + Aµ (xν (τ )) ] dτ. (17.2)
0 dτ dτ c dτ

This expression depends on the potential and so the first question we must ask is whether it is
gauge invariant, otherwise any conclusion we could obtain from it would not be physical. We
have included the mass coefficient in the first term, m0 c so that the function has the dimensions
of an action, and have included another coefficient, e, which would be the definition of charge.
Since the equations will appear as a result of an extremal principle they can not depend on
the over all magnitude of this integral. Thus, the interaction will only depend on the ratio
e
m0 c2
.
Changing the 4-potential by the gradient of a function, Aµ → Aµ + ∂µ λ, we obtain,

δIAB = IAB (Aµ + ∂µ λ) − IAB (Aµ )

e 1 dxµ
Z
= ∂µ λ dτ.
c 0 dτ
e 1 dλ
Z
= dτ
c 0 dτ
e
= [λ(B) − λ(A)] (17.3)
c

Thus we see that this quantity is not gauge invariant, so it is not a physical quantity. Nev-
ertheless, the gauge dependence is only through the values of the field at the extremes of
the integration curve, and so variations which vanish at the extreme will still give gauge
independent equations. Indeed, looking at a one parameter families of curves of the form,
xµε := xµ0 + εxµ1 , with fixed extremes, namely, xµ1 (0) = xµ1 (1) = 0, we obtain,

dxµ dxν
dIAB 1 m0 c 0
ηµν dτ1 e dxµ dxµ
Z
|ε=0 = [ dτ
+ (xν1 ∂ν Aµ 0 + Aµ 1 )] dτ, (17.4)
dxµ dxν
q
dε 0 ηµν dτ0 dτ0 c dτ dτ

where we have used that,

dAµ dxν ∂Aµ

|ε=0 = |ε=0 ν |ε=0 = xν1 ∂ν Aµ (xσ0 ).
dε dε ∂x

dxµ
As for the case of a free particle we now take a parametrization on which the norm of dτ
0

is constant and integrate by parts to obtain,

17.3. VARIATIONAL DYNAMICS OF MAXWELL’S FIELDS 245

d2 xµ
dIAB 1 m0 c
2 ηµν x
0 ν
e ν dxµ0 dAµ µ
Z
|ε=0 = [− q dτ µ dxν1 + (x1 ∂ν Aµ − x1 )] dτ − eAµ xµ1 |10
dε 0 ηµν d0dτx dτ0 c dτ dτ
d2 xµ
m0 c dτ 20 ηµν xν1 e dxµ ∂Aµ dxν0 µ
Z 1
= [− q + (xν1 ∂ν Aµ 0 − x )] dτ
0 dxµ dxν
ηµν dτ0 dτ0 c dτ ∂xν dτ 1
d2 xµ
m0 c dτ 20 ηµν xν1 e dxµ0 ν
Z 1
= [− q + (∂ν Aµ − ∂µ Aν ) x ] dτ
0 dxµ dxν
ηµν dτ0 dτ0 c dτ 1
d2 xµ
m0 c dτ 20 ηµν e dxµ0 ν
Z 1
= [− q + (∂ A
ν µ − ∂ A
µ ν ) ]x1 dτ
0 dxµ ν
0 dx0 c dτ
ηµν dτ dτ
d2 xµ
1
2 ηµν
m0 c
0
e dxµ
Z
= [− q dτ µ ν + Fνµ 0 ]xν1 dτ
0 dx dx
ηµν dτ0 dτ0 c dτ

So the condition for the curve xµ0 (τ ) to be an extremal for the above quantity under arbitrary
perturbation curves with fixed endpoints implies it must satisfy,
µ
d2 xν0 ν dx0
m0 2 = eF µ = F ν µj µ,
dτ dτ
where we have already chosen the proper time parametrization for the extremal curve. We
see that indeed we get the Lorentz force expression on the right-hand side.
Notice that the expression for the action, 17.2, depends on the potential A instead of
the Maxwell tensor F . So we see that variational principles need the potential as a basic
component of the theory and here it is no longer a helpful way of getting rid of an equation,
acquiring a more foundational character. It is interesting that defining the Lagrangian for
this theory as,
s
dxµ dxν e dxµ
L := m0 c ηµν + Aµ (xν (τ )) ,
dτ dτ c dτ
we get as the associated momentum,
∂L dxµ e
pµ := dxµ = m0 c + Aµ
∂ dτ dτ c
and we see that now the momentum is no longer a property of the particle itself, but also
depends on the electromagnetic field present at the particle position. This property is related
to the gauge character of the theory and has a natural extension to other theories where
particles are seen as fields.

17.3 Variational Dynamics of Maxwell’s Fields

We now turn to Maxwell’s fields and their dynamics as deduced from a variational principle.
So the question is now: Does there exists an action integral such that its variation will result
246 CHAPTER 17. VARIATIONAL FORMALISMS

in Maxwell’s equations?
Notice that we want to vary an object with respect to some tensorial fields and obtain
equations, another tensorial objects. These equations are partial differential equations, so the
object we vary can not be a line integral, as was conventional in classical mechanics, that
would only give us ordinary differential equations, namely equations where the derivatives are
with respect to the parameters involved in the integration. So for this case the integral should
be over space-time, or at least part of it. The integrand should then be a scalar function over
the basic fields of the theory, F , or the vector potential associated to it, A. But we have
already seen that there are at most two scalars one can construct out of the tensor F , Fµν F µν
and Fµν ⋆ F µν . The same result holds if we use A and furthermore require gauge invariance.
Furthermore, we want the action to include only terms up to second order in powers of F ,
since otherwise the variation would give non-linear equations.
Thus, not much room is left and the action integral has to be of the form,
1
Z
I := [ (Fµν F µν + RFµν ⋆ F µν ) + j µ Aµ ] d4 x. (17.5)
space−time 16π
We have also added the same factor we had for the charged particle, it is the same factor, but
now to represent the interaction between charge matter and the electromagnetic field from
the side of the electromagnetic source terms. Again this is the only scalar term we can make
which is both linear in the fields and in the sources, notice that we need to use the 4-vector
1
potential, we can not get a scalar out of j and F alone. The factor 16π is to get the correct
equations when coupled to the sources, the other factor, R, is arbitrary, and we shall see that
its value is irrelevant for that part of the action does not contribute to the variation.
Again we have to check whether this term is gauge invariant, under a gauge transformation
Aµ → Aµ + ∂µ λ, we get,

Z
δI = j µ ∂µ λ d4 x
space−time
Z
= [∂µ (j µ λ) + λ∂µ j µ ] d4 x
space−time
Z
= Nµ j µ λ d3 x,
∂(space−time)

where we have used charge conservation in the second line. The integral √ of the last line has to
be understood as the limit r → ∞ of an integral over the surface r = t2 + x2 + y 2 + z 2 . So,
again, for gauge transformations which are of compact support, or which vanish sufficiently
fast at both, space and time directions the action is gauge invariant. 1
We shall perform variations over a region D of space-time bounded by two space-like
hyperplanes, Σ1 and Σ2 , one to the future of the other, both going like parallel planes towards
infinity. Imagine now we have some field A0 which we assume are an extrema of the action
1
It is difficult to make these formalisms to be mathematically correct in the sense that some of these space-
time integrals might be infinite at the extreme points, for there are many solutions which are of compact
support on space directions, but can not be so along time directions (basically by uniqueness or energy
conservation). So they are usually formal in more than one sense.
17.3. VARIATIONAL DYNAMICS OF MAXWELL’S FIELDS 247

and look at variations along arbitrary one parameter families of fields of the form Aε :=
A0 + εA1 , and consequently F ε := F 0 + εF 1 . We shall request the variations to vanish at
both hyperplanes, A1 |Σ1 = A1 |Σ2 = 0, and to decay sufficiently fast at infinity. 2
We then get,

dI Z
1
|ε=0 = [ (F0µν 2∂µ A1ν + 2R ⋆ F0µν ∂µ A1ν ) + j ν A1ν ] d4 x
dε D 8π
Z
1 1
= [ ∂µ (F0µν A1ν + R ⋆ F0µν A1ν ) + ∂µ (F0µν + R ⋆ F0µν )A1ν ) + j ν A1ν ] d4 x
D 4π 4π
1 1
Z Z
= [− ∂µ F0µν + j ν ]A1ν d4 x + Nµ (F0µν A1ν + R ⋆ F0µν A1ν ) d3 x
D 4π 4π ∂D
where in the last line we have used that ∂µ ⋆ F0µν = 21 ∂µ εµνσρ Fσρ = ∂µ εµνσρ ∂σ Aρ = 0 . Thus,
for the field A0µ to be an action extrema under arbitrary variations of compact support we
need,

∂µ F0µν = 4πj ν
and so Maxwell’s equations are satisfied.

2
Again, mathematically, this kind of variations are difficult to define, the solutions do not decay as fast
as we would like to at infinity and so the finiteness of the integral is delicate, furthermore the set of field
solutions which vanish at the extreme surfaces does not have a nice topology.
248 CHAPTER 17. VARIATIONAL FORMALISMS
Chapter 18

Plane Waves

18.1 Definition and Properties

In this chapter we shall look for special solutions of vacuum Maxwell equations, in principle
they are very particular solutions, but in some sense they form a complete set of solutions,
so we can write any vacuum solution as a (infinite) linear combination of them. Also they
represent light rays namely electromagnetic radiation flowing in a very particular direction.
Thus, they are the classical representation of the light we see coming from distant compact
objects.
The main assumption we shall make is that they are invariant under translational motion
in a space-time plane. That is, there is a coordinate system in which the components of
Maxwell’s tensor depends on s := kµ xµ + s0 . Geometrically the level sets of this function are
the planes perpendicular to the co-vector k.
We then have,

d d ′
∂σ Fµν = ∂σ s Fµν = kσ Fµν := kσ Fµν
ds ds
where we have denoted, for brevity, by a ′ the derivative with respect to s.
The vacuum Maxwell’s equations become,

∂ µ Fµν = 0 ⇒ k µ Fµν′
=0
′
∂[σ Fµν] = 0 ⇒ k[σ Fµν] = 0

The second equation implies,

′
Fµν = 2k[µ uν] (18.1)

Exercise: Convince yourself of this assertion by taking a bases of co-vectors, {ei }, i = 0 . . . 3

with e1 = k, express Fµν in that base and apply the equation. All coefficients of tetrads
having three different vectors are non-zero, so their coefficients must vanish.

249
250 CHAPTER 18. PLANE WAVES

Notice that the vector uν is not completely determined, for we can add to it any other
vector proportional to kν and still obtain the same Fµν . But obviously can not be just
proportional to kν , otherwise we will get the trivial solution.
The first equation now says,

0 = 2k µ k[µ uν] = k µ kµ uν − k µ uµ kν (18.2)

since we don’t want kν proportional to uν we must have,

k µ kµ = 0 and k µ uµ = 0

the first says that kµ is a null vector, the second that uµ is perpendicular to kµ . Notice that
we still have the freedom of adding to u a vector in the k direction for k is null.
Thus, we see that the general solution depends on two vectors, the vector k determining
the plane, and we see that the plane has to be null, and another vector, u(s) perpendicular
to it and which is just defined up to a multiple of k. So there are just two free functions left.
We can represent the solutions in the following figure, 18.1 the tensor F is constant along
the null planes and changes along the s function s from plane to plane, representing waves
which move to the right in the k direction.

s=2

s=1

s=0

Figure 18.1: Level surfaces of s := kµ xµ + s0 for a non-null vector k.

To obtain the tensor F we must now integrate u along s to obtain another vector, v,
which will also be perpendicular to k (appropriately choosing the integrations constants) and
defined up to a multiple of k. So we will have,

Fµν = 2k[µ vν]

18.1. DEFINITION AND PROPERTIES 251

k
t

s=2

s=1

s=0

Figure 18.2: Null plane waves s := kµ xµ + s0

252 CHAPTER 18. PLANE WAVES

18.2 Invariants and Energy Momentum Tensor

We can now compute the invariants of these solutions, we get

Fµν F µν = 4k[µ vν] k [µ v ν] = 0,

for k is null and v perpendicular to it.
Likewise,

Fµν ⋆ F µν = 2εµνσρ kµ vν kσ vρ = 0,
due to the total antisymmetry of the Levi-Civita tensor. So we see that these solutions are
very special indeed.
It is interesting to compute the energy-momentum tensor of this solution,

−1 µ σ −1 µ ν σ
T µν = (k v − k σ v µ )(k ν vσ − kσ v ν ) = k k v vσ . (18.3)
2π 2π
Thus, the momentum any observer with 4-velocity u would assign to this solution is,

−1 σ
pµ = v vσ k ν uν k µ .
2π
So it is a null vector, pµ pµ = 0. Furthermore, notice that v σ vσ ≤ 0, vanishing only if v is
proportional to k in which case F vanishes. So the waves always carry energy-momentum,
and it is in the form of a null four-vector. If k is future directed, then we say the plane wave
is moving into the future.
Energy flux, which is proportional to Poynting vector, is given by the space component of
µ
p , namely, writing kµ = ω(uµ + k̂µ )
ω σ
pµ⊥ = v vσ k ν uν k̂ µ
2π

Exercise: Check that v σ vσ ≤ 0 by choosing a Cartesian base so that k µ = ω(t̂µ + x̂µ ) and so
we have, t̂µ vµ = −x̂µ vµ . Write now the metric in this base and compute v σ vσ .

18.3 Gauge Potentials

Notice that since v = v(s), given a 4-vector potential we must have,

Fµν = 2k[µ vν] = 2∂[µ Aν] ,

so,

A′ν = vν .
18.4. HELICITY 253

That is the vector potential is a first integral of the vector v. Since v is perpendicular to k
we can choose the integration constants so that A is also perpendicular to it,

∂µ Aµ = kµ v µ = 0.
So this 4-potential is in Lorentz gauge! The freedom we have in adding to v a vector propor-
tional to k is just the freedom to add to A the differential of a function which depends only
on s, and those are just (plane) solutions to the wave equation, so it is just the freedom one
has on the choice of Lorentz gauge 1

18.4 Helicity
We have seen that plane waves are defined by two vectors, k and v, the individual size of
them is irrelevant. If we scale one and scale the other with the inverse factor the Maxwell
tensor they define will not change. Both lie at the symmetry plane. The vector k determines
the symmetry plane of the wave and the propagation direction, so its direction is well-defined.
On the contrary the direction of v is not, for we can change it by a multiple of k and still
would produce the same F . Thus, the object it defines is not a direction, but rather a plane,
the 2-plane perpendicular to k of vectors of the form αv + βk, for some vectors v and k. This
plane is called the helicity plane of the wave.

t
k ũ

u
αk

Figure 18.3: The helicity plane, one dimension suppressed u ≡ ũ := u + αk.

1
One can of course choose other gauge functions solutions to the wave equation which do not satisfy the
plane symmetry, but we are restricting attention here only to the ones that do.
254 CHAPTER 18. PLANE WAVES

18.5 Observer dependent expressions

Here, for completeness will give the expressions for the plane waves for arbitrary observer
with 4-velocity t in terms of the electric and magnetic fields he sees. Notice that since,

Fµν F µν = −2Eµ E µ + 2Bµ B µ = 0,

the electric and magnetic vectors have the same magnitude for all observers.
We also have,

Fµν ⋆ F µν = 2E µ Bµ = 0,
so the vectors are perpendicular to each other. Furthermore, since F and ⋆ F are per-
pendicular to k, both electric and magnetic vectors are also perpendicular to it. That is
Eµ k µ = (Fµν tν )k µ = (Fµν k µ )tν = 0, the same for Bµ using ⋆ F .
Choosing an observer, t we have,

kµ = ω(tµ + k̂µ )
with k̂ perpendicular to t. This is the propagation direction this observer assigns to the wave
in his simultaneity surface. Since E~ and B~ will be perpendicular to both, t and k they will
be also perpendicular to k̂. Given any vector ṽ we can find another,

(ṽ · t)
v = ṽ − k ,
k·t
which belongs to same equivalent class and furthermore is perpendicular to t. This vector is
~ indeed,
proportional to E,

Eµ := Fµν tν = 2k[µ vν] tν = −vµ kν tν = −ωvµ

q
the magnetic field is perpendicular to it and to k̂, and of the same magnitude, η(t, k) −η(v, v),
so is completely defined by these conditions, indeed,

Bµ = −εµνσρ tν k σ v ρ = ε̃µσρ k σ v ρ = ωε̃µσρ k̂ σ v ρ .

We see that the direction of E ~ can be associated to the helicity or polarization plane of
the wave, which for any given observer acquires a very concrete sense. Indeed, the electric
field points in the direction along line at which the helicity plane intersects the simultaneity
hyperplane.

18.6 Covariant Boundary Conditions for Superconduc-

tors and Total Reflection
Consider a time-like hyper-surface S with normal vector (space-like) n̂. On one side of it we
shall consider a perfect superconductor, that is on its surface the tangential electric field and
18.6. COVARIANT BOUNDARY CONDITIONS FOR SUPERCONDUCTORS AND TOTAL REFLE

~
B

~
E

Figure 18.4: Space view of plane waves.

the normal magnetic field must both vanish. Since we can take any time direction to split
Maxwell’s tensor into its electric and magnetic parts there must be a covariant equation to
express this boundary condition. Indeed, choosing any time direction perpendicular to n̂, one
can see that the condition is,

0 = ⋆ Fµν nν := (2B[µ tν] − εµνσρ E σ tρ )nν = −tµ Bν nν + tρ ερµνσ E σ nν (18.4)

The first term is along the time direction while the second is perpendicular to it, so both must
be zero independently, and they are just the usual expressions for the boundary conditions
on superconductors.
We shall apply this to study total reflection of plane waves on a superconductor plane,
that is (for visible light frequencies) a mirror.

Exercise: Using 15.15, show that the above condition means that the energy flux across the
superconductor vanishes for all energy densities defined with respect to observers tangent to
it. That is,
T µν tµ nν = 0 with tµ nµ = 0.

Exercise: Show that there is no momentum flux for the components normal to the boundary.
Compute the momentum flux for the normal component. Does it has a definite sign?

Exercise: Show that if a superconductor body has axial symmetry then there is no angular
momentum flux for the axial component of it.

For a plane wave the above condition, 18.4 becomes:

256 CHAPTER 18. PLANE WAVES

0 = ⋆ Fµν nν |S = εµνσρ k σ uρ nν |S , (18.5)

geometrically this condition means that the vectors k, u, and n lie in a two-plane (they are not
linearly independent). Considering that we have fixed the wave propagation direction k and
the normal to the plane, n, this condition means that u must be in the two-plane generated
by these two vectors (this is indeed a two-plane for these vectors can not be proportional
to each other, k is a null vector, while n is a space-like one). Since we can add to u any
vector proportional to k, there will be one of these vectors, ũ which is parallel to n. For
any time-like direction for which n is perpendicular to it, the vector ũ represents the electric
field direction. More importantly, since u is perpendicular to k, k one must be tangent to
the superconductor hypersurface. Thus, we see that we have a wave traveling along a null
direction tangent to the superconductor.
What happens to a wave which is incident at a given angle with respect to this hypersur-
face? The above calculation means we can not have a unique plane wave and still satisfy the
boundary conditions. Physically we know that there will be an outgoing wave bouncing from
the superconductor, so we expect that with two plane waves chosen appropriately we will be
able to satisfy the conditions. Thus, we try with a Maxwell solution of the following type:
in in out out
Fµν = 2k[µ uν] + 2k[µ uν] (18.6)
Here uin depends on sin := kµin xµ , and uout on sout := kµout xµ . Since we want some
cancellation to occur at the boundary xµ nµ = 0 we need both waves to have the same space-
time dependence at it, namely the level surfaces of sin intersection S must coincide with the
level surfaces of sout intersection S. That is,

0 = kµin xµ − kµout xµ = (kµin − kµout )xµ ∀xµ such that nµ xµ = 0.

So the condition becomes,

kµin − kµout ∝ nµ or kin out

|| = k|| , (18.7)
where from now on the || subscript means tangent to the hyper-surface, that is, perpendicular
to n. Since both vectors must be null it is easy to see that their normal components either
coincide or are the opposite of each other,

0 = η(kin , kin )
= η(kin in in in
|| , k|| ) + η(k⊥ , k⊥ )

= η(kout out in in
|| , k|| ) + η(k⊥ , k⊥ )

= −η(kout out in in
⊥ , k⊥ ) + η(k⊥ , k⊥ ).

If they coincide then both wave vectors are the same, and we know there is no solution unless
the wave is tangent to the superconductor. So the general solution must have two wave vector
whose tangent components coincide and whose normal components are opposite.

kin out
⊥ = −k⊥ , (18.8)
18.6. COVARIANT BOUNDARY CONDITIONS FOR SUPERCONDUCTORS AND TOTAL REFLE

where from now on the subscript ⊥ means the part of the vector perpendicular to the hyper-
surface, that is, proportional to n. Therefore,

kout = kin in
|| − k⊥ (18.9)

Exercise: Choose a time direction and coordinates adapted to it and to the hyper-surface
and recuperate Snell’s reflection law.

Exercise: Show that the planes generated by n, and kin , coincides with the one generated
by n, and kout , that is, span{n, kin } = span{n, kout }. Once an observer is chosen, the
intersection of its simultaneity hyper-surface with this plane is a 2-plane usually called the
incidence plane.

kin kout

uout
n

uin

Figure 18.5: Plane wave reflexion.

The boundary condition then becomes,

0 = εµνσρ (kµin uin out out

ν + kµ uν )nρ = ε
µνσρ in in
kµ (uν + uout
ν )nρ (18.10)
where we have used that the tangent projections to the boundary of both wave vectors are the
same. Thus, we see that in this case we need, that uin + uout must be in the plane spanned
by the vectors kin and n which, as we mention above, is actually the same as the plane
spanned by kout and n. Adding to uin + uout a vector proportional to kin we can make it to
be proportional to n. Therefore, the tangent component to the boundary of it, (uin + uout )|| ,
vanish. That is,

uin out
|| = −u||
258 CHAPTER 18. PLANE WAVES

The perpendicularity conditions imply,

0 = η(kin , uin )
= η(kin in in in
|| , u|| ) + η(k⊥ , u⊥ )

= −η(kin out in in
|| , u|| ) + η(k⊥ , u⊥ )

= −η(kout out out in

|| , u|| ) − η(k⊥ , u⊥ )

= η(kout out out in

⊥ , u⊥ ) − η(k⊥ , u⊥ )
= η(kout out in
⊥ , (u⊥ − u⊥ ))

where in the third line we have used the above condition, in the fourth the relation between kin ||
and kout
|| , in the fifth other perpendicularity condition η(k out
, u out
) = 0, and in the sixth the
relation between kin out
⊥ and k⊥ . Since we are assuming k not to be in the plane perpendicular
to n, this implies that uout in
⊥ = u⊥ . We see then that the relation between the incoming and
outgoing field strength is:

uout = −uin || + uin

⊥. (18.11)

Exercise: Check that the above condition is invariant under the addition to uin of a vector
proportional to kin and a corresponding vector kout to uout .

Exercise: Compute F µν Fµν and F µν ⋆ Fµν for the total field.

Exercise: Use the boundary condition, 18.4, and the expression of the energy momentum
tensor in terms of ⋆ F to compute the momentum flux across the conductor boundary. Is
there any net momentum transfer for the normal component of it? Does it has a definite
sign? Then imagine the conductor is a finite body with finite mass, if the wave transfers to
it a finite momentum, how is that this is compatible with the fact that there is initially no
energy flux across it?

Exercise: Choose an observer and adapted coordinates and express the above result in terms
~ and B
of the incoming E ~ fields.

18.7 Monochromatic Waves and Fourier Decomposition

We call monochromatic waves whose dependence of the vector v on s is harmonic.

v(s) = v 0 eis
18.7. MONOCHROMATIC WAVES AND FOURIER DECOMPOSITION 259

Since s = kµ xµ +s0 we see that an observer for which kµ = ω(tµ + k̂µ ) will see a wave oscillating
with frequency ω = η(k, t). When considering plane waves moving along different directions
it is customary then to label the coefficients of them with the wave vector k,

v(s) = v k eis

and in general consider them as complex, so after a sum over different modes one takes the
real part of it.
We shall show now that any vacuum solution to Maxwell’s equations can be written as a
liner combination of monochromatic plane waves. To see this we shall pick a time direction t
and a Cartesian coordinate system compatible with it.
We recall that any smooth function f in R3 which suitable decay at infinity can be de-
composed in Fourier modes, namely,

1 ~
Z
f (~x) = 3 fˆ(~k)eik ·~x d3 k
(2π) 2 R3

where fˆ(~k) is given by,

1 ~
Z
ˆ ~k) =
f( 3 f (~x)e−ik ·~x d3 x.
(2π) 2 R3

¯
Notice that if f (~x) is a real function then we have, fˆ(~k) = fˆ(−~k), thus a real function is
represented by a complex one, but within the class having the above property.
The same hold for vectors, as long as we write them in Cartesian coordinates, so that inte-
grals can be taken, that is so that we can add vectors at different points in space. Thus, given
a vector vµ (~x) we have its Fourier components, v̂µ (~k). Now, and arbitrary linear combination
of solutions to the wave equation will have an expression,

1 + + ~ i(|~ ~ − − ~ i(−|~ ~
Z
Fµν (xσ ) = 3 [2k[µ v̂ν] (k)e k |t+k ·~x) + 2k[µ v̂ν] (k)e k |t+k ·~x) ] d3 k
(2π) 2 R3

where we now label the waves by its 3-dimensional wave number vector ~k, and since there
are two null planes with the same 3-vector, namely kµ+ = (|~k|, ~k), and kµ− = (−|~k|, ~k), we
have taken care of that by labeling those contributions v̂µ+ (~k) and v̂µ− (~k). Notice that these
contributions must satisfy, v̂ ±µ (~k)kµ± = 0, and since they can be chosen such that they are
perpendicular to the time direction, they result to be perpendicular to ~k.
Consider now arbitrary solutions to vacuum Maxwell’s equations, if one assumes some
asymptotic and regularity conditions, then we claim all such solutions can be written in this
way, namely as linear combination of plane waves. To see this recall that given F at t = 0,
with the conditions assumed above there is a unique solution to vacuum Maxwell’s equations,
for they are determined by their initial data. So we assume Ej (~x, 0) and Flm (~x, 0) are given
and have their corresponding Fourier transform, Êj (~k), and F̂lm (~k). Using the expression
260 CHAPTER 18. PLANE WAVES

above (18.7) we have,

1 Z
~
Fµν (t = 0, ~x) = 3
+ + ~
[2k[µ v̂ν] (k) + 2k[µ v̂ν] (k)]eik ·~x d3 k.
− − ~
(2π) 2 R3

In particular,

Êj (~k) = [v̂j+ (~k) − v̂j− (~k)]|~k|

where we have used that we are using the gauge in which η(t, v̂ ± ) = 0, and,

F̂lm (~k) = [2k[l+ v̂m]

+ ~ − ~
(k) + 2k[l− v̂m] (k)]
= 2k[l [v̂ + (~k) + v̂ − (~k)].
m] m]

Contracting with ~k and using that they are perpendicular we wet,

+ ~ − ~
k l F̂lm = k 2 (v̂m (k) + v̂m (k))
Thus, provided that Êj (~k) and F̂lm (~k) vanish sufficiently fast when |~k| → 0 we can invert the
above relations and find the v̂j± (~k), corresponding to this initial data.

1 kl
v̂j± = ( F̂lj ± Êj )
2|~k| |~k|
Therefore we get the whole solution in terms of an integral over plane waves.

Exercise: Why it is true that these v̂j± satisfy k j v̂j± = 0?

For computing energy fluxes and other physical quantities some time it is useful to take
time averages of harmonic functions, in that case, we are given two functions A(t) = 21 (A0 eiωt +
Ā0 e−iωt ) and B(t) = 12 (B0 eiωt + B̄0 e−iωt ), then

1 T
Z
< A(t)B(t) > := limT →∞ A(t)B(t)dt
T 0
1 T 1
Z
= limT →∞ (A0 eiωt + Ā0 e−iωt )(B0 eiωt + B̄0 e−iωt )dt
T 0 4
1ZT 1
= limT →∞ (A0 B̄0 + Ā0 B0 + A0 B0 ei2ωt + Ā0 B̄0 e−i2ωt )dt
T 0 4
1
= ℜ(A0 B̄0 ). (18.12)
2

Exercise: Compute the average energy flux of a plane wave along an arbitrary time-like
plane.
Chapter 19

Resonant Cavities and Wave Guides

19.1 Resonant Cavities

We want to consider now waves in a compact region of vacuum space, V , enclosed by a
conductor, as shown on Figure 19.1. In this region the electromagnetic field will satisfy
Maxwell vacuum equations,

~
∂E ~ ∧B
= c∇ ~ (19.1)
∂t
~
∂B ~ ∧E
= −c∇ ~ (19.2)
∂t

~ ·E
∇ ~ =0 (19.3)

∇ ~ =0
~ ·B (19.4)
In the approximation where the conductor is perfect (something which depends on the
frequency of the waves, but that here we assume valid for all frequencies) we have that at the
boundary the electric field must be perpendicular to it,

~ ∂V = 0
n̂ ∧ E| (19.5)
this is the boundary condition we have. It is therefore simple to express the equations and
their solutions in terms of the electric field. To do that we take another time derivative to
the equation for the electric field, 19.1, and use 19.2 to get,
−1 2 i
∂ E + ∆E i = 0 (19.6)
c2 t
~ ~x) = F~ (~x), and ∂t E(t,
We add to it equation 19.3. Given as initial data, E(0, ~ ~x)|t=0 =
~ ∧ B(0,
c∇ ~ ~x) = c∇ ~ ∧ G(~
~ x), satisfying ∇
~ · F~ = ∇
~ ·G
~ = 0 and the boundary condition, 19.3,
there is a unique solution to the wave equation in [0, ∞) × V satisfying the constraints, and
the boundary condition for all times. Once the electric field is computed, the magnetic field

261
262 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

n̂

Figure 19.1: A resonant cavity.

is obtained by integrating in time equation 19.2 with arbitrary initial conditions B ~ = G ~

satisfying 19.4.
We shall not prove this assertion, but it follows from the fact that with this boundary
conditions the electromagnetic energy inside the cavity is conserved (the Poynting vector is
tangent to the surface). The applied boundary condition can be expressed as the condition
that the normal to the surface outgoing plane waves are equal to the normal incoming plane
waves (normal reflection) and this is an admissible boundary condition from the point of view
of the mathematical theory behind this assertion.

19.1.1 Monochromatic solutions

To find the solutions to the above problem we consider monochromatic solutions, namely
electromagnetic fields of the form,

~ ~x) = E
E(t, ~ ω (~x)eiωt ~ ~x) = B
B(t, ~ ω (~x)eiωt (19.7)
We need to solve then the following eigenvalue-eigenfunctions system,

~ ∧E
∇ ~ ω = −iω B~ω (19.8)
c
~ ∧B
∇ ~ ω = iω E~ω (19.9)
c
~ ·E
∇ ~ω = 0 (19.10)
~ ·B
∇ ~ω = 0 (19.11)
~ ω |∂V
n̂ ∧ E = 0 (19.12)
19.1. RESONANT CAVITIES 263

Note that the divergence equations are consistent with the first two equations.
We now assert: There is a countable, and infinite set of real frequencies, {ωi }, i =
0..∞, for which this system of equations has solutions (E ~ li , B
~ li ), the li supra-index meaning
ωi ωi
that in general there will be two solutions for each ωi . Given an arbitrary initial data set
~ 0 (~x), B
(E ~ 0 (~x), satisfying ∇·E
~ 0 = 0, ∇·B
~ 0 = 0, and the boundary condition, 19.5, the resulting
solution can be expressed as a linear combination of the above monochromatic solutions.
This assertion is based in the following observations:

• The above system can be considered as a system in the Hilbert space of square integrable
fields with zero divergence (in the distributional sense),

~ B)
H = {(E, ~ ∈ L2 |∇
~ ·E
~ω = ∇
~ ·B
~ ω }.

• H is (formally) invariant under the action of operator

~ ~
! !
~ B)
~ := 0 −i∇∧ E
A(E, ~ ~ .
i∇∧ 0 B

~
• In H the operator is elliptic. For plane solutions with dependence eik ·~x and ~k · E
~ =
~k · B
~ = 0 we have,
|A(E, ~ = |~k|2 [|E|
~ B)| ~ 2 + |B|
~ 2 ].

• With the above boundary condition the operator A is formally self adjoint in H.

~¯2 · ∇ ~¯2 · ∇
Z
~ 2, B
h(E ~ 2 ), A(E
~ 1, B
~ 1 )i = (−iE ~ ∧B
~ 1 + iB ~ ∧E
~ 1 ) dV
ZV
= ~1 · ∇
[−iB ~¯2 + iE
~ ∧E ~1 · ∇ ~¯2
~ ∧B
Σ

+ i∇ ~¯2 ∧ B
~ · (E ~1 + E ~¯2 )] dV
~1 ∧ B
Z
= ~ ∧B
[−i(∇ ~ 2) · E
~ 1 + i(∇
~ ∧E
~ 2) · B
~ 1 ] dV
Σ
~¯2 ∧ B ~¯2 ] dΣ
I
+ i n̂ · [E ~1 + E
~1 ∧ B
∂V
~ 2, B
= hA(E ~ 2 ), (E
~ 1, B
~ 1 )i. (19.13)

Where, in the last step, we have used the boundary condition.

Alternatively we can take the curl of 19.8, and use 19.9, to get,

~ ∧ (∇
∇ ~ ∧E ~ ω ) = ω2E
~ (19.14)
~ ω |∂V = 0
n̂ ∧ E (19.15)
264 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

Notice that if ω 6= 0 then any solution to this eigen-equation automatically satisfies ∇~ ·E

~ = 0.
This system is often used in numerical calculations, it is of course also a well posed (hermitian
- elliptic) system.

~ ω = c∇
Exercise: Check that using that iω B ~ ∧E
~ ω that the other boundary condition,
~ ω · n̂|∂V = 0 is also satisfied.
B

We shall not prove the above assertion, but show its implication in an example.

19.1.2 Rectangular hole

Consider a region given by the interior of a .... of sides (a, b, c) at which we attach our
Cartesian coordinate system. In these coordinates the wave equation can be separated, so
that we have, that, for each component of the electric field, Φ = X(x)Y (y)Z(z), we must
have,

∂2X ∂2Y ∂2Z ω2

YZ + XZ + XY + 2 XY Z = −ω 2 XY Z (19.16)
∂x2 ∂y 2 ∂z 2 c
so that,

∂2X
2
= −kx2 X
∂x
∂2Y
= −ky2 Y
∂y 2
∂2Z
= −kz2 Z
∂z 2
and,

ω2
= kx2 + ky2 + kz2 (19.17)
c2
The solutions to the above system are, X(x) = A+ eikx x + A− e−ikx x , and similarly for the
other coordinates. In particular if we want, for instance, that X(0) = X(a) = 0, then the
solution is X(x) = A sin(kx x) with kx = πl a
.
We now impose the boundary conditions, Eω1 (x, 0, z) = Eω1 (x, b, z) = Eω1 (x, y, 0) = Eω1 (x, y, c) =
0, Eω2 (0, y, z) = Eω1 (a, y, z) = Eω1 (x, y, 0) = Eω1 (x, y, c) = 0, etc. We have then,

Eω1 = (E+1 eikx x + E−1 e−ikx x ) sin(ky y) sin(kz z)

Eω2 = (E+2 eiky y + E−2 e−iky y ) sin(kx x) sin(kz z)
Eω3 = (E+3 eikz z + E−3 e−ikz z ) sin(kx x) sin(ky y)
(19.18)
19.1. RESONANT CAVITIES 265

where for each solution we have, in principle, different wave vectors ~k = (kx , ky , kz ) each one
of them satisfying the above dispersion relation 19.16, and furthermore for Eω1 , ky = πn b
, and
kz = πmc
, etc.
It remains now to impose the divergence free condition,

(∂x X 1 )Y 1 Z 1 + X 2 (∂y Y 2 )Z 2 + X 3 Y 3 (∂z Z 3 ) = 0 (19.19)

dividing by Y 1 Z 1 we find that ∂x X 1 , X 2 , and, X 3 must proportional among each other.
Likewise for the Y and Z functions. It is not difficult to see that all the vectors ~k must
actually be the same, and furthermore, E+i = E−i , and so the general solution has the form,

E 1 cos(kx x) sin(ky y) sin(kz z)

 

~  2
E ω =  E cos(ky y) sin(kx x) sin(kz z) 

(19.20)
E 3 cos(kz z) sin(kx x) sin(ky y)
with,

kx E 1 + ky E 2 + kz E 3 = 0, (19.21)
the last requirement from the divergence free condition, and

~k = ( πl , πn , πm ). (19.22)
a b c
Since s
l2 m2 n2
ωlnm = ±π + + 2,
a2 b2 c
we see that it is real and there are a countable infinite number of them. For each vector ~k lmn
there are at most two linearly independent solutions for we have also to satisfy 19.21. To do
that we can choose, for instance, any two vectors perpendicular to ~k lmn , except for the case
where one of the integers vanish, in that case the vector direction is completely determined.
For instance if l = 0 then the vector is:

(E 1 sin(ky y) sin(kz z), 0, 0). (19.23)

In the general case it is simple to take these two vectors to be also perpendicular among each
other, we shall call them E ~ + and E ~− . 1
lmn lmn
Notice that all these eigenfunctions are orthogonal among each others (in the square
integral norm) and one can show that they form a complete set. Thus, the general solution
can be written in terms of them,

+ −
~ ~x) = ++ iωlmn t −+ −iωlmn t ~ +− iωlmn t −− −iωlmn t ~
X
E(t, [(Clmn e + Clmn e )E lmn (~x) + (Clmn e + Clmn e )E lmn (~x)]
l,n,m
(19.24)
1 ~ + + iE
Actually is better to choose a complex vector, E ~ − for performing calculations.
266 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

Notice that these are all standing waves, namely waves which do not propagate, their nodes
remain at fixed points in space. Nevertheless, they can describe general solutions which can
travel along the cavity bouncing off their walls.

Exercise: Check that given initial data (E ~ 0, B

~ 0 ) satisfying the constraint equations we can
±±
construct a solution. Find the explicit values for Clmn .

Exercise: Show that for a mode in a cavity,

Z Z
~ ω |2 dV =
|E ~ ω |2 dV
|B
V V

Hint: use −iω B ~ω = ∇

~ ∧E
~ ω and then ω2 ~
E ~ ω = 0. This is a type of energy-
+ ∆E
c2 ω
equipartition theorem.

19.2 Wave Guides

We consider here the case of a hole in a conductor which is invariant under translation along
a given direction, that is, it is an infinite tunnel carved inside a conductor, as shown in the
figure. We shall choose the z-axis in that direction, which we denote by k̂ (the Killing vector
realizing the translational symmetry).
Because this translational invariance, all the relevant equations will be confined to the
perpendicular section, which we shall call Σ. Out of all components of the electric and
magnetic fields of particular relevance will be the following two scalars, Ez := k̂ · E ~ and
~
Bz := k̂ · B.
Since each of the component of the electric and magnetic fields satisfy the wave equation,
we can see, using separation of variables, that each component must be of the form,

U(t, x, y, z) = V (x, y)e−i(ωt±kz z) , (19.25)

where V (x, y) satisfies the following eigenvalue-eigenfunctions equation,

∆2 V = −k 2 V (19.26)
with,

ω2
k 2 :=− kz2 , (19.27)
c2
where ∆Σ := ∂x2 + ∂y2 is the two-dimensional Laplacian in the guide section, Σ. We recall
that not all components are independent since the divergence of the both fields must vanish,
and we have the evolution equations which give one of the field by time integration when the
other is known.
We shall consider forward going solutions, that is with the dependence e−i(ωt−kz z) , where
we are assuming sign(ω) = sign(kz ). The treatment for the others is completely analogous.
19.2. WAVE GUIDES 267

k̂

n̂

Figure 19.2: A wave guide.

In this case Maxwell’s evolution equations become,

iω
∂y Ez − ikz Ey = Bx (19.28)
c
iω
−∂x Ez + ikz Ex = By (19.29)
c
iω
∂y Ex − ∂x Ey = Bz (19.30)
c
and
−iω
∂y Bz − ikz By = Ex (19.31)
c
−iω
−∂x Bz + ikz Bx = Ey (19.32)
c
−iω
∂y Bx − ∂x By = Ez . (19.33)
c
We further have the two constraint equations,

∂x Ex + ∂y Ey + ikz Ez = 0 (19.34)
∂x Bx + ∂y By + ikz Bz = 0 (19.35)
268 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

To study the different solutions it is helpful to classify them according as to whether they
have components along the symmetry direction, k̂.

19.2.1 Case 1: Transversal (T) waves, Ez = Bz = 0.

In this case the above equations imply,
iω
−ikz Ey = Bx , (19.36)
c
and
−iω
ikz Bx = Ey , (19.37)
c
that is,

ω2
kz2 = or k 2 = 0 (19.38)
c2
Thus these waves travel to the speed of light. Since we are assuming sign(kz ) = sign(ω),
we have, Bx = −Ey , and By = Ex , that is,

~ = k̂ ∧ E,
B ~ (19.39)
~ is completely determined once we know E.
and so B ~
The remaining two equations are:

∂y Ex − ∂x Ey = 0
∂x Ex + ∂y Ey = 0

and identical equations for B ~ which we shall not need. The first equation tell us that there
~ = −∇φ.
exists a scalar field φ(x, y) such that E ~ The second that it must satisfy,

∆Σ φ = 0, (19.40)
in the guide section. Since we are assuming the conductor is perfect at its boundary we must
have,

~ ∂Σ = (n̂ ∧ ∇φ)|
(n̂ ∧ E)| ~ ∂Σ = 0,
Thus, the boundary condition implies that on each connected component of the boundary,

φ|Σi = φi = cons (19.41)

From the uniqueness of solutions to the Laplace equations we see that in order to have non-
trivial solutions we need at least a guide with two separate conductors at different potentials,
for instance the ones in the figure ??. Otherwise, if ∂Σ is connected, then φ = φ0 in all of
∂Σ, then δφ := φ − φ0 satisfies ∆Σ δφ = 0, δφ|∂Σ = 0 and so δφ = 0 on Σ. When the guide
19.2. WAVE GUIDES 269

configuration allows for these waves they propagate at the speed of light along the k̂ direction
with the electric and magnetic fields perpendicular to it and to each other.
Notice that the following condition on the magnetic field follows from the first, namely
~ · n̂)|Σ = 0.
(B

Indeed,
~ · n̂ = (k̂ ∧ E)
B ~ · n̂ = k̂ · (E
~ ∧ n̂). (19.42)

Exercise: Make a qualitative drawing of the fields for the wave guide in the right.

Exercise: Show that the magnetic field satisfy the remaining equations.

19.2.2 Case 2: Transverse Magnetic (TM) waves, Bz = 0.

In this case the above equations imply,
iω
∂y Ez − ikz Ey = Bx , (19.43)
c
and
−iω
ikz Bx = Ey , (19.44)
c
that is,

ω −ω ω2 k2
∂y Ez = [ikz + i ( )]Ey = ikz [1 − 2 2 ]Ey = Ey . (19.45)
c kz c c kz ikz
If k 2 = 0 then ∂y Ez = 0 and the boundary condition Ez |Σ = 0 implies Ez = 0 in the whole
section, and we are back to Case 1. So we consider k 2 6= 0. In this case then Ey is completely
determined by Ez ,

ikz
Ey = ∂y Ez (19.46)
k2
Similarly, we obtain,

ikz
Ex = ∂x Ez (19.47)
k2
So,

E~ Σ = ikz ∇
~ Σ Ez . (19.48)
k2
where the label Σ means projection into the two-dimensional section Σ.
270 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

The other equations are identical as those in Case 1, so we get,

B~ = ω (k̂ ∧ E ~ Σ ) = ω (k̂ ∧ E).

~ (19.49)
ckz ckz
Thus, every field depends on Ez , this is the only quantity we need to compute now. But we
already have seen that Ez must satisfy,

∆Σ Ez = −k 2 Ez . (19.50)
One can check that this is equivalent to the divergence free equation, 19.34.
To see which is the appropriate boundary condition for this case, we compute,

~ ∂Σ = (n̂ ∧ (E
(n̂ ∧ E)| ~ Σ + k̂Ez ))|∂Σ
ikz ~
= (n̂ ∧ ( 2 ∇ Σ Ez + k̂Ez ))|∂Σ
k
ikz ~ Σ Ez )|∂Σ + (n̂ ∧ k̂)Ez |∂Σ
= (n̂ ∧ ∇ (19.51)
k2
each term points in a different direction, so they both must cancel. This is the case if we
require,

Ez |∂Σ = 0 (19.52)
then both conditions are satisfied. As before we also have,

B ~ Σ ) · n̂|∂Σ = ω k̂ · (E
~ · n̂|∂Σ = ω (k̂ ∧ E ~ ∧ n̂)|∂Σ . (19.53)
ckz ckz

~ ·E
Exercise: Check that if this equation is satisfied, then we also have, ∇ ~ =∇
~ ·B
~ = 0.

So we must solve now equation 19.50 subject to the boundary condition 19.52. We know
from the general theory of self-adjoint operators that this system has an infinite (countable)
number of eigenvalues-eigenfunctions {kn2 , Ezn } which form a complete orthonormal eigenfunc-
tion base.
Multiplying both sides of 19.50 by −Ez and integrating over σ we obtain,
Z Z
k 2 Ez2 d3~x = −Ez ∆Σ Ez d3~x
Σ Σ
Z Z
= ~ Σ Ez · ∇
∇ ~ Σ Ez d3~x − Ez ∂n Ez dΣ
ZΣ ∂Σ

= ~ Σ Ez · ∇
∇ ~ Σ Ez d3~x
Σ
≥ 0.
~ z=0
Thus, k 2 ≥ 0. The case k 2 = 0 reduces to the Case 1, for the above equation implies ∇E
2
and then the boundary condition implies Ez = 0. Thus, we take k > 0, thus k is real, and
we have traveling waves. We also have,
19.2. WAVE GUIDES 271

ω2
= k 2 + kz2 > kz2 , (19.54)
c2
so the phase speed of these waves, vp := kωz , is greater than c. Nevertheless, the group velocity,
dω 2
dkz
= c ωkz ≤ c. Since the eigenvalues are numerable, and they can at most accumulate at
infinity, there is a minimal one, kmin ≈ L1 where L is the maximal transversal length of Σ,
and therefore a minimal frequency, ωmin = kmin c bellow which there can not be transmission
of TM waves.

19.2.3 Case 3: Transverse Electric (TE) waves, Ez = 0.

From the symmetry of vacuum Maxwell equations under substitution E ~ ↔ B ~ and time
inversion we see that this situation is almost identical with the one above: All fields would
now depend on Bz and this component would satisfy

∆Σ Bz + k 2 Bz = 0 (19.55)
To complete the system we need to impose some boundary condition to the above equation.
Since for this case,
~ Σ = ikz ∇
B ~ Σ Bz .
k2
We have,

~ Σ )|∂Σ = ikz (n̂ · ∇

~ ∂Σ = (n̂ · B
(n̂ · B)| ~ Σ Bz )|∂Σ = ikz ∂n Bz |∂Σ .
k 2 k2
On the other hand, since,

E ~ Σ ) = −ω (k̂ ∧ B),
~ = −ω (k̂ ∧ B ~ (19.56)
ckz ckz
we have,

~ ∂Σ = −ω n̂ ∧ (k̂ ∧ B)
n̂ ∧ E| ~
ckz
−ω ~ k̂ − (n̂ · k̂)B]|
~ ∂Σ
= [(n̂ · B)
ckz

Since n̂ · k̂ = 0 we see that we can satisfy both boundary conditions by requiring,

∂n Bz |∂Σ = 0.
In this case we also have an infinite number of wave modes. They are obviously different from
those found on the other cases.

Exercise: By an argument similar to the one employed for the TM waves, see that k 2 ≥ 0.
272 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

19.2.4 Energy Flux

The time averaged Poynting vector for the case of a TM wave is:

~ T · k̂ =c ~ ~⋆ c ωkz ~ ωkz ~
hSi (E ∧ B ) · k̂ = |∇Σ Ez |2 = |∇Σ Ez |2 ,
8π 8π ck 4 8πk 4
and so the energy across a section Σ per unit time (averaged) is,

Z
< P >T = ~ TΣ
k̂ · hSi
Σ
ωkz Z ~
= |∇Σ Ez |2 dΣ
8πk 4 Σ
ωkz
Z Z
= [ Ēz n̂ · ~
∇ E
Σ z dl − Ēz ∆2 Ez dΣ]
8πk 4 Σ Σ
ωkz
Z
= Ēz Ez dΣ
8πk 2 Σ

Exercise: Compute the momentum flux.

Exercise: Compute the angular momentum flux.

Exercise: Compute the energy flux for a T and a T E wave.

~ = k̂ ∧ E
Answer: For a T wave, B ~ and E
~ =∇
~ Σ φ, therefore

~ T = c ~ 2
hSi |E| k̂,
8π
and

c
Z
< P >T = ~ 2 dΣ
|E|
8π ZΣ
c ~ Σ φ|2 dΣ
= |∇
8π ZΣ
c ~ dΣ
= φ̄ n̂ · E
8πZ Σ
c
= φ̄ σ dΣ
2 Σ
cX Z
= φ̄i σ dΣ
2 i Σi
cX
= φ̄i Qi
2 i
cX
= hφi(t) Qi (t)i
2 i
19.3. CAVITIES WITH CONSTANT SECTION 273

where Qi , and φi are respectively the total charges by unit length, and the potentials (con-
stants) on each of the conducting boundaries.

19.3 Cavities with constant section

A particular case of cavities is when they are pieces of a wave guide as shown in the figure
below.
To find the modes of such cavities we can use the machinery just developed above, but
take into account that we need to find standing waves, namely waves with fixed nodes at the
end of the cavities in the symmetry direction. Thus, we need to consider waves traveling in
both direction and take linear combination of them to obtain the standing ones. So in the
case of the electric field we will have,

~ + (x, y)eiωt+kz z + E
~ ~x) = E
E(t, ~ − eiωt−kz z ,
~ Σ = 0, Bz = 0), would imply that,
and the boundary conditions at z = 0 and z = z0 , (E

~ + (x, y) + E
E ~ − (x, y) (z = 0) ~ + (x, y)eikz z0 + E
E ~ − (x, y)e−ikz z0 = 0 (z = z0 ).

These two conditions imply,

kz = πn/z0 .
So, as expected, we get only discrete modes.
We can now analyze each case:

19.3.1 T modes
In that case Bz = 0 and so no extra condition is needed. The modes are given by the solution
to the 2-dimensional Laplace equation times the solutions just find above.

~ ~x)n = −∇φ(x,
~ i πnz
πcnt
E(t, y)e z0
)sin(
z0
Note that again in this case, the boundary to the cavity section must have at least two
disconnected components.

19.3.2 TE
In that case we still have Bz = 0, so we just solve for 19.50 with Dirichlet homogeneous
boundary conditions. The frequency of the resulting modes will be:
s
πn 2
ωni = c ( ) + ki2 ,
z0
and the mode, is built out of,
274 CHAPTER 19. RESONANT CAVITIES AND WAVE GUIDES

πnz
Ezni (t, ~x) = Ezi (x, y)eiωni t sin( ).
z0

19.3.3 TM
In this case we must make sure Ez vanishes at the boundary. We solve 19.2.3 with Neumann
homogeneous boundary conditions, since Bz = 0 2 at the extremes would mean EΣ = 0 there.
We again use the same kz as in the other cases and obtain the same form for the standing
waves.

Exercise: Check that all these modes are present are the same as those we found for the
case of the cubical resonant cavity.

2
One could consider Bz = cons at the extremes, but this just generates the trivial solution.
Chapter 20

Wave Propagation in Continuum

Media

20.1 The polarizability model for time dependent fields.

The leading effect on wave propagation on continuum media is due to polarizability effects,
so we shall concentrate here in this case, although much of the formulas can be easily adapted
to include magnetic effects.
To study the polarizability effects on time varying fields a dynamical model is needed. The
simplest one is to represent the atomic dipoles as charges of opposite sign with a force acting
upon them in the linear approximation, but including a viscosity term representing energy
absorption by the system. If m represents the mass, e the charge, and ~xl the displacement
from equilibrium of particle at rest position ~x, we have the equation:

m[~x¨l + γ ~x˙ l + ω02~xl ] = −eE(~

~ x + ~xl , t)
1
Here, γ, with dimensions of time , is an attenuation factor modeling different energy losses.
They are due to excitation of other atomic modes, or the crystal net where it is embedded.
There also losses due to radiation. Since the energy of the particles must decay in time, we
assume γ ≥ 0. The frequency ω0 represents the resonance frequency of the system, it is the
derivative of the acting forces at the equilibrium point with respect to position. We have
excluded the magnetic field term form Lorentz force, for we are assuming the velocity of the
charged particles to be small when compared with the speed of light. Clearly one can envision
cases on which this is not true.
If we assume a harmonic dependence on the fields,

~ x, t) = E
E(~ ~ 0 (~x)e−iωt ,

and consider long enough wave lengths so that the spatial dependence of the field can be
~ ~
ignored, ( |∇E | << |~xl |). Under these assumptions the solution is approximately given by:
~
E
−e E~ 0 (~x)e−iωt
~x(t)l = ℜ[ 2 ].
m ω0 − ω 2 − iωγ

275
276 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

besides this solution there are homogeneous ones, but they decay exponentially to the equi-
librium position (~xl = 0) on times of the order of T = γ1 , and so, after that transient they do
not play any role.

Exercise: Find the homogeneous solutions and check the above assertion.

Thus, we see the interaction results in a dipole,

e2 ~ 0 e−iωt
E
p~(~x, t)l = −e~xl ≈ ,
m ω02 − ω 2 − iωγ
which oscillates to the same frequency as the electric field, but with a different phase and an
amplitude that depends on the frequency.
In fact, in Fourier–Laplace space, summing over all contributions,

e2 N 2
~ ω (~x) = ~ ω (~x),
~pω (~x) = χe (ω)E [ω − ω 2 − iωγ]−1 E
m 0
where N is the particle number density. Thus,

~ ω (~x) = ε(ω)E
D ~ ω (~x),

with,

4πe2 N 2
ε(ω) = 1 + 4πχe (ω) = 1 + [ω0 − ω 2 − iωγ]−1 .
m
We have then:

1. The polarizability of a medium depends on the frequency, if we have different types of

dipoles,

X e2j Nj 1 X 4πe2j Nj 1
χe (ω) = 2 2
, ε(ω) = 1 + 2 2
,
j mj ωj − ω − iγj ω j mj ωj − ω − iγj ω

where Nj is the number density of dipole sites times the number of charged particles
per site.

2. If γ ≥ 0, then for ω ≥ 0),

X e2j Nj ωj2 − ω 2 + iγj ω X e2j Nj γj ω

ℑε(ω) = ℑ 2 2 2
= 2
≥ 0.
j mj |ωj − ω − iγj ω| j mj |ωj − ω 2 − iγj ω|2
This fact will be very important in what follows. We shall see that this is a necessary
condition for not having strange waves and for the theory to be causal.
~ ω and p~ω have maxi-
3. Near resonances, ω ≈ ωi , χe (ω) is almost purely imaginary, so E j j
mum phase-difference.
20.1. THE POLARIZABILITY MODEL FOR TIME DEPENDENT FIELDS. 277

4. Contrary to what many textbooks state, but fortunately do not use:

~
∂D ~
∂E
6= ε .
∂t ∂t
Indeed,

~ ~x) := √1
∞ Z
D(t, ε(ω)E ~ ω (~x)e−iωt dω
2π −∞
1 Z∞Z∞Z∞ ~ ω (~x)e−iωt dωdω̃dt̃
= ε(ω̃)e−i(ω̃−ω)t̃ E
2π −∞ −∞ −∞
1 ∞ ∞ ∞
Z Z Z
= ~ ω (~x)e−iω(t−t̃) dωdω̃dt̃
ε(ω̃)e−iω̃t̃ E
2π −∞ −∞ −∞
Z ∞
= ε(t̃)E~ ω (~x, t − t̃)dt̃
−∞

where in the second step we have used that,

1 ∞
Z
√ e−i(ω̃−ω)t̃ dt̃ = δ(ω̃ − ω).
2π −∞

Thus,

~
∂D
Z ∞ ~ ω (~x, t − t̃)
∂E ~
∂E
= ε(t̃) dt̃ 6= ε ,
∂t −∞ ∂t ∂t
unless ε(t) = ε0 δ(t). Which in general, as in our model, it is not the case. This implies
that in general the space-time equations describing these phenomena are no longer
partial differential equations, and are, at best, integro-differential ones.

5. Dielectrics: If ωj 6= 0 ∀j, so that there are no free electrons, then at low frequencies,

X e2j Nj 1
ε(ω) = 1 + 4π := ε̃,
j mj ωj2
is real, and we have the dielectric model already studied in electrostatics.

6. Conductors: If there are free electrons, then ω0 = 0 and so,

i4πe2 N 1
ε(ω) = ε̃ + ,
m ω(γ0 − iω)
where in ε̃ we have included the contributions due to the other dipoles with ωj 6= 0.
~ ω = εω E
This term can be related to the media conductivity. Indeed, substituting D ~ω
in,
278 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

∇ ~ ω = − iω D
~ ∧B ~ ω.
c
(where we have assumed no free current), we get,

2 2
~ ∧B
∇ ~ ω = −[ iω ε̃ − 4πe N ]E
~ ω = − iω [ε̃ + i4πe N ]E ~ ω + 4π J~l ,
~ ω = − iω ε̃E
c mωγ0 c mcγ0 c c
Where in the last step we have assumed Ohm’s law to hold at these frequencies,

J~l = σ0 E,
~

with

e2 N
σ0 = .
mγ0
So we see that we recuperate this way Maxwell’s equations at the given frequency. In
the low frequency limit, ε̃ does not depend on the frequency, and so we can recuperate
a partial differential equation, namely,

~ ~
~ ~x) = 1 ε̃ ∂ E (t, ~x) + 4π J~l (t, ~x) = 1 ∂ D (t, ~x) + 4π J~l (t, ~x).
~ ∧ B(t,
∇
c ∂t c c ∂t c
With the understanding that is only valid for low frequencies, ω << ωj .

7. For frequencies higher than resonance, ω >> ωj ,

ωp2
ε(ω) ≈ 1 −
ω2
with
4πe2 N
ωp2 = ,
m
where N is the total electron number density, usually N = NA Z (= # atoms ×
atomic # per unit volume).
For dielectrics, in general ωp ≤ ωj and so the above formula only says that

ε(ω) ≤ 1,
for it only holds when ω >> ωj ≥ ωp .
But for an electron gas (plasma) or in a metal, which in most aspects behaves like an
electron gas, the above expression is valid for ω ≈ ωp , and so ε(ω) ≈ 0. We shall see
later some consequences of this.
20.2. PLANE WAVES IN HOMOGENEOUS AND ISOTROPIC MEDIA. 279

20.2 Plane Waves in Homogeneous and Isotropic Me-

dia.
If the medium is isotropic then εi j = εδ i j , and so is determined by a scalar, ε. If the medium
is furthermore homogeneous,

ε(ω, ~x) = ε(ω).

we look now for solutions representing plane waves, in the frame where the medium is at rest
we then have,

~ ω e−i(ωt−~k ·~x) )
~ ~x) = ℜ(E
E(t,
~ ω e−i(ωt−~k ·~x) ),
~ ~x) = ℜ(B
B(t,

that is plane waves traveling along the ~k direction. The phase speed of these waves would be,
(keeping the phase, ωt − ~k · ~x constant),
ω ω
|~v| = = . (20.1)
|~k| k
Inserting this ansatz in Maxwell’s equations we get,

−iωµε ~
Eω = i~k ∧ B
~ω (20.2)
c
−iω ~
Bω = −i~k ∧ E
~ω (20.3)
c
~k · E
~ω = 0 (20.4)
~k · B
~ω = 0 (20.5)

Notice that when ω 6= 0 the last two equations are a consequence of the first two.
This is an eigenvalue-eigenvector problem identical to the one studied for the case of
monochromatic modes in resonant cavities. To obtain the eigenvectors we proceed in a similar
way, taking the vector product of the first equation with ~k we get,
−iωµε ~ ~
k ∧ E ω = i~k ∧ (~k ∧ B
~ ω ) = i(~k · B
~ ω )~k − (~k · ~k)B
~ ω, (20.6)
c
using now the second equation on the left and the fourth equation on right we get,

−ω 2 µε ~
B ω = −(~k · ~k)B
~ ω, (20.7)
c2
from which it follows that

ω 2µε ~ ~
=k·k (20.8)
c2
280 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

or defining n2 := µε, and k 2 = ~k · ~k,

ωn kc
k= or ω = (20.9)
c n
which is called the dispersion relation of the medium.

1. Notice that the phase speed is now,

ω c
|~v | =
= . (20.10)
k n
Although this speed can be larger than the speed of light, it is just the phase speed and
not the real propagation speeds of perturbations of compact support, which is called
the group velocity and it is given by,

dω
vg = , (20.11)
dk
and this is in general smaller or equal to c. We shall prove in the next chapter that
under very general conditions in ε(ω) all solutions propagate at speeds limited by the
vacuum speed of light.

2. Since ε can be in general complex (or also real but negative), n, and therefore ~k, are
complex. So care must be exercised when asymptotic conditions are imposed, since
there could be asymptotically growing modes.
Indeed, considering for simplicity, the case µ = 1, we see that there are three possibilities:

a) ε real, ε > 0. In this case n > 0 and so ~k can be taken to be real. In our simple
matter model this corresponds to systems with no dissipation, γ = 0, and indeed
the resulting wave propagates with no dissipation along any arbitrary k̂ direction.
We call this a transparent medium.
b) ε real, ε < 0. In this case n is purely imaginary, n = iκ, κ real. In general in this
case ~k is a complex vector, but there are solutions where ~k = iκk̂, with k̂ a real
unit vector. For instance the vector ~k = (1, 2i, 0), has ~k · ~k = −3.
c) ε complex, n = ñ + iκ, and so ~k is complex, as in the former case there are
solutions with ~k = (ñ + iκ)k̂, with k̂ a real unit vector. But there are many others.
These other solutions are not plane waves, for there is no real plane perpendicular
to a generic complex vector, so the generic solution depends on three variables,
the time and two other spatial directions. There are cases where these solutions
must be considered. In the case a real direction can be defined, namely when
~k = (ñ + iκ)k̂, the quantity ñ is also called the refractive index of the medium,
and κ the absorption coefficient .

3. Notice that already in cases b, and c, above the solutions grow exponentially in one
direction and decay exponentially in the opposite. In the case ~k = (ñ + iκ)k̂, with k̂ a
20.2. PLANE WAVES IN HOMOGENEOUS AND ISOTROPIC MEDIA. 281

real unit vector, if sign(ñ) = sign(κ), say positive, then we have a wave traveling along
k̂ and decaying along the same direction. This is acceptable because it means there is
absorption from the media, but if sign(κ) is the opposite to sign(ñ), then we should be
having a medium producing the wave and this one growing exponentially as it travels.
This is not physically acceptable, so we should always have, ℑε ≥ 0.
Indeed, if
ε = |ε|eiθ , (20.12)
then
q
n = ± |ε|eiθ/2 , (20.13)
and so

sign(ñ) = sign(κ) ⇐⇒ 0 ≤ θ/2 ≤ π/2 or π ≤ θ/2 ≤ 3π/2 ⇐⇒ 0 ≤ θ ≤ π,

(20.14)
but that is just the condition,
ℑε ≥ 0 (20.15)

π
2

n
ε
0
π
ℜ
n
3π
2

Figure 20.1: Relation between ε and n in the complex plane.

~ ω = − c ~k ∧ E
4. Since B ~ ω , and ~k · E
~ ω = 0, we have,
ω

~⋆ = − c E
~ω ∧ B
E ~ ⋆ ) = c [−(E
~ ω ∧ (~k ⋆ ∧ E ~ ω · ~k ⋆ )E
~ ⋆ + ~k ⋆ E ~ ⋆ ].
~ω · E
ω ω ω ω
ω ω
Thus, the time averaged Poynting vector is given by,

~ c ~ ~ ⋆ c2 1 ~ ⋆ ~ ~ ⋆ ~ ~ ω · (~k − ~k ⋆ ))E
~ ⋆ } + ℜ{~k}|E
~ ω |2 ]
hSi = ℜ{E ω ∧ B ω } = [ {(E ω · (k − k ))E ω − (E ω
8π 8πω 2
282 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA
⋆
In particular, along the real direction ~l := −i(~k − ~k ) we have,

2
~ · ~l = c (ℜ{~k} · ~l)|E
hSi ~ ω |2
8πω
Thus, we see that the standard expression for Poynting’s vector is only valid for trans-
parent media, κ = 0.

Exercise: Check that our model satisfies the above condition, 20.15.

Exercise: Check that |~k ∧ E|

~ 2 := (~k ⋆ ∧ E
~ ⋆ ) · (~k ∧ E)
~ = (~k · ~k ⋆ )E ~ ⋆ − (~k · E
~ ·E ~ ⋆ )(~k ⋆ · E).
~

Exercise: Check that the group velocity in the high frequency limit is smaller than c.

20.3 Reflection and Refraction across interfaces, the

Perpendicular Case.
Here we study the case of a plane wave whose symmetry plane coincides with the interface
plane. In that case the symmetry of the incoming wave is respected and the whole solution
would have such symmetry. Otherwise, by applying a rotation along the symmetry plane we
would obtain another solution for the same incoming wave, that is for the same (asymptotic)
initial conditions.
Thus, we expect to find a solution composed of three plane waves, all conforming with the
symmetry (that is with their wave vectors proportional to each other), the incoming wave,
the transmitted wave on the other side of the interface and a reflected wave on the same side
of the incoming wave.
Maxwell’s equations provide us with the jumping conditions for this problem, indeed,
performing loop and box integrations as in the electrostatic/magnetostatic cases, (assuming
µ = 1 and so B ~ = H,
~ and that the time derivatives of the fields are finite at the interfaces,
so that their contributions vanish in the limit the integrations are at the boundary), we get

n̂ ∧ [E~ ω ]|S = 0 (20.16)

~ ω ]|S = 0
n̂ · [B (20.17)
n̂ ∧ [B~ ω ]|S = 0 (20.18)
~ ω ]|S = 0
n̂ · [n2 E (20.19)

Where[·] indicates the field difference across the interface, and n2 := εµ. So we see that in
this case the only lack of continuity is in the normal component of the electric field.
20.3. REFLECTION AND REFRACTION ACROSS INTERFACES, THE PERPENDICULAR CASE.

We shall choose coordinates so that the plane z = 0 is the interface plane, to the left we
shall have a medium with refraction index n, and an incident wave on it. To the right we
shall denote the refraction index by n′ .
Since the jumps are fixed, we must have that the waves have all the same time dependence,
so the ω’s are all the same. Thus, we expect to have three waves, two on the side (left) where
the wave is coming, (the incoming one and the reflected one), and one on the side the wave
is leaving the interface.

~ I e−i(ωt−kI z) + E
~ R e−i(ωt−kR z) z < 0
(
~ ~x) = E
E(t, ~ T e−i(ωt−kT z) (20.20)
E z>0
From the conditions,

ω 2 n2 ω 2 n′2 n′2 2
kI2 = kR
2
= and k 2
T = = k
c2 c2 n2 I
where n, (n′ ) are the left, respectively right, refraction indexes. From this we see we can take
′
kR = −kI (the other sign would just produce two waves in the same direction), and kT = nn kI .
We must see now what are the relative field strengths. From the above boundary conditions
and equations 20.2-20.5 we get the following relations:

E~I + E ~R − E ~ T = 0 continuity of electric field, (20.21)

~I − E
n(E ~ R ) − n′ E
~ T = 0 continuity of magnetic field, and 20.3. (20.22)

From which we obtain,

~T = 2n ~
E EI (20.23)
n + n′
~R n − n′ ~
E = EI (20.24)
n + n′
For transparent media (ε real and positive), we have three plane waves without any dis-
persion nor dissipation and without any phase difference. In that case the energy fluxes of the
incoming wave minus the one of the reflected wave should equal the one of the transmitted
~ = −n~kˆ ∧ E
wave, indeed, recalling that B ~ we have,

hSlef t iT · n̂ = (hS~ I iT − hS ~ R iT ) · n̂
cℜn ~ 2 ~ R |2 }
= {||E I | − |E
8π
c |n − n′ |2 ~ 2 c 4ℜn|n||n′ | ~ 2
= ℜn(1 − )| E I | = |E I |
8π |n + n′ |2 8π |n + n′ |2
while,
′ 2 ′
~ T iT · n̂ = cℜn |E
hSright iT · n̂ = hS ~ T |2 = c 4|n| ℜn |E ~ I |2
8π 8π |n + n′ |2
284 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

These two expressions coincide, as it should be because of energy conservation.

Exercise: Assume the left side medium is transparent while the one at the right is dissipative.
Should the energy fluxes also coincide at the interface? Compute the energy flux as a function
of the depth in the right media.

Exercise: If the incident media is not transparent both expressions do not coincide. Nev-
ertheless energy conservation must hold, so there is an assumption which is invalid. Which
one?

Exercise: Compute Poynting’s vector for the total field on the left (incident plus reflected)
and see under which conditions holds that it is the sum of individual Poynting’s vector for
each of the two fields.

20.4 Reflection and Refraction, the General Case.

We consider now the general case of reflection and refraction, that is the phenomena that
can occur when a plane wave reaches a flat surface of discontinuity between two otherwise
homogeneous and isotropic media. We recall the general solution for plane waves,

~ ω e−i(ωt−~k ·~x) )
~ ~x) = ℜ(E
E(t,
~ ω e−i(ωt−~k ·~x) ),
~ ~x) = ℜ(B
B(t,

~~ 2
with ω 2 = k ·nk2c , and ~k · E
~ ω = 0. For some cases we shall allow ~k to be complex, and not only
′ ′
of the form ~k = a~k with a complex and ~k real. When ~k does not define a real direction the
solutions are strictly speaking no longer plane waves, for they do not have a planar symmetry
in real space.
We shall choose coordinates so that the plane z = 0 is the interface plane, to the left we
shall have a medium with refraction index n, and an incident wave on it. To the right we
shall denote the refraction index by n′ . Let us assume that the incident wave is a plane wave
with wave number, ~k I , real. Let us denote by n̂ the unit normal to the discontinuity surface.
If n̂ is parallel to ~k I we are in the case already considered, so we assume they are not parallel.
Thus, both define a plane, called the incidence plane, in our figure (20.1), the (x, z) plane.
As before we shall propose an ansatz consisting in a transmitted wave, with wave number
vector ~k T , and a reflected wave, with wave number vector ~k R . The direction perpendicular to
the incident plane is common in both sides, and so will remain a symmetry direction, in our
scheme that coincides with the ŷ direction, so we expect the wave vectors will not have any
component along it, kIy = kT y = kRy = 0.
20.4. REFLECTION AND REFRACTION, THE GENERAL CASE. 285

kT kI
n

kI

Figure 20.2: The incidence plane, 3 dimensional view.

z
~kT

θT

~kR
θI θR
~k I

Figure 20.3: The incidence plane.

286 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

In order to be able to satisfy the jump conditions at the interface we need the waves at both
sides to have the same functional dependence both in time and in the directions tangent to
the interface, that is at the plane z = 0. This means that ωI = ωR = ωT and kIx = kT x = kRx .
Furthermore, we have that
2 2 ′2 2
~k I · ~k I = ~k R · ~k R = n ω ~k T · ~k T = n ω . (20.25)
c2 c2

Since kIx = kRx and ~k I · ~k I = ~k R · ~k R , that is kI = kR , we have,

kRx kIx
= . (20.26)
kR kI
If the medium to the left is transparent, then (see figure 20.3)
kRx kIx
sin θR := = := sin θI , (20.27)
kR kI
so the reflection angle equals the incidence angle.
Furthermore, we have,
2
kIz = kI2 − kIx
2 2
= kR 2
− kRx 2
= kRz or kIz = ±kRz . (20.28)

We shall take kIz = −kRz for we know we need a reflected wave (traveling to the left) to have
a consistent solution.
′2 2 ′2 2 2 ′2
Since kT2 = n c2ω = nn2 n c2ω = nn2 kI2 , in the case the right medium is transparent, we have
kT x kIx n kIx n
sin θT :== = ′ := ′ sin θI , (20.29)
kT kT n kI n
which in optics is known as Snell’s law. Notice also that,

n′2 2 2 n
′2
kT2 z
= kT2
− = kT2 x
− kT2 2
= 2 kI − kIx = kI ( 2 − sin2 θI ).
kIx 2
(20.30)
n n
Thus, if n′ < n, both real, then there are incidence angles for which kT z vanishes or becomes
purely imaginary. We call the limiting angle, for which kT z = 0, the total reflection angle, θr ,
it is defined by the relation,

n′
sin θr =
.
n
For those angles with θI > θr the field decays exponentially inside the medium on the right,
and so we have total reflection. On the above equation we must take the positive root, that
is ℜkT z ≥ 0, in order to have a wave traveling to the right. If ℜkT z = 0, that is when n′ < n
′2
both real, and sin2 θI > nn2 , then one must choose, for consistency, the root with ℑkT z ≥ 0,
so that the wave decays to the right. If both ℜkT z 6= 0, and ℑkT z 6= 0 then we can only make
one choice and check for the other, that is, if we choose the root with ℜkT z ≥ 0, then we must
check that ℑkT z ≥ 0. If this does not happen, that is if ℑ( nn2 − sin2 θI ) = ℑ εε < 0, then we
′2 ′

say one of the susceptibilities is unphysical.

20.4. REFLECTION AND REFRACTION, THE GENERAL CASE. 287

We now look for the relative field strengths. We consider two separate cases, in the first
we shall assume the electric field is perpendicular to the incidence plane, that is in the ŷ
direction, in the second that it is in the incidence plane, the general case can be obtained by
decomposing the general wave into these two directions.

20.4.1 Electric field perpendicular to the incidence plane

For the first case, we have the same situation as for the case of normal incidence already
considered, except that now the magnetic field will not, in general be tangent to the interface.

EIy + ERy − ET y = 0 continuity of electric field, (20.31)

kIz (EIy − ERy ) − kT z ET y = 0 continuity of magnetic field, and 20.3. (20.32)

Where we have only considered the x component of equation 20.3, since the z component is
automatically satisfied from the continuity of the electric field and the tangential component
of the wave vector. As before we obtain,

ET y 2kIz 2 cos θI 2 cos θI sin θT

= = =
kIz + kT z sin (θI + θT )
r
EIy n′ 2

cos θI + n
− sin2 θI
r 2
n′
ERy kIz − kT z cos θI − n
− sin2 θI sin (θT − θI )
= = = , (20.33)
kIz + kT z sin (θI + θT )
r
EIy n′ 2

2
cos θI + n
− sin θI

that is, Fresnel’s law. The second formulas are valid when the incident medium is transparent.
The third formulas, in terms of incidence and transmission angles, are valid only when both
media are transparent.

0
5

-1

-4
-2
0
2 10
4

Figure 20.4: EI and ET for the case where the electric field is tangent to the interface. Case
n′
n
< 1.
288 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA
0

10

1.0

0.5

0.0

-4 -0.5
-2
0
2 -1.0
4

Figure 20.5: EI and ET for the case where the electric field is tangent to the interface. Case
n′
n
> 1.

The other components of the electric field, namely those at the incidence plane, are zero.

Exercise: Consider the equations for these other components, show that they form a linear
homogeneous system and compute its determinant.

Exercise: Compute the total averaged Poynting vector along the interface normal, both, on
the left and on the right of the interface, and show that they are given by:
2
~ >T = 1 ℜkT z |kIz | |EI |2
n̂· < S
2π |kI + kT |2
Note: On each side compute the Poynting vector corresponding to the total electromagnetic
field.

20.4.2 Electric field at the incidence plane

In this case the magnetic field will be perpendicular to the incidence plane, indeed, both,
~k I , and E
~ I are at the incidence plane, and are perpendicular to each other, and so B
~ I has
~ ~ ~ ~ ~ ~
to be perpendicular to both, B I = (0, BIy , 0). Since B = k ∧ E and k · E = 0, we have,
~k ∧ B ~ = −(~k · ~k)E
~ and therefore,

~ ~
~ = −k ∧ B .
E
k2
~ ~
Thus, we expect to have the same formulas, but replacing ~k I by kn2I , and ~k T by kn′2T .
Thus, we get,
20.4. REFLECTION AND REFRACTION, THE GENERAL CASE. 289

′
BT y 2n′2 kIz ( nn )2 cos θI sin 2θI
= = 2 =
n kIz + n kT z sin (θI + θT ) cos (θI − θT )
q
BIy ′2 2 ′ ′
( nn )2 cos θI + ( nn )2 − sin2 θI
′
q
′
BRy n′2 kIz − n2 kT z ( nn )2 cos θI − ( nn )2 − sin2 θI tan (θI − θT )
= = = , (20.34)
n kIz + n kT z tan (θI + θT )
q
BIy ′2 2 ′ ′
( nn )2 cos θI + ( nn )2 − sin2 θI

As above, the second formulas are valid when the incidence medium is transparent, the third,
in terms of incidence and transmission angles, are valid only when both media are transparent.

20.4.3 reflection Coefficients

The reflection coefficients are defined as the ratio of the reflected power (normal component
of Poynting vector) vs the incident power for each mode. That is,

cos θR |E ~⋆ |
~ R ||B ~ R |2
n cos θR |E ~ R |2
|E
R
R := ⋆ = =
~ I ||B
cos θI |E ~ | ~ I |2
n cos θI |E ~ I |2
|E
I
For transparent media they become,

sin2 (θT − θR )
R⊥ = (20.35)
sin2 (θT + θI )
tan2 (θT − θR )
R|| = (20.36)
tan2 (θT + θI )

Exercise: From this coefficients, and energy conservation find a bound for the power of the
transmitted power as a function of the incident power.

20.4.4 Total Polarization Angle or Brewster angle

Note that the numerator in the expression for BRy in the case the incidence medium is
transparent can be made to vanish for a given angle. In this case we will also have R|| = 0.
Indeed, setting BRy = 0 in the expression above and squaring each part we get,
!4 !2
n′ 2 n′
cos θI = − sin2 θI
n n
 2
n′
This is a quadratic equation for x := n
. By inspection, we see that one solution of it is,

n′
!
= tan θI
n
Thus, there is always an incidence angle, called the Brewster angle ,
290 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

n′
!
θb := arctan
n
such that the reflected wave is totally polarized in the direction tangent to the interface.

Exercise: What does represents the other solution?

From the above expression we see that this incidence angle satisfies,
n
 
cos θb = sin θb = sin θT
n′
where we have used 20.29, and where θT is the transmission angle corresponding to the inci-
dence angle θb . Therefore, we see that θT = π2 − θb so the incidence direction is perpendicular
to the transmission direction. Notice that in this case the denominator in 20.36 for R|| tends
to infinity. So this coefficient indeed vanishes.

20.4.5 Total reflection

Another interesting feature of the reflection coefficients arises in the case in which the reflection
media is less optically dense than the incidence media, n′ < n. As we saw, in this case, for
angles bigger than θr , those fields have the form,

a − ib
, a and b real
a + ib
so they absolute value is unity. Thus, R⊥ = R|| = 1. Energy conservation then implies that
the transmitted flux in the incidence direction vanishes. This does not mean that the fields
inside vanish, they decay inside, but their Poynting vector is tangent to the interface. Note
that although the magnitude of the reflected field coincide with that of the incidence one,
there is a phase change between them. It is given by,

a − ib δ b
e−iδ = , ⇐⇒ tan =
a + ib 2 a

Exercise: Find the expression for the phase difference between the parallel and perpendicular
cases.

The general behavior of the reflection coefficients can be seen on the following figures
[20.6-20.7]. Notice that R|| ≤ R⊥ so that in general the reflected light is (partially) polarized
in the normal direction to the incidence angle.

20.5 Skin effect and the surface impedance of metals

An important case is the low frequency limit of a conducting medium, that is, a metal. In
that case,
20.5. SKIN EFFECT AND THE SURFACE IMPEDANCE OF METALS 291

0.5

0.4

0.3

0.2

0.1

0.5 1.0 1.5

Figure 20.6: Reflection coefficient as a function of the incidence angle for the case n′ > n.

1.0

0.8

0.6

0.4

0.2

0.5 1.0 1.5

Figure 20.7: Reflection coefficient as a function of the incidence for the case n′ < n.
292 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

i4πσ
ε′ ≈ , |ε| >> 1,
ω
and so,
s
2πσ
n′ ≈ (1 + i),
ω
We shall now study the behavior of a plane wave traveling in a non-absorbing media when it
reaches a metallic surface. In this case we have two length scales: λ = 2π k
of a wave incident
in the metal surface, and the length scale built out of the metal resistivity, λσ = σc . We shall
consider situations in which λ >> λσ . In this case we have that,

c 1 c√ 1q √
r
δ := √ = λn = δσ λn << λ
2πσω 2π σ 2π
where n is the index of the transparent medium. Thus, using the continuity of the tangential
component of the wave vector at the media/metal
√
interface we conclude that inside the metal
we will√have a wave vector given by ~k ≈ c (1 + i)k̂ and so will decay along that direction
2πσω
2πσω
like e− c z . That is, at a distance, δ the field decreases to half its value. The fact that
we have decay does not mean there is absorption, it could mean reflection, or in general,
scattering. The skin effect is the fact that the electric fields penetrate a metal a length δ
called the metal skin thickness.
We have seen that with the conditions above the derivatives which are really important
in Maxwell’s equations inside a metal are just the normal derivatives. Thus, inside the metal
the solution would be approximately a plane wave perpendicular to the metal surface. That
is, defining ζ = n1′ ,
s
~T = µ′ ~ ~ T ∧ k̂.
E H T ∧ k̂ := ζ H (20.37)
ε′
For the case we are dealing with,
s √
ωµ′ 1 − i δσ
ζ = ζr + iζi = (1 − i) = √ |ζ| << 1,
8πσ 2 λn
where we are considering µ and σ (the resistivity) to be real.
Expression 20.37 can be considered as a boundary condition to the external fields. This is
so because they are tangential components of the fields and so are continuous across the metal
surface. Indeed, for the expert, this expression can be considered as a mixture of outgoing
and incoming modes for Maxwell’s equations. The condition is stable, and so allowed, if
the energy leaks from the interface towards the inside of the metal, that is, if the normal
component of Poynting vector is positive,
c
~ · k̂i := ~ ⋆ ) = c ζr |H
~T ∧ H ~ T |2 ≥ 0.
hS ℜ(E T
8π 8π
So 20.37 is a good boundary condition provided that,
20.6. ENERGY LOSSES IN CAVITIES AND GUIDES 293
0

10

1.0

0.5

0.0

-4 -0.5
-2
0
2 -1.0
4

Figure 20.8: EI and ET for a metal in the case where the incident electric field is tangent to
the interface.

ζr > 0.

So the correct root must be taken in the relation defining ζ.

Since ζ is small compared to unity we can approximate the outside field as the field with
trivial boundary condition where no field is allowed inside the metal. With this approximation
we can estimate, using 20.5, and the magnetic field computed, the energy lost inside the metal
by unit surface and unit time.

Exercise: Find the frequency range (or/and wave length) for which cooper satisfies the
approximations assumed above. Compute δcooper .

For cooper, we have a conductivity of, ρ = 1.68 × 10−8 Ohm m

= 1.18 × 10−17 s or σ =
5. × 1017 s−1 .
√
Defining δσ := σc = cρ, then the decay length is given by qδ = δσ λ, where λ := ωc
is the vacuum wave length of the incident wave. While |ζ| = δλσ . For cooper, we have,
δσ ≈ 10−2 nm, while visible light has a wave length of the order of λ = 100nm.

20.6 Energy losses in cavities and guides

In a good cavity the Ohmic looses will be very small, indeed, as seen above ζr ≈ 10−2 . Thus,
we can have a good measure of the rate energy is leaking by just computing the fields assuming
the cavity is perfect and thus estimate it. To have a measure independent on the field mode
strength we compute the rate of energy loss by unit energy, that is,

dE
cζr
R
~ 2 dS
|H|
dt ∂V
= ,
E ~ 2 dV
2 V |H|
R
294 CHAPTER 20. WAVE PROPAGATION IN CONTINUUM MEDIA

where we have already used that the energy is equiparted among the electric and magnetic
fields. We are also assuming, for simplicity, that the interior of the cavity is empty. Using
(19.1, 19.2) we can transform the volume integral into a surface one, obtaining,
dE R
~ 2 dS
ωζr ∂V |H| −ω
dt
=R :=
E ~ 2 ~ 2 Q
∂V [|H| − |E| ]dV
The a-dimensional quantity Q is called the quality factor of the cavity, it depends on the
geometry of the cavity and on the conductivity of its walls. It counts the cavity’s attenuation
rate in periods of the mode that has been exited.
Similar expressions are valid for wave guides, in that case the relevant quantity is the
attenuation rate per unit energy flux across the guide section. In this case the final integral
is a line integral along the section contour.
Chapter 21

Wave Packets, Kramers–Kroning

relations and Causality

21.1 Propagation of wave packets

So far we have considered single plane waves, that is waves oscillation at a given frequency,
and therefore at a given wave number, k, or at most at a finite number of them, given by the
dispersion relation,

ωn(ω) kc
k= or ω(k) = ,
c n(k)
this last being the way we are going to use it. We shall assume n(ω) real, that is a transparent
medium, otherwise the wave dissipate and the effects get confused. This relation provides the
information from the evolution equation we are solving, and, in the case of electromagnetism,
information about the medium across which the wave moves. We shall consider now a wave
packet, that is a solution made out from contributions from many monochromatic plane waves.
For simplicity, we shall consider plane waves along a single direction, x̂, we shall comment on
the fact that when waves have finite extent this is no longer the case and further dispersion
occurs. A generic Cartesian component of the electric and magnetic field have then the form,

1 ∞
Z
u(t, ~x) = √ û(k)e−i(ω(k)t−kx) dk,
2π −∞

where û(k) is the number density of waves between wave number k and wave number k =
k + dk. For simplicity, we just integrate over the whole k-space although we are thinking in
adding contributions from a narrow frequency band, this will be evident from the form of
û(k), which we shall assume smooth and of compact support.
For Maxwell equations, and in fact for all second order systems, frequencies come into
pairs, or alternatively we can think that for each frequency there are two wave numbers for
each wave propagation direction,

±ωn(ω) ±kc
k= or ω(k) = ,
c n(k)

295
296CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

so the solution actually looks like,

1 Z∞
u(t, ~x) = √ {û(k)− e−ik(ct/n−x) + û(k)+ eik(ct/n+x) } dk,
2π −∞

And we need to determine two functions from the initial data.

Exercise: From the expressions,

1 Z∞
u(0, ~x) = √ {û(k)− + û(k)+ }eikx dk,
2π −∞
∂u(t, ~x) ic
Z ∞
|t=0 = √ k{−û(k)− + û(k)+ }eikx dk,
∂t 2π −∞

determine û(k)+ and û(k)− in terms of the initial data.

We shall consider now propagation of wave packets, that is where the initial data is of
compact support, initially spatially centered around a point, and such that they Fourier
coefficients are also centered around a given, dominant, frequency.

21.1.1 Non-dispersive Waves

Let first consider the case
±k
ω(k) =
.
c
In this case, corresponding in Maxwell theory to the vacuum case, µε = 1, or the hight
frequency limit of generic materials, all phase speeds are the same, vp := ωk = c and so we
expect the wave packet to move undistorted, indeed, the solution is in this case,
1 ∞
Z
u(t, ~x) = √ {û(k)− e−ik(ct−x) + û(k)+ eik(ct+x) } dk,
2π −∞

and we see that the solution will be of the form,

u(t, ~x) = u− (0, x − ct) + u+ (0, x + ct),

representing two wave packets, one moving (undistorted) to the right, other, (also undistorted)
to the left. Although the original packet spreads out into two packets moving in opposite
directions these waves are considered non-dispersive, for this is just and effect of taking
different propagation directions in space.

21.1.2 Dispersive Waves

If the dispersion relation is non-linear then we should have dispersion, that is, each wave
would travel at a different speed and the original packet would distort and spread. To see
this we shall first consider an example that can be solved explicitly. Let us assume,
21.1. PROPAGATION OF WAVE PACKETS 297

a2 k 2
ω(k) = ω0 + vk + ,
2
and take as initial data,
−x2
u(0, ~x) = e 2L2 cos(k0x)
∂u(t, ~x)
|t=0 = 0. (21.1)
∂t
So the wave has initially a large peak at x = 0 and oscillates at k = k0 . Since the time
derivative is zero we expect to have two wave packets moving initially in opposite directions,
indeed, since u(k)− = u(k)+ = u(k) we get,

1 ∞ −x2
Z
û(k) = √ e−ikx e 2L2 cos(k0 x) dx
2 2π −∞
1
Z ∞
−x2
= √ [e−i(k−k0 )x + e−i(k+k0 )x ]e 2L2 dx
4 2π −∞
1 Z ∞ −(k−k0 )2 L2 /2−[−i(k−k0 ) √L + √x ]2
= √ [e 2 L 2 + (k0 → −k0 )] dx (21.2)
4 2π −∞
where in the last line we have indicated the other term as obtained from the one written
by substituting k0 → −k0 . Taking as variable y := −i(k−k
√ 0 )L + √
2
x
L 2
, dy = √dx
2L
the integral
becomes,

√
2L ∞ −(k−k0 )2 L2 /2
Z
2 2 2
û(k) = √ [e + e−(k+k0 ) L /2 ]e−y dy
4 2π −∞
√ √
2L π −(k−k0 )2 L2 /2 2 2
= √ [e + e−(k+k0 ) L /2 ]
4 2π
L −(k−k0 )2 L2 /2 2 2
= [e + e−(k+k0 ) L /2 ] (21.3)
4
√
So û(k) has two peaks, one at k0 and the other at −k0 , both have with δk = L2 . We shall
compute the time evolution of this data. For simplicity, we shall concentrate in the term
corresponding to the peak at k = k0 , which we shall call u− (t, ~x), the other is obtained by
substituting k0 → −k0 on the expressions we shall derive.
We have,

L ∞ Z
2 2
u− (t, ~x) = √ e−(k−k0 ) L /2 e−i(ωt−kx) dk
4 2π −∞
L
Z ∞
a2 k 2 2 2
= √ e−i[(ω0 +vk+ 2 )t−kx]−(k−k0 ) L /2 dk
4 2π −∞
to compute this integral we also use the same trick as before and complete squares. Calling
R the exponent inside the integral we have,
298CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

B2 B √
R = A + 2kB − k 2 C = (A + ) − ( √ + Ck)2
C C
which we use with

−i L2 ia2 t
A = −iω0 t − k02 L2 /2 B= (vt − x) + k0 L2 ) C= (1 + 2 ),
2 2 L
B
√ √
Defining l := √C + Ck, dl = Cdk, we get,

L 1 ∞
Z
2 2
u− (t, ~x) = √ √ eA+B /C e−l dl
4 2π C −∞
√
L π 2
= √ √ eA+B /C
4 2π C
(vt+a2 k0 t−x)2
− 4 2
2L2 (1+ a t )
e L4
= q e−iγ
ia2 t
1+ L2
vg t−x 2
−( 2L(t) )
e
= q e−iγ
ia2 t
1+ L2
(21.4)
q
dω a4 t2
where we have defined vg := |
dk k=k0
:= v + a2 k0 , L(t) := L 1 + L4
, and

a2 t 2a2 t2 ω0 (tv−x)2
(v + ω0 /k0 − x)k0 + 2L2
( L2 + k02 L2 + L2
)
γ := 4 2
1 + aLt4

is an unimportant phase factor. We see from this example that:

• The peak moves as a whole with speed vg . This is called the group velocity of the
packet. In general is defined as vg := dω |
dk k=k0
, where k0 is some point considered as the
dominant wave number of the given packet. Note that the more spread is the initial
packet in physical space the more peaked is in Fourier space, and therefore the better
defined is the group velocity.

• The initial peak had width L and its Fourier transform one of width L1 . This is a general
effect and it is phrased saying that the product of both widths is constant, ∆x∆k ≥ 1.

under evolution, ∆k = L1 . But in

• In Fourier space the width of the peak does not change q
4 2
physical space it does gets enlarged, ∆x(t) = L(t) = L 1 + aLt4 . The peak gets smaller
in high as the square root of L(t). So that peak squared (intensity) times its area keeps
about constant. The solution disperses as its components move with different speeds
(no dissipation is present in this model).
L2 a2 t a2
• For long times, t >> a2
, L(t) ≈ L
so a dispersion speed can be defined by L
.
21.2. GROUP VELOCITY 299

21.2 Group Velocity

To see how the group velocity appears in a generic case let us assume we have an initial data,
u(x), whose Fourier transform, û(k) has a peak at k0 , with a width ∆k0 . Expanding in Taylor
series the dispersion relation,

dω 1 d2 ω
ω(k) = ω(k0) + |k=k0 (k − k0 ) + |k=k̃ (k − k0 )2
dk 2d k
2

where the last term represents the error, we can approximate the solution as,

1 ∞ 1 d2 ω
Z
dω 2
u(t, x) = √ û(k)e−i([ω(k0 )+ dk |k=k0 (k−k0 )+ 2 d2 k |k=k̃ (k−k0 ) ]t−kx) dk
2π −∞
−i(ω(k0 )− dω | k )t Z ∞
e dk k=k0 0 1 d2 ω 2
= √ û(k)e−i(vg t−x)k e−i( 2 d2 k |k=k̃ (k−k0 ) )t dk.
2π −∞
(21.5)

Thus, for short times such that

d2 ω 2
maxk∈[k0 −∆k0 ,k0 +∆k0 ] | ∆k0 t << 1,
d2 k
dω
u(t, x) = e−i(ω(k0 )− dk |k=k0 k0 )t u(x − vg t),
and we see that apart from a phase factor the peak propagates initially with the group speed
at the main frequency. At longer times scales we start to see dispersion as we did in the
2
example above. In it, 21 dd2ωk = a2 , ∆k0 = L1 and so the dispersion start to be important when
a2 t
L2
≈ 1.

21.3 Kramers–Kroning Relations

We have studied a simple model of matter to obtain a specific relation between the displace-
ment vector and the electric field in Fourier space,

~ ω (~x) = ε(ω, ~x)E

D ~ ω (~x),

in particular for the simplest model we had,

ωp2
εω = 1 + (21.6)
ω02 − ω 2 − iγω
From this expression we can compute the actual temporal dependence between the fields
in time, as we already saw,
Z ∞
~ ~x) =
D(t, ~ − t̃, ~x) dt̃,
ε(t̃, ~x)E(t
−∞
300CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

with
1 ∞ Z
ε(t̃) = √ ε(ω)e−iωt̃ dω
2π −∞
where we omit the spatial dependence from now on.
For the case at hand now
∞ e−iωt̃
Z
ε(t̃) = δ(t̃) + ωp2 dω,
−∞ (ω+ − ω)(ω− − ω)
with,
q
−iγ ± 4ω02 − γ 2 iγ q
ω± := = − ± ν0 ν0 := ω02 − γ 2 /4, ω02 > γ 2 /4 (21.7)
2
r 2
4ω02
1± 1− γ2
ω± := −iγ , ω02 < γ 2 /4 (21.8)
2
the two roots of the denominator in 21.6. Notice that both roots are complex (recall that
γ ≥ 0). If ω02 > γ 2 /4 both are equidistant from the ℜ[ω] = 0 axis and have ℑ[ω± ] = −iγ/2.
If ω02 < γ 2 /4, the discriminant, ν0 , becomes pure imaginary and both roots are in the line
ℜ[ω] = 0. In the limit ω0 → 0 one of them goes to −iγ while the other goes to zero.
Since ε(ω) is a meromorphic function on the complex plane we can compute the above inte-
gral using techniques from complex analysis: Since ℜ[−iω t̃] = ℑ[ω]t̃ we have that ℜ[−iω t̃] < 0
whenever t̃ < 0 and ℑ[ω] > 0. So for t̃ < 0 we can deform the integration along the ℑ[ω] = 0
axis into the path C+ (ρ) shown in the figure 21.1, since no pole is encounter along the defor-
mation the value of the integral along the deformed path does not change. But when ρ → ∞
the argument on the exponent has real part going to −∞ and so the integral goes to zero.
Thus, it must vanish at all paths and so at the original one.

C+ (ρ)

ω+ ω−

Figure 21.1: Integration path for t̃ < 0.

We reach the conclusion that ε(t̃) = 0 ∀t̃ < 0. Thus,

Z ∞
~ ~x) =
D(t, ~ − t̃, ~x) dt̃,
ε(t̃, ~x)E(t
0
21.3. KRAMERS–KRONING RELATIONS 301

and the relation is causal.

If t̃ > 0 then ℜ[−iω t̃] < 0 whenever ℑ[ω] < 0, so we can deform the integral along the
path C− (ρ) shown in the figure 21.2

ρ
ω+ ω−

C− (ρ)

Figure 21.2: Integration path for t̃ > 0.

This time the path crossed the two points where the ε(ω) has poles, and so we pick these
contributions. Noticing that the limiting path (ρ → ∞) does not contribute we get,

1 e−iω+ t̃ e−iω− t̃
ε(t̃) = δ(t̃) + ωp2 (−i2π) √ [ + ]
2π ω− − ω+ ω+ − ω−
iπωp2 1
= δ(t̃) + √ [e−iω+ t̃ + e−iω− t̃ ] t̃ ≥ 0
ν0 2π
(21.9)

In particular if ω02 > γ 2 /4,

i2πωp2 1 γ t̃
ε(t̃) = δ(t̃) + √ θ(t̃)e− 2 sin(ν0 t̃),
ν0 2π
and so,
Z ∞
~ ~x) = E(t,
D(t, ~ ~x) + ~ − t̃, ~x) dt̃,
G(t̃, ~x)E(t
0

with,

i2ωp (~x)2 π 1 γ t̃
G(t̃, ~x) = √ θ(t̃)e− 2 sin(ν0 t̃).
ν0 2π
Thus, we not only see that the relation is causal, but also that it depends on the past
behavior on a time intervals of the order of γ −1 , which in normal materials is of the order of
10−10 seconds.
302CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

In the case of metals, where the presence of free electrons imply in our simple model that
ω0 = 0, we obtain, in this limit ω− → −iγ while ω+ → 0, that is the case where we have a
simple pole at the origin. In terms of G(t) this indicates G(t) → 4πσ as t → ∞. Indeed, in
the case G(t) = 4πσ,
Z ∞
~
D(t) ~ + 4πσ
= E(t) ~ − t̃) dt̃,
E(t
0
and so

Z ∞
~
∂t D(t) ~ + 4πσ
= ∂t E(t) ~ − t̃) dt̃
∂t E(t
0
Z ∞
~ − 4πσ
= ∂t E(t) ~ − t̃) dt̃
∂t̃ E(t
0
~ − 4πσ[E(−∞)
= ∂t E(t) ~ ~
− E(t)] dt̃
~ + 4πσ E(t)
= ∂t E(t) ~

~
where in the last step we have assumed no initial electric field is present (E(−∞) = 0). We
see that this way recuperate Maxwell’s vacuum equations with an Ohmic current given by,
4πσ ~
J~ = E.
c
From the above calculation it is clear that always that the poles of ε(ω) are in the lower
~ and E.
half plane we shall have a causal relation between D, ~
We shall explore now the reverse, namely what are all the implications and restrictions
that the causality conditions imposes on ε(ω). They are called the Kramers-Kroning relations.
They appear in many places in physics and more generally in information theory, every time
we have a relation between two quantities which is causal, and linear,
Z ∞
~
D(t) ~ +
= E(t) ~ − t̃) dt̃,
G(t̃)E(t
0

with G(t̃) smooth and bounded. We shall study properties of this function. We shall follow
the simpler argument given in [3] which does not use Cauchy’s theorem and so does not
assume the involved functions to be meromorphic. Its Fourier transform is:
Z ∞
Ĝ(ω) := ε(ω) − 1 = G(t)eiωt dt. (21.10)
0

For simplicity in the argument we now extend G(t) to all of t and write is as,

G(t) := G E (t) θ(t), (21.11)

O

with θ(t) the step function whose value is 1 for t > 0, and otherwise vanishes. For G± (t) we
mean an extension of G(t) to negative values of t as G± (t) = ±G± (|t|), that is an even and
an odd extension.
With this extension it is easy to compute Ĝ(ω) using the convolution formula for the
Fourier transform of a product in equation (21.11):
21.3. KRAMERS–KRONING RELATIONS 303

1 ∞
Z
Ĝ(ω) = Ĝ± (ω̃)θ̂(ω̃ − ω) dω̃.
2π −∞
Now, the Fourier transform of the step function is given by:
i
θ̂(ω) = πδ(ω) − .
ω
Thus, using it in the previous expression we arrive at the following relation:

1 −i Z ∞
Ĝ± (ω)
Ĝ(ω) = Ĝ± (ω) + PV dω̃.
2 2π −∞ ω̃ − ω

Consider first the case in of the even extension GE (t), notice that its Fourier transform is
real, so we get:

1 −1 ∞ Ĝ (ω) Z
ℜ[Ĝ(ω)] = Gˆ+ (ω)
+
ℑ[Ĝ(ω)] = PV dω̃.
2 2π −∞ ω̃ − ω

Substituting we arrive at the first Kramers–Kroning relation,

−1 ∞ ℜ[Ĝ(ω)]
Z
ℑ[Ĝ(ω)] = PV dω̃.
π −∞ ω̃ − ω

Exercise: : Using the odd extension get the other relation, namely,

1 Z ∞
ℑ[Ĝ(ω)]
ℜ[Ĝ(ω)] = PV dω̃.
π −∞ ω̃ − ω

So, we see that these relations assert that knowing the real part we can determine the
imaginary one and vice versa That is there is a relation between the dissipation part of the
relation (the imaginary part) and the propagation part (the real part). It is easy to see that,
∞ 1
Z
PV dx = 0,
0 x2 − ω 2
and so,
4πiσ
ε(ω) = ε0 +
ω
is a solution to the above relations.

Exercise: Check the assertion above.

Furthermore, one can check that if we choose any ℑ[ε(ω)] defined in ω ∈ [0, +∞], odd and
analytic, then the first relation defines a ℜ[ε(ω)] even and analytic. If we plug this function
in the second relation we get the original ℑ[ε(ω)], but corrected with a possible pole at the
origin, that is a different σ, and ε0 . We can think of these relations as one being the inverse
304CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

of the other, except that each one of them have a non-zero kernel, and so they are not strict
inverses. So they are linear transforms in the same sense as the Fourier one.
From the experimental point of view these relations are very important for, for example
the knowledge of ℑ[ε(ω)] from absorption experiments (energy losses) allows to infer ℜ[ε(ω)],
and so ε(ω). Or vise-versa, some times it is difficult to perform absorption experiments
because the media is too opaque but reflection experiments are easy and so one obtains the
information that way.
Consider the extreme (limiting) case of a unique absorption line,

ℑ[ε(ω)] = Kδ(ω − ω0 ) + ℑ[ε̃(ω)]

with ε̃(ω) some analytic extra part. In this case we get,

2K ω0
ℜ[ε(ω)] = 1 + + ℜ[ε̃(ω)]
π ω02 − ω 2
This effect can be easily seen in the following experimental result, 21.3

Figure 21.3: Measured extinction curves of F and M color centers in a KBr crystal at dif-
ferent temperatures (a); the corresponding changes of the refractive index are computed via
Kramers-Kronig relation (b). The curves correspond to 1) 250 K, 2) 170 K, 3) 100 K, and
4) 35 K, with peaks becoming sharper at low temperature. Adapted from Peiponen et al.
(1984).
21.4. EINSTEIN’S CAUSALITY 305

21.4 Einstein’s Causality

We have seen that there are two kinds of velocities in wave propagation, the phase velocity,

ω(~k) c
vp := = ,
k n(k)
the speed at which individual waves of wave number ~k moves, and the group velocity,

∂ω ∂ ck k ∂n
~vg := = ( ) = vp (k̂ − )
∂~k ∂~k n(k) n ∂~k
the velocity at which a peaked wave packet moves in first approximation. For many substances
n(~k) > 1 and | ∂n |, and both velocities are smaller than the speed of light. But there are cases
∂~
k
in which this is not so, nevertheless, under very generic conditions, which we spell in detail
below, and which have clear physical meaning, the systems remains Einstein causal, that is,
no propagation speed exceeds c. The scope of this proof is not so wide, as is stands only
works for wave packets where all the waves have the same propagation direction, but it is
simple and adds a lot of understanding on the subject.

Theorem 21.1 Let n(ω) be real analytic in ω, so it admits a unique extension to the complex
plane which we also denote with n(ω). Let such an extension satisfies:

• n(ω)−1 is analytic for ℑ[ω] ≥ 0.

• n(ω) → 1 as |ω| → ∞ in all directions with ℑ[ω] > 0.

Then the propagation is Einstein causal.

Exercise: Show that the ε(ω) of our model satisfies this conditions.

Answer: In our simple model,

e2 N 1
ε(ω) = 1 + 4π 2
, γ ≥ 0,
m ω0 − ω 2 − iγω
so it is real analytic. n−1 (ω) could have singularities only at points where ε(ω) vanishes, for
that we need the denominator to be real. That is,

ℑ[ω02 − ω 2 − iγω] = −2ℑ[ω]ℜ[ω] − γℜ[ω] = 0,

that is,
−γ
ℜ[ω] = 0 or ℑ[ω] = .
2
306CHAPTER 21. WAVE PACKETS, KRAMERS–KRONING RELATIONS AND CAUSALITY

At ℜ[ω] = 0 the denominator is positive (ℜ[ω02 − ω 2 − iγω] = ω02 − ℜ[ω]2 + ℑ[ω]2 + γℑ[ω]) and
so ε(ω) can not vanish. Therefore, we could only have poles at ℑ[ω] < 0.

We now prove the theorem. Let u(0, x) be a smooth and bounded, but otherwise arbitrary,
function with support on x < 0. We shall proof that it vanishes for x > ct. Thus, there is no
propagation faster than light. Its Fourier transform, is
1 ∞
Z
u(0, x)e−ikx dx.
û(k) :=
2π −∞
Since ℜ[−ikx] = ℑ[k]x < 0 the above function is analytic for ℑ[k] > 0.
The solution then is,

1 Z∞
u(t, x) = û(k)e−i(ω(k)t−kx) dk
2π −∞
1 ∞
Z
= û(k)e−ik(ct/n−x) dk
2π −∞
Since û(k) and n(k)−1 are analytic on the upper half of the plane 1
we could deform the
integration path as shown in the figure 21.4. Then,
Z
u(t, x) = lim û(k)e−ik(ct/n−x) dl.
ρ→∞ C+ (ρ)

But since n(k) → 1 as k → ∞, then, in the limit ρ → ∞ the argument of the exponential has
real negative part for ct − x < 0 and goes to −∞ as k → ∞. Thus, the integral vanishes and
we have Einstein Causality.

C+ (ρ)

Figure 21.4: Path deformation for Einstein Causality argument.

1
The relation between ω and k is analytic also in this region, so n(k)−1 is analytic if n(ω)−1 is.
Chapter 22

Electromagnetic Radiation produced

by charge currents.

22.1 Introduction
We want to address now the problem of determining the electromagnetic field produced by
a system of moving charges. The question in its complete generality does not have a simple
satisfactory solution since:

• The charges are affected by the radiation they themselves produce. In taking into
consideration this fact one is lead into considering also the evolution equations for the
sources. The resulting system is non-linear and so it is extremely difficult to split
fields into internal and charge generated ones. To overcome this problem we shall
assume the charge motion is given beforehand and so not affected by the radiation they
themselves produce. This is to be understood as a first step in solving a very complex
phenomenon. Once this step is taken one can consider the deviation of the source motion
this radiation produces and in some cases improve this first approximation. In other
cases this assumption is not justified and other methods/approximations must be used
to get anywhere

• Given a fixed charge distribution in space-time there are many, in fact infinitely many,
Maxwell’s solutions having them as sources, one for each initial data condition we can
give satisfying the constraint equations. Of course their radiation content is different,
so the question is: Which one to choose? The problem is only enhanced if we allow the
charges to self interact in a non-linear fashion as discussed above. If we fix the charge
motion, then we can also overcome this problem by requiring in the past the charges to
be stationary. In that case we can take initial data at those earlier times which are also
stationary. This condition fixes uniquely the initial data, and we have a problem we can
solve for. Notice that there is a catch on this: If we assume evolution equations also for
the sources, and the whole evolution systems has unique solutions for given data, then
the stationary data gives a stationary solution and no radiation!

To avoid many of the problems referred above we shall pretend now we have a given,

307
308CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

~ ~x)) satisfying the continuity equation,

fixed, charge distribution in space-time, (ρ(t, ~x), J(t,
such that for t < t0 ,

ρ(t, ~x) = ρ0 (~x), ~ ~x) = J~0 (~x).

J(t,
Notice that necessarily,
~ · J~0 (~x) = 0.
∇
~ in the Lorentz gauge,
We shall solve Maxwell’s equations in terms of the potentials, (φ, A),

~ ·A
∂t φ + c∇ ~ = 0.
The equations to solve are then,

✷φ = 4πρ
4π ~
✷A~ =
c
J.

Notice that the gauge condition imposed allows to know φ(t, ~x) once we know A(t, ~ ~x) and
have initial data values for φ. Indeed, integrating in time (at fixed ~x) equation 22.1 we have,
Z t
φ(t, ~x) = φ0 (~x) − c ~ · A(
∇ ~ t̃, ~x) dt̃
t0

Since at t = t0 the sources, and the electromagnetic field, are summed to be stationary we
must have,

∆φ0 (~x) = 4πρ0 (~x),

and so, with appropriate boundary conditions, φ0 (~x) is uniquely determined, as is φ(t, ~x) once
~ ~x) is obtained. Thus, we can concentrate on the second equation,
A(t,
4π ~
✷A~ = c
J.
Assertion: Retarded Green Function
′
~ − |~x−~x | , ~x′ )
~ ~x) = 1 J(t
Z
A(t, c
d3~x′ , (22.1)
c lR3 |~x − ~x′ |
with the components of J~ expressed in Cartesian coordinates.
Proof:
We have seen that the solution to the wave equation,

✷ψ(t, ~x) = 4πf (t, ~x),

with initial data,

ψ(t0 , ~x) = ψ0 (~x), ∂t ψ(t, ~x)|t=t0 = ψ1 (~x),

22.1. INTRODUCTION 309

was,

∂
ψ(t, ~x) = [(t − t0 )Mc(t−t0 ) (ψ0 (~x))] + (t − t0 )Mc(t−t0 ) (ψ1 (~x))
∂t Z
t−t0
+ 4π t̃Mct̃ (f (t − t̃, ~x) dt̃,
0

with
c
Z
Mct (g(~x)) = g(~x + ctn̂) dΩ2 ,
4π S2

and n̂ the vectors covering a unit sphere, S 2 .

Since the general solution given above consists of two parts, an inhomogeneous one and
a homogeneous due to the initial data, we can consider first the inhomogeneous one, namely
consider the case J~0 (~x) = A
~ 0 (~x) = 0. In this case only the integral term remains, (ψ0 (~x) =
ψ1 (~x) = 0), and so applying the above formula to each component of A ~ we get,
Z t−t0
~ ~x) = 4π
A(t, ~ − t̃, ~x)) dt̃.
t̃Mct̃ (J(t
0

Since we are considering the case J~0 = 0 we can extend the integral to t0 = −∞ without
′
adding anything. Changing variables to ~x′ = ~x+ct̃n̂, we have, t̃ = |~x−~x | , (ct̃)2 d(ct̃)dΩ2 = d3~x′ ,
c
and the integral becomes,

Z ∞ Z
~ ~x) =
A(t, ct̃ ~ − t̃, ~x + ct̃n̂) dt̃ dΩ2
J(t
0 S2
1 1 ∞
Z Z
= ~ − t̃, ~x′ ) (ct̃)2 d(ct̃) dΩ2
J(t
c 0 ct̃ S 2
′

1 Z
J~(t − |~x−c~x | , ~x′ ) 3
= d ~x,
c lR3 |~x − ~x′ |

We consider now the general case, J(t,~ ~x) = J~0 (~x); t < t0 . We first notice that the
above expression is a solution to the wave equation for all values of t, if we define now
′
~ ~x) = J~0 (~x); t < t0 . Second we notice that if t < t0 then t − |~x−~x | < t0 and so,
J(t, c

~ ~x) = 1 J~0 (~x′ ) 3

Z
A(t, d ~x, t < t0 ,
c lR3 |~x − ~x′ |
so it is time independent and a solution to

4π ~~ x) =
J 0 (~x).
∆A(~
c
The stationary equation for the vector potential, so in fact this is the solution we wanted for
the general case!
310CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

(t, ~
x)

|~ x′ | = ct
x−~

t = t0

Figure 22.1: The Current Integral.

22.2 Asymptotic behavior of the fields

We are interested in describing the behavior of the radiation field of a source of compact
support, so we consider |~x′ | ≤ R where R is length scale of the source.

22.2.1 Large Space Directions

′
At large spatial distances, (t = cons), |~x| >> R, we have that t − |~x−c~x | < t0 and we are back
to the static situation. In this case, since A~ 0 (~x) satisfies Poisson’s equation and ∇
~ ·A
~ 0 = 0,

~ ~x)| ≈ O( 1
|A(t, ).
|~x|2

So, far away along spatial directions we see the electrostatic/magnetostatic fields of the far
in the past stationary distribution: A Coulombic electric field and a magnetostatic field, each
one with its own multipole distribution, obtained by expanding |~x − ~x′ | in the denominator
of Poisson’s formula.

22.2.2 Large Null Directions

If we move away from the sources along a null direction, t − |~xc | = τ = cons, then the
dependence on t and ~x of the time dependence of J~ in the integral can not be ignored. In
order to keep only leading terms, we only approximate |~x − ~x′ | by |~x| in the denominator,
′
~ − |~x−~x | , ~x′ ) as,
(other terms would give fields decaying faster than 1 ), and expand J(t
|~
x| c
22.2. ASYMPTOTIC BEHAVIOR OF THE FIELDS 311

′ ′
~ − |~x − ~x | , ~x′ ) = J(τ,
J(t ~ ~x′ )[ n̂ · ~x ] + O(|~x|2 )
~ ~x′ ) + ∂t J(τ,
c c
where n̂ = ~x . The second term will be important only if the first vanishes, or TRc ≈ 1, where
|~
x|
~
T is the characteristic time on which J~ varies, (∂t J~ ≈ JT ). Correspondingly T c would be
the wave length λ of the emitted radiation, for it should oscillate at the same characteristic
frequency as the source does. Thus, we see that the higher up terms are only important when
the wave length of the emitted radiation is of the order of the size of the source.
For longer wave length then,

1 |~x| ′ 3 ′
Z
~ ~x) ≈
A(t, J~(t − , ~x ) d ~x
c|~x| lR 3 c
−1 ~ ′ · J~(t − |~x| , ~x′ ) d3~x′
Z
≈ 3
~x′ ∇
c|~x| lR c
1 |~x| ′ 3 ′
Z
≈ 3
~x′ ∂t ρ(t − , ~x ) d ~x
c|~x| lR c
1 |~x|
≈ ∂t p~(t − )
c|~x| c

where p~(t) is the total dipole momentum of the source at time t.

The corresponding magnetic field is,

~ ~x) = ∇
B(t, ~ ∧ A(t,
~ ~x)
−1 |~x|
≈ 2 n̂ ∧ ∂t2 p~(t − ) (22.2)
c |~x| c

Since

~ ~x) = −1 ∂t A(t,
E(t, ~ ~x) − ∇φ(t,
~ ~x),
c
~
we must first compute ∇φ(t, ~x). To do that we use that it is also a solution to the wave
equation and so satisfies Poisson’s equation, but in this case we must keep terms up to second
order,

1 |~x| ′ |~x| ′ n̂ · ~x′ 3 ′

Z
φ(t, ~x) ≈ [ρ(t − , ~
x ) + ∂t ρ(t − , ~x ) ] d ~x
|~x| lR3 c c c

the first term is a constant, indeed it is just the charge at time τ = t− |~xc | , charge conservation
then implies that it does not depend on τ and therefore it does not depend on ~x, nor on t.
~ only to order O( 1 ) so we do not take it into account. The second
Thus, it contributes to ∇φ
|~
x|2
1
term gives a O( ) contribution,
|~
x|
312CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

1 |~x|
φ(t, ~x) ≈ n̂ · ∂t ~p(t − ).
c|~x| c

Thus,

~ ~x) ≈ −1 [∂ 2 p~(t − |~x| ) − n̂(n̂ · ∂ 2 p~(t − |~x| ))] = −1 (δ i j − n̂i n̂j )∂ 2 pj (t − |~x| )
E(t,
c2 |~x| t c t
c c2 |~x| t
c

~ ~x) = O( 1 ), and so to first order E

Notice that n̂· E(t, ~ and B
~ are perpendicular to each other
|~
x|2
and to the direction from which we are looking at the source, and are equal in magnitude.
Thus, we each a very important conclusion:
Electromagnetic radiation produced by a compactly supported source, when seeing from
far away looks like a plane wave traveling from the source to the observer, the magnitude
being proportional to the second time derivative of the total dipole moment as computed at
the retarded time.

22.2.3 The nature of the approximation

′
From the figure above we see that the first approximation t − |~x−c~x | ≈ t − |~xc | changes the cone
integral into a constant time surface integral, t′ = t − |~x| = cons The second approximation,
c
′ ′
|~
x−~x | |~
x| n̂·~
x
t− c
≈ t− c
+
, makes the integration region into a null plane tangent to the cone
c
that crosses the coordinate origin (assumed to lie inside the source’s support), at t′ = t − |~xc | .
As we recede farther and farther away from the source, the cone near the source becomes
flatter and flatter and so the approximation better and better.

22.2.4 The power output

The Poynting vector of the radiation to the approximation considered is given by,

~ = c ~ ~ 1 1 ¨(t − |~x| )|2 = 1 1 |p~¨(t − |~x| )|2 sin2 (θ),

|S| |E ∧ B| = |n̂ ∧ p
~
4π 4πc3 |~x|2 c 4πc3 |~x|2 c

where θ is the angle between the sight direction and the second time derivative of the dipole
vector. The energy flux by solid angle is obtained multiplying the norm of the Poynting vector
(which is in the n̂ direction) by the surface element, |~x|2 dΩ2 , so,

|p~¨(t − |~xc | )|2

dP = sin2 (θ) dΩ2
4πc 3
22.2. ASYMPTOTIC BEHAVIOR OF THE FIELDS 313

(t, ~
x)

~
x

|~
x|
x′
~ t=t− c

|~ x′ | = ct
x−~

Figure 22.2: First approximation, the integration region is at the surface t = t − |~xc | .
314CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

(t, ~
x)

|~
x| ′
n̂·~
x
t=t− c
− c

~
x

~
x’ |~
x|
t=t− c

|~ ~ ′ | = ct
x−x

Figure 22.3: Second approximation, the integration region is now along a null plane tangent
to the cone.
22.2. ASYMPTOTIC BEHAVIOR OF THE FIELDS 315

Since
Z Z 2π Z π
sin2 (θ) dΩ2 = sin3 (θ) dφ dθ
S2 0
Z0 π
= −2π sin2 (θ) d(cos(θ))
0
Z −1
= −2π (1 − x2 )dx
1
Z 1
= 2π (1 − x2 )dx
−1
1
= 2π(x − x3 )|1−1
3
8π
= .
3

Thus, the total power output is

2|p~¨(t − |~xc | )|2

P (t) = (22.3)
3c3

22.2.5 Quadrupole, and Magnetic Moment contributions

If the second order time derivative of the momentum vanishes then we must look at higher
orders. In the next order we have,

1 |~x| ′
Z
~ ~x) ≈
A(t, ∂t J~(t − ), ~x )n̂ · ~x′ d3~x
2
c |~x| lR 3 c
1 −1 ~ ~ ′ 1
Z
≈ 2 [ ∂t (∇ · J)~x (n̂ · ~x′ ) − n̂ ∧ (~x′ ∧ ∂t J)] ~ d3~x′
3
c |~x| lR 2 2
|~
x|
1 (Q̈(t − c ), ~x′ ) · n̂ |~x| ′
≈ 2 [ − cn̂ ∧ m ~˙ (t − ), ~x )]
c |~x| 6 c

Exercise: Compute B ~ and E~ for this case and then the total power output. First make a
guess of the result (up to constant numbers) based on the previous result and on dimensional
analysis.

Exercise: A rugby player kicks the ball trying to make a conversion. The ball acquires a net
charge due to air friction. The kick in not very good and the ball spins along the three main
momentum axis. Estimate how much power gives away as radiation.
316CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

22.2.6 The intermediate radiation zone

If R << λ and R << |~x|, but |~x| ≈ λ, we have to be a bit more careful than in the presiding
calculation, for now terms of the form λ 2 can be as important as terms of the form 1 .
|~
x| |~
x|
One can check that to arrive to the approximation,

~ ~x) ≈ 1 p~˙(t − |~x| ),

A(t,
c|~x| c
we have only used R << |~x|, to approximate |~x − ~x′ | ≈ |~x| in the denominator, and R << λ
to discard the next order,

1 x′ 3 ′
′ n̂ · ~
Z
∂t
~
J (τ, ~
x )[ ] d ~x .
|~x| lR3 c
So these approximations are still valid, but we must now be careful in not dropping factors
~ and B.
higher than O(|~x|) in our calculations for E ~ Thus,

~ ~x) = 1 ∇
B(t, ~ ∧ ( 1 ~p˙(t − |~x| ))
c |~x| c
−1 |~x| 1 |~x|
= 2
(n̂ ∧ ~p˙(t − )) − 2 (n̂ ∧ ~p¨(t − ))
c|~x| c c |~x| c
−1 |~x| c ˙ |~x|
= 2 n̂ ∧ [~p¨(t − )+ ~p(t − )]
c |~x| c |~x| c

To compute the electric field we need the scalar potential, this time we shall compute it
using the gauge condition. Since we are in the Lorentz gauge we have,

~ · A.
∂t φ = −c∇ ~

Therefore,

~ · [ 1 ~p(t0 )] − ∇
φ(t, ~x) = φ0 (~x) + ∇ ~ · [ 1 p~(t − |~x| )],
|~x| |~x| c
where we have chosen the integration initial time so that all fields are static at that earlier
times. Thus, we have,

~ ~x) = −∇φ(t,
~ 1 ~
E(t, ~x) − ∂t A(t, ~x)
c
~ · [ 1 p~(t0 − |~x| )]) + ∇(
~ 0 (~x) + ∇
= −∇(φ ~ ∇~ · [ 1 ~p(t − |~x| )]) − 1 p~¨(t − |~x| ).
|~x| c |~x| c c2 |~x| c

Using now that 1

p~(t − |~xc | ) satisfies the wave equation we get,
c|~
x|
22.3. SPECTRAL DECOMPOSITION 317

~ ~x) = −∇[φ
E(t, ~ · ( 1 ~p0 )] + ∇(
~ 0 (~x) + ∇ ~ ∇~ · [ 1 ~p(t − |~x| )]) − ∆( 1 p~(t − |~x| ))
|~x| |~x| c |~x| c
~ · 1 p~0 ] + ∇
~ 0 (~x) + ∇
= −∇[φ ~ ∧ (∇~ ∧ ( 1 ~p(t − |~x| ))
|~x| |~x| c

Exercise: Check the above assertion about 1

~p(t − |~xc | ) satisfying the wave equation when
|~
x|
|~x| > R > 0.

Performing all the derivatives of the above expression one gets,

~ ~x) = −∇(φ
E(t, ~ 0 (~x) + ∇~ · [ 1 p~0 ])
c|~x|
1 1 ˙ + 3n̂(n̂ · ~p˙)] − 1 [−p~¨ + n̂(n̂ · p~¨)]
+ [−~
p + 3n̂(n̂ · p
~ )] + [−p~
|~x|3 c|~x|2 c2 |~x|

where all dipoles are computed at retarded time t − |~xc | .

Exercise: Check that when λ ≈ |~x| all terms above are of the same magnitude.

22.3 Spectral decomposition

If one considers harmonic time dependence for the sources, J~(t, ~x) = J~ω (~x)e−iωt , then

1 ~ω (~x′ )e−iω(t− |~x−c~x | )

J
Z
~ ~x) =
A(t, ′ d3~x′ ,
c lR 3 ~x − ~x |
′
J~ω (~x′ )ei c |~x−~x | 3 ′
ω
e−iωt
Z
= d ~x ,
c lR3 |~x − ~x′ |
~ ω (~x)e−iωt
= A

Far from the sources,

v
2~x · ~x′
q u
|~x − ~x′ | = |~x|2 + |~x′ |2 − 2~x · ~x′ ≈ ≈ |~x| − n̂ · ~x′ ,
u
|~x|t1 − 2
|~x|
and so,

−i c |~
ω
x|
~ ω (~x) ≈ e
′
Z
A ~ω (~x′ )e−i ωc (n̂·~x ) d3~x′
J
c|~x| lR3
318CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

If ωc R = R
cT
= R
2πλ
<< 1, then

−i ωc (n̂·~
′
x)
∞
X(−i ωc n̂ · ~x′ )l
e =
l=0 l!
and only the first term will be important, it gives,

−i c |~
ω
x|
~ ω (~x) ≈ e ~ω (~x′ ) d3~x′ = −iω e−i ωc |~x| p~ω ,
Z
A J
c|~x| lR3 c|~x|
where we have used that ∇ ~ · J~ω = iωρω . From this expression we can compute the rest of the
fields and other multipoles as needed.

Example: Linear antena

We shall assume a linear antenna as in figure 22.4. We shall assume a current distribution
as follows,

Figure 22.4: Dipolar antenna configuration.

~ ~x) = I(z)ẑe−iωt |z|

J(t, I(z) = I0 (1 − )
d
We then get for the charge density,

±iI0
.ρω (z) =
ωd
Thus, the momentum will point in the ẑ direction and will have magnitude,
d iI0 d
Z
pω = zρω (z) dz = ,
−d ω
22.4. LIENARD–WIECHERT POTENTIALS. 319

The radiated power will be,

dPω I 2 d2 ω 2
= 0 3 sin2 (θ),
dΩ 8πc
and the total power,

I02 ωd 2 2πc
Pω = ( ) valid when λ := >> d.
3c c ω

Figure 22.5: Dipole radiation. Power radiated by solid angle.

22.4 Lienard–Wiechert potentials.

We want now to consider point like charges moving along an arbitrary world line and find the
electromagnetic fields they produce.
So, we want to consider,
′
Z
j µ (t − |~x−c~x | , ~x′ ) 3 ′
✷Aµ = 4πj µ ⇐⇒ Aµ (t, ~x) =
lR3 |~x − ~x′ |
d ~x

where the source current becomes in the limit a distribution with support on a world–line. This
is a complicated limiting process: The charge distribution in this case is a four–dimensional
distribution, having support only on a world line, y µ (τ ), and the formula we have for com-
puting the potential is also a four–dimensional distribution, having support only in the past
light–cone of the observation point. We shall proceed in two ways, first we shall use an ar-
gument by Landau and Lifshitz to find the potential in a preferred frame and then get an
320CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

Figure 22.6: Quadrupolar Radiation, case a = 1, b = 1. Power radiated by solid angle.

Figure 22.7: Quadrupolar Radiation, case a = 1, b = −1. Power radiated by solid angle.
22.4. LIENARD–WIECHERT POTENTIALS. 321

equivalent covariant expression valid in any system. After using this expression for finding the
Maxwell tensor y present a different argument leading to the same expression but performing
the distributional calculus needed.
The simplest way of doing this computation is as follows, we look into the past of our
observation point along null directions, those directions form the past like cone of this point.
Since its velocity is smaller than the velocity of light and is coming from the infinite past,
a particle world line would intersect the cone in just a single point. So the contribution to
the potential integral would come only from that point, (ct̃(τ0 ), ~y (τ0 )). Given the observer’s
point, (t, ~x), and the particle world line, y µ (τ ) we can find the intersection point of it with
the past like cone, and so we have, τ0 = τ0 (t, ~x), and correspondingly t̃ = t̃(τ0 (t, ~x)), and
~y = ~y (τ0 (t, ~x)). We can now go to a frame for which the particle at that past point is at rest,
so at that point, (ct̃(τ0 ), ~y (τ0 )), its four–velocity will be, uµ = (1, 0), and the corresponding
current vector j µ = euµ = (e, 0). At this point the proper time of the particle coincides with
0
its coordinate time, ct̃(τ0 ) = τ0 , and therefore at that instance, dx̃ dτ
= 1.

(t, ~
x)

|~
x−~x′ |
t= c
lµ

uµ

x′
x−~
~

xµ (τ )

Figure 22.8: Lienard–Wiechert setting.

In this frame the integral can be done easily and the four–potential is given by,
322CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

e ~ ~x) = 0
φ(t, ~x) = , A(t,
|~x − ~y |
so the field corresponds to the static field of the particle as if it continues at rest at the
position ~y it had at the past at time t̃ = t − |~x−c ~y | . Since the potential is a four-vector,
all we have to do now is to write this expression on a covariant way and then it would
be valid in any other frame. At our disposal we only have the four-velocity vector, uµ ,
which in this frame is uµ = (1, 0, 0, 0), the corresponding current j µ = euµ and the four-
vector connecting the observation point with the position of the charge along the light cone,
lµ = xµ − y µ = (c(t − t̃), ~x − ~y ). This last one is a null vector, thus,

c2 (t − t̃)2 − |~x − ~y |2 = 0, or |~x − ~y | = c(t − t̃),

since y µ lies into the past of xµ (otherwise we should have taken the other root). But c(t− t̃) =
uµ lµ , so,

µ jµ
A = ν ,
u lν
is the required expression, valid now on any frame we please. For a generic frame, tµ ,
~ in terms of three–dimensional quantities we have associated to it, namely using
uµ = γ(1, β),
coordinates where tµ = (1, 0, 0, 0),

e ~ ~x) = e~v
φ(t, ~x) = A(t, (22.4)
(|~x − ~y | − β~ · (~x − ~y )) c(|~x − ~y | − β~ · (~x − ~y ))

where ~v is the three velocity of the particle at point ~y and time t̃ = t − |~x−c ~y | .
We can now build the Maxwell tensor, Fµν := 2∂[µ Aν] . In taking time derivatives of the
potential we will have normal ones, on the dependence of the vector lµ connecting the emission
point and the observer, and on the dependence of lµ and uµ on τ . So first we compute the
space-time derivatives of τ .
∂ t̃
Differentiating the relation, c(t − t̃) = |~x − ~y | one gets, c(1 − ∂t ) = − (~x−~y )·~v ∂t
∂ t̃
. So,
|~
x−~y |

∂ t̃ 1 1
= = .
∂t 1− (~
x −~y )·~v 1 − n̂ · β~
|~
x−~y |c

Differentiating with respect to the space coordinates, the same relation, we get,
ni 1 ni 1
∂i t̃ = − =− .
c 1 − (~x−~y )·~v c 1 − n̂ · β~
|~
x−~y |c

Since dt̃
dτ
= γc , we have,

lµ
∂µ τ = , (22.5)
(uσ lσ )
22.4. LIENARD–WIECHERT POTENTIALS. 323

Something which it is simplest to see in the frame co-moving with the particle. Consequently,
lµ
∂µ uν = u̇ν ∂µ τ = aν ,
(uσ lσ )
similarly,
uσ lµ
∂µ lσ = δµ σ − uσ ∂µ τ = δµ σ − ,
uρ lρ
it follows that

F µν = 2∂ [µ Aν]
u̇[ν ∂ µ] τ u̇σ lσ u[ν ∂ µ] τ u[ν uσ ∂ µ] lσ
= 2e[ σ − − ]
u lσ (uσ lσ )2 (uσ lσ )2
2e [µ ν] aρ lρ ν] 1
= l [(a − σ u ) − σ uν] ]
(u lσ )
σ 2 u lσ u lσ
2e [µ ν] 1
=: l [s + σ uν] ],
(u lσ )
σ 2 u lσ
ρ
where, sν := aν − uaσρ llσ uν .
We easily see that,
F µν ⋆ Fµν = 0. (22.6)
So the magnetic and electric fields are always perpendicular to each other.
Notice that the Maxwell field is composed of two parts, one that depends on the vector
u and the other on the acceleration vector aµ . The term which depends only on the velocity
µ

vector corresponds to the Coulombic field of the particle as if it were following a straight line
with the velocity given by the value it had at time t̃. Indeed, in the frame where the particle
is momentarily at rest, namely the one defined by, uµ , the electric field will correspond to
that field,
e e ~x − ~y
F µν uν = (l µ
− (u ρ
lρ )u µ
) = (0, ),
(uρ lρ )3 |~x − ~y |2 |~x − ~y |
while the corresponding magnetic field would vanish. Notice that for that part of the field,

−2e2
F µν Fµν = < 0,
(lσ uσ )4
and so, since 22.6 also holds, there is a frame for which the field has only an electric part,
namely the one referred above.
We now analyze the term which depends on the acceleration. This term represents a pure
electromagnetic wave, indeed, note that,
aρ lρ ν
sν lν = (aν − u )lν = 0,
uσ lσ
and so, since lµ is a null vector, we have
324CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

F µν Fµν = 0.
Thus, both scalar invariants vanish, and we have a wave traveling along the vector lµ . Since
the vectors l and s appears in F , as the vectors k and v for plane waves, it is important
to see them in geometrical terms. The null vector lµ is such that lµ = u−1 ρ l ∂µ u, where u =
ρ
c(ttip − t) − |~xtip −~x| is the function whose zero level set is the past cone of the point (ttip , ~xtip ).
So as kµ was normal to the plane u = kµ xµ , here lµ is normal to the light cone. Since sρ lρ = 0
we see that s is the unique linear combination of the acceleration a and the velocity u which
is in the tangent plane of the cone. Since we can add to s any vector in the direction of
l without changing the value of F , and therefore of T , s represents an equivalent class of
tangent vectors at the cone. We can choose a particular one only when we choose a particular
time direction. These are the different electric fields.
As an example we compute now the radiation term of the electromagnetic field correspond-
ing to an arbitrary observer tµ . For this observer the four–velocity is given by, uµ = γ(tµ + β µ )
and the corresponding acceleration by,
1 duµ γ2 µ
aµ = = (ã + γ(β σ ãσ )uµ )
c2 dτ c2
where ãµ is the coordinate acceleration with null time-like component (ãµ tµ = 0). 1 The γ
factors appear from the difference between proper time and coordinate time, each derivative
contributing with one power.
The null vector lµ in the new frame has the same form, (since it is a null vector it remains
so)

lµ = R̃(tµ + nµ ),
where now nµ is a unit vector in the simultaneity surface perpendicular to tµ pointing in the
direction from the observer to the particle in the retarded position, and R̃ = lµ tµ the distance
between observer and particle in that simultaneity surface, as it was R = lµ uµ in the particle
rest frame. Contracting it with uµ we have,

uµ lµ = R = R̃(1 + nµ βµ )γ
Contracting it with aµ we get,
γ2 µ γ3 ρ µ
aµ lµ = R̃n ãµ + u lρ β ãµ .
c2 c2
aρ lρ µ
The vector sµ := aµ − uσ lσ
u becomes then,

γ2 µ nσ ãσ
sµ = (ã − uµ ).
c2 γ(1 + nρ βρ )
2 σ
Finally, using that sµ tµ = − γc2 1+n
n ãσ µ
ρ β , and l tµ = R̃, we have,
ρ

1 d dt d ~ γ2
This comes about because aµ := dτ (γ(tµ + β µ )) = dτ dt (γ(1, β)) = ρ
c2 (γβ ãρ u
µ
+ ãµ ). The vector ãµ is
not the space component of the acceleration vector aµ .
22.4. LIENARD–WIECHERT POTENTIALS. 325

Eµ := Fµν tν
e
= [lµ sν tν − sµ lν tν ]
(u lσ )2
σ

eR̃γ 2 nσ ãσ nσ ãσ

= 2 σ 2 [−(tµ + nµ ) + (−ãµ + uµ )]
c (u lσ ) 1 + nρ βρ γ(1 + nρ βρ )
eR̃γ 2 nσ ãσ nσ ãσ
= 2 σ 2 [−(tµ + nµ ) + (−ãµ + (tµ + βµ )]
c (u lσ ) 1 + nρ βρ 1 + nρ βρ
e
= 2 [−nµ nσ ãσ − ãµ (1 + nρ βρ ) + nσ ãσ βµ ]
c R̃(1 + n βρ )
ρ 3

−e
= 2 [(nµ − βµ )nσ ãσ + ãµ (1 + nρ βρ )]
c R̃(1 + nρ βρ )3
(22.7)

which is the usual expression. Note that it is perpendicular to nµ . The magnetic field is
perpendicular to it and to nµ . So both fields can be written in a more compact vector
calculus expression, in which we have also added the Coulombic part.

~ ~x) = e v2 |~x − ~y |
E(t, [(1 − )(~x − ~y − ~v )
(|~x − ~y | − ~v ·(~xc−~y ) )3 c 2 c
1 |~x − ~y |
+ 2 (~x − ~y ) ∧ ((~x − ~y − ~v ) ∧ ~v˙ )] (22.8)
c c
~ ~x) = n̂ ∧ E(t,
B(t, ~ ~x), (22.9)

where we have used that

~ − (1 − n̂ · β)~
(n̂ · ~a)(n̂ − β) ~ a = (n̂ · ~a)(n̂ − β)
~ − n̂ · (n̂ − β)~
~ a = n̂ ∧ ((n̂ − β)
~ ∧ ~a).

Once again we see that the magnetic and electric fields are perpendicular to each other, that
is, F µν ⋆ Fµν = 0.
The first term on the electric field decays at large distances (null or space–like) as O( 1 2 )
|~
x|
and depends only on the particle velocity. This is in fact a Coulomb field corresponding to
the particle as if it would be moving at constant speed from the past to the point where it
should be now. The second term decays as O( 1 ), and represents radiation coming out of
|~
x|
the particle due to its acceleration in the past.

Exercise: Check that the first term corresponds to the Coulomb field of a charged particle
moving at constant speed in the corresponding frame. Hint: realize that ~x − ~y − |~x−c ~y | ~v =
~x − ~y − (t − t̃)~v = ~x − (~y + (t − t̃)~v), is the vector connecting the observation point with the
point where the particle would have been, were it moving at constant velocity (~v).
326CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

22.4.1 Energy Momentum Tensor

The energy-momentum tensor for the radiation field is

−1 µσ ν e2
T µν = F F σ= l µ l ν sρ sρ
4π 4π(l uσ )
σ 4

ρ σ
where, sν = (aν − uaσρ llσ uν ), and therefore, since sρ sρ = aρ aρ − ( auρllσρ )2 ≥ 0 a space-like vector. 2
Using this formula we can compute Poynting’s vector and from it deduce the radiation
power as a function of emission angle and then the total power radiated away.

22.4.2 Power radiated

In the frame where the particle is at rest it is easy to compute the power radiated by solid
angle,
dP ~ · n̂
:= R2 S
dΩ
~
where R is the distance to the surface where Poynting’s vector (Tµν uµ = (e, S/c)) is integrated,
2
(actually R is just part of the surface element, and n̂ the unit normal to it). Contracting
the energy momentum tensor with uµ and n̂ we get, (we are assuming lµ = uσ lσ (uµ + nµ ),
uσ lσ = R = |~x − ~x|).

dP e2 c ρ e2 c ρ aρ lρ e2 e2
= s sρ = (a aρ − ( σ )2 ) = (~
a · ~
a − (~
a · n̂) 2
) = |~a|2 sin(θ)2
dΩ 4π 4π lσ u 4πc3 4πc3
where θ is the angle between ~a and n̂. Using now that,
8π
Z
sin(θ)2 dΩ =
S2 3
we can compute the total power output radiated by the particle,

2e2 2
P =
|~a| .
3c3
Where we have expressed it in terms of time units (not in terms of x0 = ct), to show that it
coincides with the dipole formula 22.3 if we consider the dipole to be a particle moving with
the given acceleration.
This is the power output the particle will see in its own frame. But this quantity is
independent of the frame chosen to describe it. This is clear from its definition, but it is
instructive to see the following argument: From the figure 22.9 we see what would be this
integral for two different observers. Using energy–momentum conservation we shall see that
2
To see this write the metric as η µν = −2l(µ k ν + q µν , where k µ = R̃1 (tµ − nµ ) and notice (by explicitly
writing it in Cartesian coordinates, taking, say nµ = (0, 1, 0, 0)) that q µν is the identity matrix when restricted
to the y, z coordinates. Since sµ lµ = 0, sµ sµ is just the square of the acceleration components in the direction
perpendicular to nmu .
22.4. LIENARD–WIECHERT POTENTIALS. 327

this integral is the same for both surfaces. First note that due to the particular form of the
energy–momentum tensor, −Tµν uµ nν = 41 Tµν k µ k ν , where k µ := uµ − nµ , is a unique null
vector, k µ kµ = 0, normal to the null surface ruled by lµ , and to the two-dimensional surface,
and normalized such that k µ lµ = 2R. Thus,

µ ν −1 µ ν −1 e2 c2
Tµν u n = Tµν k k = sρ sρ .
4 4 4π(l uσ )
σ 2

So this is the quantity to be integrated at a cut, C, on the outgoing light cone of the particle
at a given moment, obtained by intersecting any plane perpendicular to uµ .
I I
P (τ ) = Tµ ν uµ nν dS = j ν nν dS
C cut

where we have defined the current j ν = Tµ ν uµ . Notice that j µ lµ = 0, so this is indeed a vector
tangent to the light cone of the particle, and so we can use Gauss theorem to transform the
integral into a volume integral, between two different cuts, C and C. ˜
Thus,
I Z √ I
P (τ ) = j ν nν dS = ∂µ (j µ s) dΩ dr + j ν nν dS
C S C˜

But,

√ √ √ √ √
∂µ (j µ s) = ∂µ (Tµ ν uµ s) = ∂µ (Tµ ν )uµ sTµ ν ∂µ (uµ ) s + j µ ∂µ s = 0

where the first term vanishes because of Maxwell’s equations (energy conservation), the second
because we have extended the 4-velocity vector of the particle at time τ into a global frame,
and the final one because the T µν has only components in the lµ direction, and along that
direction the determinant of the cone metric is constant. Thus, we see that the power output
can be obtained by performing an integration on any 2-surface at the future light cone of the
particle which includes the tip of it in its interior.
It is instructive for applications to write the above form in different forms. Since,

γ2
0 = aµ uµ = (ãµ uµ − γ(β σ ãσ )),
c2
we have,

γ4 γ4
aµ aµ = (ãµ ãµ
+ γ 2
(β σ
ãσ ) 2
− 2γ(β σ
ã )ã
σ µ u µ
) = (ãµ ãµ − γ 2 (β σ ãσ )2 ),
c4 c4

and therefore,
328CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

uµ

C n̂

C˜

uµ

xµ (τ )

Figure 22.9: Different cuts where the flux integration can be carried out.
22.4. LIENARD–WIECHERT POTENTIALS. 329

−2e2 c ρ
P = a aρ
3
2e2 γ 4
= [~a · ~a + γ 2 (β~ · ~a)2 ]
3c 3

2e2 γ 6 2
= [|~a| (1 − β 2 ) + (β~ · ~a)2 ]
3c3
2e2 γ 6 2 ~v
= [|~a| − | ∧ ~a|2 ], (22.10)
3c 3 c
in terms of 3-velocities and 3-accelerations, or when the particles have a constant rest mass,

−2e2 c dpµ dpµ 2e2 d~p 2 1 dE 2 2e2 d~p 2 dp

P = = [| | − ( ) ] = [| | − β 2 ( )2 ]
3m dτ dτ
2 3cm dτ
2 2
c dτ 3cm dτ
2 dτ
that is in terms of the applied external force to the particle, here E := γmc2 , and p~ := γm~v .
2
Exercise: Check the above formula. Hint: from pµ pµ = Ec2 − p2 = m2 c2 , we get E 2 =
2 dp
(p2 + m2 c2 )c2 = (m2 β 2 c2 γ 2 + m2 c2 )c2 = m2 γ 2 c4 . Thus, dE
dτ
= pcE dτ dp
= β dτ .

22.4.3 Co–linear acceleration

In the case the acceleration is along the velocity, as is the case in linear particle accelerators,
dp
~ dp
| dτ | = | dτ |, and so

2e2 2 dp 2 2e2 dp 2 2e2 dE 2

P = (1 − β )( ) = ( ) = ( ) ,
3cm2 dτ 3c3 m2 dt 3c3 m2 dx
where in the last equality we have expressed the force as the change of energy per unit length.
Thus, the rate of power radiated to power supplied (~v · F~ = ~v · dE
dx
= dE
dt
), in the acceleration
is,
2
P 2e2 dE 2 e 2 dE
dE = 3 2 ≈ mc 2 ,
dt
3c m v dx 3 mc dx
where, in the last line we have made the approximation v ≈ c, for this is the generic case in
accelerators where particles reach near light speeds very quickly. For electrons the numbers
are,

P m dE
dE = 0.5 × 10−14 [ ]
dt
MeV dx
while the typical energy delivered per meter in present machines is of the order of tens of
MeVs. Thus, radiation losses in these accelerators is negligible and the main limitation they
have is their length and the amount of energy delivered per unit length.
330CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

22.4.4 Centripetal acceleration

For circular accelerators the energy delivered per unit length is not a limitation since particles
can circle it several millions times, gradually increasing their energy. In these accelerators the
main change in momentum is in its direction, and we have,

d~p
|
| ≈ γω|~p|,
dτ
v
where ω = R
is the orbital frequency, and R is the orbit radius. In this case we have,

2 e2 cβ 4 γ 4
P =
3 R2
In this case the relevant quantity is the energy lost per revolution compared with the energy
gain during the same trajectory. The first quantity is,

2πR 4π e2 β 3 γ 4 me [E(GeV )]4

δE = P = = 8.85 × 10−2( )4
v 3 R m R(m)
with v ≈ c.
For the LEP (Large Electron–Positron collider), R = 4300m and the beam reached energies
of 60 GeV, the energy losses per revolution are,

δE ≈ 300MeV.
For the LHC (Large Hadronic Collider), R = 4300m, and the proton beam reached energies
of 7 TeV. The energy loss per revolution in this case is, of about 10KeV. We see from the
form the energy loss scales with particle mass that it is more convenient to accelerate heavy
particles in a circular accelerator while it is more convenient to accelerate light particles in
linear accelerators. To have an idea of the total energy loss, one has to have into account
that the particles circle the ring about 11245 times per second, while the number of protons
at any given moment in the ring is about 3 × 1014 .

22.5 Angular distribution of radiation

To compute the total power radiated per solid angle we must compute Poynting’s vector
contracted with the normal to a surface and multiply it by the surface element of solid angle.
Thus, we shall obtain an integral of the form,
Z t2
E(t2 , t1 ) = ~ · n̂R2 dΩdt.
S
t1

Since the all quantities in this expression depend on the retarded time, τ = t − |~x − ~x′ (τ )|/c
dt
it is necessary to express it in terms of that time, since dτ = (1 − β~ · n̂) the integral becomes,
Z τ2
E(τ2 , τ1 ) = ~ · n̂R2 (1 − β~ · n̂) dΩdτ,
S
τ1
22.5. ANGULAR DISTRIBUTION OF RADIATION 331

and it is clear that the relevant quantity to look for is

dP (τ ) ~ · n̂ R2 (1 − β~ · n̂)
=S
dΩ
which is the power radiated per unit solid angle and unit retarded time.
Using the expressions above 22.8, for the electric field, and noticing that the corresponding
magnetic field has the same magnitude and is perpendicular to it, we have,

dP (τ ) ~ ∧ β)|
e2 |n̂ ∧ ((n̂ − β) ~˙ 2
= .
dΩ 4πc (1 − β~ · n̂)5
The main feature of this distribution is the relativistic factor in the denominator, for
ultra–relativistic motion that term is very small and dominates the distribution.

22.5.1 Co-linear motion

In the case the velocity is parallel to the acceleration one of the terms vanish and calling θ at
the angle between n̂ and ~v the expression can be reduced to,

dP (τ ) e2 |~a|2 − (~a · n̂)2 e2 a2 sin(θ)2 e2 a2 γ 8 (γθ)2

= = ≈ ,
dΩ 4πc3 (1 − β~ · n̂)5 4πc3 (1 − β cos(θ)2 )5 4πc3 (1 + (γθ)2 )5
where in the last equality we have included the small angle approximation (using θ << 1,
and β 2 = 1 − γ12 , so that β ≈ 1 − 2γ12 ).
1
This distribution has a zero at θ = 0 but rises to a maximum when θmax ≈ 2γ , thus we
have,

dP (τ ) e2 a2 γ 8 43
(θmax ) ≈ ,
dΩ 4πc3 55
so it grows as the eighth power of γ!, see figures 22.10 and 22.11.
30

25

20

15

10

-1.0 -0.5 0.0 0.5 1.0

Figure 22.10: Power radiated by solid angle. Approximation for small angle for γ = 2.
332CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

1 ´ 107

8 ´ 106

6 ´ 106

4 ´ 106

2 ´ 106

-1.0 -0.5 0.0 0.5 1.0

Figure 22.11: Power radiated by solid angle. Approximation for small angle for γ = 10.

1 2 3

-1

-2

-3

Figure 22.12: Power radiated by solid angle in the co-linear case, β = 0.5.
22.5. ANGULAR DISTRIBUTION OF RADIATION 333

2000

1000

-2000 -1000 1000 2000

-1000

-2000

Figure 22.13: Power radiated by solid angle in the co-linear case, β = 0.9.

Exercise: Find the angle of maximal radiation for the ultra-relativistic limit by taking a
derivative of the exact expression with respect to the angle and equating to zero. Use the
approximations for β given above. Redo the calculation using the small angle approximation
expression and check that the angles coincide.

Figure 22.14: Power radiated by solid angle, velocity and acceleration co-linear, β = 0.4

22.5.2 Circular motion

In this case the velocity is perpendicular to the acceleration direction, and we have,
334CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

Figure 22.15: Power radiated by solid angle, velocity and acceleration co-linear, β = 0.9.

dP (t′ ) e2 a2 1
= [(1 − β cos θ)2 a2 − (1 − β 2 )(~a · n̂)2 ]
dΩ 4πc3 (1 − β cos θ)5
e2 1
= [(1 − β cos θ)2 − (1 − β 2 ) sin2 θ cos2 φ]
4πc (1 − β cos θ)
3 5

2e2 a2 γ 6 a2 4γ 2 θ2 cos(φ)2
≈ [1 − ]
πc3 (1 + (γθ)2 )3 (1 + γ 2 θ2 )2

where the angles are shown in figure 22.16. One can see that there will be maximal radiation
in the ultra–relativistic case when the angle φ = 0, that is in the orbit plane. Furthermore,
the maximum is along the particle motion.

22.5.3 Alternative deduction for the potential

The way we obtained the potential vector is neat and economical but not satisfactory. In
principle the vector potential could depend also on the acceleration of the particle, or in
details on how the limit to point particles is taken. Here we shall give another derivation
based in the use of distributions and the retarded Green function. This alternative derivation
is also not rigorous, for involves the product of two distributions, the one defining the particle
times the Green function which, when expressed as an integral over space-time is also a
distribution. I do not know of any rigorous derivation so the ultimate justification for the
above expressions is the fact that they are reproduced in many experiments with extreme
precision.
We start with the four-dimensional definition of a distributional current,
∞ dy µ
Z
j µ (xσ ) = e δ(xσ − y σ (s)) ds, (22.11)
−∞ ds
22.5. ANGULAR DISTRIBUTION OF RADIATION 335

~
β

n̂

~a(τ )

y
~ (τ )

Figure 22.16: Angles definition for the circular motion.

2 4 6 8

-1

-2

Figure 22.17: Power radiated by solid angle, case velocity perpendicular to acceleration,
β = 0.5, φ = 0 (blue) and φ = π/2 (violet).
336CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

Figure 22.18: Power radiated by solid angle, case velocity perpendicular to acceleration,
β = 0.92

Figure 22.19: Power radiated by solid angle, case velocity perpendicular to acceleration,
β = 0.92, cut to see the φ dependence.
22.5. ANGULAR DISTRIBUTION OF RADIATION 337

where y µ (s) is the particle trajectory and s is any parameter used for expressing it. As always
for vectorial integrals it is only valid when expressed in Cartesian coordinates.

Exercise: Check that this definition is independent on the choice of parameter s. Check
that it satisfies, ∂µ j µ = 0 in the distributional sense.

Since this is a four-dimensional distribution which needs to be inserted in Green’s formula,

22.1, it is convenient to write it as a four-dimensional integral,

′
Z
j µ (t − |~x−c~x | , ~x′ ) 3 Z
j µ (t′ , ~x′ ) ′ |~x − ~x′ | 3
d ~
x = δ(t − t + ) d ~x dt. (22.12)
lR3 |~x − ~x′ | lR4 |~x − ~x′ | c

Inserting in the above expression the distributional form for the particle trajectory we get,

dy µ
e ∞ |~x − ~x′ | 3
Z Z
µ σ ds ′σ σ ′
A (x ) = δ(x − y (s))δ(t − t + ) d ~x d(ct) ds
c lR4 −∞ |~x − ~x′ | c
dy µ
eZ ∞ ds |~x − ~y (s)|
= δ(t̃(s) − t + ) ds
c −∞ |~x − ~y (s)| c
(22.13)

where ct̃(s) = y 0(s), and in the second equality we have performed the four dimensional
integral.
In order to perform the last integral we need to use, that

∞ g(xi )
Z X
g(x)δ(f (x)) dx = , ∀ {xi |f (xi ) = 0}
−∞ i |f ′(xi )|

but

d |~x − ~y (s)| dt̃ 1 d~y (s) dt̃ 1 d~y (s)

(t̃(s) − t + )= − · n̂ = (1 − · n̂)
ds c ds c ds ds c dt̃

where, as usual, n̂ = ~x−~y (s) .

|~
x−~y (s)|
Thus,
338CHAPTER 22. ELECTROMAGNETIC RADIATION PRODUCED BY CHARGE CURRENTS.

dy µ
e ∞ |~x − ~y (s)|
Z
µ σ ds
A (x ) = δ(t̃(s) − t + ) ds
c −∞ |~x − ~y (s)| c
dy µ
dt̃ −1
e ds
( ds )
=
c |~x − ~y (s)| (1 − 1 d~y (s) · n̂)
c dt̃
dy µ
e dt̃
=
c |~x − ~y (s)|(1 − 1 d~
y (s)
c dt̃
· n̂)
dy µ
dτ
= e
1 d~
y (s)
γ|~x − ~y (s)|(1 − c dt̃
· n̂)
dy µ
= e dydτρ
l
dτ ρ
(22.14)

which is the four-dimensional form of 22.4.

Exercise: Find F µν by first performing the derivatives in 22.13 and then performing the
space-time derivatives.
Chapter 23

Quasi static fields, different time scales

23.1 Introduction
In many situations in electromagnetism, in particular when we want to study fields near
sources, we have to different time scales: the time that takes a wave to transverse the sources,
that is τc = L/c, where L is the source’s length, and some other time scales coming from the
sources. In particular there appears one which is related to the adjustment of the sources to
the effect of electromagnetic fields through Ohmic effects. If the first time scale is very short
compared with the second, then we can consider the fields as statics at each instant of time.
If they are comparable, then radiation effects must be taken into account. If we want to study
the limit in which τc → 0, so as to get the effect of the other time scale, we can look at the
equations in the limit of c → ∞. Defining ε = 1c . We have,

~ = 1∇
∂t E ~ ∧B~ − 4π J~
ε
~ = −1 ∇
∂t B ~ ∧E~
ε
~ ·E
∇ ~ = 4πρ
~ ·B
∇ ~ = 0

to study them we seek a solution of the form

~ = E
E ~ 0 + ε2 E
~ 1 + ε4 E
~ R (ε)
~ = εB
B ~ 0 + ε3 B
~ 1 + ε5 B
~ R (ε)

we have omitted some terms because by re-scaling one can see that the solutions should depend
on even powers of ε, starting at some arbitrary power. It can be seen that the omitted term
are zero if appropriate boundary conditions are imposed. An important theorem for systems
with different time scales states that if the initial data is such that the time derivatives up to
some order are initially bounded then they remain so for subsequent times. So the requirement
that there would be solutions which are regular in the time variable is actually a restriction

339
340 CHAPTER 23. QUASI STATIC FIELDS, DIFFERENT TIME SCALES

on the set of possible initial data sets, we are restricting the solution not to have waves and so
to be quasi-static or quasi-Coulombic. So see what type of solutions we are getting we plug
the terms of the ansatz on the equation and to the lowest order we get,

~0 = ∇
∂t E ~ ∧B ~ 0 − 4π J~ (23.1)
~ 0 = −1 ∇
∂t B ~ ∧E ~0 + ∇ ~ ∧E
~2 (23.2)
ε2
~ ·E
∇ ~ 0 = 4πρ − ε2 ∇~ ·E
~2 (23.3)
~ ·B
∇ ~0 = 0 (23.4)
The smoothness requirement on the time derivatives implies,

~ ∧E
∇ ~0 = 0 (23.5)
imposing also that

~ ·E
∇ ~ 0 = 4πρ
we are back to the electrostatic equations, which now have to be thought as valid at each
instant of time. We are back to instantaneous influence. So, giving appropriate boundary
conditions we have a unique E ~ 0 and so equation 23.1 have to be thought as an equation for
~
B 0 . But actually it does not need to be solved in the present form. Taking its curl, using ,
and 23.4 we obtain,

~ ∧ J.
∆B0 = −4π ∇ ~ (23.6)
So we can solve this elliptic equation with appropriate boundary conditions and have a solution
at each time. We have recuperated the magneto-static equations. But since these equations
are now valid at each instant of time, and we can allow for source variations their field of
applications is much wider.
Consider for instance the case in which Ohm’s law is valid. Then,

J~ = σ E,
~
and we have,

∆B0 = −4π ∇ ~ ∧ J~
= −4πσ ∇~ ∧E ~
~ ∧E
= −4πε2 σ ∇ ~2
4πσ
= ∂t B0
c2
and we get a parabolic equation for the magnetic field to this order. Thus, the magnetic field
will diffuse itself in a time scale given by
4πσL2
τσ =
c2
23.1. INTRODUCTION 341

where L is the system size. This time scale is in general much larger than the radiation time
scale, related to the time the light spend in transversing the object, τc = L/c. For instance,
if we consider the earth core, then L ≈ 103 Km, σ ≈ 1017 s−1 and so, τσ ≈ 1012 s which is of
the order of tens of thousand of years.

Exercise: Find the length for which both time scales coincide for the case of cooper.

The relevant theorem for systems with different temporal scales, when applied to the case
of electromagnetism, is that if one considers smooth initial data satisfying the above equations,
then the resulting solution will be smooth (both in space and time) and will satisfy the same
equations for all times, in the sense that the error terms, E~ R (ε), and E
~ B (ε) are bounded.

Exercise: Find the equations the electromagnetic field would satisfy in a dielectric substance
(ε 6= 1) assuming the medium is homogeneous and isotropic.

23.1.1 Example: The lowest decay of a magnetic field on a con-

ducting sphere.
23.1.2 Example: The magnetic polarizability of an isotropic con-
ducting sphere.
23.1.3 Example: The skin effect on a wire
342 CHAPTER 23. QUASI STATIC FIELDS, DIFFERENT TIME SCALES
Chapter 24

Examination Questions 2

Problem 41 Why do we assert that if a symmetry of space-time is valid for Maxwell’s equa-
tions it is also valid for the wave equation?

Problem 42 Complete the argument after equation 13.7, that is the exercise that follows it.

Problem 43 Probe the Lemma in the variational principle for a particle in space-time. Do
the exercise below.

Problem 44 Find the expression for the time delay to first non-vanishing order between a
person on the longest path between two time-like events and a person along a nearby path.

Problem 45 Deduce the Doppler effect.

Problem 46 Deduce the aberration effect.

Problem 47 Show that the sum of two future directed time-like vectors is also time-like and
future directed.

Problem 48 Show that the sum of a future directed time-like vector with a future directed
null vector is always time-like and future directed.

Problem 49 Under which considerations is total momentum conserved for a set of particles?

Problem 50 Show that ∂ν ⋆ F νµ = 0 is equivalent to

1
∂[µ Fνσ] := [∂µ Fνσ + ∂σ Fµν + ∂ν Fσµ ] = 0. (24.1)
3
Problem 51 Use 24.1 and 15.5 to show that

✷Fµν = 8π∂[µ jν] .

Problem 52 Show that
⋆
F µν = 2B [µ tν] − εµν σρ E σ tρ (24.2)

343
344 CHAPTER 24. EXAMINATION QUESTIONS 2

Problem 53 Show that if at some point of space-time Fµν F µν > 0 and Fµν ⋆ F µν = 0, that
is, if for any observer at that point the electric and magnetic fields are perpendicular and the
magnetic field is bigger in norm than the electric field, then there is an observer for which the
electric field vanishes.

Problem 54 Show that if for some time-like vector uµ , Fµν uν = 0 then ⋆ Fµν = 2b[µ uν] for
some space-like vector bµ . Show that it also follows that F µν ⋆ Fµν = 0 and F µν Fµν ≤ 0

~ and B
Problem 55 Check that pµ in terms of E ~ is given by,

−1
pµ := −etµ + P~ = ((E 2 + B 2 )tµ − 2tρ ερµσν Eσ Bν ),
8π

that is, the 3-momentum is P~ = 1 ~

4π
(E ~ = 1 S,
∧ B) c
~ where S
~ is Poynting’s vector.

~ and B.
Problem 56 Compute pµ pµ in terms of E ~ Does it have a definite sign?

Problem 57 Show that ∂µ T µν = j µ Fµν if Maxwell’s equations are satisfied. Is the converse
true? If so, in which case?

Problem 58 Deduce energy conservation for a situation like figure 15.1.

Problem 59 Use energy-momentum conservation to show, that if two solutions to Maxwell

equations coincide at Σ0 then they must coincide inside a region like the one shown in figure
15.3 as long as the normal to ΣT is time-like.

Problem 60 Let Aµ be such that Fµν = 2∂[µ Aν] . Given a constant unit time-like vector t
define Aµ = φtµ + õ that is, φ = A0 = −A · t = −A0 and à · t = 0. Check that

B µ = ε̃µνσ ∂ν Aσ := tρ ερµνσ ∂ν Aσ ,

and
Eµ = −∂˜µ φ + ∂0 õ .
Check explicitly that both vectors are gauge invariant.

Problem 61 Describe the initial conditions problem for the vector potential in the Lorentz
gauge. Which fields can be given and which are specified and why.

Problem 62 Discuss for which type of problem one would preferably use the Lorentz gauge
and for which Coulomb’s one? In a radiation problem? In a quasi-stationary problem?

Problem 63 Deduce the equations of motion of a charged particle using the variational prin-
ciple.

Problem 64 Deduce Maxwell’s equations using a variational principle. Can you deduce this
way all equations or just some subset?
 345

Problem 65 Study the plane waves in terms of the Maxwell tensor and also in terms of the
four-vector potential.
Problem 66 Deduce formula 18.11 for total reflection of a superconductor.
Problem 67 Find the Fourier amplitudes for the vector potential in terms of its initial data
in space for a monochromatic wave.
Problem 68 Check that given initial data (E ~ 0, B
~ 0 ) = (Pl,n,m E ~ 0lmn (x) for
~ 0lmn (x), Pl,n,m B
the electric and magnetic fields inside a rectangular cavity, we can construct a solution as
sum of the cavity modes. Find the explicit values for E ~ + and E ~− .
lmn lmn

Problem 69 Estate the problem of finding the solutions of waves in a wave guide in all three
cases. Assuming you have solved for all the modes, write the solution in terms of arbitrary
initial data on the wave guide.
Problem 70 Show that for a mode in a cavity,
Z Z
~ 2 dV =
|E| ~ 2 dV =
|B|
V V
ω2 ~ ~ ω = 0 and −iω B
~ω = ∇
~ ∧E
~ ω.
Hint: use E
c2 ω
+ ∆E
Problem 71 Find the amplitudes for the reflected and transmitted waves when the magnetic
vector is tangent to the interface, but doing the calculation with the boundary conditions for
the electric field.
Problem 72 Deduce the Kramers-Kroning relations.
Problem 73 Deduce the causality theorem.
Problem 74 Prove the assertion on the retarded Green function.
Problem 75 Describe in geometrical terms the nature of the approximation made in the
Green function integral regarding the power series expansion of the numerator.
Problem 76 Compute up to numerical factors the total output for quadrupole radiation.
Problem 77 A rugby player kicks the ball trying to make a conversion. The ball acquires a
net charge due to air friction. The kick in not very good and the ball spins along the three
main momentum axis. Estimate how much power gives away as radiation.
Problem 78 Deduce the Lienard–Wiechert potential from the argument of going to a pre-
ferred frame.
Problem 79 Deduce the Lienard–Wiechert potential from the argument using the retarded
Green function.
Problem 80 Find the Maxwell tensor corresponding to the Lienard–Wiechert potential. First
write the most general antisymmetric covariant tensor depending on four vectors, lµ , j µ , uµ ,
and the acceleration aµ . Recall that j µ , and uµ are parallel, and aµ is perpendicular to uµ .
346 CHAPTER 24. EXAMINATION QUESTIONS 2
Bibliography

[1] General Relativity from A to B, R. Geroch (University of Chicago Press, Chicago, 1978).

[2] no se donde está, R. Geroch

[3] Concepts in thermal physics Blundell SJ, Blundell KM. New York: Oxford University
Press 2010.

347