Electrodynamics (honours)

Work and lecture notes

Heribert Weigert
heribert.weigert@uct.ac.za
typeset with LATEX on February 18, 2025, 13:15
Preliminaries
Recommended literature

Zangwill: Modern Electrodynamics [1]

Jackson: Electrodynamics [2]

Landau: The classical theory of fields [3]

Griffiths: Introduction to Electrodynamics [4]

Pollack and Stump: Electromagnetism [5]

Feynman: Feynman Lectures II [6]

Electrodynamics in honours at UCT Electrodynamics is an “old” theory dating back into

the formative years of a science later to be called physics. Yet it is strangely modern: It is in fact
fully relativistic (and was so long before Einstein); it is usually taught before quantum mechanics,
yet many of the tools usually only properly taught in quantum mechanics have essential uses
in electrodynamics (were in fact invented there to be reused and refined to formulate quantum
mechanics) and in particular, it is the precursor of Quantum Electrodynamics (QED) a key part of
our Standard Model of Particle Physics, one of our deepest microscopic theories of nature. In this
honours course, were you, the students should have at least a smattering of all of these references
at your disposal, I will attempt to build on these connections to help you develop a better sense
of the unity of physics as a science, hopefully breaking a few barriers to scientific thinking in the
process.

ED 1 • history and perspective • an introduction to vector calculus • basic principles of elec-

trostatics • solving differential equations: Green’s functions, boundary conditions, complete sets
of states • electrostatics in media • magneto-statics • magneto-statics in media • electrodynamics
and Maxwell’s equations

ED 2 • the relativistically covariant formulation of electrodynamics • gauge potentials and a

count of degrees of freedom • electrodynamics as a classical field theory • Greens functions revisited:
Fourier transforms and the use of residues

ii typeset with LATEX on February 18, 2025, 13:15

Contents
Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

Contents iii

1 Maxwell’s equations in a given coordinate system (frame) 1

1.1 The differential form of Maxwell’s equations in microscopic form (vacuum form) . 1
1.2 Interpretations I: charges, currents, and fields . . . . . . . . . . . . . . . . . . . . . 1
1.3 Interpretations II: of vectors and pseudo-vectors, index notation . . . . . . . . . . . 3
1.4 Looking back on chapter 1 and looking ahead . . . . . . . . . . . . . . . . . . . . . 10
1.5 Solutions for chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Electrostatics 14
2.1 The physics and mathematics of Coulomb’s law . . . . . . . . . . . . . . . . . . . . 14
2.1.1 Observational background – point particles and Coulomb’s law, or “how to
measure a Green’s function” . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 From electric flux to Gauss’ law . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.3 Solving Laplace’s and Poisson’s equations – the idea of Green’s functions . 22
2.1.4 Symmetries: from charge distributions to potentials . . . . . . . . . . . . . 30
2.1.5 Looking back on Sec. 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Applications: charged surfaces, wires and capacitors . . . . . . . . . . . . . . . . . 33
2.2.1 Surface and line charge densities in R3 . . . . . . . . . . . . . . . . . . . . . 33
2.2.2 Electric fields near charged surfaces – matching conditions . . . . . . . . . . 34
2.2.3 “Capacitors” in three geometries . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.4 Looking back on Sec. 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3 Properties of (rigid) static charge distributions . . . . . . . . . . . . . . . . . . . . 48
2.3.1 Forces and torques between charge distributions – energies stored in charge
distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 The electrostatic total energy of a charge distribution, UE [ρ] . . . . . . . . 50
2.3.3 The electric stress tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4 Far fields, shapes and symmetries: multipole expansions . . . . . . . . . . . . . . . 52
2.4.1 The Cartesian multipole expansions . . . . . . . . . . . . . . . . . . . . . . 52
2.4.2 Spherical multipole moments . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.4.2.1 Far fields and spherical outer multipole moments . . . . . . . . . . 65
2.4.2.2 Sperical multipole expansion inside a hollow charge distribution . 67
2.4.2.3 Spherical multipole expansions for general charge distributions . . 68
2.4.3 Cartesian (primitive) vs traceless multipole expansions . . . . . . . . . . . . 70
2.4.4 Dipole layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.5 Looking back on Sec. 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.5 Dielectric media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5.1 Induced responses: the polarization field P . . . . . . . . . . . . . . . . . . 75
2.5.2 The idea of the displacement field D . . . . . . . . . . . . . . . . . . . . . . 80
2.5.3 Linear dielectric media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.5.4 Worked examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.5.4.1 Two dielectrics and a point charge . . . . . . . . . . . . . . . . . . 86
2.5.5 Looking back on Sec. 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.6 Boundary value problems – electrostatics in finite volumes . . . . . . . . . . . . . . 88
2.6.1 The uses of Green’s theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.7 Constructing Greens functions from zero mode solutions . . . . . . . . . . . . . . . 91

typeset with LATEX on February 18, 2025, 13:15 iii

 2.7.1 Green’s functions, boundary conditions, spectral decompositions and active
cusps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.7.1.1 Green’s functions from the full spectrum: spectral theorems . . . 93
2.7.1.2 Green’s functions without knowledge of the full spectrum – active
cusps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.7.1.3 Constructing Green’s functions from zero mode solutions – engi-
neering active cusps . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.7.2 Adapting Green’s functions of ∇2 to boundary conditions, a comparison of
methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2.7.3 The method of images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.7.4 Choosing the right method for the right type of boundary value problem? . 107
2.7.5 Looking back on Sec. 2.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

3 Magnetostatics 109
3.1 Steady currents – the source for static magnetic fields . . . . . . . . . . . . . . . . 109
3.1.1 Steady currents – the concept . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.1.2 Features of physical realizations . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.2 From Ampere’s law to gauge potentials . . . . . . . . . . . . . . . . . . . . . . . . 114
3.2.1 Static field and steady currents . . . . . . . . . . . . . . . . . . . . . . . . . 114
3.2.2 Examples with simple geometries . . . . . . . . . . . . . . . . . . . . . . . . 117
3.2.3 New concepts: gauge freedom and magnetic potentials . . . . . . . . . . . . 122
3.3 Magnetic multipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.4 Magnetic fields in matter – magnetic matter, M and H . . . . . . . . . . . . . . . 133
3.4.1 Magnetization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.4.2 Bound fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.4.3 Linear magnetic media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.4.4 Summary of section 3.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.5 Magnetic force and energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.5.1 Magnetostatic force, torque, and work . . . . . . . . . . . . . . . . . . . . . 142
3.5.2 Magnetostatic total energy does not allow adiabatic assembly . . . . . . . . 144

4 Electrodynamics – introducing the time dependent terms of Maxwell’s equa-

tions 145
4.1 Introducing time dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.2 Faraday’s law of induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3 The scope of electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

5 From Galilei to Einstein – the foundations of special relativity 153

5.1 Digression: Symmetries and Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.1.1 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.1.3 Groups and inner product spaces . . . . . . . . . . . . . . . . . . . . . . . . 154
5.1.4 Matrix Lie groups and their Lie algebras . . . . . . . . . . . . . . . . . . . . 157
5.1.5 Parametrizing (s)u(n) and (s)o(n), dimensions . . . . . . . . . . . . . . . . 168
5.1.6 Orthogonal and unitary Lie-groups and algebras, a summary . . . . . . . . 170
5.1.7 Generators and commutation relations . . . . . . . . . . . . . . . . . . . . . 170
5.1.8 Special properties of SU(2) and SO(3) . . . . . . . . . . . . . . . . . . . . . 174
5.1.8.1 Pauli-matrices and su(2): . . . . . . . . . . . . . . . . . . . . . . . 174
5.1.8.2 The canonical basis of so(3) and the exponential map: . . . . . . . 175

iv typeset with LATEX on February 18, 2025, 13:15

 5.1.8.3 Aside: Reconstructing ω for elements of SO(3) . . . . . . . . . . . 178
5.1.9 Looking back on Sec. 5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.2 Galilei invariance in classical mechanics . . . . . . . . . . . . . . . . . . . . . . . . 183
5.3 What happens to observables in a change of frames? . . . . . . . . . . . . . . . . . 186
5.4 Moving on to special relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.5 The Lorentz and Poincaré groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5.1 The Poincaré group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5.2 Properties of the Lorentz group . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5.2.1 The connected components . . . . . . . . . . . . . . . . . . . . . . 198
5.5.2.2 The Lie-algebra of the Lorentz group . . . . . . . . . . . . . . . . 202
5.5.2.3 Finite rotations and boosts . . . . . . . . . . . . . . . . . . . . . . 204
5.5.2.4 From rotations and boosts to all Lorentz transformations . . . . . 210
5.5.3 Form invariance and “covariatization” . . . . . . . . . . . . . . . . . . . . . 213
5.5.3.1 Pure boosts and rotations for d = 4, a geometrical take . . . . . . 213
5.5.3.2 Rest frames and connecting boosts . . . . . . . . . . . . . . . . . . 216
5.5.3.3 Pure rotations in d = 4 . . . . . . . . . . . . . . . . . . . . . . . . 219
5.5.3.4 Aside: ϵ-tensors and Hodge duals . . . . . . . . . . . . . . . . . . 221
5.6 Relativistic effects – Lorentz contraction and time-dilation . . . . . . . . . . . . . . 227
5.7 Scalars, vectors, tensors, and invariants – ingredients to relativistic covariance . . . 229
5.7.1 Relativistic covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.7.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
5.7.2.1 Some physically relevant scalars and vectors . . . . . . . . . . . . 231
5.7.2.2 Scalar, vector and tensor valued functions . . . . . . . . . . . . . . 237
5.8 Exercises on Lorentz transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 243

6 Relativistic electrodynamics 244

6.1 Potentials and fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.2 Maxwell’s equations – potentials versus E and B . . . . . . . . . . . . . . . . . . . 246
6.2.1 Gauge potentials, field strengths, and gauge invariance . . . . . . . . . . . . 246
6.2.2 The field strength tensor in terms of E and B . . . . . . . . . . . . . . . . 247
6.2.3 Lorentz transformation properties of F µν , E, and B . . . . . . . . . . . . . 248
6.3 Lagrangian and Hamiltonian formulations . . . . . . . . . . . . . . . . . . . . . . . 250
6.3.1 Actions in a relativistic theory – free point particles . . . . . . . . . . . . . 250
6.3.2 Variational derivatives for fields . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.3.3 The Lagrangian formulation of electrodynamics . . . . . . . . . . . . . . . . 255
6.3.4 Equations of motion from local variations the traditional way . . . . . . . . 258
6.3.5 Global internal symmetries and a precursor to Noether’s theorem . . . . . . 261
6.3.6 Global space time symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . 265
6.3.7 Space time symmetries in electrodynamics . . . . . . . . . . . . . . . . . . . 274
6.3.8 The Hamiltonian formulation of Electrodynamics . . . . . . . . . . . . . . . 278
6.4 The covariant Lorentz force law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
6.5 Retarded potentials and fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
6.5.1 Retarded potentials of charge distributions . . . . . . . . . . . . . . . . . . 293
6.5.2 The associated fields: Jefimenko’s equations . . . . . . . . . . . . . . . . . . 299
6.5.3 The point charge limit: Liénard-Wiechert potentials . . . . . . . . . . . . . 301
6.5.4 The fields of a moving point charge . . . . . . . . . . . . . . . . . . . . . . . 303
6.5.5 Radiation off a point charge . . . . . . . . . . . . . . . . . . . . . . . . . . . 306

7 Electromagnetism and optics 313

typeset with LATEX on February 18, 2025, 13:15 v

 7.1 Electromagnetic waves in a dielectric . . . . . . . . . . . . . . . . . . . . . . . . . . 313
7.2 Boundary conditions between media . . . . . . . . . . . . . . . . . . . . . . . . . . 314
7.3 Waves at an interface of linear media – reflection, refraction, and energy transport 315
7.3.1 Reflection and refraction – wave vectors at an interface, ray optics and beyond316
7.3.2 Reflectivity – energy transport across an interface and the role of polariza-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
7.3.2.1 Normal incidence . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
7.3.2.2 Arbitrary angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
7.4 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
7.4.1 Frequency dependence of the phase velocity and dispersion . . . . . . . . . 327
7.4.2 Electromagnetic waves in a conductor? . . . . . . . . . . . . . . . . . . . . . 328
7.4.3 Dispersion in a model dielectric . . . . . . . . . . . . . . . . . . . . . . . . . 330
7.4.4 Dispersion in a plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331

8 Wave Guides and transmission lines 333

8.1 Waves inside metal tubes – general discussion . . . . . . . . . . . . . . . . . . . . . 333
8.2 TE waves in a rectangular wave guide . . . . . . . . . . . . . . . . . . . . . . . . . 338
8.3 The TEM mode of a coaxial cable . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
8.4 Cavity resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

A Linear Algebra on vector spaces of functions? 342

A.1 Motivation and notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342

B Distributions for pedestrians 346

B.1 Dirac’s δ-distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
B.2 Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
B.2.1 Zur Konvergenz im quadratischen Mittel . . . . . . . . . . . . . . . . . . . . 360
B.3 Lineare Integraltransformationen allgemein . . . . . . . . . . . . . . . . . . . . . . 361
B.4 Fouriertransformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
B.4.1 Das Faltungstheorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
B.4.2 Fouriertransformation und Ableitungen, Wellengleichung . . . . . . . . . . . 365
B.4.3 Sinus- und Cosinustransformationen . . . . . . . . . . . . . . . . . . . . . . 366
B.4.4 Funktionenräume und Vollständigkeitsrelationen . . . . . . . . . . . . . . . 367

C Rotations and states: of angular momentum, spherical harmonics and other

representations of the rotation group 369
C.1 Representations of the rotation group . . . . . . . . . . . . . . . . . . . . . . . . . 369
C.2 The product of two angular momentum representations – adding angular momenta 378
C.3 How representations are typically encountered . . . . . . . . . . . . . . . . . . . . . 381

D Unused tidbits 383

D.1 Point charge expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383

E Consistent self forces 384

Bibliography 385

vi typeset with LATEX on February 18, 2025, 13:15

List of short questions to the reader and todos for the author
o Make sure you fully understand the index logic in Eq. (1.3.15). . . . . . . . . . . . . . . 7
o Use Gauss’ law to show that E must transform like a vector field. . . . . . . . . . . . . . 8
o Use the Lorentz force law to work out the units of the B field. . . . . . . . . . . . . . . 9
o Check consistency with Faraday’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
o Write your own summary of chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
o Determine the units of ϵ0 from Eq. (2.1.13) . . . . . . . . . . . . . . . . . . . . . . . . . 17
o Complete the separation of variables for Laplace’s equation . . . . . . . . . . . . . . . . 23
o Adress the questions posed in emark 2.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
o Why does G0,n depend only on |r − r ′ |? . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
o Verify explicitly that G0 in Eq. (2.1.81) is invariant under translations and rotations. . . 30
o Write your own summary of section 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
o Derive the transformation behavior of use in Eq. (2.2.27). . . . . . . . . . . . . . . . . . 38
o Discuss the physics differences between Fig. 2 and Fig 3 . . . . . . . . . . . . . . . . . . 39
o Visualize the potential of the line segment in Eq. (2.2.50) . . . . . . . . . . . . . . . . . 44
o Write your own summary of section 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
o Study V ∗ as a vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
o finish this . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
o Write your own summary of section 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
o Write your own summary of section 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
o Finish this . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
o Write your own summary of section 2.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
o Insert new example with a circular wire . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
o Remove the trace contributions from the magnetic multipole moments. . . . . . . . . . . 133
o pull part of this into the lecture notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
o Verify arguments in σ ∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
o Force between two parallel wires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
o Dimensional analysis of potential corrections of Gauss’ law . . . . . . . . . . . . . . . . . 146
o How do inner products govern lengths and angles? . . . . . . . . . . . . . . . . . . . . . 154
o Write out the definition of a vector space. . . . . . . . . . . . . . . . . . . . . . . . . . . 155
o Check that Lin(V, W ) is a vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

typeset with LATEX on February 18, 2025, 13:15 vii

o Verify that U(n) and O(n) preserve the associated inner products. . . . . . . . . . . . . 158
o exp(g) = G? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
o Find the eigenvalues of a 2d and a 3d rotation. . . . . . . . . . . . . . . . . . . . . . . . 178
o Write your own summary of section 5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
o Why only contract upper with lower indices . . . . . . . . . . . . . . . . . . . . . . . . . 194
o Check the index structure of Eq. (5.4.21b) . . . . . . . . . . . . . . . . . . . . . . . . . . 195
o Compare Eq. (5.4.21a) with its counterpart for rotations . . . . . . . . . . . . . . . . . . 195
o Show that group actions induce equivalence relations . . . . . . . . . . . . . . . . . . . . 196
o Draw the invariant surfaces for a d = 1 + 2-Galilei space-time. . . . . . . . . . . . . . . . 197
o Draw the invariant surfaces for a d = 1 + 1-Minkowski space-time. . . . . . . . . . . . . 197
o Lorentz transformations connecting space time points . . . . . . . . . . . . . . . . . . . 197
o Inverse Lorentz transformaton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
o Verify that Klein’s group is a group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
o Example basis elements for so(1, 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
o Find the characteristic polynomial for boosts . . . . . . . . . . . . . . . . . . . . . . . . 205
o Verify form of boosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
o A boost in the x-direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
o Draw the world line of a particle at rest in a 1+2 space time . . . . . . . . . . . . . . . . 232
dt
o Verify that dτ = γ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
o Prove that u2 = c2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
o Verify that (uµ ) = γ(cv)t is compatible with (5.7.38). . . . . . . . . . . . . . . . . . . . . 235
o Express x2 via xµ and a frame vector uµ . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
o Show that ∂ µ ϕ(x) transforms like a vector field. . . . . . . . . . . . . . . . . . . . . . . . 238
o Find T µν in the fluid’s local rest frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
o Relate T µ0 and pµ and verify consistency . . . . . . . . . . . . . . . . . . . . . . . . . . 242

o Convert ∂µ F̃ µν = 0 to the sourceless Maxwell equations . . . . . . . . . . . . . . . . . . 248

o Are all three terms in Eq. (6.3.5) possible? . . . . . . . . . . . . . . . . . . . . . . . . . . 251
o Why list π and π † separately? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
o more on the the GR derivation for T ? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
o T µν for hydro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
o Explore the connection between work and force exposed in remark 6.11 . . . . . . . . . 282

viii typeset with LATEX on February 18, 2025, 13:15

o Rederive the field strength tensor from Lorentz force . . . . . . . . . . . . . . . . . . . . 282
o Find the worldline that maximises (z(τ ′ ) − z(τ ))2 . . . . . . . . . . . . . . . . . . . . . . 283
o Review the point charge current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
o fix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
o This implicitly assumes θ ∈ [0, π/2]. Why? . . . . . . . . . . . . . . . . . . . . . . . . . . 317
o Where does the sign come from? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
o 11.01.2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
o The fact that this is indeed a representation of the algebra requires proof. How would
you go about it? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
o need to check what raising and lowering operators do here . . . . . . . . . . . . . . . . . 375

typeset with LATEX on February 18, 2025, 13:15 ix

1 Maxwell’s equations in a given coordinate system (frame)

1.1 The differential form of Maxwell’s equations in microscopic form

(vacuum form)
Maxwell’s equations summarize all there is to know about classical electrodynamics. In differential
(i.e. local form) they read:
1
∇ · E(t, r) = ρ(t, r) Gauss’ law (1.1.1a)
ϵ0
∂
∇ × B(t, r) − µ0 ϵ0 E(t, r) =µ0 J (t, r) Ampere’s law with Maxwell’s correction
∂t
(1.1.1b)
∇ · B(t, r) =0 Gauss’ law for magnetism (1.1.1c)
∂
∇ × E(t, r) + B(t, r) =0 Faraday’s law (1.1.1d)
∂t
Equations without interpretation are meaningless. There are several levels of understanding to be
found as we start to analyze what is in front of us.
Section 1.2 will discuss the immediate physical relevance of the ingredients, Sec 1.2 will delve
deeper into the mathematical formulation and explore what type of model for space is built into
this formulation.

1.2 Interpretations I: charges, currents, and fields

The constants are called ϵ0 : permittivity of free space; µ0 : permeability of free space

Remark 1.1 – Permittivity, permeability and the speed of light:

ϵ0 and µ0 are intimately linked to the speed of light:

c2 ϵ0 µ0 = 1 (1.2.1)

This relationship emerges as we study the propagation of electromagnetic waves.

Gauss’s law: Electric fields are driven by charges

Ampere’s law: the curl of magnetic fields and the time derivative of electric fields (this is
Maxwell’s correction) are driven by electric currents.

Gauss’s law for magnetism: there are no magnetic analogues to the electric charge that might
create a divergence of a magnetic field.

Faraday’s law states that the curl of the electric field is balanced by the time derivative of the
magnetic field. No magnetic analogues of currents exist. This is consistent with the absence of
magnetic charges (also called magnetic monopoles).

typeset with LATEX on February 18, 2025, 13:15 1

Maxwell’s equation are complemented by a continuity equation which encodes the conservation of
electrical charge, and the equation for the Lorentz force on a point charge q.
The continuity equation can be viewed as a simple consequence of Maxwell’s equations – one only
needs to take the divergence of (4.1.1b) and use (4.1.1a) to derive it.

Exercise 1.1 Derive the continuity equation

Derive Eq (1.2.2) by using Eqns (4.1.1b) and (4.1.1a).
→

∂
ρ+∇·J =0 (1.2.2)
∂t
but it is in fact much deeper than that: it expresses conservation of charge. The total charge in
any volume can only change if there is a net flow of current through the volume’s surface. This
is a fundamental statement: electrical charge can not be destroyed or created, it can only be
moved. Quantum field theory –quantum electrodynamics– reinterprets this as a consequence of
gauge symmetry which couples electrically charged fermions to photons through Noether’s theorem
this providing a deep reason for a fundamental phenomenon.
The Lorentz force equation for a point particle
F = qE + qv × B (1.2.3)
is, at this stage, an independent result. Where Maxwell’s equations show how charged matter
creates electromagnetic fields, the Lorentz force equation captures the back reaction of the fields
on matter configurations.

Remark 1.2 – Conservation of charge and Noether’s theorem:

If you think back to classical mechanics, you might recall that one can indeed use the equations
of motion of a particle to derive pertinent conservation laws. Noether’s theorem lifts this onto
a different level and rephrases most of these as a consequence of a continuous symmetry of the
system (typically associated with a Lie group): As you recall translational symmetry ensures
momentum conservation, rotational symmetry ensures angular momentum conservation.
The same can be done here: charge conservation can be interpreted as a consequence of a
gauge symmetry that is an integral part of the equations governing electrodynamics as we will
discuss in detail later.

Remark 1.3 – Lorentz force and measurements:

The Lorentz force law is our main tool to measure the effects of electromagnetic fields: forces
on charges induce motion (currents) whose effects we can observe. This is not the way elec-
trodynamics was established historically, in fact the first derivation of the Lorentz force is

2 typeset with LATEX on February 18, 2025, 13:15

 commonly attributed to Oliver Heaviside in 1889, although other historians suggest an earlier
origin in an 1865 paper by James Clerk Maxwell. Hendrik Lorentz derived it a few years after
Heaviside.

Remark 1.4 – The Lorentz force law defines E and B?:

Some sources claim one may use the Lorentz force law as a tool to define the fields E and B.
One would measure both force and velocity and reconstruct the fields.
The situation is indeed straightforward in the static case where E and F are simply propor-
tional.
Access to the B field is harder since the v projection of B drops out of the equation entirely.
For a single pair of v and F we have three equations for six unknowns in E and B. To obtain
a full set of six linearly independent equations we need to probe the same fields at the same
location with different, linearly independent velocities (and thus forces).

1.3 Interpretations II: of vectors and pseudo-vectors, index notation

Most books stop interpretation here, but the list of statements above is by no means complete.
In particular it completely avoids any thought on how to change coordinate systems. Admittedly
this fits seamlessly with the way electrodynamics is experienced in practice and in a laboratory
environment: One typically deals with a static experimental setup, securely fixed in place and
is only rarely concerned with moving wires or satellites moving past a measuring apparatus at
an appreciable relative velocity – wait a moment: GPS? Yes, this complacency is not very useful.
While changes to moving reference frames will be the domain of special relativity, fixed translations
and rotations can be addressed immediately and are in fact inherent to the way Maxwell’s equations
have been written down in (1.1.1). Maxwell’s equations, as displayed in (1.1.1), contain

apparent scalars ρ (taking values in R[ mC3 ]),

V
vectors E and J (taking values in R3 with units [ m ] and [ mC
3 m ] respectively), as well as gradients
s
∇ · . . . (with units [1/m]),
Ns
and even pseudo-vectors B (taking values in R3 [T = Cm ]) and ∇ × some vector quantity.

This classification can be based on the transformation properties of these objects under three
dimensional rotations, elements of

O(n) := {R ∈ Rn×n |Rt R = 1n }

with n = 3. We discuss the case of general integer n as the arguments are completely generic.

typeset with LATEX on February 18, 2025, 13:15 3

 Concepts 1.1 – Orthogonal groups:
O(n) is the group of linear maps (in three dimensions) that leave the canonical scalar product
of Rn invariant, imposing ⟨u|v⟩ =! ⟨R · u|R · v⟩ for all u, v ∈ Rn , or (in a Cartesian basis),
!
ut · v = (Ru)t · (Rv) for all u, v ⇔ 1n = R t · R (1.3.1)

which is the condition for O(n) elements listed in the definition

O(n) := {R ∈ Rn×n |Rt R = 1n } (1.3.2)

For a given n, this set forms a group with matrix multiplication as its binary operation. We
speak of the orthogonal group in n-dimensions.
This is explored in the two exercises below.

Exercise 1.2 O(n) as a group

In this solution I choose to proceed in a very elementary manner and spell out all the steps
in great detail, mainly to revitalize (or perhaps recucitate?) your linear algebra skills.
Verify the equivalence claimed in Eq. (1.3.1) and then show that each of the sets O(n) indeed
forms a group.
→

Exercise 1.3 Justify the orthogonal in “orthonal group”

Verify that R R = 1n implies that the columns of R are mutually orthonormal. Thus a better
t 1

name for these groups would be orthonormal groups.

→

Taking the determinant of Rt R = 1n , we find that

| det(R)|2 = 1 . (1.3.3)
With all ingredients being real we conclude that
det(R) = ±1 . (1.3.4)
Since det : R → R is a continuous map (it is differentiable even) we know that the pre-images
n×n

of +1 and −1 must form two disconnected components as a subspace in O(n) ⊂ Rn×n , one cannot
continuously deform an element of O(n) with determinant +1 into one with determinant −1 and
vice versa.
Only one of these subsets can form a group by itself, the subset with det(R) = −1 cannot contain
the unit element. The component that does contain the unit element forms a subgroup called
SO(n) := {R ∈ Rn×n |Rt R = 1& det R = 1} . (1.3.5)

4 typeset with LATEX on February 18, 2025, 13:15

SO(n) is the set of proper rotations, rotations that preserve handedness (or orientation), the
det(R) = −1 component of SO(n) consists of orientation changing maps as explained in detail in
concept box 1.2.
Signature of a cyclic
permutation:
Exercise 1.4 Special orthogonal groups The signature of a cyclic
permutation in Sn
Verify that the sets SO(n) are indeed groups. depends on n and is
given by (−1)n+1 . Thus
→ in 3d it is even, but for
our four dimensional
Minkowski space it is
odd.

Concepts 1.2 – Index notation, the ϵ-symbol and handedness:

The argument starts with the Leibniz formula for determinants. Let Aij be the components
of some Kn×n matrix A, then its determinant can be written as
X
det A = ϵ(σ)A1σ(1) · · · Anσ(n) = A1j1 · · · Anjn ϵj1 ...jn (1.3.6)
σ∈Sn

where the last expression employs Einstein’s summation convention and


0 if the jk are not all different from each other
ϵj1 ...jn = (1.3.7)
signature({j , . . . , j }) otherwise.
1 n

Here signature({j1 , . . . , jn }) is the signature of the permutation σ : {1, 2, . . . , n} 7→ {j1 , . . . , jn }.

(In particular ϵ1...n = 1.)
The relation to handedness becomes clear if we note that this expression actually involves the
row vectors A1j1 through Anjn of the matrix A (i.e. the images of the basis vectors under
the linear map At : Kn → Kn ) and restrict to three dimensions where we have geometric
intuition. Then the determinant is positive if these vectors (in the order given) form a right
handed coordinate system and negative if they form a left handed coordinate system. This is
best explored if we use the three standard Cartesian coordinate vectors as rows of a matrix and
compare some standard examples (displaying only the non-zero components of the matrices)
       
1 −1 1 1
 1   −1   1  1  (1.3.8)
1 −1 1 −1

Away from two or three dimensions the sign of the determinant simply becomes the tool to
define handedness of a set of “row” vectors.
Note that det(A) = det(At ) so that we could have phrased this also in terms of columns.

typeset with LATEX on February 18, 2025, 13:15 5

 Concepts 1.3 – The ϵ-symbol as a pseudo-tensor:
The last issue is the transformation property of the ϵ symbol defined in this way. Consider the
expression on the right hand side of the Leibniz equation for the determinant and compare it
to the determinant of a matrix with columns reordered from {1, . . . , n} 7→ {i1 , . . . , in } we find
that the sign swaps according to the signature of that reordering:

Ai1 j1 · · · Ain jn ϵj1 ...jn = ϵi1 ...in A1j1 · · · Anjn ϵj1 ...jn = ϵi1 ...in det(A) (1.3.9)

We can contract this equation with ϵi1 ...in and normalize with 1/n! to compensate for the
number of identical copies in the sums to find
1 i1 ...in j1 ...jn i1 j1
ϵ ϵ A · · · Ain jn = det(A) (1.3.10)
n!
and simultaneously conclude that det(A) = det(At ).
Now specialize to a rotation (in n dimensions), an element in O(n) and note that we can read
Eq. (1.3.9) as a transformation law for the tensor ϵj1 ...jn under rotations:

ϵi1 ...in 7→ Ri1 j1 · · · Rin jn ϵj1 ...jn = ϵi1 ...in det(R) (1.3.11)

The remainder of Eq. (1.3.9) tells us that the epsilon tensor is almost invariant under rotations,
it only changes by a sign if in the process we swap handedness. This swap of sign is the
characteristic of a pseudo-tensor.
We note in preparation of later use that we can of course rewrite Eq. (1.3.9) by applying a
number of R−1 factors to remove a select number Rs on the left, replacing them with R−1
factors appearing on the right. For example, removing all but the first R from the left yields

Ri1 j1 ϵj1 i2 ...in = [R−1 ]i2 j2 · · · [R−1 ]in jn ϵi1 j2 ...jn det(R) = det(R)ϵi1 j2 ...jn Rj2 i2 · · · Rjn in
(1.3.12)

In decoding this it is crucial to track the location of indices on the R: [R−1 ]ij = Rji .

The ingredient to identify pseudo-vectors is the complement of SO(3) in O(3), the set

O(3) \ SO(3) = −1 · SO(3) , (1.3.13)

since under orientation preserving rotations, elements of SO(3), pseudo vectors behave just like
vectors.
We will step through all cases in turn:

Scalars: All constants appearing in Maxwell’s equations such as ϵ0 and µ0 remain invariant
under rotations, we call them scalars under rotations.

Vectors: Coordinate vectors and gradients transform like vectors under rotations:

∂r′i
r 7→ r ′ := Rr or, in components r′i = Rij rj (note Rij = ) (1.3.14)
∂rj

6 typeset with LATEX on February 18, 2025, 13:15

 Assignment:
and so do gradients
Make sure you fully
chain rule R−1 = Rt since R ∈ O(3) understand the index
logic in Eq. (1.3.15).
Which indices are fixed,
∂ ∂rj ∂ ∂ ∂
(∇′ )i = ′i
= ′i j
=[R−1 ]ji j =Rij j = (R∇)i . (1.3.15) which are summed over?
∂r ∂r ∂r ∂r ∂r What range are they
summed over? How do
∂rj you see that a transpose
rj = [R−1 ]ji r′i , hence = [R−1 ]ji appears? Write this
∂r′i again with random
renaming of summed
To reaffirm what we have just learned: Under a general coordinate transform r 7→ r ′ (r), gradient indices to test if you
still understand what is
operators transform with the transpose of the differential of the inverse map r ′ 7→ r(r ′ ) written there. Any
confusion unchecked at
∂ri
[R−1 ]ij := (1.3.16) this point will likely
∂r′j develop into a
debilitating problem
as
later on.
∇′ = [R−1 ]t ∇ . Only if R−1 = Rt , does this turn into ∇′ = R∇ , (1.3.17)
the transformation pattern of a vector.
Time is not affected by rotations and might well be called a scalar under spatial rotations, but
this is not a common practice and a potentially confusing statement. The confusion appears as
a threat particularly if one starts to refer to general Lorentz transformations in special relativity
as Lorentz rotations. These do mix time and space contributions.

All other quantities are functions of coordinates and time. The nomenclature captures this by
calling them fields. Since the time dependence is unaffected by rotations I will suppress it in most
of the expressions for brevity during this discussion.

ρ as a scalar field or scalar density under rotation: Our notion of the behavior of a charge
density ρ(x) under a rotation is compactly summarized by the commutative diagram

id
R R
ρ(r) ρ′ (r ′ ) ρ′ (r ′ ) = ρ(R−1 r ′ ) = ρ(r) . (1.3.18)

R3 R3
R
This states that the shape and values of any given charge density in a rotated coordinate system
r ′ = R · r can be recovered by simply looking up its values in the original coordinate system by
transforming back r = R−1 · r ′ without the need to do anything else (the identity map, id, in
the diagram).
To understand this last step a little better, let us, as we should have done from the beginning,
track units in this process:

id
R[ mC3 ] R[ mC3 ]

ρ(r) ρ′ (r ′ ) ρ′ (r ′ ) = ρ(R−1 r ′ ) = ρ(r) (1.3.19)

R3 [m] R3 [m]
R

typeset with LATEX on February 18, 2025, 13:15 7

 The only thing we could possibly to to a charge density that already carries the right units,
without changing its character as a charge density is to multiply it by a number. Any number
different from one would change the amount of charge, both locally at r ′ and in total. This
would be unphysical, hence the identity map.

J and E as vector fields under rotations:

The current density J , we expect to transform like

R
R3 [ mC3 ms ] R3 [ mC3 ms ]

J (r) J ′ (r ′ ) J ′ (r ′ ) = R · J (R−1 r ′ ) = R · J (r) (1.3.20)

R3 [m] R3 [m]
R
Unlike the rotationally scalar charge density, one needs to also rotate the direction of the
current density vector. After all, if an observer reorients oneself, the direction a set of charges
Assignment:
move in changes relative to the observer. Note that the rotation at the top of the diagram
preserves the magnitude of the current vector as is required physically.
Use Gauss’ law to show
that E must transform If, in the process, one flips handedness no special behaviour is expected.
like a vector field under
rotations. Spell out E fields are less tangible than moving charges and direct intuition might be weaker. To
clearly why this is visualize what is going on recall the a point particle of charge q sitting at r experiences a
necessary for a
consistent physical force F (r) = qE due to E, the force causes an acceleration ma(r) that has a spatial direction.
description of associated At this point we can apply the same intuitive argument as we did with J .
phenomena in our
Alternatively, we have Maxwell’s equations, the destillate of hundreds of years of experiment
world. What is the
model for space and to use to make definite statements. For the E field Gauss’ law is the tool that is easiest to
time that is underlying use. The argument is based on the idea that Maxwell’s equations must hold in any frame.
this requirement? (You
might use the more
If Gauss’ law is to be valid independent of the orientation of our coordinate system, the
complicated discussion combination ∇ · E(t, x) must behave like a scalar field so that it can match the scalar field
of B below as a model ρ(t, x)/ϵ0 on the right hand side in all frames.
for your treatment.)
This only holds if the electric field E transforms as a vector field under rotations.

B as a pseudo vector field under rotations via Ampere’s law:

It turns out Ampere’s law shows nontrivial behaviour if our coordinate transformation changes
handedness. This is, in fact, the reason for the importance of the celebrated right hand rule.
To see this let us write it as
∂
∇ × B(t, r) =µ0 J (t, r) + µ0 ϵ0 E(t, r) (1.3.21)
∂t
and note that both terms on the right hand side transform like a vector under O(3) rotations.
For consistency, we must have the same behaviour for the left hand side where we find a verified
vector ∇, a cross product (or ϵ tensor) and the B field.
It would not be unnatural to expect that B should behave like a vector field under rotation.
This turns out to be too simpleminded. To see why we need to be more sophisticated consider
for the moment the rotation of an arbitrary vector field A, in primed coordinates with x′ = R ·x
then reexpressing it in terms of the original unprimed quantities yields
′ ′ ′ ′
((∇′ × A′ (t, x′ ))i = ϵijk Rjj ∇j Rkk Ak (t, x) (1.3.22)

8 typeset with LATEX on February 18, 2025, 13:15

 which features the original ∇ and A in addition to
′ ′ Rt =R−1 ′ ′ (1.3.12) 1 ′ ′ ′ ′
ϵijk Rjj Rkk = [R−1 ]j j [R−1 ]k k ϵijk = Rii ϵi j k (1.3.23)
det(R)

so that Eq. (1.3.22) becomes

1 ′ ′
((∇′ × A′ (t, x′ ))i = Rii ((∇ × A(t, x))i (1.3.24)
det(R)

stating that the curl of (any) vector field A transforms according to (we omit the time depen-
dence as it only plays a spectator role, we also dispense with units as we have not specified any
for A.)

1
det(R) R
3
R R3
1
∇′ ×A′ (x′ ) = R·(∇×A)(R−1 x′ ) (1.3.25)
∇ × A(x) ∇′ × A′ (x′ ) det(R)

R3 R3
R

We conclude that the curl of a vector field picks up an additional factor 1/ det(R) that would
conflict with the behavior on the right hand side of Ampere’s law. Since the properties of both
∇ and the epsilon tensor are fixed, we conclude that the transformation behavior of B must
compensate:
If we assume that B does not transform as a vector but as a pseudo vector, i.e. according to

det(R)R
R3 [. . .] R3 [. . .]
Assignment:
B(x) ′
B (x ) ′ B ′ (x′ ) = det(R)R · B(R−1 x′ ) = det(R)R · B(x) Use the Lorentz force
law to work out the
units of the B field in
R3 [m] R3 [m] terms of the SI base
R
units. In other words
(1.3.26) express the derived unit
“Tesla” in terms of SI
its curl will transform like a vector and turn Ampere’s equation into a physically acceptable base units.

relationship. Assignment:
We have based our
argument on Ampere’s
Remark 1.5 – Connection to right hand rule: law, verify that B as a
pseudo vector is also
Take a straight wire along ê(3) with a current in the same direction with the B field
consistent with
according to the right hand rule. Now flip the direction of the current to run in −ê3 Faraday’s law.
direction without changing anything else in the physical system, i.e. use a coordinate
transformation
 
1
R= 1  (1.3.27)
−1

typeset with LATEX on February 18, 2025, 13:15 9

 The B fields have no ê3 component in both cases so that R · B = B – still, the right hand
rule requires that the B fields in the two cases must have opposite directions. This minus
sign is provided by the pseudo vector factor det(R) = −1.

One should note that this discussion of scalar, vector, and pseudo-vector fields is inherently built
upon the notion of an isotropic, three dimensional, Euclidean vector space that describes our world
at any given time. This fully parallels the conception of space embodied in Newtonian mechanics.
For those of you who dabble in general relativity: The space of Maxwell’s equations is flat, not
curved. Yet, we have only discussed spatial transformations. Explicitly we have only touched
rotations, but we will soon also include translations. We will indeed stay away from touching time
for a while: Maxwell’s equations, unlike Newtonian mechanics, are fully relativistic in the sense of
special relativity. We will come back to that, and how to square it with out current insights, in
Sec. 5 and beyond.

Assignment:
1.4 Looking back on chapter 1 and looking ahead
Write your own
summary of chapter 1

In chapter 1, we have gone to great lengths to gain a good understanding of the ingredients of
Maxwell’s equations, the conservation law for charges and the Lorentz force equation.

We have interpreted Maxwell’s equations as equations that describe how charges at rest and in
motion generate electromagnetic fields.
This is accompanied by the Lorentz force equation that describes the reverse proccess, the way
fields affect charges.

We have identified the “continuity equation” as one of the most fundamental conservation laws
we know about: the law of charge conservation.

We have classified the ingredients as scalars, scalar fields, vector fields and pseudo-vector fields.

We have revived our understanding of rotations and identified a mathematical way to define
handedness.

We have learned how to deduce features by assuming that our physical equations must hold in
any frame, no matter what its orientation is.

Constants, such as the permittivity ϵ0 and permeability µ0 serve as “unit converters”. They are
needed since our original methods to measure charges, electric and magnetic fields respectively
led us to introduce independent units for them. These are interrelated via Maxwell’s equations
using the unit converting constants.

Maxwell’s equations in the form given in chapter 1 are by no means the only way to describe how
charges create fields. We will later identify a fully relativistically covariant form to describe the
same content, but there are many more of them. Have a look at aside 1.1 for another example.

10 typeset with LATEX on February 18, 2025, 13:15

 Aside 1.1 – The Majorana representation of Maxwell’s equations:
Maxwell’s equations allow a number of different equivalent ways to write them, one of them,
very similar in appearance to the Hamiltonian form of the Dirac equation, is called the Ma-
jorana representation, and is built around the Riemann-Silberstein vector

G(t, r) := E(t, r) + icB(t, r) (1.4.1)

which, by construction, is complex even if E and B are real.

Maxwell’s equations now become a pair of equations
1
∇·G= ρ (1.4.2a)
ϵ0
i ∂
∇×G− G = icµJ (1.4.2b)
c ∂t

Introducing the Hermitian momentum operator p̂ = −iℏ∇ and defining the vector of Hermi-
tian matrices
j
αik := iϵijk (1.4.3)

(with j labeling the matrices and i, k the components of the matrices) these equations take
the form
ℏ
p · G = −i ρ (1.4.4a)
ϵ0
∂
(iℏ − cα · p)G = −iℏcµJ (1.4.4b)
∂t
where the operator cα · p is Hermitian and takes the role of a Hamiltonian. Indeed this
second equation is reminiscent of the original formulation of the Dirac equation (but unlike
that example with an external source). It is dynamical in the sense that it governs time
evolution. The first of the two equations is a constraint equation – a condition that must
be met at all times while G is evolving from some initial condition according to the second
equation.
In the absence of sources, ρ = 0 and J = 0, (this is the realm of plane wave solutions) we see
that the first equation turns into an obvious transversality condition.

People 1.1 – James Clerk Maxwell:

James Clerk Maxwell: (13 June 1831 – 5 November 1879) was a Scottish scientist in the
field of mathematical physics. His most notable achievement was to formulate the
classical theory of electromagnetic radiation, bringing together for the first time
electricity, magnetism, and light as manifestations of the same phenomenon. Maxwell’s
equations for electromagnetism have been called the “second great unification in physics”
after the first one realized by Isaac Newton.

typeset with LATEX on February 18, 2025, 13:15 11

 People 1.2 – Johann Carl Friedrich Gauss:
Johann Carl Friedrich Gauss (German: Gauß) (30 April 1777 – 23 February 1855) was a
German mathematician who contributed significantly to many fields, including number
theory, algebra, statistics, analysis, differential geometry, geodesy, geophysics, mechanics,
electrostatics, astronomy, matrix theory, and optics.

People 1.3 – Michael Faraday:

Michael Faraday (22 September 1791 – 25 August 1867) was an English scientist who
contributed to the fields of electromagnetism and electrochemistry. His main discoveries
include those of electromagnetic induction, diamagnetism and electrolysis.

People 1.4 – Charles-Augustin de Coulomb:

Charles-Augustin de Coulomb (4 June 1736 – 23 August 1806) was a French military
engineer and physicist. He is best known as the eponymous discoverer of what is now
called Coulomb’s law, the description of the electrostatic force of attraction and repulsion,
though he also did important work on friction.

People 1.5 – André-Marie Ampère:

André-Marie Ampère (20 January 1775 – 10 June 1836) was a French physicist and
mathematician who was one of the founders of the science of classical electromagnetism,
which he referred to as ”electrodynamics”. Inventor of numerous applications, such as the
solenoid (a term coined by him) and the electrical telegraph.

1.5 Solutions for chapter 1

Solution 1.1
 
∂
∇ · ∇ × B(t, r) − µ0 ϵ0 E(t, r) = µ0 ∇ · J (t, r)
∂t
∂
− ϵ0 ∇ · E(t, r) = ∇ · J (t, r)
∂t
∂
ρ+∇·J =0 (1.5.1)
∂t

12 typeset with LATEX on February 18, 2025, 13:15

Solution 1.2
The equivalence in Eq. (1.3.1): Going from right to left is trivial, simply multiply both sides of
1n = Rt · R from the right with v and from the left with v t (in the sense of matrix multiplication)
to obtain the left side of the equivalence statement. To obtain the right statement from the left
expand v = e(j) v j and u = e(i) ui using the Cartesian standard basis. This takes

! !
ut · v = (Ru)t · (Rv) ⇔ ui et(i) · e(j) v j = ui et(i) Rt Re(j) v j (1.5.2)

Since the vectors (and hence their components are arbitrary and can be turned on and off at will)
this is equivalent to

et(i) · e(j) = et(i) Rt Re(j) ⇔ δ ij = [Rt R]ij ⇔ 1n = R t R (1.5.3)

The step relating the equivalences in 1.5.2 to those in 1.5.3 relies on a fact that should be familiar
to you: the fact that a linear map is fully determined by its action on all basis elements. Here it
appears twice, once for each argument of the inner product.
To show that the sets form groups, note that The binary operation, matrix multiplication, is
associative for all elements in Rn←n . All we are left to show is that the set contains a neutal
element, that the set is closed under multiplication, and that for any element R it also contains its
inverse R−1 . The neutral element is the unit matrix, and since 1tn 1n = 1n , this is an element of
O(n). Taking two elements R1 and R2 their product satisfies (R1 R2 )t (R1 R2 ) = R2t R1t R1 R2 = 1n .
Since the group condition states that R−1 = Rt , we easily verify that R−1 satisfies (R−1 )t R−1 =
RRt = RR−1 = 1n .
←

Solution 1.3
For this one needs some notation. Let R◦i denote the i-th column of the matrix R. Starting from
1n = Rt R we note that it has the equivalent component equation δ ij = (Rt R)ij . For each pair
(i, j), the latter can be expressed in terms of column vectors to obtain

δ ij = (Rt R)ij = (R◦i )t R◦j (1.5.4)

The key now lies in interpreting this expression: If i ̸= j (where δ ij is zero) this states that column
i is orthognal to column j. If i = j (where δ ij is one) it states that the column vector is a unit
vector, normalized to length 1.
←

Solution 1.4
Obviously SO(n) ⊂ O(n), so that we need not worry about properties associated with Rt R =
1n , only the new condition, det(R) = 1 needs attention. Clearly det(1n ) = 1, so the neutral
element is indeed in the smaller set. The set is closed under multipliation since det(R1 R2 ) =
det(R1 ) det(R2 ) = 1 · 1 = 1 for all R1 , R2 in the set. Lastly, inverses are available since det(R−1 ) =
det(Rt ) = det(R) for all R in the set.
←

typeset with LATEX on February 18, 2025, 13:15 13

2 Electrostatics

2.1 The physics and mathematics of Coulomb’s law

2.1.1 Observational background – point particles and Coulomb’s law, or
“how to measure a Green’s function”
The historical development of electrodynamics began with a quantitative understanding of static,
i.e. time-independent electric fields, followed and paralleled by a development of magneto-statics,
and a final unification of both in the time-dependent case culminating in the formulation of
Maxwell’s equations. In this chapter I will try to retrace the formulation of the basic concepts for
electrostatics.
The core statement here is based on the observation that some material objects may have a property
known as an electric charge that, if present, will result in a force between two such charged objects.
The force may be either repulsive or attractive. We will begin with the notion of point charges,
charged objects whose spatial extend is idealized to be zero (for practical purposes it is enough
that the charges are much smaller than any distances we can resolve in our experiments).
The key experimental observation then is that the magnitude of the force between two electric
point charges q1 and q2 is proportional to the product of the amounts of charge involved

|F1←2 | ∼ |q1 q2 | , (2.1.1)

decreases inversely proportional to the square of their mutual distance

1
|F1←2 | ∼ (2.1.2)
|r1 − r2 |2
and is directed along the line of separation, i.e.
r1 − r2
F1←2 = kq1 q2 . (2.1.3)
|r1 − r2 |3

The electromagnetic force is a two-body force in the sense that in the presence of a whole set of n
point charges, the force all other particles exert on the i-th particle is given by a superposition of
all pairwise forces
n n
X X ri − rj
Fi← = Fi←j = kqi qj (2.1.4)
|ri − rj |3
j̸=i j̸=i

There are two conceptually different things to clean up in these expressions:

1. We need to adapt the direction of the force to the empirically observed repulsion between unlike
charges (which already automatically induces attraction between like charges).
If we consider F12 to be the force exerted by particle 2 onto particle 1 (henceforth denoted by
F1←2 ), this forces the proportionality constant k to be a positive quantity.
2. If we insist that the left hand side be a force, the qi be measured in Coulomb and the ri to
carry length units, then k must be dimensionful. This is discussed in remark 2.1.

14 typeset with LATEX on February 18, 2025, 13:15

 Remark 2.1 – a proportionality constant – old SI units:
The product(s) kqi qj with the yet unknown and possibly dimensionful constant k allows us
to choose units for the electromagnetic charges. We determine charge by measuring the force
exerted by two current-carrying conductors on each other. According to the old definition
of SI units, a current of one Ampere flowing through two parallel, rectilinear, infinitely long
conductors placed at a distance of one meter from each other, causes a force of 2·10−7 Newton
per meter of their length between them. The product of current and time gives the quantity
of charge:

1 Coulomb (C) = 1 Ampere second (As) (2.1.5)

By this (arbitrary) definition, the constant of proportionality k takes a dimension as well as

a fixed numerical value; one sets
1 As
k= ϵ0 = 8.854 · 10−12 (2.1.6)
4πϵ0 Vm
(I have taken the freedom to rely on the notions of Ampere and Volt.)
This now outdated definition still showcases neatly that units often are an arbitrary choice.
Our reconstruction effort highlights that one should invoke all sub-disciplines of physics in
order to efficiently analyze dimensions. There is no dividing boundary between classical
mechanics and electrodynamics in this sense.
Notation and
comment::
For any vector a,
With all constants fixed the force one point charge exerts on another, â := a/|a|, denotes the
associated unit vector,
r̂12 its “direction”. All
F1←2 = kq1 q2 where r12 := r1 − r2 , (2.1.7)
|r12 |2 distance information in
Eq. (2.1.7) resides firmly
is called the Coulomb force, the expression for it is referenced as Coulomb’s (inverse square) law. in the denominator.

Remark 2.2 – A little history and Coulomb’s torsion balance:

Electric and magnetic phenomena were known already in antiquity, but were thoutht of as
clearly distinct phenomena. In historical times, the compass (the navigation instrument not
the drawing tool) led to direct interest in magnetic phenomena. You can find a short but
intriguing historical survey at britannica.com.
Coulomb’s experiment measures

r̂12
F1←2 = kq1 q2 (2.1.8)
|r12 |2

typeset with LATEX on February 18, 2025, 13:15 15

and can separately access the 1/r2 falloff and the com-
pound numerator kq1 q2 . Importantly at the time one
can distinguish attractive from repulsive situations. The
tool used to do so is a torsion balance, the same tool
Cavendish would use later to measure gravitational
force. A schematic of the force balance and an image
of the original instrument are shown on the right.
To determine the constant k (and thus extract the ϵ0 of the modern formulation) one needs a
means to measure charge.
Coulomb had no means to measure charge directly, but by invoking an additional assumption
he worked around that: He assumed that touching a charged metal sphere with an otherwise
identical but uncharged one would equally distribute charge over both and thus half the charge
to be used in a second experiment. Repeating the process several times allows to both test
this additional hypothesis and to isolate k. The fewer iterations one can perform the weaker
the experimental evidence for the veracity of the assumption.
Take note that the experimental setup does not deal with point charges: It involves decidedly
finite sized spheres to carry the charges. Our introcution of point charges is an addition
conceptual notion. Its validity is granted by the differential equations we will rederive later,
while making sure that we can go through all the steps without postulating the notion of
point charges.
In a paper from 2006 [7], Alberto Martı́nez (UT Austin) reports on a modern replication
of Coulomb’s experiment. The account provides ample insight in the experimental pitfalls
encountered in the recreation.

Aside 2.1 – Range of the Coulomb force:

The infinite range of the Coulomb force will translate into a massless force carrier, the photon
in Quantum ElectroDynamics (QED). The finite range of the weak force on the other hand is
directly related directly to the large masses of the W and Z bosons.

Electric fields: If we want to consider the i-th charge as a test charge, we can divide out qi
and define the electric field of the charge distribution of the remaining charges (in the static
case) via

F
E(ri ) = (2.1.9)
qi

This makes sense, since the resulting quantity is a property of the remaining charges only. The
static electric field E(r) generated by a collection of n static (non-moving) point charges qi is
given by
n
1 X r − ri
E(r) := qi (2.1.10)
4πϵ0 i |r − ri |3

16 typeset with LATEX on February 18, 2025, 13:15

 Electric charge densities: Introducing a three dimensional δ-distribution
d3 k ik·(x−x′ )
Z
(3) ′
δ (x − x ) := e (2.1.11)
(2π)3
we can reinterpret Eq. (2.1.4) as
r − r′
Z
1
E(r) := d3 r ′ ρ(r ′ )
4πϵ0 |r − r ′ |3
where the integral extends over all space, and ρ is the static point charge distribution
n
X
ρ(r′ ) = qi δ (3) (r ′ − ri ) (2.1.12)
i

Remark 2.3 – δ-distributions tend to be dimensionful:

Note that the exponent in Eq. (2.1.11) must be dimensionless for the expression to make
mathematical sense. This implies that [k] = [1/length] so that [d3 k] = [1/length3 ] and our
charge Assignment:
particular δ-distribution has units of an inverse volume so that [ρ] = [ volume ]. A volume
Determine the units of
charge density! ϵ0 from Eq. (2.1.13).
You will need input for
the units of ρ and E.
This leaves us with a plausible ansatz for a general formula for the electric field generated by Check your result for
any static charge density ρ(r) given by consistency. For this
you need other
1
Z
r − r′ equations involving
E(r) := d3 r ′ ρ(r ′ ) (2.1.13) these quantites. List
4πϵ0 |r − r ′ |3 suitable ones you have
used.

Remark 2.4 – Point charges to charge distributions – superposition:

The step from Eq. (2.1.10) to Eq. (2.1.13), i.e. from the field of a collection of point
charges to charge distributions relies heavily on the notion of additivity of fields generated
by individual charges. As the consequence the compound field is the sum, or more general
the superposition of of the fields generated by the constituent charges. This is known as the
superposition principle. Above this has been justified by the notion that the forces involved
are two point forces, that there is no intrinsic contribution that involves three or more
charges to be present to appear. Such contributions would, by necessity not be additive.
With this as a background, the generalization is fully justified, but it will lead to a host of
consequences, the first of which will be a linear differential equation relating the fields and
the charge. Note that the solutions of such equations automatically allow superposition of
solutions.

Electric potentials:
At this point we may also observe that the key structure in the integrand can be rewritten as a
gradient
r − r′ 1
′
= −∇r (2.1.14)
|r − r |3 |r − r ′ |

typeset with LATEX on February 18, 2025, 13:15 17

so that we may introduce the (apparently scalar) static electromagnetic potential of our charge
distribution
Z
1 1
Φ(r) := d3 r ′ ρ(r ′ ) (2.1.15)
4πϵ0 |r − r ′ |

and immediately employ it to rewrite E(r) as

E(r) = −∇r Φ(r) . (2.1.16)

Having expressed E as a pure gradient, we conclude that

(∇ × ∇)i Φ = ϵijk ∇j ∇k Φ = −ϵikj ∇j ∇k Φ; partial

derivatives commute were Φ is differentiable

∇ × E(r) = −∇ × ∇Φ(r) = 0 , (2.1.17)

so that qΦ(r) has the interpretation of a potential energy of a point charge q in the presence
of the field E(r) = −∇r Φ(r). Consequently, the work needed to move a charge between two
points (adiabatically – we talk electrostatics here!) is independent of the path taken to move
the charge between the two points. (See reminder 2.1.)

Reminder 2.1 – Conservative and irrotational vector fields:

A force F on a particle in R3 [m] is called conservative if it can be expressed as a pure gradient
of a function ϕ(r), i.e. as

F = −∇Φ (2.1.18)

This automatically implies that its curl vanishes:

∇ × F = −∇ × ∇Φ = 0 (2.1.19)

It also implies that the work done against that force is path independent if the space where
the force is a pure gradient is simply connected or that Φ is not multivalued. For a path
γ : [a, b] → R3 we have
Z Z
Wγ = dγ · F = − dγ · ∇Φ = −(Φ(γ(b)) − Φ(γ(a)) (2.1.20)
γ γ

Imagine now that γ(a) and γ(b) are two points on a closed contour. We can then move γ(b)
towards γ(a) along the contour such that the integral covers more and more of that closed
path. If Φ is single valued, the right hand side of Eq. (2.1.20) will then approach zero. If by
contrast Φ is not single valued, the values will not generally cancel.
We can use Stokes theorem to track down the origin of the problem. We now take γ to be
closed path and S a surface such that its boundary ∂S = γ. Then
Z Z
Wγ = dγ · F = dΣ · (∇ × F ) (2.1.21)
S

18 typeset with LATEX on February 18, 2025, 13:15

 If the curl of F (∇ × F ) vanishes everywhere on S, this gives safely zero, if S extends through
a region where F has a curl (and thus, in that region cannot be written as −∇Φ exclusively)
we may get a nonzero contribution.
This situation automatically leads to a multivalued Φ so that the terms on the right hand side
of Eq. (2.1.20) do not cancel as we let the endpoint approach each other when we “close the
path.”

We can now summarize Coulomb’s law in its integral form:

Observation 1 – The integral form of Coulombs law:

A static (time independent) charge distribution ρ(r) generates a conservative force arising
from a potential
Z
1 1
Φ(r) := d3 r ′ ρ(r ′ ) (2.1.22)
4πϵ0 |r − r ′ |

This potential corresponds to a curl free static electric field E(r) through

E(r) = −∇r Φ(r) . (2.1.23)

The electric field at r implies a force on a stationary point charge q at r via Fq←ρ (r) = qE(r).

In the next section we will translate this into a differen1tial form known as Gauss’ law, starting
from a flux of electric field through a closed surface.

2.1.2 From electric flux to Gauss’ law

Let us consider a charge distribution ρ(r) (any physically plausible distribution will have compact
support) and let V be a volume that may or may not fully contain the support of the charge
distribution. We then may consider the field flux through the surface ∂V of that volume (this will
possibly cut through the charge distribution) defined as
Z
Flux(E, ∂V ) := dΣ · E (2.1.24)
∂V

where dΣ is the oriented surface measure of the surface.

Remark 2.5 – Some books have a penchant for spherical volumes:

There is no reason to restrict the type or shape of the volumes we talk about for the arguments
played out below, but you might recall that many introductory books seem to have a preference
for spherical volumes of some radius r0 for which

dΣ = r̂r02 dΩ = r̂r02 sin θdθdϕ (2.1.25)

typeset with LATEX on February 18, 2025, 13:15 19

 We will see below that the definition above with the more general volume shape leads to
the same result for the flux, provided the two volumes fully enclose the same total charge
distribution.

Reminder 2.2 – Gauss’ theorem:

Let V be a compact volume in R3 with a smooth, orientable, closed surface ∂V with an
oriented measure dΣ. (dΣ has a direction normal to ∂V and points outwards). For a smooth
vector field A in R3 , we have
Z Z
3
d r ∇ · A = dΣ · A , (2.1.26)
V ∂V

i.e. the volume integral over the divergence of the vector field equals the net flux of the vector
field through the volume’s surface.

With this at hand we return to Eq. (2.1.13) and use Gauss’ theorem for integrals to turn the surface
integral into a volume integral over the gradient of E:
Z Z
dΣ · E = d3 r ∇r · E(r) (2.1.27)
∂V V

In both of these expressions we may trade E for the integral expression in terms of a charge density
in Eq. (2.1.13) or, even better for our purpose, write E as a gradient of a potential (2.1.22) and
insert our integral expression for Φ in terms of ρ directly (2.1.23):

ρ(r ′ )
Z  
1 1
∇ · E(r) = −∇2 Φ(r) = d3 r′ ∇2r − (2.1.28)
4π |r − r ′ | ϵ0

At first glance this looks like a non-local relationship between the divergence of E and the charge
density with an integral kernel
 
2 1 1
∇r − (2.1.29)
4π |r − r ′ |

governing the nonlocality. The expression in brackets will play a major role in what follows so we
introduce a shorthand, we define
1 1
G0 (r − r ′ ) := − (2.1.30)
4π |r − r ′ |

The object G0 plays an important role in the study of differential equations of the form

∇2 f (r) = g(r) (2.1.31)

and we will, for the moment, simply assume that G0 satisfies the equation

∇2r G0 (r − r ′ ) = δ (3) (r − r ′ ) . (2.1.32)

20 typeset with LATEX on February 18, 2025, 13:15

For if this holds, Eq. (2.1.28) represents in fact a local relationship between the divergence of E and
the charge density ρ. Provided (2.1.32) holds, we are, in fact looking at Gauss’ law of electrostatics:
ρ(r)
∇r · E(r) = (2.1.33)
ϵ0
Since this equation does not constrain the curl of E, ∇ × E, we need to supplement it with the
knowledge gained earlier, namely that in the static case (this also applies the absence of currents!),
this curl must vanish:
∇×E =0 . (2.1.34)
Let us collect these insights:

Observation 2 – Differential counterpart to Coulomb’s law:

What we have learned from measuring forces, i.e. Coulomb’s law has now been translated into
a counterpart of paired differential equations, Gauss’ law and Faraday’s law for electrostatics

ρ(r)
∇r · E(r) = , and ∇×E =0 , (2.1.35)
ϵ0
that together fully determine E from ρ in the static case. In the static case, the divergence
of E is directly equal to its source, the charge distribution, its curl vanishes.
If we introduce the potential Φ, we can replace the preceding pair of equations by an equivalent
pair,
ρ
∇2 Φ = − , and E = −∇Φ . (2.1.36)
ϵ0
(The curllessness condition for E follows automatically.) The first of these is known as Pois-
son’s equation.
If you are interested in electric flux, we need to integrate Gauss’s law over a volume. The
result connects the electric flux through a (closed) surface directly to the total charge contained
within it:
ρ(r ′ )
Z Z Z
dΣ · E = d3 r ∇r · E(r) = d3 r . (2.1.37)
ϵ0
∂V V V

Remark 2.6 – Linear differential equations and the superposition principle:

As discussed earlier, Eqns. (2.1.35) constitute a set of (coupled) linear differential equations. The
solutions to such equation always can be superposed: if ρ1 and ρ2 are two charge densities and E1 and
E2 are the associated curl free solutions to Coulomb’s law of the individual charge distributions, then
E1 + E2 is the solution associated with the sum of the charge densities ρ1 + ρ2 . For nonlinear systems
of equations this would not be the case, the observed superposition principle forces the equations to
be linear!

Physically this is a consequence of the pure two body nature of the electric force. The force on one

typeset with LATEX on February 18, 2025, 13:15 21

 particle depends linearly on each of the other particles in the system and contains no such thing as
dependence on the product of several charges or even more complicated functional dependencies.

Mathematically this confronts us with a vector space of solutions, a vector space that for any-
thing more complicated than a single harmonic oscillator turns out to be infinite dimensional.
Mathematically we encounter a setting in which we need to deal with infinite dimensional linear
algebra.

2.1.3 Solving Laplace’s and Poisson’s equations – the idea of Green’s

functions
We have now seen the Coulomb’s law can be translated into differential equations recorded in
observation 2, but we have simply quoted outside authority for it by invoking Eq. (2.1.32). Now
that we know that Poisson’s equation,
ρ
∇2 Φ = − , (2.1.38)
ϵ0
is of interest we need to acquire the tools to deal with it and find ways to obtain its solutions. In
the process we will justify Eq. (2.1.32) and close the gap in our argument.
Poisson’s equation is an inhomogeneous partial differential equation for nonzero charge densities.
Its homogeneous counterpart,
∇2 Φ = 0 , (2.1.39)
is known as Laplace’s equation. This latter equation has nontrivial solutions that form a vector
space: Given a set of Φi that solve Laplace’s equation we immediately conclude that any linear
combination formed from them with constant coefficients ai also is a solution:
X X
∇2 ai Φi = ai ∇2 Φi = 0 (2.1.40)
i i

simply because the differential operator is linear. (Eq. (2.1.40) already includes the statement
that with Φ also −Φ is a solution. 0 trivially is one as well. This takes care of all vector space
requirements.)

Remark 2.7 – a complete set of solutions to Laplaces’s equation:

There are many ways to tackle ∇2 Φ = 0. Most straightforwardly one may choose a coordinate system
and the perform a separation of variables ansatz. Note that the coordinate system chosen for this
will determine the bounday conditions satisfied by the solution set that emerges at the end of the
process.

We can choose any meaningful coordinate system, but us recall the procedure for the simplest possible
case, Cartesian coordinates. We write our ansatz as Φ(r) = f (x)g(y)h(z) and obtain the equation

f ′′ (x)g(y)h(z) + f (x)g ′′ (y)h(z) + f (x)g(y)h′′ (z) = 0 (2.1.41)

It is straightforward to isolate the z dependence from the others:

h′′ (z) f ′′ (x)g(y) + f (x)g ′′ (y)

=− (2.1.42)
h(z) f (x)g(y)

22 typeset with LATEX on February 18, 2025, 13:15

 (This of course assumes that h(z) ̸= 0 and f (x)g(y) ̸= 0.)

Since the right hand side does not depend on z and the left does not depend on x and y, both sides
must be constant and equal to each other. Call that constant α for the moment. We now have two
equations

h′′ (z) = αh(z) and f ′′ (x)g(y) + f (x)g ′′ (y) = −αf (x)g(y) (2.1.43)

we can continue to split x and y, which I urge you to do explicitly to eventually arrive at plane wave
solutions

f (x)g(y) = eik⊥ ·r⊥ (2.1.44)

t 2 −1 t 2
where k⊥ = (kx , ky ) ∈ R [m ] and likewise r⊥ = (x, y) ∈ R [m]. Returning to Eq. (2.1.43) we
2
conclude that α = k⊥ . Inserting this into the equation for h gives

h′′ (z) = k⊥
2
h(z) (2.1.45)

which leaves us with two linearly independent solutions solutions governing the z dependence:

h± (z) = e±|k⊥ |z . (2.1.46)

To summarize: the set of functions

{e+|k⊥ |z eik⊥ ·r⊥ , e−|k⊥ |z eik⊥ ·r⊥ |k⊥ ∈ R2 [m−1 ]} (2.1.47)

forms a complete set of linearly independent functions, spanning the nullspace of the hermitian
operator ∇2 . We may single out a different direction than z or use different coordinates altogether
and obtain a different spanning set. What remains unchanged is the space spanned by the complete
sets of states obtain following any of these procedures diligently and correctly, namely

ker(∇2 ) := span{e+|k⊥ |z eik⊥ ·r⊥ , e−|k⊥ |z eik⊥ ·r⊥ |k⊥ ∈ R2 [m−1 ]} (2.1.48)

A general solution can thus be written as a Fourier superposition of the form

Z 2
d k⊥  
Φ0 [f+ , f− ](r) = 2
f+ (k⊥ )e+|k⊥ |z eik⊥ ·r⊥ + f− (k⊥ )e−|k⊥ |z eik⊥ ·r⊥ (2.1.49)
(2π)

I have used functional notation to indicate that this indeed depends on the superposition coefficients
f± (k⊥ ). A unique solution to Laplace’s equation emerges only after we impose specific boundary
conditions that fully determine the f± (k⊥ ). I will often refer to these solutions as zero mode solutions
of the Laplacian ∇2 since they correspond to the zero eigenvalue of the differential operator.
Any other separation ansatz leads to a different, if equivalent representation of the zero mode solutions. Assignment:
The boundary conditions satisfied by the individual solutions will also change. It is not a trivial Complete the separation
enterprise to show in which sense the spaces spanned by solutions obtained with different algorithms of variables for
Laplace’s equation by
are the same. This question is not even generically relevant, as most systems of interest have intrinsic
filling in the gaps for
symmetries that dictate the choice of a symmetry adapted factorization ansatz taylored to the problem the separation in x and
at hand. y. Why are there no ±
alternatives showing up
as in the solution for the
z component?

Some lessons learned from solving ordinary differential equations carry over: Assume we have
obtained one arbitrary solution Φ[ρ](r) to the inhomogeneous equation (Poisson’s equation). (The
notation emphasizes that the solution depends on the charge density driving it.) Then the most

typeset with LATEX on February 18, 2025, 13:15 23

 general solution to Eq. (2.1.38) reads

Φ[ρ, f+ , f− ](r) = Φ[ρ](r) + Φ0 [f+ , f− ](r) (2.1.50)

since evidently
ρ
∇2 (Φ[ρ](r) + Φ0 [f+ , f− ](r)) = ∇2 Φ[ρ](r) = − . (2.1.51)
ϵ0
and the difference between any two solutions to Poisson’s equation must lie in ker(∇2 ):
ρ−ρ
∇2 (Φ1 [ρ](r) − Φ2 [ρ](r)) = − =0 (2.1.52)
ϵ0

Our goal must now be to find one solution to the inhomogeneous equation (Poisson’s equation)
and we will go about this in a constructive manner. Let us return to our discussion of Gauss’ law
and the assumption we had to make to arrive at it: Suppose we know a function G0 (r − r ′ ) that
satisfies (2.1.32)
Assignment:
Answer the “Which ∇2r G0 (r − r ′ ) = δ (3) (r − r ′ ) ,
one?” and “Why?”
questions in remark 2.8. then it is trivial to show that
Provide an example for
ρ(r ′ )
Z
ker(A) = A(V ).
Prove Eq. (2.1.58). Φ(r) := − d3 r′ G0 (r − r ′ ) (2.1.53)
(Hint: Subtract any two
ϵ0
solutions.)
solves the equation ∇2 Φ(r) = −ρ(r)/ϵ0 as we have already done.

Remark 2.8 – purpose of a Green’s function or “this is linear algebra”:

The setting described in this section is in direct analogy to the problem of a set of coupled equations in
linear algebra. In the following it will help to visualize the analogy if you choose a basis for the vectors
and linear maps encountered and think of the vector indices as the counterpart to the coordinates in
the differential equation setting.

Consider a finite dimensional F vector space V and a linear map A ∈ Lin(V ). Consider the equation

Ax = b (2.1.54)

where b ∈ V is a given vector (the inhomogeneity, the ingredient analogous to ρ) and we are interested
in finding all x ∈ V that solve this equation. In other words, we want to find the space of solutions

Sol(A, b) := {x ∈ V |Ax = b} (2.1.55)

in a preferably explicit and easily usable form.

There are a few results that we should recall from linear algebra in this setting:

This equation will have solutions only if b ∈ A(V ) (the image space of A).

The associated homogeneous equation

Ax = 0 (2.1.56)

has a nonzero solution only if A is not invertible. (It always has the trivial solution x = 0, but
that is hardly useful.) The space of solutions to the homogeneous equation is precisely the kernel

24 typeset with LATEX on February 18, 2025, 13:15

 of A:

Sol(A, 0) = ker(A) = {x ∈ V |Ax = 0} (2.1.57)

If x0 is a solution to Eq. (2.1.54), the solution space of the equation, is

Sol(A, b) = x0 + ker(A) := {x0 + x|x ∈ ker(A)} (2.1.58)

since the difference of any two solutions of the inhomogeneous equation is a solution to the homo-
geneous equation:

Ax1 = b & Ax2 = b ⇒ Ax1 − Ax2 = A(x1 − x2 ) = 0 ⇔ (x1 − x2 ) ∈ ker(A). (2.1.59)

The insight into the structure of the solution space presented in Eq. (2.1.58) gives rise to a nice
dimension formula for the space. (Which one?) Be careful, however: the idea of dimension still
makes sese, but unlike ker(A), Sol(A, b) is not a vector space. (Why?)

If ker(A) = {0}, i.e. if A is invertible, then the equation has a unique solution, given by

x = A−1 b (2.1.60)

If ker(A) ̸= {0}, we need to work harder, but we still need to construct only one inhomogeneous
solution, despite there being many.
If we have an inner product at our proposal, and our operator is hermitian –like ∇2 is– we can
find the orthogonal complement to ker(A), ker(A)⊥ = A(V ). Because A is hermitian, both are
invariant subspaces of A: ker(A) is associated with all zero eigenvalues and ker(A)⊥ is associated
with all nonzero ones. This allows us to consider A|ker(A)⊥ , the restriction of A to that complement,
where it becomes an invertible map ker(A)⊥ → ker(A)⊥ . In our Green’s function example we are
(somewhat sneakily) following this route.
Now let us assume b ∈ A(V ) so that a solution exists, denote the basis elements of ker(A) ker(A)⊥
by v(k) and w(i) respectively. Then we have solutions only of b ∈ ker(A)⊥ = A(V ) where it can be
written as b = w(i) bi and the solutions take the form

x = w(i) [A|−1 ]i [b]j + v(k) αk

ker(A)⊥ j
(2.1.61)

(A summation convention over appropriate index ranges applies. The αk ∈ F parametrize the
freedom of choice brought about by the nontrivial kernel.) The extra effort that has gone into
finding a single inhomogeneous solution is the analog of what we do when we construct a Green’s
function.
To make this more visually evident, compare equation (2.1.61) to
Z
Φ(r) = d3 r′ G0 (r − r ′ )(−ρ(r ′ )/ϵ0 ) (2.1.62)
Z 2
d k⊥  +|k⊥ |z ik⊥ ·r⊥ −|k⊥ |z ik⊥ ·r⊥

+ f + (k ⊥ )e e + f − (k ⊥ )e e
(2π)2

You should mentally align j ↔ r ′ and k ↔ k⊥ . The sums of the finite dimensional case are
replaced by integrals for the solutions to our differential equations.
We will see in remarks 2.9 and 2.10 that the complement of the kernel of ∇2 is spanned by plane
wave solutions eik·x with k ∈ R3 [m−1 ] \ {0}.

typeset with LATEX on February 18, 2025, 13:15 25

How do we go about finding such a Green’s function and with it the desired solution to the
inhomogeneous equation? We return to Eq. (2.1.53) and recall that we know already a lot if our
charge density is that of a point charge located at r0 . In this case ρ(r) = qδ (3) (r − r0 ) and we can
isolate G0 (r − r ′ ) directly in Eq. (2.1.53):

qδ (3) (r ′ − r0 )
Z
q
Φpoint (r) = − d3 r′ G0 (r − r ′ ) = − G0 (r − r0 ) (2.1.63)
ϵ0 ϵ0
q 1
But we already know the potential generated by a point charge, it is − 4πϵ0 |r−r0 |
and so we have
an expression for G0 :
1 1
G0 (|r − r ′ |) := − (2.1.64)
4π |r − r ′ |

as anticipated notationally in Eq. (2.1.81).

Note that the solution to the inhomogeneous equation we obtain in this manner has specific bound-
ary conditions: it falls off to zero at spatial infinity. By adding homogeneous solutions we can
modify this to accommodate different requirements:

ρ(r ′ )
Z
Φ[ρ, f+ , f− ](r) = − d3 r′ G0 (r − r ′ ) + Φ0 [f+ , f− ](r) (2.1.65)
ϵ0

The argument used above to determine G0 did invoke the idea of point charges, and strictly
speaking, we have no experimental justification to extrapolate Coulomb’s measurements to such
an extent. We therefore need to get independent confirmation that G0 is the right tool to use. For
this reason and since we will need to be more flexible in working with Green’s functions in later
applications we will develop methods to recognize and construct Green’s functions in different ways
in remarks 2.9 and 2.10.

Remark 2.9 – A Green’s function of the Laplace operator:

To understand why Eq. (2.1.32) holds we can either verify that the left hand side operates in
precisely the same way on test functions as the right hand side, i.e. that
Z
1 !
d2 r ′ ∇2r f (r ′ ) = −4πf (r) (2.1.66)
|r − r ′ |

for any test function f (r ′ ) (with compact support) or we can attempt to relate the factors
1
∇2r |r−r ′ | appearing under the integral on left hand side to a known representation of a delta

distribution. Here I will go the latter route and aim for the plane wave representation of delta
distributions alluded to in Eq. (2.1.11), i.e. we will employ Fourier transforms. To tackle the
1
Fourier transformation of |r−r ′ | we exploit two ideas:

1) inverse powers of lengths can be written as integrals over Gaussians and

2) the Fourier transform of a Gaussian is a Gaussian.

26 typeset with LATEX on February 18, 2025, 13:15

 To start, observe that
Z∞
1 1 1 ds 1 −s(r−r′ )2
= 1 = s2 e (2.1.67)
|r − r ′ | [(r − r ′ )2 ] 2 Γ( 21 ) s
0

To see this, you only need to rescale the integration variable (by setting s′ := s(r − r ′ )2 ) and
recall that the resulting integral is in fact a representation of a Γ function
Z∞ Z∞
dα n −α dβ −n − β1
Γ(n) = α e = β e . (2.1.68)
α β
0 0

The second variant (obtained by a substituting α = 1/β in the first one) will prove useful in
a minute.
The exponential in Eq. (2.1.67) is a Gaussian of width squared 1/(2s) and has a Fourier
representation in terms of a Gaussian with inverse width
′ 2
d3 k −ik·(r−r′ )− 1 k2
Z
′ 2 1 (r−r ) 2s 3
e−s(r−r ) = e− 2 1/(2s) = ( )− 2 e 2 2s . (2.1.69)
2π (2π)3

(Check normalization at r = r ′ .)
Now we are left with
Z∞
d3 k −ik·(r−r′ )
Z
1 1 ds 1 s − 3 − k2 /4
= e s2 ( ) 2 e s . (2.1.70)
|r − r ′ | Γ( 12 ) (2π)3 s π
0
| {z }
3 ∞ 1
ds
s−1 e− s
R
π 2 (k2 /4)−1 s
0

All that is left to be done is to√query your favourite symbolic algebra program to retrieve a
few constants, namely Γ( 21 ) = π and Γ(1) = 1 to arrive at

d3 k −ik·(r−r′ ) 1
Z
1
= 4π e . (2.1.71)
|r − r ′ | (2π)3 k2

With this Fourier representation at hand, Eq. (2.1.32) becomes evident:

d3 k −ik·(r−r′ ) 1 d3 k −ik·(r−r′ )
Z Z
1
∇2r = ∇ 2
r 4π e = −4π e = −4πδ (3) (r − r ′ )
|r − r ′ | (2π)3 k2 (2π)3
(2.1.72)

We emphasize once more that this Greens function produces solutions for potentials and fields
that vanish in a particular way as we approach spatial infinities. This is a specific choice of
Green’s function that is useful for situations in which localized charge distributions are fully
sufficient to describe the situation.

typeset with LATEX on February 18, 2025, 13:15 27

Take note of units:
Conversely it is less obvious if we can use this specific Green’s function to also describe
Note that s in situations with extended surfaces of constant potential in the physical setup (capacitor plates
Eq. (2.1.67) is carrying
units of inverse square for example). At a minimum it would require to add in some linear combination of solutions
length. to the homogeneous equation (zero mode solutions as we will call them). Mathematically this
By grouping powers into issue falls under the heading “boundary value problems,” a topic we will return to in Sec. 2.6
a dimensionless,
“logarighmic” measure
ds/s and a remainder,
the argument of the Γ
function is exposed as
the power on the only
dimensionful factor in Remark 2.10 – Other dimensions? – Running the argument backwards:
the expression. This
way of doing it is Instead of identifying a coordinate space candidate for our Green’s function, as we have done
typically easier to leading up to Eq. (2.1.64), we might of course just have used a Fourier ansatz for our Green’s
remember than the function in terms of a Fourier-amplitude G̃(p) to constuct an expression for it. In hindsight
formula you see in
generic sources.
this is rather obvious. The plane waves eik·x with k ∈ R3 [m−1 ] are eigenstates of the Laplacian
Note also that rescaling
with eigenvalue −k2 viz
to α changes to a
dimensionless ∇2 e−ik·x = −k2 e−ik·x . (2.1.73)
parameter. The
logarithmic measure With the exception of k = 0, these do not belong to the kernel of our operator. (As it will turn
also keeps things tidy
when replacing α with β out, this set of measure zero will not interere with our argument.) The point is that the plane
in Eq. (2.1.68). wave solutions diagonalize our differential operator and allow us to find its inverse simply by
inverting the eigenvalues. In other words, we expect the Fourier transform of G0,n (r − r ′ ) to
be 1/(−k2 ). Let us work through this step by step.
The ansatz is rather obvious, we write
Z n
d k −ik·(r−r′ )
G0,n (r − r ′ ) = e G̃0,n (k) . (2.1.74)
(2π)n

G̃0,n (k) is then determined by imposing

Z n Z n
d k −ik·(r−r′ ) ! d k −ik·(r−r′ )
∇2 G0,n (r − r ′ ) = e (−k 2
) G̃0,n (k) = e . (2.1.75)
(2π)n (2π)n
1
We conclude that G̃(k) = −k 2 will do the job, for any integer dimension n of space we might

want to consider. In this case we have to reconstruct the coordinate space form from this
integral expression.
The Green’s functions defined through a Fourier repesentation in n dimensional space,
′
dn k e−ik·(r−r )
Z
′
G0,n (|r − r |) := , (2.1.76)
(2π)n −k2

have different (unit) dimensions depending on n: [G0,n ] = [k n−2 ] = [r2−n ], leading to the 1/r
we have seen for n = 3, a dimensionless expression (in fact a logarithm as we will see below) in
2d and r like growth in 1d! Even if this result of naive dimensional analysis were modified by
dimensionless functions (it turns out not to be since |r − r ′ | is the only dimensionfull quantity
available and we cannot form nontrivial dimensionless ratios) it is clear that we get drastically
different behavior in different dimensions. (It turns out the naive analysis already yields the

28 typeset with LATEX on February 18, 2025, 13:15

 full structure, at least for n ̸= 2.) The tools to translate to coordinate space are already at our
disposal, we just use the tricks we learned earlier to perform the transformation backwards.
It is worth paying attention to the claim in Eq. (2.1.76) that the Fourier transform leads to a
coordinate expression that only depends on the modulus of the coordinate difference |r − r ′ |,
not the difference vector itself. (Why?)
An explicit calculation yields (it pays to check dimensions in each step: [k] = [1/r], [s] = [r2 ])
′ Z∞
dn k e−ik·(r−r ) dn k −ik·(r−r′ )
Z Z
2
′
G0,n (r − r ) := − =− e ds e−sk
(2π)n k2 (2π)
0
Z∞ Z n Z∞  1  n2
d k −ik·(r−r′ ) k2
− 12 1/2s
′ 2
1 (r−r )
=− ds e e =− ds 2s
e− 2 2s
(2π)n 2π
0 0
1 n n−2
=− [π(r − r ′ )2 ]1− 2 Γ( ) (2.1.77)
4π 2
1
n = 1: Γ( 1−2
2 ) = −2π , so that
2

1
GL,1 (r − r ′ ) = |r − r ′ | (2.1.78)
2
This is growing linearly from r − r ′ = 0.
There are systems in nature that behave effectively like one dimensional systems under
certain conditions. One might guess that these involve charges to be confined to move
only along a single direction. A different, quite prominent example is a system invariant
under all translations parallel to a given plane. What remains nontrivial in this case is the
behavior in the direction perpendicular to it – effectively a one dimensional system.

n = 2 appears to sit on a pole of Γ( n−2

2 ) – does that mean the calculation is meaningless?
Let us expand in n around n = 2. To this end, I need to introduce a scale to keep the
dimension of GL,n at its naive n = 2 value [GL,2 ] = [r0 ]

1 n n−2
G0,2 (r − r ′ ) :=[− [π(r − r ′ )2 µ2 ]1− 2 Γ( )]reg,n→2
4π 2
1 ln((r − r ′ )2 πµ2 ) + γE
=[− ]subtract + + O(n − 2) (2.1.79)
2π(n − 2) 4π

In this expansion, the infinite term is constant and does not contribute in the defining equa-
tion. The scale µ is a regularization artifact, varying it changes the overall normalization of
the potentials calculated using it and will drop out of physical quantities such as potential
differences.
Nevertheless, we see a result that diverges both at long distances and short distances that
does need careful physical assessment in any given example. You should be made aware
that practitioners refer to the short distance limit as the “ultraviolet” or UV limit, and to
the long distance limit a the “infrared” or IR limit.

typeset with LATEX on February 18, 2025, 13:15 29

Assignment: 1
n = 3 was our starting point, and all we need is Γ( 3−2
2 ) = π to arrive at
2

Show that G0,n , as

defined via the Fourier 1 3 3−2 1 1
representation of G0,3 (r − r ′ ) := − [π(r − r ′ )2 µ2 ]1− 2 Γ( )=− (2.1.80)
Eq. (2.1.76), depends 4π 2 4π |r − r ′ |
only on |r − r ′ |. (Hint:
try to test the right This is the only case in which the long distance limit (the IR limit |r −r ′ | → ∞) approaches
hand side for its a finite constant – chosen to be zero in this case.
behavior under
r − r ′ 7→ R(r − r ′ ),
R ∈ O(n). Consider a
coordinate change for
the integration
variable.)
Remark 2.11 – From point particles to Green’s functions:
Let us review our argument this far: The idea to use point particles as a reference instead of
much more familiar macroscopic charged objects has immediately allowed us to abstract away
from the specific properties of these objects and hone into the spatial falloff of electromagnetic
forces. We have identified the superposition principle as a key element in our theory and made
it an observable testable hypothesis. These together have allowed to identify electric fields
and potentials with the Green’s function of the Laplace equation. We have built a series of
mathematical objects based on measurement.

2.1.4 Symmetries: from charge distributions to potentials

Spatial symmetries, be they translational or rotational, of charge distributions imprint directly
onto the associated potentials. This sounds very plausible since “it is the charge distribution
alone” that induces the potential via Eq. (2.1.22):

ρ(r ′ )
Z
1 1
Φ(r) = d3 r′ ′
.
4π |r − r | ϵ0

But what is it precisely in the equation that makes this statement valid? The key point lies in the
other objects featuring in the equation, the integral kernel
1 1
G0 (|r − r ′ |) := − (2.1.81)
4π |r − r ′ |

and the integral measure. Since the integration is over all space, there are no boundaries to
influence the outcome. This kernel is

translation invariant under simultaneous translations of r and r ′ and

Assignment:
Verify explicitly that G0 O(3) invariant under simultaneous transformation of r and r ′ . (With respect to both handedness
in Eq. (2.1.81) is
flips and rotations!)
invariant under
translations and
rotations. since it is a function of the distance between r and r ′ alone.
The integral measure is concerned with r ′ only and responds with a unit Jacobian to both trans-
lations and O(3) transformation of the integration variable r ′ .
Let us explore how this plays out:

30 typeset with LATEX on February 18, 2025, 13:15

 Assume the charge distribution is translationally invariant (constant) in direction â. Then
ρ(r) = ρ(r + a) for all a = aâ. We obtain

ρ(r) = ρ(r + a)
Z Z
−ϵ0 Φ(r) = d3 r′ G0 (|r − r ′ |)ρ(r ′ ) = d3 r′ G0 (|r − r ′ |)ρ(r ′ + a)

r̃ := r ′ + a, unit Jacobian for variable change

Z Z
= d r̃G0 (|r − (r̃ − a)|)ρ(r̃) = d3 r̃G0 (|(r + a) − r̃|)ρ(r̃)
3

= − ϵ0 Φ(r + a) (2.1.82)

Assume the charge distribution is rotationally invariant under rotations around an axis n̂:
ρ(r) = ρ(R(θn̂)r) for all θ ∈ R. Then, in full analogy with the previous calculation
Z Z
−ϵ0 Φ(r) = d3 r′ G0 (|r − r ′ |)ρ(r ′ )= d3 r′ G0 (|r − r ′ |)ρ(R(θn̂)r ′ )

|r − R−1 (θn̂)r̃| = |R(r − R−1 (θn̂)r̃)|

Z Z
= d3 r̃ G0 (|r − R−1 (θn̂)r̃|)ρ(r̃)= d3 r̃ G0 (|R(θn̂)r − r̃|)ρ(r̃)

= − ϵ0 Φ(R(θn̂)r) (2.1.83)

These arguments would fail if the Green’s function were not translation and rotation invariant.

There are, however, physical situations were we are forced to modify the Green’s funtion to accom-
modate boundary conditions with reduced symmetry as we will do explicitly later, for example in
Sec. 2.7.1. Assignment:
Write your own
2.1.5 Looking back on Sec. 2.1 summary of section 2.1
and hand it in with your
homework.

Section 2.1 has run a long race, starting from empirical observations of electric forces encoded in
Coulomb’s law.

A whole host of abstraction steps were introduced, starting with the idea of a (curlless) electric
field. This was followed by observing (mathematically) that it derives from a potential field, with
the space structure of Coulomb’s force field determining how a charge in one place affects charges
in another.

The idea of an electric flux then triggered an interest in the divergence of the electric field and
eventually led us to the differential equivalent of Coulomb’s law: phrased either via the pair of
Gauss’ and Faraday’s laws or Poisson’s equations.

In a last step we developed methods to solve Laplace’s and Poisson’s equations using Green’s
functions as the tool to obtain a generic solution to the inhomogenous source problem.

typeset with LATEX on February 18, 2025, 13:15 31

We have observed that the Green’s functions we have used have the symmetries of an infinite
Euclidean space. This allows symmetries of charge distributions to directly transfer over to the
associated potentials.
The methods we have employed to perform Fourier transforms of arbitrary powers are a useful
calculational method. Variants and generalizations thereof are used in the evalutions of Feynman
integrals in quantum field theory.

People 2.1 – Pierre-Simon, marquis de Laplace:

Pierre-Simon, marquis de Laplace (23 March 1749 – 5 March 1827) was an influential
French scholar whose work was important to the development of mathematics, statistics,
physics, and astronomy. Laplace formulated Laplace’s equation, and pioneered the
Laplace transform which appears in many branches of mathematical physics, a field that
he took a leading role in forming. The Laplacian differential operator, widely used in
mathematics, is also named after him.

People 2.2 – Siméon Denis Poisson:

Siméon Denis Poisson (21 June 1781 – 25 April 1840), was a French mathematician,
geometer, and physicist. He obtained many important results, but within the elite
Académie des Sciences he also was the final leading opponent of the wave theory of light
and was proven wrong on that matter by Augustin-Jean Fresnel.

32 typeset with LATEX on February 18, 2025, 13:15

2.2 Applications: charged surfaces, wires and capacitors
2.2.1 Surface and line charge densities in R3
Up to now we have encountered point charges and generic 3-d charge densities. This is not (quite)
enough to cope with charged metal plates or wires – we need surface and line charge densities.
To understand what is involved let us look at a planar charged surface in the (1,2)-plane through
the origin.
Let me contrast two ways of thinking about it:

The constraint route: Think of this just like you would think of a surface of constraint in
classical mechanics, characterized by the zeroes of a single constraint function f (r), i.e. the Recall:
! Z∞
solution to the constraint equation f (r) = 0, which for our example simply is f (r) = ê(3) · r =: dx ψ(x)δ(f (x)) :=
!
r3 = 0. −∞
n
′
X ψ(xi )
In our case f (r) = 1 where f (r) = 0, but in the general case we need to account for such f ′ (xi )|
i=1
contributions

ρ(r) = σ(r)δ(f (r))|n̂ · ∇r f (r)| (2.2.1)

where n̂ is the surface normal and the units of σ are those of a surface charge density, i.e.
[σ(r)] = [ mC2 ].

The embedding route: An alternative construction is based on parametrizations of the surface

and embedding it in 3-d space. For our simple example, we may describe the surface as

{R3 [length] ∋ w(s1 , s2 ) := s1 e(1) + s2 e(2) |s1 , s2 ∈ R[m]]} (2.2.2)

Please note that the spanning vectors e(1) and e(2) in this setting must be dimensionless.
Trivially
Z
d2 s δ (3) (r − w(s1 , s2 )) = δ(r3 ) (2.2.3)

for our example. This is of course dimensionful: [δ(r3 )] = [length−1 ]. (Note [d2 s] = [length2 ]
and [δ (3) (r)] = [length−3 ].)
Clearly this procedure is not limited to such a simple surface.
The general formula then is
Z
ρ(r) = σ(r) d2 s δ (3) (r − w(s1 , s2 )) (2.2.4)

Note that again [σ(r)] = [ mC2 ] as required of a surface charge density. We do not need to account
for Jacobian factors due to reparametrization invariance of the d2 s integral, even if it is over a
finite patch (as needed for real life capacitor plates for example).

For charged wires we can again walk both routes, but it is the embedding route that wins the
simplicity contest: Imagine a wire along a path w(s) ∈ R3 [m], parametrized by a parameter
s ∈ R[m], a coordinate along the path.

typeset with LATEX on February 18, 2025, 13:15 33

 The constraint route: We obviously need two constraints, best chosen orthogonally to each
other and the path, f i (r) = 0

′
ρ(r) = λ(r)δ(f 1 (r))δ(f 2 (r))|n̂1 · ∇r f 1 (r)||n̂2 · ∇r f 2 (r)| (2.2.5)

where λ(r) has dimensions of a line density.

The embedding route: leads straight to

Z
ρ(r) = λ(r) ds δ (3) (r − w(s)) (2.2.6)

where [λ] = [C/m].

With these tools we are ready to talk about charged 2-d surfaces and 1-d wires using the very same
tools as we have developed for 3-d charge densities and there remains no doubt that all of these
are governed by the same mathematical equations.

2.2.2 Electric fields near charged surfaces – matching conditions

What then do these equations tell us about the electric field as we cross a charged surface? To
this end, choose a trial patch on the surface P , parametrize it in the manner indicated above so
that the charge distribution on the surface can be written as
Z
ρ(r) = σ(r) d2 s δ 3 (r − w(s1 , s2 )) (2.2.7)
P

Now enclose the surface in a thin, layer like trial volume Vϵ , extending a short distance ϵ to both
sides of the surface so that the boundary ∂P lies in the boundary ∂Vϵ . Then consider the flux
through the boundary of Vϵ :
Z Z Z Z Z
ρ(r) σ(r)
dΣ · E = d3 r ∇r · E(r) = d3 r = d3 r d2 s δ (3) (r − w(s1 , s2 ))
ϵ0 ϵ0
∂Vϵ Vϵ Vϵ Vϵ P
Z
σ(w)
= d2 s (2.2.8)
ϵ0
P

where the last expression is evidently independent of ϵ. For consistency, so must be all other
expressions equated here. To obtain an interesting observation from this, split the volume boundary
integral into the two surface–parallel parts and the pieces that cross the charged surface and let
ϵ go to zero. Then the flux through the shrinking “side walls” goes to zero and we are left with
the flux through the parallel parts. Decomposing dS = d2 sn̂ into a surface element and a normal
direction, we find
Z Z
lim dΣ · E = d2 s n̂ · (E+ − E− ) (2.2.9)
ϵ→0
∂Vϵ P

34 typeset with LATEX on February 18, 2025, 13:15

where E± are the limits of E towards the surface along the positive (negative) n axis. We have
Z Z
σ(w(s1 , s2 ))
d2 s n̂ · (E+ − E− ) = d2 s (2.2.10)
ϵ0
P P

Since this must hold for arbitrary surface patches P , the integrands must coincide

σ(s)
n̂ · (E+ − E− )(s) = (2.2.11)
ϵ0
on all points along the charged surface. This gives a jump of the perpendicular or normal component
of the E field across the surface that is directly proportional to the charge density on the surface.
To explore what happens to the parallel component as we cross a charged surface we need a
structure that sees the parallel components. Here we can draw a trial line L on our charged
surface and employ Stokes theorem for a trial surface that cuts our charged surface along the trial
line and extends on both sides into 3d-space perpendicularly by a small distance ϵ.

Reminder 2.3 – Stokes’ theorem:

Let S be a smooth orientable 2-d surface in R3 with surface measure dS, ∂S its (orientable)
boundary (a curve in R3 , with line measure ds oriented such that Ŝ and ds
ˆ form a right helix.
Then, for any smooth vector field A in R 3

Z Z
dS · (∇ × A) = ds · A (2.2.12)
S ∂S

Then the contour integral over the E field probes the difference of the parallel components of the
E field if we let ϵ go to zero, the contribution along the surface normal vanishes, only the parallel
contributions l̂ · E± remain:
Z Z
lim ds · E = dll̂ · (E+ − E− ) (2.2.13)
ϵ→0 ∂S L

On the other hand, via Stokes’ theorem

∇ × E = 0 (electrostatics)
Z Z
ds · E = dS · (∇ × E)=0 (2.2.14)
∂S S

independently of ϵ. We find
Z
dll̂ · (E+ − E− ) = 0 (2.2.15)
L

and since the trial line is arbitrary

n̂ × (E+ − E− ) = 0 (2.2.16)

where again n̂ is the surface normal at the point where we probe E± .

typeset with LATEX on February 18, 2025, 13:15 35

2.2.3 “Capacitors” in three geometries
We will now consider a set of (partially infinite) homogeneously charged surfaces to build “ca-
pacitors” and look at the merits of two different ways of calculating potentials and fields in these
configurations: Direct calculation using Green’s functions and flow arguments supplemented by
symmetry.

Planar geometry – a single charged surface: The potential of a very large (not quite
infinite) homogeneously charged plate of area A. Let a be the surface area of A and consider
A to be covered by a constant charge density σ = q/a. (This is “half a capacitor” if you wish.)
We normalize the surface charge density such that
(
1 if r∥ ∈ A
Z Z
3 3 3 2
q = d rσδ(r − r0 )ΘA (r∥ ) = σ d r where ΘA (r∥ ) (2.2.17)
0 otherwise
A
| {z }
=a

Once we have defined σ we can of course imagine letting a → ∞ while σ remains constant. This
is somewhat dangerous as we casually adopt an infinite total charge.
In this limit ΘA (r) → 1 in all space and we obtain

′
!
d3 k e−ik·(r−r )
Z Z Z
1 3 ′ 1 ′ 1 3 ′
Φ(r) := d r ρplate (r ) = d r 4π σδ(r3 − r03 )
4πϵ0 |r − r ′ | 4πϵ0 (2π)3 k2
3 3 3
!
dk 3 e−ik (r −r0 )
Z
1 1 σ 3
= σ 4π = − σG0,1 (r3 − r03 ) = |r − r03 | (2.2.18)
4πϵ0 2π (k 3 )2 ϵ0 2ϵ0

This shows a linear growth perpendicular to the surface with a cusp at the surface. Obviously
there are a few issues with the infinite surface limit.
Since there is only a gradient in the 3-direction, only E 3 will be nonzero – and, as it turns out,
piecewise constant. It changes sign at the surface
∂ σ r3 − r03 σ
E3 = − Φ(r) = − =− sign(r3 − r03 ) (2.2.19)
∂r 3 2ϵ0 |r3 − r03 | 2ϵ0

Note that E 3 jumps precisely by ϵσ0 as required according Eq. (2.2.11). The calculation shows
beautifully how the 1-dimensional linear growth emerges from a superposition of bona fide
3-dimensional r−1 falloffs.

Fig. 1: Box shaped trial volumes with one pair of faces par-
allel to the surface. Left: An aligned box crossing the surface.
Middle and right: a box touching the surface with different
“heights”.

We may derive most of this result by using flux integrals and symmetry. As many textbooks
blithely state: “Since the plate is infinite, we expect Ê = e(3) .” The largely unspoken argument

36 typeset with LATEX on February 18, 2025, 13:15

 behind this is based on translational symmetry along the surface. For the potential this yields Comment:
ea·∇r Φ(r) is simply a
a·∇r !
Φ(r + a) = e Φ(r) = Φ(r) for all a parallel to the surface ⇔ Φ(r) = Φ(n̂ · r) compact way to write
the Taylor series of
(2.2.20) Φ(r + a) around r
assuming that it
since all derivatives along the surface must vanish. converges.

This has strong consequences for the associated E field. Writing n̂ = e(3) for the plate normal

= 0 (indep of all transverse components

E(r) =−∇Φ(n̂ · r) = − (1 − n̂ ⊗ n̂) · ∇Φ(n̂ · r) −n̂ ⊗ n̂ · ∇Φ(n̂ · r)

Comment::
split ∇ into parts transverse and parallel to n̂
(2.2.21) n̂ · ∇ is simply the
directional derivative in
or the direction of n!

E(r) = n̂ En (n̂ · r) where En (n̂ · r) = −n̂ · ∇Φ(n̂ · r) (2.2.22)

Any component of the E field parallel to the plate vanishes completely, (1 − n̂ ⊗ n̂) · E = 0.
There is one more symmetry in this problem that is present even if the plate is of finite size: a
reflection symmetry about the plane of the plate:

Φ(r⊥ , n̂ · r − z0 ) = Φ(r⊥ , −(n̂ · r − z0 )) ⇒ Φ(r⊥ , |r3 − z0 |) (2.2.23)

Combining the two symmetry criteria we get

d ′ d|r3 − z0 |
E(r) = −n̂ n̂ · ∇Φ(|r3 − z0 |) = −n̂ Φ(|r 3
− z0 |) = − n̂ Φ (|r 3
− z 0 |)
d|r3 − z0 | | dr
3
{z }
sign(r 3 −z0 )
(2.2.24)

so that En changes direction at the plate

En (r3 − z0 ) = −Φ′ (|r3 − z0 |)sign(r3 − z0 ) with En (r3 − z0 ) = −En (−(r3 − z0 ))

(2.2.25)

Remark 2.12 – Symmetry arguments compared:

One may attempt the symmetry argument on the level of the potential or the field. In
both cases one obtains that the function in question may only depend on n̂ · r, not on the
parallel components. However, only if we apply the argument on the level of the potential
do we learn that (1 − n̂ ⊗ n̂) · E = 0.
Attempting to argue purely on the level of the E fields all one can say is that E∥ :=
(1 − n̂ ⊗ n̂) · E, immediately adjacent to the surface, are not affected by the charge on the
surface and hence not constrained by it. A nonzero value would need a reason beyond the
surface charge we are explicitly considering. Unnecessarily incomplete and imprecise.

typeset with LATEX on February 18, 2025, 13:15 37

 Now calculate the total flux through a rectangular box-like trial volume placed with two sides
parallel to the charged surface as shown in for a few examples in Fig 1 with the faces parallel to
the surface displaced by an̂ and bn̂ from the surface (a, b ∈ R[m]). Parametrize the points in
the projection of the box onto the surface by w(s). Since only the sides parallel to the surface
contribute due to translation invariance along the surface we have
R 2
d s (En (w(s) + an̂) − En (w(s) + bn̂))
(En (w(s) + an̂) − En (w(s) + bn̂)) =
Aparallel surface
(
0 if the box avoids the surface
= σ (2.2.26)
ϵ0 if the box cuts through the surface

but is otherwise independent of a and b separately. This means the electric field is constant up
to a jump at the plate. Note that the average fields are equal to the fields at any point along
these surfaces (translation invariance parallel to the plates).
With both the integration area and the distance arbitrary, we get that n̂ · E is separately
Assignment:
constant on either side of the surface, with the jump across the surface given by Eq. (2.2.11).
Derive the behavior of
E under a reflection at The sum of the normal components on opposite sides is fixed by the reflection symmetry however:
a plane through a point
that is not the origin En (w(s) − an̂) + En (w(s) + an̂) = −En (w(s) + an̂) + En (w(s) + an̂) = 0 (2.2.27)
using translations and a
suitable O(3) element.

Now we have two pairs of equations for two unknowns and can recover the result of the Green’s
function calculation performed earlier. Since the electric field implies a force, these values are
physical, unlike the potential, where an overall constant remains unmeasurable.

Planar geometry – a capacitor: To construct a capacitor, superimpose two such charge dis-
tributions with parallel surfaces and opposite charge densities ±σ and obtain potentials (fields)
directly
σ  3 
Φ(r) = |r − (−d/2)| − |r3 − (d/2)| (2.2.28)
2ϵ0
σ
E(r) = − e3 θ(r3 − (−d/2))θ((d/2) − r3 ) (2.2.29)
ϵ0
The potentials become constant outside the plates, the field will vanish there as a consequence.

The potential difference (the voltage) between the two charged plates is

σ σd
∆V := Φ(d/2) − Φ(−d/2) = (d + d) = (2.2.30)
2ϵ0 ϵ0
If we interpret sigma as charge per area, σ = q/A, introduce the capacitance C as the charge
Q
per voltage difference C = ∆V , then
q aϵ0
C= σd
= (2.2.31)
ϵ0
A

38 typeset with LATEX on February 18, 2025, 13:15

 Greens-functions-examples.nb

Fig. 2: Potential Φ/(σ/ϵ0 ) (blue) and normal component of E/(σ/ϵ0 ) (yellow) for a single infinite size plate
(left) and for an infinite size parallel plate capacitor with oppositely charged plates (right). Both horizontal are
plotted in units of an arbitrary length scale d.

Remark 2.13 – Finite size or edge effects:

The argument used to arrive at Eq. (2.2.31) strictly applies only to a capacitor made from
infinite size plates as it assumes translation invariance parallel to the plates. Physical
objects are by necessity finite in size, thus breaking translation invariance. One might
expect that locally i.e. “deep inside the capacitor” the situation is not very different from
the infinite size case. In other words: Ignoring finite size effects amounts to assuming there
are no edge effects and we can just cut out a piece from the infinite parallel-plates.
A real capacitor with finite size plates is much harder to calculate analytically, although a
numerical treatment is straight forward. The situation is illustrated in Fig. 3
A single finite plate generates a potential that at large r must converge to that of a point
charge (in the limit where |r|/|a| ≪ 1 if a characterizes the size of the plates) and thus must
differ strongly from the infinite size realization (Fig. 3 top row). The approximation used
above for a finite size capacitor is not good if d/a ∼ 1 (Fig. 3 middle row) but becomes
1
much better if d/a = 100 ≪ 1 (Fig. 3 bottom row) at least for the potential “inside”,
i.e. between the plates: we observe a nicely uniform linear rise in the bulk. Outside the
capacitor the potential quickly falls to zero in all directions – the net charge, after all, is
zero.
The behavior outside the plates is very different from the idealized case of infinite plates Assignment:
shown in Fig. 2. Discuss how this is consistent in a physical sense.
Discuss in some detail
Just by looking at the plots in Fig.. 3 one cannot fail to notice that the falloff behavior the physics differences
between the situations
is very different as one moves away from the the plate configuration in different directions
depicted in Fig. 2 and
and strongly depends on d/a. Linear, 1/r and stronger falloff are all in evidence in the Fig 3. Recreate the
different configurations. What do you observe and why does it occur? plots in Fig. 3.
(Mathematica might be
a good all in one
solution for this.)

Cylindrical geometry: Envision a cylindrical surface of infinite length, with the symmetry
axis aligned with the 3 axis of our coordinate system. Let the surface carry a homogeneous
charge density q/a where a is a unit area.
In this case the symmetry based flux argument is stronger and one can derive all the full fields

typeset with LATEX on February 18, 2025, 13:15 39

Greens-functions-examples.nb

Fig. 3: An illustration of edge effects on capactior plates. The units on all plots are in multiples of plate with a
or plate separation d. They show potentials in the presence of plates with constant surface charge distributions
plotted over the x, y plane at z = 0. Top row: Potential of a single charged plate. Middle row: Two plates
separeated by a distance d = a. Bottom row: Two plates separated by a distance d = a/100. See remark 2.13
for a detailed discussion.

40 typeset with LATEX on February 18, 2025, 13:15

 and potentials. We consider integrals through cylindrical cans centered around the symmetry
axis inside the charge cylinder and cutting through the cylinder as shown and label its free

Fig. 4: Charged cylinder and concentric trial vol-

umes V with radii small and larger than the charged
cylinder.

Reminder:
parameters, the radius and length ρ and h respectively.
\
∂
Recall that êq := ∂q r(q)
With cylindrical coordinates
       
r cos(φ) cos(φ) − sin(φ) 0
r =  r sin(φ)  with er =  sin(φ)  eφ =  cos(φ)  ez = 0 (2.2.32)
z 0 0 1

the symmetry implies that the potential Φ(r) remains invariant under both rotations around
the z axis and translations along the z axis, i.e.
!
rotations: Φ(R(ω ẑ)r) = Φ(r) (2.2.33)
!
translations: Φ(r + aẑ) = Φ(r) (2.2.34)

The translations tell us that V is independent of z, the rotations tell us the same for any
angular dependence. Taking a gradient, and projecting onto the unit directions of the cylindrical
coordinate system according to

E(r) = −∇r Φ(r) = Er er + Eφ eφ + Ez ez (2.2.35)

where Ei = ei · E = −ei · ∇Φ(r = rer ) we find immediately that Eφ (r) = Ez (r) = 0 and
conclude that

E(r) = Er (r)er (2.2.36)

We choose a trial cylinder of radius ρ and height h

Z (
0 inside
dS · E = Er (ρ)2πρh = QV (2.2.37)
ϵ0 outside

where QV is the total charge inside the can shaped trial volume V .
This can be solved for Er to give
(
0 inside
Er (r) = Q (2.2.38)
ϵ0 2πrh outside

The potential per unit length (normalized to zero at the charged cylinder) can then be found
by integration:

Q
Φ(r) = θ(r − R) ln(r2 /R2 ) (2.2.39)
ϵ0 4πh

typeset with LATEX on February 18, 2025, 13:15 41

Can we do this directly using Green’s functions?
To attempt this we must first settle on an expression for the surface charge density in this case:

σ(r) = σδ(r − R) (2.2.40)

Where the constant σ carries units of charge per area, δ(r − R) units of inverse length, so that
σ(r) indeed carries charge per volume.
This is normalized such that a concentric cylindrical trial volume V yields an enclosed charge
of
Z (
0 V inside
dzdrrdφσ(r) = (2.2.41)
V σ2πRh =: QV V encloses cylinder

We start from the Fourier-representation once more and will use ⊥ to denote coordinates per-
pendicular to the symmetry axis z.
We can once more integrate out the symmetry direction ẑ and, in this case, reduce dimensionality
of the Green’s-function by one. The last trivial step is using the surface constraint to arrive at

V
σR/4πϵ0

r Fig. 5: Numerical result of evaluating (2.2.43). The units on the

R horizontal axis correspond to choosing µ = 1/R.

′
!
d3 k e−ik·(r−r )
Z Z Z
1 1 1
Φ(r) := d3 r′ ρcylinder (r ′ ) = d3 r′ 4π σδ(r′ − R)
4πϵ0 |r − r ′ | 4πϵ0 (2π)3 k2
(2.2.42)

Only the z ′ integral in Eq. (2.2.42) is unconstrained and produces a factor 2πδ(kz ). This allows
us to perform the kz integral by setting all instances of kz to zero. This incidentally also removes
rz from the expression altogether. After these manipulations we are left with a two dimensional
Fourier integral
′
!
d k⊥ e−ik⊥ ·(r−r )⊥
Z Z 2
1 2 ′
(2.2.42) = σ d r 4π 2 δ(r′ − R)
4πϵ0 (2π)2 k⊥
| {z }
=ln((r−r ′ )2⊥ µ2 )

Z2π
σ
= R dϕ ln((r2 + R2 − 2rR cos(ϕ))µ2 ) (2.2.43)
4πϵ0
0

For a numerical evaluation, this poses no difficulty, see Fig. 5. (The numerical integration can be
done in Mathematica, Mathemacica v9 will not really succeed with the analytical integration.)

42 typeset with LATEX on February 18, 2025, 13:15

 To proceed analytically the calculus of residues is the most promising route: We interpret the
ϕ integral as an integral over the unit circle in the complex plane and rewrite cos(ϕ) via

1  it 1  eit 7→z 1 1
cos(t) = e + it −−−−→ (z + ) ,
2 e 2 z

so that
I
σ 1 dz 1 1
(2.2.42) = R ln(r2 + R2 − 2rR (z + )) (2.2.44)
4πϵ0 i z 2 z
|z|=1

This is still a little too cumbersome for a direct attack, but differentiating in r turns the logarithm
in the integrand into a rational function. Hoping it will be easy to integrate back up again after
we have done the resulting integral using residues, we differentiate to find:

R(R − 2rz + Rz 2 )
I I
d 1 dz 1 1 1
R ln(r2 + R2 − 2rR (z + )) = dz (2.2.45)
dr i z 2 z i z(R − rz)(r − Rz)
|z|=1 |z|=1

This has three single poles at

R
z=0 with residue , always enclosed by the unit circle
r
r R
z= with residue − , enclosed only if r < R
R r
R R
z= with residue , enclosed only if r > R
r r
where, as indicated, only one of the latter two poles contributes at a given value of r. We
can account for this by introducing θ-functions to implement the conditions. Our intermediate
integral becomes

2πi R R
(2.2.45) = (1 − θ(R − r) + θ(r − R)) = 4π θ(r − R) . (2.2.46)
i r r
Integrating back up in r gives us the potential up to a constant (this reinstates µ), which can
be chosen to set the potential to zero on the cylinder’s surface (by integrating from R to r)

Z    Zr
σ 1 dz 1 1 σ R
Φ(r) := 2 2
R ln r + R − 2rR z+ = θ(r − R) dr′ ′
4πϵ0 i z 2 z ϵ0 r
|z|=1 R
σ r
= R θ(r − R) ln( ) (2.2.47)
ϵ0 R

What happens if we build a capacitor by adding another, concentrically aligned cylinder of

larger radius but opposite sign charge density?

typeset with LATEX on February 18, 2025, 13:15 43

 Since the theory is linear the potentials are additive. Clearly this will not alter the fields inside
the larger outside cylinder, they remain fully determined by the inner cylinder, we obtain

σ1 σ2
Φ(r) = R1 θ(r − R1 ) ln(r/R1 ) + R2 θ(r − R2 ) ln(r/R2 ) (2.2.48)
ϵ0 ϵ0

By choosing the charge density appropriately we may cancel the electric fields outside the larger
cylinder.

Remark 2.14 – A charged line segment:

A finite length charged line segment of length h provides a good approximation for the far
field of a charged cylinder of the same height with h ≫ R.
This example is easy do deal with by direct integration. Assuming a constant line charge
density along the segment, we have

ρlseg (r) = λδ (2) (r⊥ )θ(z − (−h/2))θ(h/2 − z) (2.2.49)

where as before r⊥ = (1 − ẑ ⊗ ẑ) · r = rx x̂ + ry ŷ, so that

Z Zh/2
1 3 ′ 1 λ 1
Φ(r) = d r ′
ρlseg (r ′ ) = dz ′
4πϵ0 |r − r | 4πϵ0 |r − z ′ ẑ|
−h/2

p
We resolve |r − z ′ ẑ| = |r⊥ + (z − z ′ )ẑ| = 2 + (z − z ′ )2 to find
r⊥

Zh/2
λ 1
= dz ′ p
4πϵ0 2
r⊥ + (z − z ′ )2
−h/2
   z′ =h/2
λ
q
= − ln z − z ′ + r⊥
2 + (z − z ′ )2 µ (2.2.50)
4πϵ0 z ′ =−h/2

Assignment: where µ is an arbitrary constant carrying inverse length units.

Visualize the potential
(Why do I include µ in writing this and why does this result not depend on µ after all?)
of the line segment in
./Potential-of-finite-wire.nb
Eq. (2.2.50). Consider
and discuss the merits
of controur plots vs field
line plots.

Spherical geometry:

We now consider a single spherical shell of radius R, centered around the origin with surface
charge density σ.

The Green’s function calculation is readily done, the integrals are outright trivial compared to
what we have done above. The main final difference is that there is no reduction in dimension-
ality.

44 typeset with LATEX on February 18, 2025, 13:15

 Z∞ Zπ Z2π
δ(r′ − R)
Z
1 3 ′ 1 σ
Φ(r) := d r ′
ρsphere (r ′ ) = ′2
r dr ′
sin θdθ dφ √
4πϵ0 |r − r | 4πϵ0 r2 − 2rr′ cos θ + r′2
0 0 0
Z1 " √ #t=+1
2 2
σR 1 σR r2 − 2rRt + R2
= dt √ = −
2ϵ0 r2 − 2rRt + R 2 2ϵ0 rR
−1 t=−1
=|r+R| =|r−R|
zp }| { zp }| {
2
σR (r + R)2 − (r − R)2
=
2ϵ0 rR

|r + R| − |r − R| =(r + R − (r − R))θ(r − R) + (r + R + (r − R))θ(R − r)

=2Rθ(r − R) + 2rθ(R − r)

σR2
 
2 2
= θ(r − R) + θ(R − r)
2ϵ0 r R
 
Q 1 1
= θ(r − R) + θ(R − r) (2.2.51)
4πϵ0 r R

with Q = σ4πR2 the total charge on the sphere.

The potential is constant inside the charged sphere. Outside the sphere it is identical to that of a
point charge of charge Q at the origin of the sphere. The corresponding E-field is nonvanishing
only outside the sphere where it coincides with that of a point charge at the center of the sphere
(here the origin):
Q r̂
E(r) = −∇Φ(r) = − θ(r − R) 2 (2.2.52)
4πϵ0 r
Note that there are no δ-contributions at r = R from differentiating the θ-functions, they cancel
between the two terms.
The symmetry+flux argument starts again with assuring that the potential will have no angular
dependence at all

Φ(r) = Φ(r) where r = |r| is the radial coordinate in spherical coordinates

(2.2.53)

The reason is rotational symmetry around any axis passing through the origin. There is no
translational symmetry.
The calculational part is best done in spherical coordinates
 
r cos(φ) sin(θ)
r =  r sin(φ) sin(θ)  (2.2.54a)
r cos(θ)

with
     
cos(φ) sin(θ) − sin(φ) cos(φ) cos(θ)
er =  sin(φ) sin(θ)  , eφ =  cos(φ)  and eθ =  sin(φ) cos(θ)  . (2.2.54b)
cos(θ) 0 − sin(θ)

typeset with LATEX on February 18, 2025, 13:15 45

 Taking gradients and expanding in spherical unit vectors,

E(r) = −∇r Φ(r) = er Er + eφ Eφ + eθ Eθ (2.2.55)

the coordinate independence assures that Eφ = Eθ = 0 so that we arrive at

E(r) = −∇r Φ(r) = er Er (r) (2.2.56)

Now we place spherical trial volumes of radius r around the origin and find
Z (
0 inside
dS · E = Er (r)4πr2 = Q (2.2.57)
V ϵ0 outside

where Q is the total charge on the charged sphere. Eqns. (2.2.56) and (2.2.57) together are
equivalent to Eq. (2.2.52).

Remark 2.15 – Symmetries of source distributions and dimensionality:

Reviewing the reduction in dimensionality of the Green’s function in the above, we note that
this is translational symmetry at work: Each linearly independent direction of translational
symmetry reduces the dimensionality of the Green’s function at hand by one.

Assignment:
Write your own
2.2.4 Looking back on Sec. 2.2
summary of section 2.2

Sec. 2.2 collects a few classic examples of charge distributions, all with excessive symmetries:

Planar charge distributions: translational symmetry parallel to the plane.

Infinite straight wires and cylinders: translation symmetry along the cylinder axis and rotation
symmetry around the symmetry axis ω.

Charged spherical surfaces or balls: rotational symmetry under rotations about any rotation
axis throuh the center of the charge distribution.

We have attacked all of these problems to derive potentials and electric fields, each in two ways:

1. By integration, using Green’s functions: The symmetries eliminate the dependence of an as-
sociated variable and thus simplify integration considerably. The presence of a translation
symmetry simplifies calculations that use Fourier representations of said Green’s function: it
leads to δ-distributions in the conjugate momentum and effectively reduces dimensionality by
one for each independent symmetry direction.
2. By flux arguments and Poisson test volumes. This again relies heavily on symmetry: One uses
symmetry arguments to eliminate variable dependence from the potential and derives properties
of the associated electric field from these. Then symmetry adapted test volumes give access to
Flux integrals that directly probe the value of selected components of the electric field.

46 typeset with LATEX on February 18, 2025, 13:15

As a prerequisite we have introduced the idea of parametrizing surfaces and lines and have demon-
strated how that provides access to the associated 3-d embedded volume charge densities. This is
a tool that will become useful in many contexts. Our next encounter with it will be in the context
of spacially well defined objects made of dielectric matter.

typeset with LATEX on February 18, 2025, 13:15 47