Lectures on

Electromagnetic Field Theory

Weng Cho CHEW1

Fall 2020, Purdue University

1
Updated: December 3, 2020
Contents

Preface xi

Acknowledgements xii

1 Introduction, Maxwell’s Equations 1

1.1 Importance of Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 A Brief History of Electromagnetics . . . . . . . . . . . . . . . . . . . 4
1.2 Maxwell’s Equations in Integral Form . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Static Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Coulomb’s Law (Statics) . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Electric Field E (Statics) . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.3 Gauss’s Law (Statics) . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.4 Derivation of Gauss’s Law from Coulomb’s Law (Statics) . . . . . . . 12
1.4 Homework Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Maxwell’s Equations, Differential Operator Form 17

2.1 Gauss’s Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Some Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.2 Gauss’s Law in Differential Operator Form . . . . . . . . . . . . . . . 21
2.1.3 Physical Meaning of Divergence Operator . . . . . . . . . . . . . . . . 21
2.2 Stokes’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Faraday’s Law in Differential Operator Form . . . . . . . . . . . . . . 25
2.2.2 Physical Meaning of Curl Operator . . . . . . . . . . . . . . . . . . . . 26
2.3 Maxwell’s Equations in Differential Operator Form . . . . . . . . . . . . . . . 26
2.4 Homework Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Historical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Constitutive Relations, Wave Equation, Electrostatics, and Static Green’s

Function 29
3.1 Simple Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Emergence of Wave Phenomenon,
Triumph of Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Static Electromagnetics–Revisted . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

i
ii Electromagnetic Field Theory

3.3.2 Poisson’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.3.3 Static Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3.4 Laplace’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Homework Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Magnetostatics, Boundary Conditions, and Jump Conditions 39

4.1 Magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.1 More on Coulomb Gauge . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Boundary Conditions–1D Poisson’s Equation . . . . . . . . . . . . . . . . . . 41
4.3 Boundary Conditions–Maxwell’s Equations . . . . . . . . . . . . . . . . . . . 43
4.3.1 Faraday’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3.2 Gauss’s Law for Electric Flux . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.3 Ampere’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.4 Gauss’s Law for Magnetic Flux . . . . . . . . . . . . . . . . . . . . . . 48

5 Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s

Theorem 49
5.1 Derivation of Biot-Savart Law . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Shielding by Conductive Media . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2.1 Boundary Conditions–Conductive Media Case . . . . . . . . . . . . . 52
5.2.2 Electric Field Inside a Conductor . . . . . . . . . . . . . . . . . . . . . 53
5.2.3 Magnetic Field Inside a Conductor . . . . . . . . . . . . . . . . . . . . 54
5.3 Instantaneous Poynting’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . 56

6 Time-Harmonic Fields, Complex Power 61

6.1 Time-Harmonic Fields—Linear Systems . . . . . . . . . . . . . . . . . . . . . 62
6.2 Fourier Transform Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Complex Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

7 More on Constitute Relations, Uniform Plane Wave 69

7.1 More on Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.1.1 Isotropic Frequency Dispersive Media . . . . . . . . . . . . . . . . . . 69
7.1.2 Anisotropic Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.1.3 Bi-anisotropic Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.1.4 Inhomogeneous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.1.5 Uniaxial and Biaxial Media . . . . . . . . . . . . . . . . . . . . . . . . 72
7.1.6 Nonlinear Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.2 Wave Phenomenon in the Frequency Domain . . . . . . . . . . . . . . . . . . 73
7.3 Uniform Plane Waves in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

8 Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 79

8.1 Plane Waves in Lossy Conductive Media . . . . . . . . . . . . . . . . . . . . . 79
8.1.1 Highly Conductive Case . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.1.2 Lowly Conductive Case . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2 Lorentz Force Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.3 Drude-Lorentz-Sommerfeld Model . . . . . . . . . . . . . . . . . . . . . . . . 82
Contents iii

8.3.1 Cold Collisionless Plasma Medium . . . . . . . . . . . . . . . . . . . . 83

8.3.2 Bound Electron Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.3.3 Damping or Dissipation Case . . . . . . . . . . . . . . . . . . . . . . . 85
8.3.4 Broad Applicability of Drude-Lorentz-Sommerfeld Model . . . . . . . 86
8.3.5 Frequency Dispersive Media . . . . . . . . . . . . . . . . . . . . . . . . 88
8.3.6 Plasmonic Nanoparticles . . . . . . . . . . . . . . . . . . . . . . . . . . 89

9 Waves in Gyrotropic Media, Polarization 91

9.1 Gyrotropic Media and Faraday Rotation . . . . . . . . . . . . . . . . . . . . . 91
9.2 Wave Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.2.1 Arbitrary Polarization Case and Axial Ratio1 . . . . . . . . . . . . . . 96
9.3 Polarization and Power Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

10 Spin Angular Momentum, Complex Poynting’s Theorem, Lossless Condi-

tion, Energy Density 101
10.1 Spin Angular Momentum and Cylindrical Vector Beam . . . . . . . . . . . . 102
10.2 Momentum Density of Electromagnetic Field . . . . . . . . . . . . . . . . . . 103
10.3 Complex Poynting’s Theorem and Lossless Conditions . . . . . . . . . . . . . 104
10.3.1 Complex Poynting’s Theorem . . . . . . . . . . . . . . . . . . . . . . . 104
10.3.2 Lossless Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.4 Energy Density in Dispersive Media . . . . . . . . . . . . . . . . . . . . . . . 107

11 Transmission Lines 111

11.1 Transmission Line Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
11.1.1 Time-Domain Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.1.2 Frequency-Domain Analysis–the Power of Phasor Technique Again! . . 116
11.2 Lossy Transmission Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

12 More on Transmission Lines 121

12.1 Terminated Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . . 121
12.1.1 Shorted Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
12.1.2 Open Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
12.2 Smith Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
12.3 VSWR (Voltage Standing Wave Ratio) . . . . . . . . . . . . . . . . . . . . . . 128

13 Multi-Junction Transmission Lines, Duality Principle 133

13.1 Multi-Junction Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . 133
13.1.1 Single-Junction Transmission Lines . . . . . . . . . . . . . . . . . . . . 135
13.1.2 Two-Junction Transmission Lines . . . . . . . . . . . . . . . . . . . . . 136
13.1.3 Stray Capacitance and Inductance . . . . . . . . . . . . . . . . . . . . 139
13.2 Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
13.2.1 Unusual Swaps2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
13.3 Fictitious Magnetic Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
1 This section is mathematically complicated. It can be skipped on first reading.
2 This section can be skipped on first reading.
iv Electromagnetic Field Theory

14 Reflection, Transmission, and Interesting Physical Phenomena 145

14.1 Reflection and Transmission—Single Interface Case . . . . . . . . . . . . . . . 145
14.1.1 TE Polarization (Perpendicular or E Polarization)3 . . . . . . . . . . . 146
14.1.2 TM Polarization (Parallel or H Polarization)4 . . . . . . . . . . . . . . 148
14.2 Interesting Physical Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . 149
14.2.1 Total Internal Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . 150

15 More on Interesting Physical Phenomena, Homomorphism, Plane Waves,

and Transmission Lines 155
15.1 Brewster’s Angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
15.1.1 Surface Plasmon Polariton . . . . . . . . . . . . . . . . . . . . . . . . . 158
15.2 Homomorphism of Uniform Plane Waves and Transmission Lines Equations . 160
15.2.1 TE or TEz Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
15.2.2 TM or TMz Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

16 Waves in Layered Media 165

16.1 Waves in Layered Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
16.1.1 Generalized Reflection Coefficient for Layered Media . . . . . . . . . . 166
16.1.2 Ray Series Interpretation of Generalized Reflection Coefficient . . . . 167
16.1.3 Guided Modes from Generalized Reflection Coefficients . . . . . . . . 168
16.2 Phase Velocity and Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . 168
16.2.1 Phase Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
16.2.2 Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
16.3 Wave Guidance in a Layered Media . . . . . . . . . . . . . . . . . . . . . . . . 173
16.3.1 Transverse Resonance Condition . . . . . . . . . . . . . . . . . . . . . 173

17 Dielectric Waveguides 175

17.1 Generalized Transverse Resonance Condition . . . . . . . . . . . . . . . . . . 175
17.2 Dielectric Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
17.2.1 TE Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
17.2.2 TM Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
17.2.3 A Note on Cut-Off of Dielectric Waveguides . . . . . . . . . . . . . . . 184

18 Hollow Waveguides 185

18.1 General Information on Hollow Waveguides . . . . . . . . . . . . . . . . . . . 185
18.1.1 Absence of TEM Mode in a Hollow Waveguide . . . . . . . . . . . . . 186
18.1.2 TE Case (Ez = 0, Hz 6= 0, TEz case) . . . . . . . . . . . . . . . . . . . 187
18.1.3 TM Case (Ez 6= 0, Hz = 0, TMz Case) . . . . . . . . . . . . . . . . . . 189
18.2 Rectangular Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
18.2.1 TE Modes (Hz 6= 0, H Modes or TEz Modes) . . . . . . . . . . . . . . 190
3 These polarizations are also variously know as TE , or the s and p polarizations, a descendent from the
z
notations for acoustic waves where s and p stand for shear and pressure waves respectively.
4 Also known as TM polarization.
z
Contents v

19 More on Hollow Waveguides 193

19.1 Rectangular Waveguides, Contd. . . . . . . . . . . . . . . . . . . . . . . . . . 194
19.1.1 TM Modes (Ez 6= 0, E Modes or TMz Modes) . . . . . . . . . . . . . 194
19.1.2 Bouncing Wave Picture . . . . . . . . . . . . . . . . . . . . . . . . . . 195
19.1.3 Field Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
19.2 Circular Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
19.2.1 TE Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
19.2.2 TM Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

20 More on Waveguides and Transmission Lines 205

20.1 Circular Waveguides, Contd. . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
20.1.1 An Application of Circular Waveguide . . . . . . . . . . . . . . . . . 206
20.2 Remarks on Quasi-TEM Modes, Hybrid Modes, and Surface Plasmonic Modes 209
20.2.1 Quasi-TEM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
20.2.2 Hybrid Modes–Inhomogeneously-Filled Waveguides . . . . . . . . . . . 210
20.2.3 Guidance of Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
20.3 Homomorphism of Waveguides and Transmission Lines . . . . . . . . . . . . . 212
20.3.1 TE Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
20.3.2 TM Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
20.3.3 Mode Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

21 Cavity Resonators 219

21.1 Transmission Line Model of a Resonator . . . . . . . . . . . . . . . . . . . . . 219
21.2 Cylindrical Waveguide Resonators . . . . . . . . . . . . . . . . . . . . . . . . 221
21.2.1 βz = 0 Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
21.2.2 Lowest Mode of a Rectangular Cavity . . . . . . . . . . . . . . . . . . 224
21.3 Some Applications of Resonators . . . . . . . . . . . . . . . . . . . . . . . . . 225
21.3.1 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
21.3.2 Electromagnetic Sources . . . . . . . . . . . . . . . . . . . . . . . . . . 227
21.3.3 Frequency Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

22 Quality Factor of Cavities, Mode Orthogonality 233

22.1 The Quality Factor of a Cavity–General Concept . . . . . . . . . . . . . . . . 233
22.1.1 Analogue with an LC Tank Circuit . . . . . . . . . . . . . . . . . . . . 234
22.1.2 Relation to Bandwidth and Pole Location . . . . . . . . . . . . . . . . 236
22.1.3 Wall Loss and Q for a Metallic Cavity . . . . . . . . . . . . . . . . . . 237
22.1.4 Example: The Q of TM110 Mode . . . . . . . . . . . . . . . . . . . . . 239
22.2 Mode Orthogonality and Matrix Eigenvalue Problem . . . . . . . . . . . . . . 240
22.2.1 Matrix Eigenvalue Problem (EVP) . . . . . . . . . . . . . . . . . . . . 240
22.2.2 Homomorphism with the Waveguide Mode Problem . . . . . . . . . . 241
22.2.3 Proof of Orthogonality of Waveguide Modes5 . . . . . . . . . . . . . . 242
5 This may be skipped on first reading.
vi Electromagnetic Field Theory

23 Scalar and Vector Potentials 245

23.1 Scalar and Vector Potentials for Time-Harmonic Fields . . . . . . . . . . . . . 245
23.2 Scalar and Vector Potentials for Statics, A Review . . . . . . . . . . . . . . . 246
23.2.1 Scalar and Vector Potentials for Electrodynamics . . . . . . . . . . . . 247
23.2.2 More on Scalar and Vector Potentials . . . . . . . . . . . . . . . . . . 249
23.3 When is Static Electromagnetic Theory Valid? . . . . . . . . . . . . . . . . . 250
23.3.1 Quasi-Static Electromagnetic Theory . . . . . . . . . . . . . . . . . . . 255

24 Circuit Theory Revisited 257

24.1 Kirchhoff Current Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
24.2 Kirchhoff Voltage Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
24.3 Inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
24.4 Capacitance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
24.5 Resistor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
24.6 Some Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
24.7 Energy Storage Method for Inductor and Capacitor . . . . . . . . . . . . . . . 264
24.8 Finding Closed-Form Formulas for Inductance and Capacitance . . . . . . . . 265
24.9 Importance of Circuit Theory in IC Design . . . . . . . . . . . . . . . . . . . 267
24.10Decoupling Capacitors and Spiral Inductors . . . . . . . . . . . . . . . . . . . 269

25 Radiation by a Hertzian Dipole 271

25.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
25.2 Approximation by a Point Source . . . . . . . . . . . . . . . . . . . . . . . . . 273
25.2.1 Case I. Near Field, βr  1 . . . . . . . . . . . . . . . . . . . . . . . . 275
25.2.2 Case II. Far Field (Radiation Field), βr  1 . . . . . . . . . . . . . . 276
25.3 Radiation, Power, and Directive Gain Patterns . . . . . . . . . . . . . . . . . 276
25.3.1 Radiation Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

26 Radiation Fields 283

26.1 Radiation Fields or Far-Field Approximation . . . . . . . . . . . . . . . . . . 284
26.1.1 Far-Field Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 285
26.1.2 Locally Plane Wave Approximation . . . . . . . . . . . . . . . . . . . 286
26.1.3 Directive Gain Pattern Revisited . . . . . . . . . . . . . . . . . . . . . 289

27 Array Antennas, Fresnel Zone, Rayleigh Distance 293

27.1 Linear Array of Dipole Antennas . . . . . . . . . . . . . . . . . . . . . . . . . 293
27.1.1 Far-Field Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 294
27.1.2 Radiation Pattern of an Array . . . . . . . . . . . . . . . . . . . . . . 295
27.2 When is Far-Field Approximation Valid? . . . . . . . . . . . . . . . . . . . . . 298
27.2.1 Rayleigh Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
27.2.2 Near Zone, Fresnel Zone, and Far Zone . . . . . . . . . . . . . . . . . 300
Contents vii

28 Different Types of Antennas—Heuristics 303

28.1 Resonance Tunneling in Antenna . . . . . . . . . . . . . . . . . . . . . . . . . 304
28.2 Horn Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
28.3 Quasi-Optical Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
28.4 Small Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

29 Uniqueness Theorem 317

29.1 The Difference Solutions to Source-Free Maxwell’s Equations . . . . . . . . . 317
29.2 Conditions for Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
29.2.1 Isotropic Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
29.2.2 General Anisotropic Case . . . . . . . . . . . . . . . . . . . . . . . . . 321
29.3 Hind Sight Using Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . 322
29.4 Connection to Poles of a Linear System . . . . . . . . . . . . . . . . . . . . . 323
29.5 Radiation from Antenna Sources and Radiation Condition . . . . . . . . . . . 324

30 Reciprocity Theorem 327

30.1 Mathematical Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
30.2 Conditions for Reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
30.3 Application to a Two-Port Network and Circuit Theory . . . . . . . . . . . . 331
30.4 Voltage Sources in Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . 333
30.5 Hind Sight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
30.6 Transmit and Receive Patterns of an Antennna . . . . . . . . . . . . . . . . . 335

31 Equivalence Theorems, Huygens’ Principle 339

31.1 Equivalence Theorems or Equivalence Principles . . . . . . . . . . . . . . . . 339
31.1.1 Inside-Out Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
31.1.2 Outside-in Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
31.1.3 General Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
31.2 Electric Current on a PEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
31.3 Magnetic Current on a PMC . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
31.4 Huygens’ Principle and Green’s Theorem . . . . . . . . . . . . . . . . . . . . 343
31.4.1 Scalar Waves Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
31.4.2 Electromagnetic Waves Case . . . . . . . . . . . . . . . . . . . . . . . 346

32 Shielding, Image Theory 349

32.1 Shielding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
32.1.1 A Note on Electrostatic Shielding . . . . . . . . . . . . . . . . . . . . . 349
32.1.2 Relaxation Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
32.2 Image Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
32.2.1 Electric Charges and Electric Dipoles . . . . . . . . . . . . . . . . . . 352
32.2.2 Magnetic Charges and Magnetic Dipoles . . . . . . . . . . . . . . . . . 353
32.2.3 Perfect Magnetic Conductor (PMC) Surfaces . . . . . . . . . . . . . . 354
32.2.4 Multiple Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
32.2.5 Some Special Cases—Spheres, Cylinders, and Dielectric Interfaces . . 356
viii Electromagnetic Field Theory

33 High Frequency Solutions, Gaussian Beams 361

33.1 Tangent Plane Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . 362
33.2 Fermat’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
33.2.1 Generalized Snell’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . 365
33.3 Gaussian Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
33.3.1 Derivation of the Paraxial/Parabolic Wave Equation . . . . . . . . . . 366
33.3.2 Finding a Closed Form Solution . . . . . . . . . . . . . . . . . . . . . 367
33.3.3 Other solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369

34 Scattering of Electromagnetic Field 371

34.1 Rayleigh Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
34.1.1 Scattering by a Small Spherical Particle . . . . . . . . . . . . . . . . . 373
34.1.2 Scattering Cross Section . . . . . . . . . . . . . . . . . . . . . . . . . . 375
34.1.3 Small Conductive Particle . . . . . . . . . . . . . . . . . . . . . . . . . 378
34.2 Mie Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
34.2.1 Optical Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
34.2.2 Mie Scattering by Spherical Harmonic Expansions . . . . . . . . . . . 381
34.2.3 Separation of Variables in Spherical Coordinates6 . . . . . . . . . . . . 381

35 Spectral Expansions of Source Fields—Sommerfeld Integrals 383

35.1 Spectral Representations of Sources . . . . . . . . . . . . . . . . . . . . . . . . 383
35.1.1 A Point Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
35.2 A Source on Top of a Layered Medium . . . . . . . . . . . . . . . . . . . . . . 389
35.2.1 Electric Dipole Fields–Spectral Expansion . . . . . . . . . . . . . . . . 389
35.3 Stationary Phase Method—Fermat’s Principle . . . . . . . . . . . . . . . . . . 392
35.4 Riemann Sheets and Branch Cuts7 . . . . . . . . . . . . . . . . . . . . . . . . 396
35.5 Some Remarks8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396

36 Computational Electromagnetics, Numerical Methods 399

36.1 Computational Electromagnetics, Numerical Methods . . . . . . . . . . . . . 401
36.2 Examples of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 401
36.3 Examples of Integral Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 402
36.3.1 Volume Integral Equation . . . . . . . . . . . . . . . . . . . . . . . . . 402
36.3.2 Surface Integral Equation . . . . . . . . . . . . . . . . . . . . . . . . . 404
36.4 Function as a Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
36.5 Operator as a Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
36.5.1 Domain and Range Spaces . . . . . . . . . . . . . . . . . . . . . . . . 406
36.6 Approximating Operator Equations with Matrix Equations . . . . . . . . . . 407
36.6.1 Subspace Projection Methods . . . . . . . . . . . . . . . . . . . . . . . 407
36.6.2 Dual Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
36.6.3 Matrix and Vector Representations . . . . . . . . . . . . . . . . . . . . 408
36.6.4 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
6 May be skipped on first reading.
7 This may be skipped on first reading.
8 This may be skipped on first reading.
Contents ix

36.6.5 Differential Equation Solvers versus Integral Equation Solvers . . . . . 410

36.7 Solving Matrix Equation by Optimization . . . . . . . . . . . . . . . . . . . . 410
36.7.1 Gradient of a Functional . . . . . . . . . . . . . . . . . . . . . . . . . . 411

37 Finite Difference Method, Yee Algorithm 415

37.1 Finite-Difference Time-Domain Method . . . . . . . . . . . . . . . . . . . . . 415
37.1.1 The Finite-Difference Approximation . . . . . . . . . . . . . . . . . . . 416
37.1.2 Time Stepping or Time Marching . . . . . . . . . . . . . . . . . . . . . 418
37.1.3 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
37.1.4 Grid-Dispersion Error . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
37.2 The Yee Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
37.2.1 Finite-Difference Frequency Domain Method . . . . . . . . . . . . . . 427
37.3 Absorbing Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 428

38 Quantum Theory of Light 431

38.1 Historical Background on Quantum Theory . . . . . . . . . . . . . . . . . . . 431
38.2 Connecting Electromagnetic Oscillation to Simple Pendulum . . . . . . . . . 434
38.3 Hamiltonian Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
38.4 Schrödinger Equation (1925) . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
38.5 Some Quantum Interpretations–A Preview . . . . . . . . . . . . . . . . . . . . 443
38.5.1 Matrix or Operator Representations . . . . . . . . . . . . . . . . . . . 444
38.6 Bizarre Nature of the Photon Number States . . . . . . . . . . . . . . . . . . 445

39 Quantum Coherent State of Light 447

39.1 The Quantum Coherent State . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
39.1.1 Quantum Harmonic Oscillator Revisited . . . . . . . . . . . . . . . . . 448
39.2 Some Words on Quantum Randomness and Quantum Observables . . . . . . 450
39.3 Derivation of the Coherent States . . . . . . . . . . . . . . . . . . . . . . . . . 451
39.3.1 Time Evolution of a Quantum State . . . . . . . . . . . . . . . . . . . 453
39.4 More on the Creation and Annihilation Operator . . . . . . . . . . . . . . . . 454
39.4.1 Connecting Quantum Pendulum to Electromagnetic Oscillator9 . . . . 457
39.5 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460

9 May be skipped on first reading.

x Electromagnetic Field Theory
Preface

This set of lecture notes is from my teaching of ECE 604, Electromagnetic Field Theory, at
ECE, Purdue University, West Lafayette. It is intended for entry level graduate students.
Because different universities have different undergraduate requirements in electromagnetic
field theory, this is a course intended to “level the playing field”. From this point onward,
hopefully, all students will have the fundamental background in electromagnetic field theory
needed to take advance level courses and do research at Purdue.
In developing this course, I have drawn heavily upon knowledge of our predecessors in
this area. Many of the textbooks and papers used, I have listed them in the reference list.
Being a practitioner in this field for over 40 years, I have seen electromagnetic theory im-
pacting modern technology development unabated. Despite its age, the set of Maxwell’s
equations has endured and continued to be important, from statics to optics, from classi-
cal to quantum, and from nanometer lengthscales to galactic lengthscales. The applications
of electromagnetic technologies have also been tremendous and wide-ranging: from geophysi-
cal exploration, remote sensing, bio-sensing, electrical machinery, renewable and clean energy,
biomedical engineering, optics and photonics, computer chip, computer system, and quantum
computer designs, quantum communication and many more. Electromagnetic field theory is
not everything, but is remains an important component of modern technology developments.
The challenge in teaching this course is on how to teach over 150 years of knowledge in
one semester: Of course this is mission impossible! To do this, we use the traditional wisdom
of engineering education: Distill the knowledge, make it as simple as possible, and teach the
fundamental big ideas in one short semester. Because of this, you may find the flow of the
lectures erratic. Some times, I feel the need to touch on certain big ideas before moving on,
resulting in the choppiness of the curriculum.
Also, in this course, I exploit mathematical homomorphism as much as possible to simplify
the teaching. After years of practising in this area, I find that some complex and advanced
concepts become simpler if mathematical homomorphism is exploited between the advanced
concepts and simpler ones. An example of this is on waves in layered media. The problem
is homomorphic to the transmission line problem: Hence, using transmission line theory, one
can simplify the derivations of some complicated formulas.
A large part of modern electromagnetic technologies is based on heuristics. This is some-
thing difficult to teach, as it relies on physical insight and experience. Modern commercial
software has reshaped this landscape, as the field of math-physics modeling through numer-
ical simulations, known as computational electromagnetic (CEM), has made rapid advances
in recent years. Many cut-and-try laboratory experiments, based on heuristics, have been

xi
xii Lectures on Electromagnetic Field Theory

replaced by cut-and-try computer experiments, which are a lot cheaper.

An exciting modern development is the role of electromagnetics and Maxwell’s equations
in quantum technologies. We will connect Maxwell’s equations to them toward the end of
this course. This is a challenge, as it has never been done before at an entry level course to
my knowledge.

Weng Cho CHEW

December 3, 2020 Purdue University

Acknowledgements
I like to thank Dan Jiao for sharing her lecture notes in this course, as well as Andy Weiner
for sharing his experience in teaching this course. Mark Lundstrom gave me useful feedback
on Lectures 38 and 39 on the quantum theory of light. Also, I am thankful to Dong-Yeop Na
for helping teach part of this course. Robert Hsueh-Yung Chao also took time to read the
lecture notes and gave me very useful feedback.
Lecture 1

Introduction, Maxwell’s
Equations

In the beginning, this field is either known as electricity and magnetism or optics. But later,
as we shall discuss, these two fields are found to be based on the same set equations known as
Maxwell’s equations. Maxwell’s equations unified these two fields, and it is common to call
the study of electromagnetic theory based on Maxwell’s equations electromagnetics. It has
wide-ranging applications from statics to ultra-violet light in the present world with impact
on many different technologies.

1.1 Importance of Electromagnetics

We will explain why electromagnetics is so important, and its impact on very many different
areas. Then we will give a brief history of electromagnetics, and how it has evolved in the
modern world. Next we will go briefly over Maxwell’s equations in their full glory. But we
will begin the study of electromagnetics by focussing on static problems which are valid in
the long-wavelength limit.
It has been based on Maxwell’s equations, which are the result of the seminal work of
James Clerk Maxwell completed in 1865, after his presentation to the British Royal Society
in 1864. It has been over 150 years ago now, and this is a long time compared to the leaps
and bounds progress we have made in technological advancements. Nevertheless, research in
electromagnetics has continued unabated despite its age. The reason is that electromagnetics
is extremely useful, and has impacted a large sector of modern technologies.
To understand why electromagnetics is so useful, we have to understand a few points
about Maxwell’s equations.
ˆ Maxwell’s equations are valid over a vast length scale from subatomic dimensions to
galactic dimensions. Hence, these equations are valid over a vast range of wavelengths,
going from static to ultra-violet wavelengths.1
1 Current lithography process is working with using ultra-violet light with a wavelength of 193 nm.

1
2 Electromagnetic Field Theory

ˆ Maxwell’s equations are relativistic invariant in the parlance of special relativity [1]. In
fact, Einstein was motivated with the theory of special relativity in 1905 by Maxwell’s
equations [2]. These equations look the same, irrespective of what inertial reference
frame one is in.

ˆ Maxwell’s equations are valid in the quantum regime, as it was demonstrated by Paul
Dirac in 1927 [3]. Hence, many methods of calculating the response of a medium to
classical field can be applied in the quantum regime also. When electromagnetic theory
is combined with quantum theory, the field of quantum optics came about. Roy Glauber
won a Nobel prize in 2005 because of his work in this area [4].

ˆ Maxwell’s equations and the pertinent gauge theory has inspired Yang-Mills theory
(1954) [5], which is also known as a generalized electromagnetic theory. Yang-Mills
theory is motivated by differential forms in differential geometry [6]. To quote from
Misner, Thorne, and Wheeler, “Differential forms illuminate electromagnetic theory,
and electromagnetic theory illuminates differential forms.” [7, 8]

ˆ Maxwell’s equations are some of the most accurate physical equations that have been
validated by experiments. In 1985, Richard Feynman wrote that electromagnetic theory
had been validated to one part in a billion.2 Now, it has been validated to one part in
a trillion (Aoyama et al, Styer, 2012).3

ˆ As a consequence, electromagnetics has had a tremendous impact in science and tech-

nology. This is manifested in electrical engineering, optics, wireless and optical commu-
nications, computers, remote sensing, bio-medical engineering etc.

2 This means that if a jet is to fly from New York to Los Angeles, an error of one part in a billion means

an error of a few millmeters.

3 This means an error of a hairline, if one were to fly from the earth to the moon.
Introduction, Maxwell’s Equations 3

Figure 1.1: The impact of electromagnetics in many technologies. The areas in blue are
prevalent areas impacted by electromagnetics some 20 years ago [9], and the areas in brown
are modern emerging areas impacted by electromagnetics.
4 Electromagnetic Field Theory

Figure 1.2: Knowledge grows like a tree. Engineering knowledge and real-world applica-
tions are driven by fundamental knowledge from math and sciences. At a university, we do
science-based engineering research that can impact wide-ranging real-world applications. But
everyone is equally important in transforming our society. Just like the parts of the human
body, no one can claim that one is more important than the others.

Figure 1.2 shows how knowledge are driven by basic math and science knowledge. Its
growth is like a tree. It is important that we collaborate to develop technologies that can
transform this world.

1.1.1 A Brief History of Electromagnetics

Electricity and magnetism have been known to mankind for a long time. Also, the physical
properties of light have been known. But electricity and magnetism, now termed electromag-
netics in the modern world, has been thought to be governed by different physical laws as
opposed to optics. This is understandable as the physics of electricity and magnetism is quite
different of the physics of optics as they were known to humans then.
For example, lode stone was known to the ancient Greek and Chinese around 600 BC to
400 BC. Compass was used in China since 200 BC. Static electricity was reported by the
Greek as early as 400 BC. But these curiosities did not make an impact until the age of
telegraphy. The coming about of telegraphy was due to the invention of the voltaic cell or
the galvanic cell in the late 1700’s, by Luigi Galvani and Alesandro Volta [10]. It was soon
discovered that two pieces of wire, connected to a voltaic cell, can transmit information.
Introduction, Maxwell’s Equations 5

So by the early 1800’s this possibility had spurred the development of telegraphy. Both
André-Marie Ampére (1823) [11, 12] and Michael Faraday (1838) [13] did experiments to
better understand the properties of electricity and magnetism. And hence, Ampere’s law and
Faraday law are named after them. Kirchhoff voltage and current laws were also developed
in 1845 to help better understand telegraphy [14, 15]. Despite these laws, the technology of
telegraphy was poorly understood. For instance, it was not known as to why the telegraphy
signal was distorted. Ideally, the signal should be a digital signal switching between one’s
and zero’s, but the digital signal lost its shape rapidly along a telegraphy line.4

It was not until 1865 that James Clerk Maxwell [17] put in the missing term in Am-
pere’s law, the displacement current term, only then the mathematical theory for electricity
and magnetism was complete. Ampere’s law is now known as generalized Ampere’s law.
The complete set of equations are now named Maxwell’s equations in honor of James Clerk
Maxwell.

The rousing success of Maxwell’s theory was that it predicted wave phenomena, as they
have been observed along telegraphy lines. But it was not until 23 years later that Heinrich
Hertz in 1888 [18] did experiment to prove that electromagnetic field can propagate through
space across a room. This illustrates the difficulty of knowledge dissemination when new
knowledge is discovered. Moreover, from experimental measurement of the permittivity and
permeability of matter, it was decided that electromagnetic wave moves at a tremendous
speed. But the velocity of light has been known for a long while from astronomical obser-
vations (Roemer, 1676) [19]. The observation of interference phenomena in light has been
known as well. When these pieces of information were pieced together, it was decided that
electricity and magnetism, and optics, are actually governed by the same physical law or
Maxwell’s equations. And optics and electromagnetics are unified into one field!

4 As a side note, in 1837, Morse invented the Morse code for telegraphy [16]. There were cross pollination

of ideas across the Atlantic ocean despite the distance. In fact, Benjamin Franklin associated lightning with
electricity in the latter part of the 18-th century. Also, notice that electrical machinery was invented in 1832
even though electromagnetic theory was not fully understood.
6 Electromagnetic Field Theory

Figure 1.3: A brief history of electromagnetics and optics as depicted in this figure.

In Figure 1.3, a brief history of electromagnetics and optics is depicted. In the begin-
ning, it was thought that electricity and magnetism, and optics were governed by different
physical laws. Low frequency electromagnetics was governed by the understanding of fields
and their interaction with media. Optical phenomena were governed by ray optics, reflection
and refraction of light. But the advent of Maxwell’s equations in 1865 revealed that they
can be unified under electromagnetic theory. Then solving Maxwell’s equations becomes a
mathematical endeavor.
The photo-electric effect [20,21], and Planck radiation law [22] point to the fact that elec-
tromagnetic energy is manifested in terms of packets of energy, indicating the corpuscular
nature of light. Each unit of this energy is now known as the photon. A photon carries an en-
ergy packet equal to ~ω, where ω is the angular frequency of the photon and ~ = 6.626×10−34
J s, the Planck constant, which is a very small constant. Hence, the higher the frequency,
the easier it is to detect this packet of energy, or feel the graininess of electromagnetic en-
ergy. Eventually, in 1927 [3], quantum theory was incorporated into electromagnetics, and the
quantum nature of light gives rise to the field of quantum optics. Recently, even microwave
photons have been measured [23,24]. They are difficult to detect because of the low frequency
of microwave (109 Hz) compared to optics (1015 Hz): a microwave photon carries a packet of
energy about a million times smaller than that of optical photon.
The progress in nano-fabrication [25] allows one to make optical components that are
subwavelength as the wavelength of blue light is about 450 nm. As a result, interaction of
light with nano-scale optical components requires the solution of Maxwell’s equations in its
full glory.
Introduction, Maxwell’s Equations 7

In the early days of quantum theory, there were two prevailing theories of quantum in-
terpretation. Quantum measurements were found to be random. In order to explain the
probabilistic nature of quantum measurements, Einstein posited that a random hidden vari-
able controlled the outcome of an experiment. On the other hand, the Copenhagen school
of interpretation led by Niels Bohr, asserted that the outcome of a quantum measurement is
not known until after a measurement [26].
In 1960s, Bell’s theorem (by John Steward Bell) [27] said that an inequality should be
satisfied if Einstein’s hidden variable theory was correct. Otherwise, the Copenhagen school
of interpretation should prevail. However, experimental measurement showed that the in-
equality was violated, favoring the Copenhagen school of quantum interpretation [26]. This
interpretation says that a quantum state is in a linear superposition of states before a mea-
surement. But after a measurement, a quantum state collapses to the state that is measured.
This implies that quantum information can be hidden incognito in a quantum state. Hence,
a quantum particle, such as a photon, its state is unknown until after its measurement. In
other words, quantum theory is “spooky”. This leads to growing interest in quantum infor-
mation and quantum communication using photons. Quantum technology with the use of
photons, an electromagnetic quantum particle, is a subject of growing interest. This also has
the profound and beautiful implication that “our karma is not written on our forehead when
we were born, our future is in our own hands!”

1.2 Maxwell’s Equations in Integral Form

Maxwell’s equations can be presented as fundamental postulates.5 We will present them in
their integral forms, but will not belabor them until later.
 d

E · dl = − B · dS Faraday’s Law (1.2.1)
 C
d
dt
 S

H · dl = D · dS + I Ampere’s Law (1.2.2)

C
 dt S

D · dS = Q Gauss’s or Coulomb’s Law (1.2.3)

 S

B · dS = 0 Gauss’s Law (1.2.4)

S
The units of the basic quantities above are given as:
E: V/m H: A/m
D: C/m2 B: W/m2
I: A Q: C

where V=volts, A=amperes, C=coulombs, and W=webers.

5 Postulates in physics are similar to axioms in mathematics. They are assumptions that need not be

proved.
8 Electromagnetic Field Theory

1.3 Static Electromagnetics

1.3.1 Coulomb’s Law (Statics)
This law, developed in 1785 [28], expresses the force between two charges q1 and q2 . If these
charges are positive, the force is repulsive and it is given by
q1 q 2
f1→2 = r̂12 (1.3.1)
4πεr2

Figure 1.4: The force between two charges q1 and q2 . The force is repulsive if the two charges
have the same sign.

where the units are: f (force): newton

q (charge): coulombs
ε (permittivity): farads/meter
r (distance between q1 and q2 ): m
r̂12 = unit vector pointing from charge 1 to charge 2

r2 − r1
r̂12 = , r = |r2 − r1 | (1.3.2)
|r2 − r1 |
Since the unit vector can be defined in the above, the force between two charges can also be
rewritten as

q1 q2 (r2 − r1 )
f1→2 = , (r1 , r2 are position vectors) (1.3.3)
4πε|r2 − r1 |3

1.3.2 Electric Field E (Statics)

The electric field E is defined as the force per unit charge [29]. For two charges, one of charge
q and the other one of incremental charge ∆q, the force between the two charges, according
to Coulomb’s law (1.3.1), is

q∆q
f= r̂ (1.3.4)
4πεr2
Introduction, Maxwell’s Equations 9

where r̂ is a unit vector pointing from charge q to the incremental charge ∆q. Then the
electric field E, which is the force per unit charge, is given by

f
E= , (V/m) (1.3.5)
4q
This electric field E from a point charge q at the orgin is hence
q
E= r̂ (1.3.6)
4πεr2
Therefore, in general, the electric field E(r) at location r from a point charge q at r0 is given
by

q(r − r0 )
E(r) = (1.3.7)
4πε|r − r0 |3
where the unit vector

r − r0
r̂ = (1.3.8)
|r − r0 |

Figure 1.5: Emanating E field from an electric point charge as depicted by depicted by (1.3.7)
and (1.3.6).

If one knows E due to a point charge, one will know E due to any charge distribution
because any charge distribution can be decomposed into sum of point charges. For instance, if
10 Electromagnetic Field Theory

there are N point charges each with amplitude qi , then by the principle of linear superposition,
the total field produced by these N charges is
N
X qi (r − ri )
E(r) = (1.3.9)
i=1
4πε|r − r i |3

where qi = %(ri )∆Vi is the incremental charge at ri enclosed in the volume ∆Vi . In the
continuum limit, one gets

%(r0 )(r − r0 )
E(r) = dV (1.3.10)
V 4πε|r − r0 |3

In other words, the total field, by the principle of linear superposition, is the integral sum-
mation of the contributions from the distributed charge density %(r).
Introduction, Maxwell’s Equations 11

1.3.3 Gauss’s Law (Statics)

This law is also known as Coulomb’s law as they are closely related to each other. Apparently,
this simple law was first expressed by Joseph Louis Lagrange [30] and later, reexpressed by
Gauss in 1813 (Wikipedia).
This law can be expressed as

D · dS = Q (1.3.11)
S

where D: electric flux density with unit C/m2 and D = εE.

dS: an incremental surface at the point on S given by dS n̂ where n̂ is the unit normal
pointing outward away from the surface.
Q: total charge enclosed by the surface S.

Figure 1.6: Electric flux (courtesy of Ramo, Whinnery, and Van Duzer [31]).

The left-hand side of (1.3.11) represents a surface integral over a closed surface S. To
understand it, one can break the surface into a sum of incremental surfaces ∆Si , with a
local unit normal n̂i associated with it. The surface integral can then be approximated by a
summation
 X X
D · dS ≈ Di · n̂i ∆Si = Di · ∆Si (1.3.12)
S i i

where one has defined ∆Si = n̂i ∆Si . In the limit when ∆Si becomes infinitesimally small,
the summation becomes a surface integral.
12 Electromagnetic Field Theory

1.3.4 Derivation of Gauss’s Law from Coulomb’s Law (Statics)

From Coulomb’s law and the ensuing electric field due to a point charge, the electric flux is
q
D = εE = r̂ (1.3.13)
4πr2
When a closed spherical surface S is drawn around the point charge q, by symmetry, the
electric flux though every point of the surface is the same. Moreover, the normal vector n̂
on the surface is just r̂. Consequently, D · n̂ = D · r̂ = q/(4πr2 ), which is a constant on a
spherical of radius r. Hence, we conclude that for a point charge q, and the pertinent electric


flux D that it produces on a spherical surface,

D · dS = 4πr2 D · n̂ = 4πr2 Dr = q (1.3.14)

S
Therefore, Gauss’s law is satisfied by a point charge.

Figure 1.7: Electric flux from a point charge satisfies Gauss’s law.

Even when the shape of the spherical surface S changes from a sphere to an arbitrary
shape surface S, it can be shown that the total flux through S is still q. In other words, the
total flux through sufaces S1 and S2 in Figure 1.8 are the same.
This can be appreciated by taking a sliver of the angular sector as shown in Figure 1.9.
Here, ∆S1 and ∆S2 are two incremental surfaces intercepted by this sliver of angular sector.
The amount of flux passing through this incremental surface is given by dS · D = n̂ · D∆S =
n̂ · r̂Dr ∆S. Here, D = r̂Dr is pointing in the r̂ direction. In ∆S1 , n̂ is pointing in the r̂
direction. But in ∆S2 , the incremental area has been enlarged by that n̂ not aligned with
D. But this enlargement is compensated by n̂ · r̂. Also, ∆S2 has grown bigger, but the flux
at ∆S2 has grown weaker by the ratio of (r2 /r1 )2 . Finally, the two fluxes are equal in the
limit that the sliver of angular sector becomes infinitesimally small. This proves the assertion
that the total fluxes through S1 and S2 are equal. Since the total flux from a point charge q
through a closed surface is independent of its shape, but always equal to q, then if we have a
total charge Q which can be expressed as the sum of point charges, namely.
X
Q= qi (1.3.15)
i
Introduction, Maxwell’s Equations 13

Figure 1.8: Same amount of electric flux from a point charge passes through two surfaces S1
and S2 .

Then the total flux through a closed surface equals the total charge enclosed by it, which is
the statement of Gauss’s law or Coulomb’s law.

1.4 Homework Examples

Example 1
Field of a ring of charge of density %l C/m.
Question: What is E along z axis? Hint: Use symmetry.

Example 2
Field between coaxial cylinders of unit length.
Question: What is E?
Hint: Use symmetry and cylindrical coordinates to express E = ρ̂Eρ and appply Gauss’s law.

Example 3
Fields of a sphere of uniform charge density.
Question: What is E?
Hint: Again, use symmetry and spherical coordinates to express E = r̂Er and appply Gauss’s
law.
14 Electromagnetic Field Theory

Figure 1.9: When a sliver of angular sector is taken, same amount of electric flux from a point
charge passes through two incremental surfaces ∆S1 and ∆S2 .

Figure 1.10: Electric field of a ring of charge (courtesy of Ramo, Whinnery, and Van Duzer
[31]).

Figure 1.11: Figure for Example 2 for a coaxial cylinder.

Introduction, Maxwell’s Equations 15

Figure 1.12: Figure for Example 3 for a sphere with uniform charge density.
16 Electromagnetic Field Theory
Lecture 2

Maxwell’s Equations,
Differential Operator Form

Maxwell’s equations were originally written in integral forms as has been shown in the previous
lecture. Integral forms have nice physical meaning and can be easily related to experimental
measurements. However, the differential operator forms1 can be easily converted to differential
equations or partial differential equations where a whole sleuth of mathematical methods and
numerical methods can be deployed. Therefore, it is prudent to derive the differential operator
form of Maxwell’s equations.

2.1 Gauss’s Divergence Theorem

The divergence theorem is one of the most important theorems in vector calculus [31–34].
First, we will need to prove Gauss’s divergence theorem, namely, that:
 
dV ∇ · D = D · dS (2.1.1)
V S

In the above, ∇ · D is defined as


D · dS
∆S
∇ · D = lim (2.1.2)
∆V →0 ∆V
The above implies that the divergence of the electric flux D, or ∇·D is given by first computing
the flux coming (or oozing) out of a small volume ∆V surrounded by a small surface ∆S and
taking their ratio as shown on the right-hand side. As shall be shown, the ratio has a limit
and eventually, we will find an expression for it. We know that if ∆V ≈ 0 or small, then the
1 We caution ourselves not to use the term “differential forms” which has a different meaning used in

differential geometry for another form of Maxwell’s equations.

17
18 Electromagnetic Field Theory

above,


∆V ∇ · D ≈ D · dS (2.1.3)
∆S

First, we assume that a volume V has been discretized2 into a sum of small cuboids, where
the i-th cuboid has a volume of ∆Vi as shown in Figure 2.1. Then

N
X
V ≈ ∆Vi (2.1.4)
i=1

Figure 2.1: The discretization of a volume V into a sum of small volumes ∆Vi each of which
is a small cuboid. Stair-casing error occurs near the boundary of the volume V but the error
diminishes as ∆Vi → 0.

2 Other terms used are “tesselated”, “meshed”, or “gridded”.

Maxwell’s Equations, Differential Operator Form 19

Figure 2.2: Fluxes from adjacent cuboids cancel each other leaving only the fluxes at the
boundary that remain uncancelled. Please imagine that there is a third dimension of the
cuboids in this picture where it comes out of the paper.

Then from (2.1.2),


∆Vi ∇ · Di ≈ Di · dSi (2.1.5)
∆Si

By summing the above over all the cuboids, or over i, one gets
X X 
∆Vi ∇ · Di ≈ Di · dSi ≈ D · dS (2.1.6)
i i ∆Si S

It is easily seen that the fluxes out of the inner surfaces of the cuboids cancel each other,
leaving only fluxes flowing out of the cuboids at the edge of the volume V as explained in
Figure 2.2. The right-hand side of the above equation (2.1.6) becomes a surface integral
over the surface S except for the stair-casing approximation (see Figure 2.1). However, this
approximation becomes increasingly good as ∆Vi → 0. Moreover, the left-hand side becomes
a volume integral, and we have
 
dV∇ · D = D · dS (2.1.7)
V S

The above is Gauss’s divergence theorem.

2.1.1 Some Details

Next, we will derive the details of the definition embodied in (2.1.2). To this end, we evaluate
the numerator of the right-hand side carefully, in accordance to Figure 2.3.
20 Electromagnetic Field Theory

Figure 2.3: Figure to illustrate the calculation of fluxes from a small cuboid where a corner
of the cuboid is located at (x0 , y0 , z0 ). There is a third z dimension of the cuboid not shown,
and coming out of the paper. Hence, this cuboid, unlike that shown in the figure, has six
faces.

Accounting for the fluxes going through all the six faces, assigning the appropriate signs
in accordance with the fluxes leaving and entering the cuboid, one arrives at

D · dS ≈ −Dx (x0 , y0 , z0 )∆y∆z + Dx (x0 + ∆x, y0 , z0 )∆y∆z
∆S
−Dy (x0 , y0 , z0 )∆x∆z + Dy (x0 , y0 + ∆y, z0 )∆x∆z
−Dz (x0 , y0 , z0 )∆x∆y + Dz (x0 , y0 , z0 + ∆z)∆x∆y (2.1.8)

Factoring out the volume of the cuboid ∆V = ∆x∆y∆z in the above, one gets

D · dS ≈ ∆V {[Dx (x0 + ∆x, . . .) − Dx (x0 , . . .)] /∆x
∆S
+ [Dy (. . . , y0 + ∆y, . . .) − Dy (. . . , y0 , . . .)] /∆y
+ [Dz (. . . , z0 + ∆z) − Dz (. . . , z0 )] /∆z} (2.1.9)

Or that

D · dS ∂Dx ∂Dy ∂Dz
≈ + + (2.1.10)
∆V ∂x ∂y ∂z
In the limit when ∆V → 0, then

D · dS ∂Dx ∂Dy ∂Dz
lim = + + =∇·D (2.1.11)
∆V →0 ∆V ∂x ∂y ∂z
Maxwell’s Equations, Differential Operator Form 21

where
∂ ∂ ∂
∇ = x̂ + ŷ + ẑ (2.1.12)
∂x ∂y ∂z
D = x̂Dx + ŷDy + ẑDz (2.1.13)

The above is the definition of the divergence operator in Cartesian coordinates. The diver-
gence operator ∇· has its complicated representations in cylindrical and spherical coordinates,
a subject that we would not delve into in this course. But they can be derived, and are best
looked up at the back of some textbooks on electromagnetics.
Consequently, one gets Gauss’s divergence theorem given by
 
dV ∇ · D = D · dS (2.1.14)
V S

2.1.2 Gauss’s Law in Differential Operator Form

By further using Gauss’s or Coulomb’s law implies that
 
D · dS = Q = dV % (2.1.15)
S

We can esplace the left-hand side of the above by (2.1.14) to arrive at

 
dV ∇ · D = dV % (2.1.16)
V V

When V → 0, we arrive at the pointwise relationship, a relationship at an arbitrary point in

space. Therefore,

∇·D=% (2.1.17)

2.1.3 Physical Meaning of Divergence Operator

The physical meaning of divergence is that if ∇·D 6= 0 at a point in space, it implies that there
are fluxes oozing or exuding from that point in space [35]. On the other hand, if ∇ · D = 0,
it implies no flux oozing out from that point in space. In other words, whatever flux that
goes into the point must come out of it. The flux is termed divergence free. Thus, ∇ · D is a
measure of how much sources or sinks exist for the flux at a point. The sum of these sources
or sinks gives the amount of flux leaving or entering the surface that surrounds the sources
or sinks.
Moreover, if one were to integrate a divergence-free flux over a volume V , and invoking
Gauss’s divergence theorem, one gets

D · dS = 0 (2.1.18)
S

In such a scenerio, whatever flux that enters the surface S must leave it. In other words, what
comes in must go out of the volume V , or that flux is conserved. This is true of incompressible
22 Electromagnetic Field Theory

fluid flow, electric flux flow in a source free region, as well as magnetic flux flow, where the
flux is conserved.

Figure 2.4: In an incompressible flux flow, flux is conserved: whatever flux that enters a
volume V must leave the volume V .

2.2 Stokes’s Theorem

The mathematical description of fluid flow was well established before the establishment of
electromagnetic theory [36]. Hence, much mathematical description of electromagnetic theory
uses the language of fluid. In mathematical notations, Stokes’s theorem is
 
E · dl = ∇ × E · dS (2.2.1)
C S

In the above, the contour C is a closed contour, whereas the surface S is not closed.3
First, applying Stokes’s theorem to a small surface ∆S, we define a curl operator4 ∇× at
a point to be measured as

E · dl
∆C
(∇ × E) · n̂ = lim (2.2.2)
∆S→0 ∆S

In the above, ∇ × E is a vector. Taking ∆C E · dl as a measure of the rotation of the field
E around a small loop ∆C, the ratio of this rotation to the area of the loop ∆S has a limit
when ∆S becomes infinitesimally small. This ratio is related to ∇ × E.
3 In other words, C has no boundary whereas S has boundary. A closed surface S has no boundary like

when we were proving Gauss’s divergence theorem previously.

4 Sometimes called a rotation operator.
Maxwell’s Equations, Differential Operator Form 23

Figure 2.5: In proving Stokes’s theorem, a closed contour C is assumed to enclose an open
surface S. Then the surface S is tessellated into sum of small rects as shown. Stair-casing
error vanishes in the limit when the rects are made vanishingly small.

First, the surface S enclosed by C is tessellated (also called meshed, gridded, or discretized)
into sum of small rects (rectangles) as shown in Figure 2.5. Stokes’s theorem is then applied
to one of these small rects to arrive at


Ei · dli = (∇ × Ei ) · ∆Si (2.2.3)
∆Ci

where one defines ∆Si = n̂∆S. Next, we sum the above equation over i or over all the small
rects to arrive at

X X
Ei · dli = ∇ × Ei · ∆Si (2.2.4)
i ∆Ci i

Again, on the left-hand side of the above, all the contour integrals over the small rects cancel
each other internal to S save for those on the boundary. In the limit when ∆Si → 0, the
left-hand side becomes a contour integral over the larger contour C, and the right-hand side
becomes a surface integral over S. One arrives at Stokes’s theorem, which is

 
E · dl = (∇ × E) · dS (2.2.5)
C S
24 Electromagnetic Field Theory

Figure 2.6: We approximate the integration over a small rect using this figure. There are four
edges to this small rect.

Next, we need to prove the details of definition (2.2.2) using Figure 2.6. Performing the
integral over the small rect, one gets

E · dl = Ex (x0 , y0 , z0 )∆x + Ey (x0 + ∆x, y0 , z0 )∆y
∆C
− Ex (x0 , y0 + ∆y, z0 )∆x − Ey (x0 , y0 , z0 )∆y

Ex (x0 , y0 , z0 ) Ex (x0 , y0 + ∆y, z0 )
= ∆x∆y −
∆y ∆y

Ey (x0 , y0 , z0 ) Ey (x0 , y0 + ∆y, z0 )
− +
∆x ∆x
(2.2.6)

We have picked the normal to the incremental surface ∆S to be ẑ in the above example,
and hence, the above gives rise to the identity that

∆S
E · dl ∂ ∂
lim = Ey − Ex = ẑ · ∇ × E (2.2.7)
∆S→0 ∆S ∂x ∂y
Picking different ∆S with different orientations and normals n̂ where |hatn = x̂ or n̂ = ŷ, one
gets
∂ ∂
Ez − Ey = x̂ · ∇ × E (2.2.8)
∂y ∂z
∂ ∂
Ex − Ez = ŷ · ∇ × E (2.2.9)
∂z ∂x
Maxwell’s Equations, Differential Operator Form 25

The above gives the x, y, and z components of ∇ × E. It is to be noted that ∇ × E is a

vector. In other words, one gets
   
∂ ∂ ∂ ∂
∇ × E = x̂ Ez − Ey + ŷ Ex − Ez
∂y ∂z ∂z ∂x
 
∂ ∂
+ẑ Ey − Ex (2.2.10)
∂x ∂y

where
∂ ∂ ∂
∇ = x̂ + ŷ + ẑ (2.2.11)
∂x ∂y ∂z

2.2.1 Faraday’s Law in Differential Operator Form

Faraday’s law is experimentally motivated. Michael Faraday (1791-1867) was an extraordi-
nary experimentalist who documented this law with meticulous care. It was only decades
later that a mathematical description of this law was arrived at.
Faraday’s law in integral form is given by5
 
d
E · dl = − B · dS (2.2.12)
C dt S

Assuming that the surface S is not time varying, one can take the time derivative into the
integrand and write the above as
 
∂
E · dl = − B · dS (2.2.13)
C S ∂t

One can replace the left-hand side with the use of Stokes’ theorem to arrive at
 
∂
∇ × E · dS = − B · dS (2.2.14)
S S ∂t

The normal of the surface element dS can be pointing in an arbitrary direction, and the
surface S can be made very small. Then the integral can be removed, and one has

∂
∇×E=− B (2.2.15)
∂t
The above is Faraday’s law in differential operator form.
In the static limit, ∂B
∂t = 0, giving

∇×E=0 (2.2.16)
5 Faraday’s law is experimentally motivated. Michael Faraday (1791-1867) was an extraordinary exper-

imentalist who documented this law with meticulous care. It was only decades later that a mathematical
description of this law was arrived at.
26 Electromagnetic Field Theory

2.2.2 Physical Meaning of Curl Operator

The curl operator ∇× is a measure of the rotation or the circulation of a field at a point in
space. On the other hand, ∆C E · dl is a measure of the circulation of the field E around the
loop formed by C. Again, the curl operator has its complicated representations in other co-
ordinate systems like cylindrical or spherical coordinates, a subject that will not be discussed
in detail here.
It is to be noted that our proof of the Stokes’s theorem is for a flat open surface S, and not
for a general curved open surface. Since all curved surfaces can be tessellated into a union of
flat triangular surfaces according to the tiling theorem, the generalization of the above proof
to curved surface is straightforward. An example of such a triangulation of a curved surface
into a union of flat triangular surfaces is shown in Figure 2.7.

Figure 2.7: An arbitrary curved surface can be triangulated with flat triangular patches. The
triangulation can be made arbitrarily accurate by making the patches arbitrarily small.

2.3 Maxwell’s Equations in Differential Operator Form

With the use of Gauss’ divergence theorem and Stokes’ theorem, Maxwell’s equations can be
written more elegantly in differential operator forms. They are:

∂B
∇×E=− (2.3.1)
∂t
∂D
∇×H= +J (2.3.2)
∂t
∇·D=% (2.3.3)
∇·B=0 (2.3.4)

These equations are point-wise relations as they relate the left-hand side and right-hand side
field values at a given point in space. Moreover, they are not independent of each other. For
instance, one can take the divergence of the first equation (2.3.1), making use of the vector
Maxwell’s Equations, Differential Operator Form 27

identity that ∇ · (∇ × E) = 0, one gets

∂∇ · B
− = 0 → ∇ · B = constant (2.3.5)
∂t
This constant corresponds to magnetic charges, and since they have not been experimentally
observed, one can set the constant to zero. Thus the fourth of Maxwell’s equations, (2.3.4),
follows from the first (2.3.1).
Similarly, by taking the divergence of the second equation (2.3.2), and making use of the
current continuity equation that

∂%
∇·J+ =0 (2.3.6)
∂t
one can obtain the second last equation (2.3.3). Notice that in (2.3.3), the charge density %
can be time-varying, whereas in the previous lecture, we have “derived” this equation from
Coulomb’s law using electrostatic theory.
The above logic follows if ∂/∂t 6= 0, and is not valid for static case. In other words, for
statics, the third and the fourth equations are not derivable from the first two. Hence all
four Maxwell’s equations are needed for static problems. For electrodynamic problems, only
solving the first two suffices.
Something is amiss in the above. If J is known, then solving the first two equations implies
solving for four vector unknowns, E, H, B, D, which has 12 scalar unknowns. But there are
only two vector equations or 6 scalar equations in the first two equations. Thus one needs
more equations. These are provided by the constitutive relations that we shall discuss next.

2.4 Homework Examples

Example 1
If D = (2y 2 + z)x̂ + 4xy ŷ + xẑ, find:

1. Volume charge density ρv at (−1, 0, 3).

2. Electric flux through the cube defined by

0 ≤ x ≤ 1, 0 ≤ y ≤ 1, 0 ≤ z ≤ 1.

3. Total charge enclosed by the cube.

Example 2

Suppose E = x̂3y + ŷx, calculate E · dl along a straight line in the x-y plane joining (0,0)
to (3,1).
28 Electromagnetic Field Theory

2.5 Historical Notes

There are several interesting historical notes about Maxwell.

ˆ It is to be noted that when James Clerk Maxwell first wrote his equations down, it was
in many equations and very difficult to digest [17, 37, 38]. It was Oliver Heaviside who
distilled those equations and presented them in the present form found in textbooks.
Putatively, most cannot read his treatise [37] beyond the first 50 pages [39].
ˆ Maxwell wrote many poems in his short lifespan (1831-1879) and they can be found
at [40].

ˆ Also, the ancestor of James Clerk Maxwell married from the Clerk family into the
Maxwell family. One of the conditions of marriage was that all the descendants of the
Clerk family would be named Clerk Maxwell. That was why Maxwell is addressed as
Professor Clerk Maxwell.
Lecture 3

Constitutive Relations, Wave

Equation, Electrostatics, and
Static Green’s Function

Constitutive relations are important for defining the electromagnetic material properties of
the media involved. Also, wave phenomenon is a major triumph of Maxwell’s equations.
Hence, we will study that derivation of this phenomenon here. To make matter simple, we
will reduce the problem to electrostatics to simplify the math to introduce the concept of the
Green’s function.
As mentioned previously, for time-varying problems, only the first two of the four Maxwell’s
equations suffice. The latter two are derivable from the first two. But the first two equations
have four unknowns E, H, D, and B. Hence, two more equations are needed to solve for
these unknowns. These equations come from the constitutive relations.

3.1 Simple Constitutive Relations

The constitution relation between electric flux D and the electric field E in free space (or
vacuum) is

D = ε0 E (3.1.1)

When material medium is present, one has to add the contribution to D by the polarization
density P which is a dipole density.1 Then [31, 32, 41]

D = ε0 E + P (3.1.2)
1 Note that a dipole moment is given by Q` where Q is its charge in coulomb and ` is its length in m. Hence,

dipole density, or polarization density as dimension of coulomb/m2 , which is the same as that of electric flux
D.

29
30 Electromagnetic Field Theory

The second term P above is due to material property, and the contribution to the electric
flux due to the polarization density of the medium. It is due to the little dipole contribution
due to the polar nature of the atoms or molecules that make up a medium.
By the same token, the first term ε0 E can be thought of as the polarization density
contribution of vacuum. Vacuum, though represents nothingness, has electrons and positrons,
or electron-positron pairs lurking in it [42]. Electron is matter, whereas positron is anti-
matter. In the quiescent state, they represent nothingness, but they can be polarized by an
electric field E. That also explains why electromagnetic wave can propagate through vacuum.
For many media, they are approximately linear media. Then P is linearly proportional to
E, or P = ε0 χ0 E, or

D = ε0 E + ε0 χ0 E
= ε0 (1 + χ0 )E = εE, ε = ε0 (1 + χ0 ) = ε0 εr (3.1.3)

where χ0 is the electric susceptibility . In other words, for linear material media, one can
replace the vacuum permittivity ε0 with an effective permittivity ε = ε0 εr where εr is the
relative permittivity. Thus, D is linearly proportional to E. In free space,2

10−8
ε = ε0 = 8.854 × 10−12 ≈ F/m (3.1.4)
36π
The constitutive relation between magnetic flux B and magnetic field H is given as

B = µH, µ = permeability H/m (3.1.5)

In free space or vacuum,

µ = µ0 = 4π × 10−7 H/m (3.1.6)

As shall be explained later, this is an assigned value giving it a precise value as shown above.
In other materials, the permeability can be written as

µ = µ0 µr (3.1.7)

The above can be derived using similar argument for permittivity, where the different perme-
ability is due to the presence of magnetic dipole density in a material medium. In the above,
µr is termed the relative permeability.

3.2 Emergence of Wave Phenomenon,

Triumph of Maxwell’s Equations
One of the major triumphs of Maxwell’s equations is the prediction of the wave phenomenon.
This was experimentally verified by Heinrich Hertz in 1888 [18], some 23 years after the
2 It is to be noted that we will use MKS unit in this course. Another possible unit is the CGS unit used in

many physics texts [43]

Constitutive Relations, Wave Equation, Electrostatics, and Static Green’s Function 31

completion of Maxwell’s theory in 1865 [17]. To see this, we consider the first two Maxwell’s
equations for time-varying fields in vacuum or a source-free medium.3 They are
∂H
∇ × E = −µ0 (3.2.1)
∂t
∂E
∇ × H = −ε0 (3.2.2)
∂t
Taking the curl of (3.2.1), we have

∂
∇ × ∇ × E = −µ0 ∇×H (3.2.3)
∂t
It is understood that in the above, the double curl operator implies ∇ × (∇ × E). Substituting
(3.2.2) into (3.2.3), we have

∂2
∇ × ∇ × E = −µ0 ε0 E (3.2.4)
∂t2
In the above, the left-hand side can be simplified by using the identity that a × (b × c) =
b(a · c) − c(a · b),4 but be mindful that the operator ∇ has to operate on a function to its
right. Therefore, we arrive at the identity that

∇ × ∇ × E = ∇∇ · E − ∇2 E (3.2.5)

Since ∇ · E = 0 in a source-free medium, we have

∂2
∇2 E − µ0 ε0 E=0 (3.2.6)
∂t2
where
∂2 ∂2 ∂2
∇2 = ∇ · ∇ = + +
∂x2 ∂y 2 ∂z 2
The above is known as the Laplacian operator. Here, (3.2.6) is the wave equation in three
space dimensions [32, 44].
To see the simplest form of wave emerging in the above, we can let E = x̂Ex (z, t) so that
∇ · E = 0 satisfying the source-free condition. Then (3.2.6) becomes

∂2 ∂2
Ex (z, t) − µ0 ε 0 Ex (z, t) = 0 (3.2.7)
∂z 2 ∂t2
Eq. (3.2.7) is known mathematically as the wave equation in one space dimension. It can
also be written as
∂2 1 ∂2
2
f (z, t) − 2 2 f (z, t) = 0 (3.2.8)
∂z c0 ∂t
3 Since the third and the fourth Maxwell’s equations are derivable from the first two when ∂/∂t 6= 0.
4 For mnemonics, this formula is also known as the “back-of-the-cab” formula.
32 Electromagnetic Field Theory

where c20 = (µ0 ε0 )−1 . Eq. (3.2.8) can also be factorized as

  
∂ 1 ∂ ∂ 1 ∂
− + f (z, t) = 0 (3.2.9)
∂z c0 ∂t ∂z c0 ∂t

or
  
∂ 1 ∂ ∂ 1 ∂
+ − f (z, t) = 0 (3.2.10)
∂z c0 ∂t ∂z c0 ∂t

The above can be verified easily by direct expansion, and using the fact that

∂ ∂ ∂ ∂
= (3.2.11)
∂t ∂z ∂z ∂t

The above implies that we have either

 
∂ 1 ∂
+ f+ (z, t) = 0 (3.2.12)
∂z c0 ∂t

or
 
∂ 1 ∂
− f− (z, t) = 0 (3.2.13)
∂z c0 ∂t

Equation (3.2.12) and (3.2.13) are known as the one-way wave equations or the advective
equations [45]. From the above factorization, it is seen that the solutions of these one-way
wave equations are also the solutions of the original wave equation given by (3.2.8). Their
general solutions are then

f+ (z, t) = F+ (z − c0 t) (3.2.14)
f− (z, t) = F− (z + c0 t) (3.2.15)

We can verify the above by back substitution into (3.2.12) and (3.2.13). Eq. (3.2.14) consti-
tutes a right-traveling wave function of any shape while (3.2.15) constitutes a left-traveling
wave function of any shape. Since Eqs. (3.2.14) and (3.2.15) are also solutions to (3.2.8), we
can write the general solution to the wave equation as

f (z, t) = F+ (z − c0 t) + F− (z + c0 t) (3.2.16)

This is a wonderful result since F+ and F− are arbitrary functions of any shape (see Figure
3.1); they can be used to encode information for communication!
Constitutive Relations, Wave Equation, Electrostatics, and Static Green’s Function 33

Figure 3.1: Solutions of the wave equation can be a single-valued function of any shape. In
the above, F+ travels in the positive z direction, while F− travels in the negative z direction
as t increases.

Furthermore, one can calculate the velocity of this wave to be

c0 = 299, 792, 458m/s ' 3 × 108 m/s (3.2.17)

p
where c0 = 1/µ0 ε0 .
Maxwell’s equations (3.2.1) implies that E and H are linearly proportional to each other.
Thus, there is only one independent constant in the wave equation, and the value of µ0 is
defined neatly to be 4π × 10−7 henry m−1 , while the value of ε0 has been measured to be
about 8.854 × 10−12 farad m−1 . Now it has been decided that the velocity of light is used
as a standard and is defined to be the integer given in (3.2.17). A meter is defined to be
the distance traveled by light in 1/(299792458) seconds. Hence, the more accurate that unit
of time or second can be calibrated, the more accurate can we calibrate the unit of length
or meter. Therefore, the design of an accurate clock like an atomic clock is an important
research problem.
The value of ε0 was measured in the laboratory quite early. Then it was realized that
electromagnetic wave propagates at a tremendous velocity which is the velocity of light.5 This
5 The velocity of light was known in astronomy by (Roemer, 1676) [19].
34 Electromagnetic Field Theory

was also the defining moment which revealed that the field of electricity and magnetism and
the field of optics were both described by Maxwell’s equations or electromagnetic theory.

3.3 Static Electromagnetics–Revisted

We have seen static electromagnetics previously in integral form. Now we look at them in
differential operator form. When the fields and sources are not time varying, namely that
∂/∂t = 0, we arrive at the static Maxwell’s equations for electrostatics and magnetostatics,
namely [31, 32, 46]

∇×E=0 (3.3.1)
∇×H=J (3.3.2)
∇·D=% (3.3.3)
∇·B=0 (3.3.4)

Notice the the electrostatic field system is decoupled from the magnetostatic field system.
However, in a resistive system where

J = σE (3.3.5)

the two systems are coupled again. This is known as resistive coupling between them. But if
σ → ∞, in the case of a perfect conductor, or superconductor, then for a finite J, E has to
be zero. The two systems are decoupled again.
Also, one can arrive at the equations above by letting µ0 → 0 and 0 → 0. In this case,
the velocity of light becomes infinite, or retardation effect is negligible. In other words, there
is no time delay for signal propagation through the system in the static approximation.
Finally, it is important to note that in statics, the latter two Maxwell’s equations are not
derivable from the first two. Hence, all four equations have to be considered when one seeks
the solution in the static regime or the long-wavelength regime.

3.3.1 Electrostatics
We see that Faraday’s law in the static limit is

∇×E=0 (3.3.6)

One way to satisfy the above is to let E = −∇Φ because of the identity ∇ × ∇ = 0.6
Alternatively, one can assume that E is a constant. But we usually are interested in solutions
that vanish at infinity, and hence, the latter is not a viable solution. Therefore, we let

E = −∇Φ (3.3.7)
6 One an easily go through the algebra in cartesian coordinates to convince oneself of this.
Constitutive Relations, Wave Equation, Electrostatics, and Static Green’s Function 35

3.3.2 Poisson’s Equation

As a consequence of the above,

∇ · D = % ⇒ ∇ · εE = % ⇒ −∇ · ε∇Φ = % (3.3.8)

In the last equation above, if ε is a constant of space, or independent of r, then one arrives
at the simple Poisson’s equation, which is a partial differential equation
%
∇2 Φ = − (3.3.9)
ε
Here, the Laplacian operator

∂2 ∂2 ∂2
∇2 = ∇ · ∇ = + +
∂x2 ∂y 2 ∂z 2

For a point source, we know from Coulomb’s law that

q
E= r̂ = −∇Φ (3.3.10)
4πεr2
From the above, we deduce that7
q
Φ= (3.3.11)
4πεr
Therefore, we know the solution to Poisson’s equation (3.3.9) when the source % represents a
point source. Since this is a linear equation, we can use the principle of linear superposition
to find the solution when the charge density %(r) is arbitrary.
A point source located at r0 is described by a charge density as

%(r) = qδ(r − r0 ) (3.3.12)

where δ(r − r0 ) is a short-hand notation for δ(x − x0 )δ(y − y 0 )δ(z − z 0 ). Therefore, from (3.3.9),
the corresponding partial differential equation for a point source is

qδ(r − r0 )
∇2 Φ(r) = − (3.3.13)
ε
The solution to the above equation, from Coulomb’s law in accordance to (3.3.11), has to be
q
Φ(r) = (3.3.14)
4πε|r − r0 |

whereas (3.3.11) is for a point source at the origin, (3.3.14) is for a point source located
and translated to r0 . The above is a coordinate independent p form of the solution. Here,
r = x̂x + ŷy + ẑz and r0 = x̂x0 + ŷy 0 + ẑz 0 , and |r − r0 | = (x − x0 )2 + (y − y 0 )2 + (z − z 0 )2 .
7 One can always take the gradient or ∇ of Φ to verify this. Mind you, this is best done in spherical

coordinates.
36 Electromagnetic Field Theory

3.3.3 Static Green’s Function

Let us define a partial differential equation given by
∇2 g(r − r0 ) = −δ(r − r0 ) (3.3.15)
The above is similar to Poisson’s equation with a point source on the right-hand side as in
(3.3.13). Such a solution, a response to a point source, is called the Green’s function.8 By
comparing equations (3.3.13) and (3.3.15), then making use of (3.3.14), we deduced that the
static Green’s function is
1
g(r − r0 ) = (3.3.16)
4π|r − r0 |
An arbitrary source can be expressed as

%(r) = dV 0 %(r0 )δ(r − r0 ) (3.3.17)
V

The above is just the statement that an arbitrary charge distribution %(r) can be expressed
as a linear superposition of point sources δ(r − r0 ). Using the above in (3.3.9), we have

1
∇2 Φ(r) = − dV 0 %(r0 )δ(r − r0 ) (3.3.18)
ε V

We can let 
1
Φ(r) = dV 0 g(r − r0 )%(r0 ) (3.3.19)
ε V
By substituting the above into the left-hand side of (3.3.18), exchanging order of integration
and differentiation, and then making use of equation (3.3.15), it can be shown that (3.3.19)
indeed satisfies (3.3.9). The above is just a convolutional integral. Hence, the potential Φ(r)
due to an arbitrary source distribution %(r) can be found by using convolution, namely,

1 %(r0 )
Φ(r) = dV 0 (3.3.20)
4πε V |r − r0 |
In a nutshell, the solution of Poisson’s equation when it is driven by an arbitrary source %, is
the convolution of the source with the static Green’s function, a point source response.

3.3.4 Laplace’s Equation

If % = 0, or if we are in a source-free region, then
∇2 Φ =0 (3.3.21)
which is the Laplace’s equation. Laplace’s equation is usually solved as a boundary value
problem. In such a problem, the potential Φ is stipulated on the boundary of a region with a
certain boundary condition, and then the solution is sought in the region so as to match the
boundary condition.
Examples of such boundary value problems are given at the end of the lecture.
8 George Green (1793-1841), the son of a Nottingham miller, was self-taught, but his work has a profound

impact in our world.

Constitutive Relations, Wave Equation, Electrostatics, and Static Green’s Function 37

3.4 Homework Examples

Example 1

Fields of a sphere of radius a with uniform charge density ρ:

Figure 3.2: Figure of a sphere with uniform charge density for the example above.

Assuming that Φ|r=∞ = 0, what is Φ at r ≤ a? And Φ at r > a.

Example 2

A capacitor has two parallel plates attached to a battery, what is E field inside the capacitor?
First, one guess the electric field between the two parallel plates. Then one arrive at a
potential Φ in between the plates so as to produce the field. Then the potential is found so
as to match the boundary conditions of Φ = V in the upper plate, and Φ = 0 in the lower
plate. What is the Φ that will satisfy the requisite boundary condition?

Figure 3.3: Figure of a parallel plate capacitor. The field in between can be found by solving
Laplace’s equation as a boundary value problem [31].

Example 3

A coaxial cable has two conductors. The outer conductor is grounded and hence is at zero
potential. The inner conductor is at voltage V . What is the solution?
38 Electromagnetic Field Theory

For this, one will have to write the Laplace’s equation in cylindrical coordinates, namely,

1 ∂2Φ
 
2 1 ∂ ∂Φ
∇ Φ= ρ + 2 =0 (3.4.1)
ρ ∂ρ ∂ρ ρ ∂φ2

In the above, we assume that the potential is constant in the z direction, and hence, ∂/∂z = 0,
and ρ, φ, z are the cylindrical coordinates. By assuming axi-symmetry, we can let ∂/∂φ = 0
and the above becomes
 
1 ∂ ∂Φ
∇2 Φ = ρ =0 (3.4.2)
ρ ∂ρ ∂ρ

Show that Φ = A ln ρ + B is a general solution to Laplace’s equation in cylindrical coor-

dinates inside a coax. What is the Φ that will satisfy the requisite boundary condition?

Figure 3.4: The field in between a coaxial line can also be obtained by solving Laplace’s
equation as a boundary value problem (courtesy of Ramo, Whinnery, and Van Duzer [31]).
Lecture 4

Magnetostatics, Boundary
Conditions, and Jump
Conditions

In the previous lecture, Maxwell’s equations become greatly simplified in the static limit. We
have looked at how the electrostatic problems are solved. We now look at the magnetostatic
case. In addition, we will study boundary conditions and jump conditions at an interface,
and how they are derived from Maxwell’s equations. Maxwell’s equations can be first solved
in different domains. Then the solutions are pieced (or sewn) together by imposing boundary
conditions at the boundaries or interfaces of the domains. Such problems are called boundary-
value problems (BVPs).

4.1 Magnetostatics
From Maxwell’s equations, we can deduce that the magnetostatic equations for the magnetic
field and flux when ∂/∂t = 0, which are [31, 32, 46]

∇×H=J (4.1.1)
∇·B=0 (4.1.2)

These two equations are greatly simplified, and hence, are easier to solve compared to the
time-varying case. One way to satisfy the second equation is to let

B=∇×A (4.1.3)

because of the vector identity

∇ · (∇ × A) = 0 (4.1.4)

39
40 Electromagnetic Field Theory

The above is zero for the same reason that a · (a × b) = 0. In this manner, Gauss’s law in
(4.1.2) is automatically satisfied.
From (4.1.1), we have
 
B
∇× =J (4.1.5)
µ

Then using (4.1.3) into the above,

 
1
∇× ∇×A =J (4.1.6)
µ

In a homogeneous medium,1 µ or 1/µ is a constant and it commutes with the differential ∇

operator or that it can be taken outside the differential operator. As such, one arrives at

∇ × (∇ × A) = µJ (4.1.7)

We use the vector identity that (see back-of-cab formula in the previous lecture)

∇ × (∇ × A) = ∇(∇ · A) − (∇ · ∇)A
= ∇(∇ · A) − ∇2 A (4.1.8)

where ∇2 is a shorthand notation for ∇ · ∇. As a result, we arrive at [47]

∇(∇ · A) − ∇2 A = µJ (4.1.9)

By imposing the Coulomb gauge that ∇ · A = 0, which will be elaborated in the next section,
we arrive at the simplified equation

∇2 A = −µJ (4.1.10)

The above is also known as the vector Poisson’s equation. In cartesian coordinates, the above
can be viewed as three scalar Poisson’s equations. Each of the Poisson’s equation can be
solved using the Green’s function method previously described. Consequently, in free space

µ J(r0 ) 0
A(r) = dV (4.1.11)
4π V R

where

R = |r − r0 | (4.1.12)

is the distance between the source point r0 and the observation point r. Here, dV 0 = dx0 dy 0 dz 0 .
It is also variously written as dr0 or d3 r0 .
1 Its prudent to warn the reader of the use of the word “homogeneous”. In the math community, it usually

refers to something to be set to zero. But in the electromagnetics community, it refers to something “non-
heterogeneous”.
Magnetostatics, Boundary Conditions, and Jump Conditions 41

4.1.1 More on Coulomb Gauge

Gauge is a very important concept in physics [48], and we will further elaborate it here. First,
notice that A in (4.1.3) is not unique because one can always define

A0 = A − ∇Ψ (4.1.13)

Then

∇ × A0 = ∇ × (A − ∇Ψ) = ∇ × A = B (4.1.14)

where we have made use of that ∇ × ∇Ψ = 0. Hence, the ∇× of both A and A0 produce the
same B; hence, A is non-unique.
To find A uniquely, we have to define or set the divergence of A or provide a gauge
condition. One way is to set the divergence of A to be zero, namely that

∇·A=0 (4.1.15)

Then

∇ · A0 = ∇ · A − ∇2 Ψ 6= ∇ · A (4.1.16)

The last non-equal sign follows if ∇2 Ψ 6= 0. However, if we further stipulate that ∇ · A0 =

∇ · A = 0, then −∇2 Ψ = 0. This does not necessary imply that Ψ = 0, but if we impose that
condition that Ψ → 0 when r → ∞, then Ψ = 0 everywhere.2 By so doing, A and A0 are
equal to each other, and we obtain (4.1.10) and (4.1.11).
The above is akin to the idea that given a vector a, just by stipulating that b × a = c is
not enough to determine a. We need to stipulate what b · a is as well. Here, a, b, and c are
independent vectors.

4.2 Boundary Conditions–1D Poisson’s Equation

To simplify the solutions of Maxwell’s equations, they are usually solved in a homogeneous
medium. As mentioned before, a complex problem can be divided into piecewise homogeneous
regions first, and then the solution in each region sought. Then the total solution must satisfy
boundary conditions at the interface between the piecewise homogeneous regions.
What are these boundary conditions? Boundary conditions are actually embedded in the
partial differential equations that the potential or the field satisfy. Two important concepts
to keep in mind are:

ˆ Differentiation of a function with discontinuous slope will give rise to step discontinuity.
ˆ Differentiation of a function with step discontinuity will give rise to a Dirac delta func-
tion. This is also called the jump condition, a term often used by the mathematics
community [49].
2 It is a property of the Laplace boundary value problem that if Ψ = 0 on a closed surface S, then Ψ = 0

everywhere inside S. Earnshaw’s theorem [31] is useful for proving this assertion.
42 Electromagnetic Field Theory

Take for example a one dimensional Poisson’s equation that

d d
ε(x) Φ(x) = −%(x) (4.2.1)
dx dx
where ε(x) represents material property that has the form given in Figure 4.1. One can
actually say something about Φ(x) given %(x) on the right-hand side. If %(x) has a delta
d
function singularity, it implies that ε(x) dx Φ(x) has a step discontinuity. If %(x) is finite
d
everywhere, then ε(x) dx Φ(x) must be continuous everywhere.
d
Furthermore, if ε(x) dx Φ(x) is finite everywhere, it implies that Φ(x) must be continuous
everywhere.

Figure 4.1: A figure showing a charge sheet at the interface between two dielectric media.
Because it is a surface charge sheet, the volume charge density %(x) is infinite at the sheet
location x0 .

To see this in greater detail, we illustrate it with the following example. In the above,
%(x) represents a charge distribution given by %(x) = %s δ(x − x0 ). In this case, the charge
distribution is everywhere zero except at the location of the surface charge sheet, where the
charge density is infinite: it is represented mathematically by a delta function3 in space.
To find the boundary condition of the potential Φ(x) at x0 , we integrate (4.2.1) over an
infinitesimal width around x0 , the location of the charge sheet, namely
 x0 +∆    x0 +∆  x0 +∆
d d
dx ε(x) Φ(x) = − dx%(x) = − dx%0 δ(x − x0 ) (4.2.2)
x0 −∆ dx dx x0 −∆ x0 −∆

Since the integrand of the left-hand side is an exact derivative, we get

x0 +∆
d 
ε(x) Φ(x) = −%s (4.2.3)
dx x0 −∆
3 This function has been attributed to Dirac who used in pervasively, but Cauchy was aware of such a

function.
Magnetostatics, Boundary Conditions, and Jump Conditions 43

whereas on the right-hand side, we pick up the contribution from the delta function. Evalu-
ating the left-hand side at their limits, one arrives at

d − d ∼
ε(x+
0) Φ(x+
0 ) − ε(x0 ) Φ(x−
0 ) = −%s , (4.2.4)
dx dx

where x± d
0 = lim∆→0 x0 ± ∆. In other words, the jump discontinuity is in ε(x) dx Φ(x) and the
amplitude of the jump discontinuity is proportional to the amplitude of the delta function.
Since E = −∇Φ, or that

d
Ex (x) = − Φ(x), (4.2.5)
dx

The above implies that

− −
ε(x+ +
0 )Ex (x0 ) − ε(x0 )Ex (x0 ) = %s (4.2.6)

or

−
Dx (x+
0 ) − Dx (x0 ) = %s (4.2.7)

where

Dx (x) = ε(x)Ex (x) (4.2.8)

The lesson learned from above is that boundary condition is obtained by integrating the
pertinent differential equation over an infinitesimal small segment. In this mathematical
way of looking at the boundary condition, one can also eyeball the differential equation and
ascertain the terms that will have the jump discontinuity whose derivatives will yield the
delta function on the right-hand side.

4.3 Boundary Conditions–Maxwell’s Equations

As seen previously, boundary conditions for a field is embedded in the differential equation
that the field satisfies. Hence, boundary conditions can be derived from the differential
operator forms of Maxwell’s equations. In most textbooks, boundary conditions are obtained
by integrating Maxwell’s equations over a small pill box [31,32,47]. To derive these boundary
conditions, we will take an unconventional view: namely to see what sources can induce jump
conditions on the pertinent fields. Boundary conditions are needed at media interfaces, as
well as across current or charge sheets. As shall be shown, each of the Maxwell’s equations
induces a boundary condition at the interface between two media or two regions separated
by surface sources.
44 Electromagnetic Field Theory

4.3.1 Faraday’s Law

Figure 4.2: This figure is for the derivation of boundary condition induced by Faraday’s law.
A local coordinate system can be used to see the boundary condition more lucidly. Here, the
normal n̂ = ŷ and the tangential component t̂ = x̂.

For this, we start with Faraday’s law, which implies that

∂B
∇×E=− (4.3.1)
∂t
The right-hand side of this equation is a derivative of a time-varying magnetic flux, and it is
a finite quantity. One quick answer we could ask is that if the right-hand side of the above
equation is everywhere finite, could there be any jump discontinuity on the field E on the
left hand side? The answer is no. To see this quickly, one can project the tangential field
component and normal field component to a local coordinate system. In other words, one
can think of t̂ and n̂ as the local x̂ and ŷ coordinates. Then writing the curl operator in this
local coordinates, one gets
 
∂ ∂
∇×E= x̂ + ŷ × (x̂Ex + ŷEy )
∂x ∂y
∂ ∂
= ẑ Ey − ẑ Ex (4.3.2)
∂x ∂y

In simplifying the above, we have used the distributive property of cross product, and eval-
uating the cross product in cartesian coordinates. The cross product gives four terms, but
only two of the four terms are non-zero as shown above.
∂ ∂
Since the right-hand side of (4.3.1) is finite, the above implies that ∂x Ey and ∂y Ex have
to be finite. In order words, Ex is continuous in the y direction and Ey is continuous in the
x direction. Since in the local coordinate system, Ex = Et , then Et is continuous across the
boundary. The above implies that

E1t = E2t (4.3.3)

Magnetostatics, Boundary Conditions, and Jump Conditions 45

or the tangential components of the electric field is continuous at the interface. To express
this in a coordinate independent manner, we have

n̂ × E1 = n̂ × E2 (4.3.4)

where n̂ is the unit normal at the interface, and n̂ × E always bring out the tangential
component of a vector E (convince yourself).

4.3.2 Gauss’s Law for Electric Flux

From this Gauss’s law, we have

∇·D=% (4.3.5)

where % is the volume charge density.

Figure 4.3: A figure showing the derivation of boundary condition for Gauss’s law. Again, a
local coordinate system can be introduced for simplicity.

Expressing the above in local coordinates (x, y, z as shown in Figure 4.3, then

∂ ∂ ∂
∇·D= Dx + Dy + Dz = % (4.3.6)
∂x ∂y ∂z
The boundary condition for the electric flux can be found by singularity matching.
If there is a surface layer charge at the interface, then the volume charge density must be
infinitely large or singular; hence, it can be expressed in terms of a delta function, or % = %s δ(z)
in local coordinates. By looking at the above expression, the only term that can produce a
∂
δ(z) is from ∂z Dz . In other words, Dz has a jump discontinuity at z = 0; the other terms do
not. Then
∂
Dz = %s δ(z) (4.3.7)
∂z
46 Electromagnetic Field Theory

Integrating the above from 0 − ∆ to 0 + ∆, we get

0+∆

Dz (z) = %s (4.3.8)
0−∆

or in the limit when ∆ → 0,

Dz (0+ ) − Dz (0− ) = %s (4.3.9)

where 0+ = lim∆→0 0 + ∆, and 0− = lim∆→0 0 − ∆. Since Dz (0+ ) = D2n , Dz (0− ) = D1n ,

the above becomes

D2n − D1n = %s (4.3.10)

In other words, a charge sheet %s can give rise to a jump discontinuity in the normal component
of the electric flux D. Expressed in coordinate independent form, it is

n̂ · (D2 − D1 ) = %s (4.3.11)

Using the physical notion that an electric charge has electric flux D exuding from it,
Figure 4.4 shows an intuitive sketch as to why a charge sheet gives rise to a discontinuous
normal component of the electric flux D.

Figure 4.4: A figure intuitively showing why a sheet of charge gives rise to a jump discontinuiy
in the normal component of the electric flux D.

4.3.3 Ampere’s Law

Ampere’s law, or the generalized one, stipulates that
∂D
∇×H=J+ (4.3.12)
∂t
Again if the right-hand side is everywhere finite, then H is a continuous field everywhere.
However, if the right-hand side has a delta function singularity, due to a current sheet in 3D
space, and that ∂D
∂t is regular or finite everywhere, then the only place where the singularity
can be matched on the left-hand side is from the derivative of the magnetic field H or ∇ × H.
In a word, H is not continuous. For instance, we can project the above equation onto a local
coordinates just as we did for Faraday’s law.
Magnetostatics, Boundary Conditions, and Jump Conditions 47

Figure 4.5: A figure showing the derivation of boundary condition for Ampere’s law. A local
coordinate system is used for simplicity.

To be general, we also include the presence of a current sheet at the interface. A current
sheet, or a surface current density becomes a delta function singularity when expressed as a
volume current density; Thus, rewriting (4.3.12) in a local coordinate system, assuming that
J = x̂Jsx δ(z),4 then singularity matching,
 
∂ ∂
∇ × H = x̂ Hz − Hy = x̂Jsx δ(z) (4.3.13)
∂y ∂z
The displacement current term on the right-hand side is ignored since it is regular or finite,
and will not induce a jump discontinuity on the field; hence, we have the form of the right-
hand side of the above equation. From the above, the only term that can produce a δ(z)
∂
singularity on the left-hand side is the − ∂z Hy term. Therefore, by singularity matching, we
conclude that
∂
− Hy = Jsx δ(z) (4.3.14)
∂z
In other words, Hy has to have a jump discontinuity at the interface where the current
sheet resides. Or that
Hy (z = 0+ ) − Hy (z = 0− ) = −Jsx (4.3.15)
The above implies that
H2y − H1y = −Jsx (4.3.16)
But Hy is just the tangential component of the H field. In a word, the current sheet Jsx
induces a jump discontinuity on the y component of the magnetic field. Now if we repeat the
same exercise with a current with a y component, or J = ŷJsy δ(z), at the interface, we have
H2x − H1x = Jsy (4.3.17)
4 The form of this equation can be checked by dimensional analysis. Here, J has the unit of A m−2 , δ(z)

has unit of m−1 , and Jsx , a current sheet density, has unit of A m−1 .
48 Electromagnetic Field Theory

Now, (4.3.16) and (4.3.17) can be rewritten using a cross product as

ẑ × (ŷH2y − ŷH1y ) = x̂Jsx (4.3.18)

ẑ × (x̂H2x − x̂H1x ) = ŷJsy (4.3.19)

The above two equations can be combined as one, written in a coordinate independent form,
to give

n̂ × (H2 − H1 ) = Js (4.3.20)

where in this case here, n̂ = ẑ. In other words, a current sheet Js can give rise to a jump
discontinuity in the tangential components of the magnetic field, n̂ × H. This is illustrated
intuitively in Figure 4.6.

Figure 4.6: A figure intuitively showing that with the understanding of how a single line
current source generates a magnetic field (right), a cluster of them forming a current sheet
will generate a jump discontinuity in the tangential component of the magnetic field H (left).

4.3.4 Gauss’s Law for Magnetic Flux

Similarly, from Gauss’s law for magnetic flux, or that

∇·B=0 (4.3.21)

one deduces that

n̂ · (B2 − B1 ) = 0 (4.3.22)

or that the normal magnetic fluxes are continuous at an interface. In other words, since mag-
netic charges do not exist, the normal component of the magnetic flux has to be continuous.
The take-home message here is that the boundary conditions are buried in the differential
operators. If there singular terms in Maxwell’s equations, then via the differential operators,
the boundary conditions can be deduced. These boundary conditions are also known as jump
condition if a current or a source sheet is present.
Lecture 5

Biot-Savart law, Conductive

Media Interface, Instantaneous
Poynting’s Theorem

Biot-Savart law, like Ampere’s law was experimentally determined in around 1820 and it
is discussed in a number of textbooks [31, 32, 48]. This is the cumulative work of Ampere,
Oersted, Biot, and Savart. At this stage of the course, one has the mathematical tool to derive
this law from Ampere’s law and Gauss’s law for magnetostatics. In addition, we will study
the boundary conditions at conductive media interfaces, and introduce the instantaneous
Poynting’s theorem.

49
50 Electromagnetic Field Theory

5.1 Derivation of Biot-Savart Law

Figure 5.1: A current element used to illustrate the derivation of Biot-Savart law. The
current element generates a magnetic field due to Ampere’s law in the static limit. This law
was established experimentally, but here, we will derive this law based on our mathematical
knowledge so far.

From Gauss’ law and Ampere’s law in the static limit, and using the definition of the Green’s
function, we have derived that

µ J(r0 ) 0
A(r) = dV (5.1.1)
4π V R

where R = |r − r0 |. When the current element is small, and is carried by a wire of cross
sectional area ∆a as shown in Figure 5.1, we can approximate the integrand as

J(r0 )dV 0 ≈ J(r0 )∆V 0 = (∆a)∆l ˆlI/∆a (5.1.2)

| {z } | {z }
∆V J(r0 )

In the above, ∆V = (∆a)∆l and ˆlI/∆a = J(r0 ) since J has the unit of amperes/m2 . Here, ˆl
is a unit vector pointing in the direction of the current flow. Hence, we can let the current
element be

J(r0 )∆V 0 = I∆l0 (5.1.3)

where the vector ∆l0 = ∆lˆl, and 0 indicates that it is located at r0 . Therefore, the incremental
vector potential due to an incremental current element J(r0 )∆V 0 is

J(r0 )∆V 0 µ I∆l0

 
µ
∆A(r) = = (5.1.4)
4π R 4π R
Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s Theorem 51

Since B = ∇ × A, we derive that the incremental B flux, ∆B due to the incremental current
I∆l0 is

µI ∆l0 −µI 0 1
∆B = ∇ × ∆A(r) = ∇× = ∆l × ∇ (5.1.5)
4π R 4π R

where we have made use of the fact that ∇ × af (r) = −a × ∇f (r) when a is a constant vector.
The above can be simplified further making use of the fact that1

1 1
∇ = − 2 R̂ (5.1.6)
R R

a unit vector pointing in the r − r0 direction. We have also made use of the fact
where R̂ isp
that R = (x − x0 )2 + (y − y 0 )2 + (z − z 0 )2 . Consequently, assuming that the incremental
length becomes infinitesimally small, or ∆l → dl, we have, after using (5.1.6) in (5.1.5), that

µI 0 1
dB = dl × 2 R̂
4π R
µIdl0 × R̂
= (5.1.7)
4πR2

Since B = µH, we have

Idl0 × R̂
dH = (5.1.8)
4πR2

or


I(r0 )dl0 × R̂
H(r) = (5.1.9)
4πR2

which is Biot-Savart law, first determined experimentally, now derived using electromagnetic
field theory.

1 This is best done by expressing the ∇ operator in spherical coordinates.

52 Electromagnetic Field Theory

5.2 Shielding by Conductive Media

5.2.1 Boundary Conditions–Conductive Media Case

Figure 5.2: The schematics for deriving the boundary condition for the current density J at
the interface of two conductive media.

In a conductive medium, J = σE, which is just a statement of Ohm’s law. From the current
continuity equation, which is derivable from Ampere’s law and Gauss’ law for electric flux,
one gets
∂%
∇·J=− (5.2.1)
∂t
If the right-hand side is everywhere finite, it will not induce a jump discontinuity in the
current. Moreover, it is zero for static limit. Hence, just like the Gauss’s law case, the above
implies that the normal component of the current Jn is continuous, or that J1n = J2n in the
static limit. In other words,
n̂ · (J2 − J1 ) = 0 (5.2.2)
Hence, using J = σE, we have
σ2 E2n − σ1 E1n = 0 (5.2.3)
The above has to be always true in the static limit irrespective of the values of σ1 and σ2 .
But Gauss’s law implies the boundary condition that
ε2 E2n − ε1 E1n = %s (5.2.4)
The above equation is incompatible with (5.2.3) unless %s 6= 0. Hence, surface charge density
or charge accumulation is necessary at the interface, unless σ2 /σ1 = ε2 /ε1 . This is found
in semiconductor materials which are both conductive and having a permitivity: interfacial
charges appear at the interface of two semi-conductor materials.
Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s Theorem 53

5.2.2 Electric Field Inside a Conductor

The electric field inside a perfect electric conductor (PEC) has to be zero by the explanation
as follows. If medium 1 is a perfect electric conductor, then σ → ∞ but J1 = σE1 . An
infinitesimal E1 will give rise to an infinite current J1 . To avoid this ludicrous situation, E1
has to be 0. This implies that D1 = 0 as well.

Figure 5.3: The behavior of the electric field and electric flux at the interface of a perfect
electric conductor and free space (or air).

Since tangential E is continuous, from Faraday’s law, it is still true that

E2t = E1t = 0 (5.2.5)

or n̂ × E = 0. But since

n̂ · (D2 − D1 ) = %s (5.2.6)

and that D1 = 0, then

n̂ · D2 = %s (5.2.7)

So surface charge density has to be nonzero at a PEC/air interface. Moreover, normal D2 6= 0,

tangential E2 = 0: Thus the E and D have to be normal to the PEC surface. The sketch
of the electric field in the vicinity of a perfect electric conducting (PEC) surface is shown in
Figure 5.3.
The above argument for zero electric field inside a perfect conductor is true for electro-
dynamic problems. However, one does not need the above argument regarding the shielding
of the static electric field from a conducting region. In the situation of the two conducting
54 Electromagnetic Field Theory

objects example below, as long as the electric fields are non-zero in the objects, currents will
keep flowing. They will flow until the charges in the two objects orient themselves so that
electric current cannot flow anymore. This happens when the charges produce internal fields
that cancel each other giving rise to zero field inside the two objects. Faraday’s law still
applies which means that tangental E field has to be continuous. Therefore, the boundary
condition that the fields have to be normal to the conducting object surface is still true for
elecrostatics. A sketch of the electric field between two conducting spheres is show in Figure
5.4.

Figure 5.4: The behavior of the electric field and flux outside two conductors in the static
limit. The two conductors need not be PEC, and yet, the fields are normal to the interface.

5.2.3 Magnetic Field Inside a Conductor

We have seen that for a finite conductor, as long as σ 6= 0, the charges will re-orient themselves
until the electric field is expelled from the conductor; otherwise, the current will keep flowing
until E = 0 or ∂t E = 0. In a word, static E is zero inside a conductor.
But there are no magnetic charges nor magnetic conductors in this world. Thus this
physical phenomenon does not happen for magnetic field: In other words, static magnetic
field cannot be expelled from an electric conductor. However, a magnetic field can be expelled
from a perfect conductor or a superconductor. You can only fully understand this physical
phenomenon if we study the time-varying form of Maxwell’s equations.
In a perfect conductor where σ → ∞, it is unstable for the magnetic field B to be
nonzero. As time varying magnetic field gives rise to an electric field by the time-varying
form of Faraday’s law, a small time variation of the B field will give rise to infinite current
flow in a perfect conductor. Therefore to avoid this ludicrous situation, and to be stable,
B = 0 in a perfect conductor or a superconductor.
Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s Theorem 55

So if medium 1 is a perfect electric conductor (PEC), then B1 = H1 = 0. The boundary

condition for magnetic field from Ampere’s law

n̂ × (H2 − H1 ) = n̂ × H2 = Js (5.2.8)

which is the jump condition for the magnetic field. In other words, a surface current Js has
to flow at the surface of a PEC in order to support the jump discontinuity in the tangential
component of the magnetic field. From Gauss’s law, n̂ · B is always continuous at an interface
because of the absence of magnetic charges. The magnetic flux B1 is expelled from the perfect
conductor making n̂ · B1 = 0 zero both inside and outside the PEC (due to the continuity of
normal B at an interface). And hence, there is no normal component of the B1 field at the
interface. Therefore, the boundary condition for B2 becomes, for a PEC,

n̂ · B2 = 0 (5.2.9)

The B field in the vicinity of a perfect conductor surface is as shown in Figure 5.5.

Figure 5.5: Sketch of the magnetic flux B around a perfect electric conductor. It is seen that
n̂ · B = 0 at the surface of the perfect conductor.

When a superconductor cube is placed next to a static magnetic field near a permanent
magnet, eddy current will be induced on the superconductor. The eddy current will expel the
static magnetic field from the permanent magnet, or in a word, it will produce a magnetic
dipole on the superconducting cube that repels the static magnetic field. Since these two
magnetic dipoles are of opposite polarity, they repel each other, and cause the superconducting
cube to levitate on the static magnetic field as shown in Figure 5.6.
56 Electromagnetic Field Theory

Figure 5.6: Levitation of a superconducting disk on top of a static magnetic field due to
expulsion of the magnetic field from the superconductor.. This is also known as the Meissner
effect (figure courtesy of Wikimedia).

5.3 Instantaneous Poynting’s Theorem

Before we proceed further with studying energy and power, it is habitual to add fictitious
magnetic current M and fictitious magnetic charge %m to Maxwell’s equations to make them
symmetric mathematically.2 To this end, we have

∂B
∇×E=− −M (5.3.1)
∂t
∂D
∇×H= +J (5.3.2)
∂t
∇·D=% (5.3.3)
∇ · B = %m (5.3.4)

Consider the first two of Maxwell’s equations where fictitious magnetic current is included
and that the medium is isotropic such that B = µH and D = εE. Next, we need to consider
only the first two equations (since in electrodynamics, by invoking charge conservation, the
third and the fourth equations are derivable from the first two). They are

∂B ∂H
∇×E=− − Mi = −µ − Mi (5.3.5)
∂t ∂t
∂D ∂E
∇×H= +J=ε + Ji + σE (5.3.6)
∂t ∂t
where Mi and Ji are impressed current sources. They are sources that are impressed into
the system, and they cannot be changed by their interaction with the environment [50].
Also, for a conductive medium, a conduction current or induced current flows in addition
to impressed current. Here, J = σE is the induced current source in the conductor. Moreover,
J = σE is similar to ohm’s law. We can show from By dot multiplying (5.3.5) with H, and
2 Even though magnetic current does not exist, electric current can be engineered to look like magnetic

current as shall be learnt.

Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s Theorem 57

dot multiplying (5.3.6) with E, we can show that

∂H
H · ∇ × E = −µH · − H · Mi (5.3.7)
∂t
∂E
E · ∇ × H = εE · + E · Ji + σE · E (5.3.8)
∂t
Using the identity, which is the same as the product rule for derivatives, we have3

∇ · (E × H) = H · (∇ × E) − E · (∇ × H) (5.3.9)

Therefore, from (5.3.7), (5.3.8), and (5.3.9) we have

 
∂H ∂E
∇ · (E × H) = − µH · + εE · + σE · E + H · Mi + E · Ji (5.3.10)
∂t ∂t
To elucidate the physical meaning of the above, we first consider σ = 0, and Mi = Ji = 0,
or the absence of conductive loss and the impressed current sources. Then the above becomes
 
∂H ∂E
∇ · (E × H) = − µH · + εE · (5.3.11)
∂t ∂t
Rewriting each term on the right-hand side of the above, we have
 
∂H 1 ∂ ∂ 1 ∂
µH · = µ H·H= µ|H|2 = Wm (5.3.12)
∂t 2 ∂t ∂t 2 ∂t
 
∂E 1 ∂ ∂ 1 ∂
εE · = ε E·E= ε|E|2 = We (5.3.13)
∂t 2 ∂t ∂t 2 ∂t

where |H(r, t)|2 = H(r, t) · H(r, t), and |E(r, t)|2 = E(r, t) · E(r, t). Then (5.3.11) becomes
∂
∇ · (E × H) = − (Wm + We ) (5.3.14)
∂t
where
1 1
Wm = µ|H|2 , We = ε|E|2 (5.3.15)
2 2
Equation (5.3.14) is reminiscent of the current continuity equation, namely that,
∂%
∇·J=− (5.3.16)
∂t
which is a statement of charge conservation. In other words, time variation of current density
at a point is due to charge density flow into or out of the point.
The vector quantity

Sp = E × H (5.3.17)
3 The cyclical identity that a · (b × c) = c · (a × b) = b · (c × a) is useful for the derivation.
58 Electromagnetic Field Theory

is called the Poynting’s vector , and (5.3.14) becomes

∂
∇ · Sp = − Wt (5.3.18)
∂t

where Wt = We + Wm is the total energy density stored in the electric and magnetic fields.
The above is similar to the current continuity equation in physical interpretation mentioned
above. Analogous to that current density is charge density flow, power density is energy
density flow. It is easy to show that Sp has a dimension of watts per meter square, or power
density, and that Wt has a dimension of joules per meter cube, or energy density.
Now, if we let σ 6= 0, then the term to be included is then σE · E = σ|E|2 which has
the unit of S m−1 times V2 m−2 , or W m−3 where S is siemens. We arrive at this unit by
2
noticing that 12 VR is the power dissipated in a resistor of R ohms with a unit of watts. The
reciprocal unit of ohms, which used to be called mhos is now called siemens. With σ 6= 0,
(5.3.18) becomes

∂ ∂
∇ · Sp = − Wt − σ|E|2 = − Wt − Pd (5.3.19)
∂t ∂t

Here, ∇·Sp has physical meaning of power density oozing out from a point, and −Pd = −σ|E|2
has the physical meaning of power density dissipated (siphoned) at a point by the conductive
loss in the medium which is proportional to −σ|E|2 .
Now if we set Ji and Mi to be nonzero, (5.3.19) is augmented by the last two terms in
(5.3.10), or

∂
∇ · Sp = − Wt − Pd − H · Mi − E · Ji (5.3.20)
∂t

The last two terms can be interpreted as the power density supplied by the impressed currents
Mi and Ji . Therefore, (5.3.20) becomes

∂
∇ · Sp = − Wt − Pd + Ps (5.3.21)
∂t

where

Ps = −H · Mi − E · Ji (5.3.22)

Here, Ps is the power supplied by the impressed current sources. These terms are positive if
H and Mi have opposite signs, or if E and Ji have opposite signs. The last terms reminds us
of what happens in a negative resistance device or a battery.4 In a battery, positive charges
move from a region of lower potential to a region of higher potential (see Figure 5.7). The
positive charges move from one end of a battery to the other end of the battery. Hence, they
are doing an “uphill climb” driven by chemical processes within the battery.
4A negative resistance has been made by Leo Esaki [51], winning him a share in the Nobel prize.
Biot-Savart law, Conductive Media Interface, Instantaneous Poynting’s Theorem 59

Figure 5.7: Figure showing the dissipation of energy as the current flows around a loop. A
battery can be viewed as having a negative resistance.

In the above, one can easily work out that Ps has the unit of W m−3 which is power
supplied density. One can also choose to rewrite (5.3.21) in integral form by integrating it
over a volume V and invoking the divergence theorem yielding
   
d
dS · Sp = − Wt dV − Pd dV + Ps dV (5.3.23)
S dt V V V

The left-hand side is

 
dS · Sp = dS · (E × H) (5.3.24)
S S

which represents the power flowing out of the surface S.

60 Electromagnetic Field Theory
Lecture 6

Time-Harmonic Fields, Complex

Power

The analysis of Maxwell’s equations can be greatly simplified by assuming the fields to be
time harmonic, or sinusoidal (cosinusoidal). Electrical engineers use a method called phasor
technique [32, 52], to simplify equations involving time-harmonic signals. This is also a poor-
man’s Fourier transform [53]. That is one begets the benefits of Fourier transform technique
without knowledge of Fourier transform. Since only one time-harmonic frequency is involved,
this is also called frequency domain analysis.1 Phasors are represented in complex numbers.
From this we will also discuss the concept of complex power.

1 It is simple only for linear systems: for nonlinear systems, such analysis can be quite unwieldy. But rest

assured, as we will not discuss nonlinear systems in this course.

61
62 Electromagnetic Field Theory

6.1 Time-Harmonic Fields—Linear Systems

Figure 6.1: A commemorative stamp showing the contribution of Euler (courtesy of Wikipedia
and Pinterest).

To learn phasor technique, one makes use the formula due to Euler (1707–1783) (Wikipedia)
ejα = cos α + j sin α (6.1.1)
√
where j = −1 is an imaginary number.2
From Euler’s formula one gets
cos α = <e ejα


(6.1.2)
Hence, all time harmonic quantity can be written as
V (x, y, z, t) = V 0 (x, y, z) cos(ωt + α) (6.1.3)
= V 0 (r)<e(ej(ωt+α) ) (6.1.4)
0 jα jωt


= <e V (r)e e (6.1.5)
 
jωt
= <e V (r)e (6.1.6)
e
Now V (r) = V 0 (r)ejα is a complex number called the phasor representation or phasor of
V (r, t)e a time-harmonic quantity.3 Here, the phase α = α(r) can also be a function of
position r, or x, y, z. Consequently, any component of a field can be expressed as
h i
Ex (x, y, z, t) = Ex (r, t) = <e E x (r)ejωt (6.1.7)
√
e
2 But lo and behold, in other disciplines, −1 is denoted by “i”, but “i” is too close to the symbol for
current. So the preferred symbol for electrical engineering for an imaginary number is j: a quirkness of
convention, just as positive charges do not carry current in a wire.
3 We will use under tilde to denote a complex number or a phasor here, but this notation will be dropped

later. Whether a variable is complex or real is clear from the context.

Time-Harmonic Fields, Complex Power 63

The above can be repeated for y and z components. Compactly, one can write
h i
E(r, t) = <e E(r)ejωt (6.1.8)
e
h i
H(r, t) = <e H(r)ejωt (6.1.9)
e
where E and H are complex vector fields. Such phasor representations of time-harmonic fields
simplify
e Maxwell’s
e equations. For instance, if one writes
 
B(r, t) = <e B(r)ejωt (6.1.10)
e
then
∂ ∂ h i
B(r, t) = <e B(r)ejωt
∂t ∂t  e 
∂ jωt
= <e B(r)e
∂t
 e 
= <e B(r)jωejωt (6.1.11)
e
Therefore, a time derivative can be effected very simply for a time-harmonic field. One just
needs to multiply jω to the phasor representation of a field or a signal. Therefore, given
Faraday’s law that

∂B
∇×E=− −M (6.1.12)
∂t
assuming that all quantities are time harmonic, then
h i
E(r, t) = <e E(r)ejωt (6.1.13)
h e i
M(r, t) = <e M(r)ejωt (6.1.14)
f
using (6.1.11), and (6.1.14), into (6.1.12), one gets
h i
∇ × E(r, t) = <e ∇ × E(r)ejωt (6.1.15)
e
and that
h i h i h i
<e ∇ × E(r)ejωt = −<e B(r)jωejωt − <e M(r)ejωt (6.1.16)
e e f
Since if

<e A(r)ejωt = <e B(r)ejωt , ∀t

   
(6.1.17)
64 Electromagnetic Field Theory

then A(r) = B(r), it must be true from (6.1.16) that

∇ × E(r) = −jωB(r) − M(r) (6.1.18)
e e f
∂
Hence, finding the phasor representation of an equation is clear: whenever we have ∂t , we
replace it by jω. Applying this methodically to the other Maxwell’s equations, we have
∇ × H(r) = jωD(r) + J(r) (6.1.19)
e e e
∇ · D(r) = %e (r) (6.1.20)
e e
∇ · B(r) = %m (r) (6.1.21)
e e
In the above, the phasors are functions of frequency. For instance, H(r) should rightly be
written as H(r, ω), but the ω dependence is implied. e
e

6.2 Fourier Transform Technique

In the phasor representation, Maxwell’s equations has no time derivatives; hence the equations
are simplified. We can also arrive at the above simplified equations using Fourier transform
technique. To this end, we use Faraday’s law as an example. By letting
∞
1
E(r, t) = E(r, ω)ejωt dω (6.2.1)
2π
−∞
∞
1
B(r, t) = B(r, ω)ejωt dω (6.2.2)
2π
−∞
∞
1
M(r, t) = M(r, ω)ejωt dω (6.2.3)
2π
−∞

Substituting the above into Faraday’s law given by (6.1.12), we get

∞ ∞ ∞
jωt ∂ jωt
∇× dωe E(r, ω) = − dωe B(r, ω) − dωejωt M(r, ω) (6.2.4)
∂t
−∞ −∞ −∞

Using the fact that

∞ ∞ ∞
∂ jωt ∂
dωe B(r, ω) = dω ejωt B(r, ω) = dωejωt jωB(r, ω) (6.2.5)
∂t ∂t
−∞ −∞ −∞

and that
∞ ∞
jωt
∇× dωe E(r, ω) = dωejωt ∇ × E(r, ω) (6.2.6)
−∞ −∞
Time-Harmonic Fields, Complex Power 65

Furthermore, using the fact that

∞ ∞
jωt
dωe A(ω) = dωejωt B(ω), ∀t (6.2.7)
−∞ −∞

implies that A(ω) = B(ω), and using (6.2.5) and (6.2.6) in (6.2.4), and the property (6.2.7),
one gets

∇ × E(r, ω) = −jωB(r, ω) − M(r, ω) (6.2.8)

These equations look exactly like the phasor equations we have derived previously, save
that the field E(r, ω), B(r, ω), and M(r, ω) are now the Fourier transforms of the field E(r, t),
B(r, t), and M(r, t). Moreover, the Fourier transform variables can be complex just like
phasors. Repeating the exercise above for the other Maxwell’s equations, we obtain equations
that look similar to those for their phasor representations. Hence, Maxwell’s equations can
be simplified either by using phasor technique or Fourier transform technique. However, the
dimensions of the phasors are different from the dimensions of the Fourier-transformed fields:
E(r) and E(r, ω) do not have the same dimension on closer examination.
e

6.3 Complex Power

Consider now that in the phasor representations, E(r) and H(r) are complex vectors, and
their cross product, E(r) × H∗ (r), which still has the
e unit ofepower density, has a different
physical meaning. First, consider the instantaneous Poynting’s vector
e e

S(r, t) = E(r, t) × H(r, t) (6.3.1)

where all the quantities are real valued. Now, we can use phasor technique to analyze the
above. Assuming time-harmonic fields, the above can be rewritten as
h i h i
S(r, t) = <e E(r)ejωt × <e H(r)ejωt
1 h jωt ie 1 h
e i
jωt ∗
= Ee + (Ee ) × Hejωt + (Hejωt )∗ (6.3.2)
2 e e 2 e e
where we have made use of the formula that
1
<e(Z) = (Z + Z ∗ ) (6.3.3)
2
Then more elaborately, on expanding (6.3.2), we get
1 1 1 1
S(r, t) = E × He2jωt + E × H∗ + E∗ × H + E∗ × H∗ e−2jωt (6.3.4)
4e e 4e e 4e e 4e e
Then rearranging terms and using (6.3.3) yield
1 h i 1 h i
S(r, t) = <e E × H∗ + <e E × He2jωt (6.3.5)
2 e e 2 e e
66 Electromagnetic Field Theory

where the first term is independent of time, while the second term is sinusoidal in time. If we
define a time-average quantity such that
 T
1
Sav = hS(r, t)i = lim S(r, t)dt (6.3.6)
T →∞ T 0

then it is quite clear that the second term of (6.3.5) time-averages to zero, and

1 h i
Sav = hS(r, t)i = <e E × H∗ (6.3.7)
2 e e
Hence, in the phasor representation, the quantity

S = E × H∗ (6.3.8)
e e e
is termed the complex Poynting’s vector . The power flow associated with it is termed complex
power.

Figure 6.2: A simple circuit example to illustrate the concept of complex power in circuit
theory.

To understand what complex power is , it is fruitful if we revisit complex power [50, 54]
in our circuit theory course. The circuit in Figure 6.2 can be easily solved by using phasor
technique. The impedance of the circuit is Z = R + jωL. Hence,

V = (R + jωL)I (6.3.9)
e e
where V and I are the phasors of the voltage and current for time-harmonic signals. Just as
in the electromagnetic
e e case, the complex power is taken to be

P = V I∗ (6.3.10)
e ee
But the instantaneous power is given by

Pinst (t) = V (t)I(t) (6.3.11)

Time-Harmonic Fields, Complex Power 67

where V (t) = <e{V ejωt } and I(t) = <e{I ejωt }. As shall be shown below,
e e
1 h i
Pav = hPinst (t)i = <e P (6.3.12)
2 e
It is clear that if V (t) is sinusoidal, it can be written as
h i
V (t) = V0 cos(ωt) = <e V ejωt (6.3.13)
e
where, without loss of generality, we assume that V = V0 . Then from (6.3.9), it is clear that
V (t) and I(t) are not in phase. Namely that e
h i
I(t) = I0 cos(ωt + α) = <e I ejωt (6.3.14)
e
where I = I0 ejα . Then
e
Pinst (t) = V0 I0 cos(ωt) cos(ωt + α)
= V0 I0 cos(ωt) [cos(ωt) cos(α) − sin(ωt) sin α]
= V0 I0 cos2 (ωt) cos α − V0 I0 cos(ωt) sin(ωt) sin α (6.3.15)

It can be seen that the first term does not time-average to zero, but the second term, by
letting cos(ωt) sin(ωt) = 0.5 sin(2ωt), does time-average to zero. Now taking the time average
of (6.3.15), we get
1 1 h i
Pav = hPinst i = V0 I0 cos α = <e V I ∗ (6.3.16)
2 2
1 h i
ee
= <e P (6.3.17)
2 e
On the other hand, the reactive power
1 h i 1 h i 1 1
Preactive = =m P = =m V I ∗ = =m V0 I0 e−jα = − V0 I0 sin α
 
(6.3.18)
2 e 2 ee 2 2
One sees
 that amplitude of the time-varying term in (6.3.15) is precisely proportional to
=m P .4
The
e reason for the existence of imaginary part of P is because V (t) and I(t) are out of
phase or V = V0 , but I = I0 ejα . The reason why they eare out of phase is because the circuit
has a reactive
e part to it.
e Hence the imaginary part of complex power is also called the reactive
power [35,50,54]. In a reactive circuit, the plots of the instantaneous power is shown in Figure
6.3. It is seen that when α 6= 0, the instantaneous power can be negative. This means that
the power is flowing from the load to the source instead of flowing from the source to the
load. This happens only when the reactive power is nonzero or when a reactive component
like an inductor or capacitor exists in the circuit. When a power company delivers power
4 Because that complex power is proportional to V I ∗ , it is the relative phase between V and I that matters.

Therefore, α above is the relative phase between the

e ephasor current and phasor voltage.e e
68 Electromagnetic Field Theory

to our home, the power is complex because the current and voltage are not in phase. Even
though the reactive power time-averages to zero, the power company still needs to deliver it
to and from our home to run our washing machine, dish washer, fans, and air conditioner
etc, and hence, charges us for it. Part of this power will be dissipated in the transmission
lines that deliver power to our home. In other words, we have to pay to the use of imaginary
power!

Figure 6.3: Plots of instantaneous power for when the voltage and the current is in phase
(α = 0), and when they are out of phase (α 6= 0). In the out-of-phase case, there is an
additional time-varying term that does not contribute to time-average power as shown in
(6.3.15).
Lecture 7

More on Constitute Relations,

Uniform Plane Wave

As mentioned before, constitutive relations are important for us to solve only the first two
of four Maxwell’s equations. Assuming that J is known or zero, then the first two vector
equations have four vector unknowns: E, H, D, and B, which cannot be determined by
solving only two equations. The addition two equations come from te constitutive relations.
Constitutive relations are useful because they allow us to incorporate material properties into
the solutions of Maxwell’s equations. The material properties can be frequency dispersive,
anisotropic, bi-anisotropic, inhomogeneous, lossy, conductive, nonlinear as well as spatially
dispersive. The use of phasors or frequency domain method will further simplify the charac-
terization of different media.,
The use of phasors allows us to study wave phenomena in these media quite easily. Hence,
we will also study uniform plane wave in such media, including a lossy conductive media.

7.1 More on Constitutive Relations

As have been said, Maxwell’s equations are not solvable until the constitutive relations are
included. Here, we will look into depth more into various kinds of constitutive relations. Now
that we have learned phasor technique, we can study a more general constitutive relationship
compared to what we have seen earlier.

7.1.1 Isotropic Frequency Dispersive Media

First let us look at the simple linear constitutive relation previously discussed for dielectric
media where [31], [32][p. 82], [48]

D = ε0 E + P (7.1.1)

69
70 Electromagnetic Field Theory

We have a simple model where

P = ε 0 χ0 E (7.1.2)

where χ0 is the electric susceptibility. When used in the generalized Ampere’s law, P, the
polarization density, plays an important role for the flow of the displacement current through
space. The polarization density is due to the presence of polar atoms or molecules that
become little dipoles in the presence of an electric field. For instance, water, which is H2 O,
is a polar molecule that becomes a small dipole when an electric field is applied.
We can think of displacement current flow as capacitive coupling between the dipoles
yielding polarization current flow through space. Namely, for a source-free medium,

∂D ∂E ∂P
∇×H= = ε0 + (7.1.3)
∂t ∂t ∂t

Figure 7.1: As a series of dipoles line up end to end, one can see a current flowing through
the line of dipoles as they oscillate back and forth in their polarity. This is similar to how
displacement current flows through a capacitor.

For example, for a sinusoidal oscillating field, the dipoles will flip back and forth giving rise
to flow of displacement current just as how time-harmonic electric current can flow through
a capacitor as shown in Figure 7.1.
The linear relationship above can be generalized to that of a linear time-invariant system
[52], or that at any given r [35][p. 212], [48][p. 330].

P(r, t) = ε0 χe (r, t) ~ E(r, t) (7.1.4)

where ~ here implies a convolution. In the frequency domain or the Fourier space, the above
linear relationship becomes

P(r, ω) = ε0 χ0 (r, ω)E(r, ω), (7.1.5)

or

D(r, ω) = ε0 [1 + χ0 (r, ω)] E(r, ω) = ε(r, ω)E(r, ω) (7.1.6)

where ε(r, ω) = ε0 [1 + χ0 (r, ω)] at any point r in space. There is a rich variety of ways
at which χ0 (ω) can manifest itself. Such a permittivity ε(r, ω) is often called the effective
permittivity. Such media where the effective permittivity is a function of frequency is termed
dispersive media, or frequency dispersive media.
More on Constitute Relations, Uniform Plane Wave 71

7.1.2 Anisotropic Media

For anisotropic media [32][p. 83]

D = ε0 E + ε0 χ0 (ω) · E
 
= ε0 I + χ0 (ω) · E = ε(ω) · E (7.1.7)

In the above, ε is a 3×3 matrix also known as a tensor in electromagnetics. The above implies
that D and E do not necessary point in the same direction: the meaning of anisotropy. (A
tensor is often associated with a physical notion like the relation between two physical fields,
whereas a matrix is not.)
Previously, we have assume that χ0 to be frequency independent. This is not usually the
case as all materials have χ0 ’s that are frequency dependent. (This will become clear later.)
Also, since ε(ω) is frequency dependent, we should view it as a transfer function where the
input is E, and the output D. This implies that in the time-domain, the above relation
becomes a time-convolution relation as in (7.1.4).
Similarly for conductive media,

J = σE, (7.1.8)

This can be used in Maxwell’s equations in the frequency domain to yield the definition of
complex permittivity. Using the above in Ampere’s law in the frequency domain, we have

∇ × H(r) = jωεE(r) + σE(r) = jωε(ω)E(r) (7.1.9)

e
where the complex permittivity ε(ω) = ε − jσ/ω. Notice that Ampere’s law in the frequency
domain with complex permittivity e in (7.1.9) is no more complicated than Ampere’s law for
nonconductive media. The algebra for complex numbers is no more difficult than the algebra
for real numbers. This is one of the strengths of phasor technique.
For anisotropic conductive media, one has

J = σ(ω) · E, (7.1.10)

Here, again, due to the tensorial nature of the conductivity σ, the electric current J and
electric field E do not necessary point in the same direction.
The above assumes a local or point-wise relationship between the input and the output
of a linear system. This need not be so. In fact, the most general linear relationship between
P(r, t) and E(r, t) is
 ∞  ∞
P(r, t) = χ(r − r0 , t − t0 ) · E(r0 , t0 )dr0 dt0 (7.1.11)
−∞ −∞

The above is a general convolutional relationship in both space and time. In the Fourier
transform space, by taking Fourier transform in both space and time, the above becomes

P(k, ω) = χ(k, ω) · E(k, ω) (7.1.12)

72 Electromagnetic Field Theory

where
 ∞
χ(k, ω) = χ(r, t) exp(jk · r − jωt)drdt (7.1.13)
−∞

(The dr integral above is actually a three-fold integral with dr = dxdydz.) Such a medium is
termed spatially dispersive as well as frequency dispersive [35][p. 6], [55]. In general1
ε(k, ω) = 1 + χ(k, ω) (7.1.14)
where
D(k, ω) = ε(k, ω) · E(k, ω) (7.1.15)
The above can be extended to magnetic field and magnetic flux yielding
B(k, ω) = µ(k, ω) · H(k, ω) (7.1.16)
for a general spatial and frequency dispersive magnetic material. In optics, most materials
are non-magnetic, and hence, µ = µ0 , whereas it is quite easy to make anisotropic magnetic
materials in radio and microwave frequencies, such as ferrites.

7.1.3 Bi-anisotropic Media

In the previous section, the electric flux D depends on the electric field E and the magnetic
flux B, on the magnetic field H. The concept of constitutive relation can be extended to
where D and B depend on both E and H. In general, one can write
D = ε(ω) · E + ξ(ω) · H (7.1.17)
B = ζ(ω) · E + µ(ω) · H (7.1.18)
A medium where the electric flux or the magnetic flux is dependent on both E and H is
known as a bi-anisotropic medium [32][p. 81].

7.1.4 Inhomogeneous Media

If any of the ε, ξ, ζ, or µ is a function of position r, the medium is known as an inhomogeneous
medium or a heterogeneous medium. There are usually no simple solutions to problems
associated with such media [35].

7.1.5 Uniaxial and Biaxial Media

Anisotropic optical materials are often encountered in optics. Examples of them are the
biaxial and uniaxial media, and discussions of them are often found in optics books [56–58].
They are optical materials where the permittivity tensor can be written as
 
ε1 0 0
ε =  0 ε2 0  (7.1.19)
0 0 ε3
1 In the following, one should replace the 1 with an identity operator, but it is generally implied.
More on Constitute Relations, Uniform Plane Wave 73

When ε1 6= ε2 6= ε3 , the medium is known as a biaxial medium. But when ε1 = ε2 6= ε3 , then

the medium is a uniaxial medium.
In the biaxial medium, all three components of the electric field feel different permittivity
constants. But in the uniaxial medium, the electric field in the xy plane feels the same
permittivity constant, but the electric field in the z direction feels a different permittivity
constant. As shall be shown later, different light polarization will propagate with different
behavior through such a medium.

7.1.6 Nonlinear Media

In the previous cases, we have assumed that χ0 is independent of the field E. The relationships
between P and E can be written more generally as

P = ε0 χ0 (E) · E (7.1.20)

where the relationship can appear in many different forms. For nonlinear media, the rela-
tionship can be nonlinear as indicated in the above. It can be easily shown that the principle
of linear superposition does not hold for the above equation, a root test of linearity. Non-
linear permittivity effect is important in optics. Here, the wavelength is short, and a small
change in the permittivity or refractive index can give rise to cumulative phase delay as
the wave has to propagate many wavelengths through a nonlinear optical medium [59–61].
Kerr optical nonlinearity, discovered in 1875, was one of the earliest nonlinear phenomena
observed [32, 56, 59].
For magnetic materials, nonlinearity can occur in the effective permeability of the medium.
In other words,

B = µ(H) · H (7.1.21)

This nonlinearity is important even at low frequencies, and in electric machinery designs
[62, 63], and magnetic resonance imaging systems [64]. The large permeability in magnetic
materials is usually due to the formation of magnetic domains which can only happen at low
frequencies.

7.2 Wave Phenomenon in the Frequency Domain

We have seen the emergence of wave phenomenon in the time domain. Given the simplicity of
the frequency domain method, it will be interesting to ask how this phenomenon presents itself
for time-harmonic field or in the frequency domain. In the frequency domain, the source-free
Maxwell’s equations are [32][p. 429], [65][p. 107]

∇ × E(r) = −jωµ0 H(r) (7.2.1)

∇ × H(r) = jωε0 E(r) (7.2.2)

Taking the curl of (7.2.1) and then substituting (7.2.2) into its right-hand side, one obtains

∇ × ∇ × E(r) = −jωµ0 ∇ × H(r) = ω 2 µ0 ε0 E(r) (7.2.3)

74 Electromagnetic Field Theory

Again, using the identity that

∇ × ∇ × E = ∇(∇ · E) − ∇ · ∇E = ∇(∇ · E) − ∇2 E (7.2.4)
and that ∇ · E = 0 in a source-free medium, (7.2.3) becomes
(∇2 + ω 2 µ0 ε0 )E(r) = 0 (7.2.5)
This is known as the Helmholtz wave equation or just the Helmholtz equation.2
For simplicity of seeing the wave phenomenon, we let E = x̂Ex (z), a field pointing in the
x direction, but varying only in the z direction. Evidently, ∇ · E(r) = ∂Ex (z)/∂x = 0. Then
with ∂/∂x = 0 and ∂/∂y = 0, (7.2.5) simplifies to
 2 
d 2
+ k 0 Ex (z) = 0 (7.2.6)
dz 2
√
where k02 = ω 2 µ0 ε0 = ω 2 /c20 where c0 = 1/ µ0 ε0 is the velocity of light. The general solution
to (7.2.6) is of the form
Ex (z) = E0+ e−jk0 z + E0− ejk0 z (7.2.7)
One can convert the above back to the time domain using phasor technique, or by using that
Ex (z, t) = <e[Ex (z, ω)ejωt ], yielding
Ex (z, t) = |E0+ | cos(ωt − k0 z + α+ ) + |E0− | cos(ωt + k0 z + α− ) (7.2.8)
where we have assumed that
E0± = |E0± |ejα± (7.2.9)
The physical picture of the above expressions can be appreciated by rewriting
 
ω
cos(ωt ∓ k0 z + α± ) = cos (c0 t ∓ z) + α± (7.2.10)
c0
where we have used the fact that k0 = cω0 . One can see that the first term on the right-hand
side of (7.2.8) is a sinusoidal plane wave traveling to the right, while the second term is a
sinusoidal plane wave traveling to the left, both with velocity c0 . The above plane wave is
uniform and a constant in the xy plane and propagating in the z direction. Hence, it is also
called a uniform plane wave in 1D.
Moreover, for a fixed t or at t = 0, the sinusoidal functions are proportional to cos(∓k0 z +
α± ). This is a periodic function in z with period 2π/k0 which is the wavelength λ0 , or that
2π ω 2πf
k0 = = = (7.2.11)
λ0 c0 c0
One can see that because c0 is a humongous number in electromagnetics, λ0 can be very
large. You can plug in the frequency of your local AM 920 station, operating at 920 KHz, to
see how big λ0 is.
The above analysis still holds true if 0 and µ0 are replaced by  and µ but are real
√
numbers. In this case, the velocity c of the wave is now given by c = 1/ µ. This velocity is
the velocity of the phase of a time-harmonic signal and hence, is known as phase velocity.
2 For a comprehensive review of this topic, one may read the lecture notes [44].
More on Constitute Relations, Uniform Plane Wave 75

7.3 Uniform Plane Waves in 3D

By repeating the previous derivation for a homogeneous lossless medium, the vector Helmholtz
equation for a source-free medium is given by [44]

∇ × ∇ × E − ω 2 µεE = 0 (7.3.1)

By the same derivation as before for the free-space case, since ∇ · E = 0 due to source-free
medium, one has

∇2 E + ω 2 µεE = 0 (7.3.2)

The general solution to (7.3.2) is hence

E = E0 e−jkx x−jky y−jkz z = E0 e−jk·r (7.3.3)

where k = x̂kx + ŷky + ẑkz , r = x̂x + ŷy + ẑz and E0 is a constant vector. And upon
substituting (7.3.3) into (7.3.2), it is seen that

kx2 + ky2 + kz2 = ω 2 µε (7.3.4)

This is called the dispersion relation for a plane wave.

In general, kx , ky , and kz can be arbitrary and even complex numbers as long as this
relation is satisfied. To simplify the discussion, we will focus on the case where kx , ky , and
kz are all real numbers. When this is the case, the vector function in (7.3.3) represents a
uniform plane wave propagating in the k direction. As can be seen, when k · r = constant, it
is represented by all points of r that represents a flat plane (see Figure 7.2). This flat plane
represents the constant phase wave front. By increasing the constant, we obtain different
planes for progressively changing phase fronts.3

Figure 7.2: A figure showing the geometrical meaning of k · r equal to a constant. It is a flat
plane that defines the wavefront of a plane wave.
3 In the exp(jωt) time convention, this phase front is decreasing, whereas in the exp(−iωt) time convention,

this phase front is increasing.

76 Electromagnetic Field Theory

Further, since ∇ · E = 0, we have

∇ · E = ∇ · E0 e−jkx x−jky y−jkz z = ∇ · E0 e−jk·r
= (−x̂jkx − ŷjky − ẑjkz ) · E0 e−jk·r
= −j(x̂kx + ŷky + ẑkz ) · E = 0 (7.3.5)
or that
k · E0 = k · E = 0 (7.3.6)
Thus, E is orthogonal to k for a uniform plane wave.
The above exercise shows that whenever E is a plane wave, and when the ∇ operator
operates on such a vector function, one can do the substitution that ∇ → −jk. Hence, in a
source-free homogenous medium,
∇ × E = −jωµH (7.3.7)
the above equation simplifies to
−jk × E = −jωµH (7.3.8)
or that
k×E
H= (7.3.9)
ωµ
Similar to (7.3.3), we can define
H = H0 e−jkx x−jky y−jkz z = H0 e−jk·r (7.3.10)
Then using (7.3.3) in (7.3.9), it is clear that
k × E0
H0 = (7.3.11)
ωµ
We can assume that E0 and H0 are real vectors. Then E0 , H0 and k form a right-handed
system, or that E0 ×H0 point in the direction of k. (This also implies that E, H and k form a
right-handed system.) Such a wave, where the electric field and magnetic field are transverse
to the direction of propagation, is called a transverse electromagnetic (TEM) wave.

Figure 7.3: The E, H, and k together form a right-hand coordinate system, obeying the
right-hand rule. Namely, E × H points in the direction of k.
More on Constitute Relations, Uniform Plane Wave 77

Figure 7.3 shows that k · E = 0, and that k × E points in the direction of H as shown in
(7.3.9). Figure 7.3 also shows, as k, E, and H are orthogonal to each other.
Since in general, E0 and H0 can be complex vectors, because they are phasors, we need to
show the more general case. From (7.3.9), one can show, using the “back-of-the-cab” formula,
that
k k
E × H∗ = E · E∗ = |E|2 (7.3.12)
ωµ ωµ
But E × H∗ is the direction of power flow, and it is in fact in the k direction assuming that
k is a real vector. This is also required by the Poynting’s theorem.
Furthermore, we can show in general that

|k||E|
r
ε 1
|H| = = |E| = |E| (7.3.13)
ωµ µ η

where the quantity

r
µ
η= (7.3.14)
ε

is call the intrinsic impedance. For vacuum or free-space, it is about 377 Ω ≈ 120π Ω.
Notice that the above analysis holds true as long as ε is real, and it can be frequency
dispersive, since we are considering a mono-chromatic or time-harmonic field. Besides, for
a mono-chromatic signal, the analysis in Section 7.2 still applies except that the velocity of
√
light is now given by c = 1/ µ0 ε. As we shall see, this velocity is the phase velocity of the
mono-chromatic wave. In the above, when kx , ky , and kz are not all real, the wave is known
as an inhomogeneous wave.4

4 The term inhomogeneous plane wave is used sometimes, but it is a misnomer since there is no more a

planar wave front in this case.

78 Electromagnetic Field Theory
Lecture 8

Lossy Media, Lorentz Force Law,

Drude-Lorentz-Sommerfeld
Model

In the previous lecture, we see the power of phasor technique or the frequency domain analysis.
The analysis of a frequency dispersive medium where ε is frequency dependent, is similar to
that of free space or vacuum. The two problems are mathematically homomorphic to each
other. In this lecture, we will generalize to the case where ε becomes a complex number, called
the complex permittivity. Using phasor technique, this way of solving Maxwell’s equations is
still homomorphic to that of solving Maxwell’s equations in free space. The analysis is greatly
simplified as a result!

8.1 Plane Waves in Lossy Conductive Media

Previously, we have derived the plane wave solution for a lossless homogeneous medium.
Since the algebra of complex numbers is similar to that of real numbers,the derivation can
be generalized to a conductive medium by invoking mathematical homomorphism. In other
words, in a conductive, one only needs to replace the permittivity with a complex permittivity,
as repeated here. When conductive loss is present, σ 6= 0, and J = σE. Then generalized
Ampere’s law becomes
 
σ
∇ × H = jωεE + σE = jω ε + E (8.1.1)
jω

A complex permittivity can be defined as ε = ε − j ωσ . Eq. (8.1.1) can be rewritten as

e
∇ × H = jωεE (8.1.2)
e

79
80 Electromagnetic Field Theory

This equation is of the same form as source-free Ampere’s law in the frequency domain for a
lossless medium where ε is completely real. Using the same method as before, a wave solution

E = E0 e−jk·r (8.1.3)

will have the dispersion relation which is now given by

kx2 + ky2 + kz2 = ω 2 µε (8.1.4)

e
Since ε is complex now, kx , ky , and kz cannot be all real. Equation (8.1.4) has been derived
previously
e by assuming that k is a real vector. When k = k0 − jk00 is a complex vector, some
of the previous derivations for real k vector may not be correct here for complex k vector. It
is also difficult to visualize a complex k vector that is suppose to indicate the direciton with
which the wave is propagating. Here, the wave can decay and oscillate in different directions.
So again, for physical insight, we look at the simplified case where

E = x̂Ex (z) (8.1.5)

√
so that ∇ · E = ∂x Ex (z) = 0, and let k = ẑk = ẑω µε. This wave is constant in the xy
plane, and hence, is a plane wave. Furthermore, in thisemanner, we are requiring that the
wave decays and propagates (or oscillates) only in the z direction. For such a simple plane
wave,

E = x̂Ex (z) = x̂E0 e−jkz (8.1.6)

√
where k = ω µε, since k · k = k 2 = ω 2 µε is still true.
Faraday’s law
e gives rise to e
s
k×E kEx (z) ε
H= = ŷ = ŷ e Ex (z) (8.1.7)
ωµ ωµ µ
√
where k vector is defined shortly after (8.1.5) above, and k = ω µε, a complex number. It
is seen that H = ŷHy , and that e
s
µ
Ex /Hy = (8.1.8)
ε
e

8.1.1 Highly Conductive Case

When the medium is highly conductive, σ → ∞, and ε = −j ωσ ≈ −j ωσ . In other words, when
| ωσ |  ε, the conduction current dominates over displacement
e current. Thus, the following
approximation can be made, namely,
r
q jσ p
k = ω µε ' ω −µ = −jωµσ (8.1.9)
e ω
Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 81

√
Taking −j = √1 (1 − j), we have for a highly conductive medium that
2

r
ωµσ
k ' (1 − j) = k 0 − jk 00 (8.1.10)
2

For a plane wave, e−jkz , it then becomes

0 00
e−jkz = e−jk z−k z
(8.1.11)

By converting the above phasor back to the time domain, this plane wave decays exponentially
as well as oscillates in the z direction. The reason being that a conductive medium is lossy,
and it absorbs energy from the plane wave. This is similar to resistive loss we see in the
resistive circuit. The penetration depth of this wave is then
r
1 2
δ= = (8.1.12)
k 00 ωµσ

This distance δ, the penetration depth, is called the skin depth of a plane wave propagating
in a highly lossy conductive medium where conduction current dominates over displacement
current, or that σ  ωε. This happens for radio wave propagating in the saline solution of
the ocean, the Earth, or wave propagating in highly conductive metal, like your induction
cooker.

8.1.2 Lowly Conductive Case

When the conductivity is low, namely, when the displacement current is larger than the
σ
conduction current, then ωε  1, we have

r  s  
σ jσ
k =ω µ ε−j = ω µε 1 −
ω ωε
 
√ 1 σ
≈ ω µε 1 − j = k 0 − jk 00 (8.1.13)
2 ωε

The above is the approximation to k = k 0 − jk 00 for a low conductivity medium where

σ
conduction current is much smaller than displacement current.1 The term ωε is called the
loss tangent of a lossy medium. It is the ratio of the conduction current to the displacement
current in a lossy conductive medium.
In general, in a lossy medium ε = ε0 − jε00 , and ε00 /ε0 is called the loss tangent of the
medium. It is to be noted that in the optics and physics community, e−iωt time convention
is preferred. In that case, we need to do the switch j → −i, and a loss medium is denoted by
ε = ε0 + iε00 .
1 We have made use of the approximation that (1 + x)n ≈ 1 + nx when x is small, which can be justified

by Taylor series expansion.

82 Electromagnetic Field Theory

8.2 Lorentz Force Law

The Lorentz force law is the generalization of the Coulomb’s law for forces between two
charges. Lorentz force law includes the presence of a magnetic field. It is given by
F = qE + qv × B (8.2.1)
The first term on the right-hand side is the electric force similar to the statement of Coulomb’s
law, while the second term is the magnetic force called the v × B force. This law can be also
written in terms of the force density f which is the force on the charge density, instead of
force on a single charge. By so doing, we arrive at
f = %E + %v × B = %E + J × B (8.2.2)
where % is the charge density, and one can identified the current J = %v.
Lorentz force law can also be derived from the integral form of Faraday’s law, if one
assumes that the law is applied to a moving loop intercepting a magnetic flux [66]. In other
words, Lorentz force law and Faraday’s law are commensurate with each other.

8.3 Drude-Lorentz-Sommerfeld Model

In the previous lecture, we have seen how loss can be introduced by having a conduction
current flowing in a medium. Now that we have learnt the versatility of the frequency domain
method and phasor technique, other loss mechanism can be easily introduced.
First, let us look at the simple constitutive relation where
D = ε0 E + P (8.3.1)
We have a simple model where
P = ε0 χE (8.3.2)
where χ is the electric susceptibility. To see how χ(ω) can be derived, we will study the
Drude-Lorentz-Sommerfeld model. This is usually just known as the Drude model or the
Lorentz model in many textbooks although Sommerfeld also contributed to it. These models,
the Drude and Lorentz models, can be unified in one equation as shall be shown.

Figure 8.1: Polarization of an atom in the presence of an electric field. Here, it is assumed
that the electron is weakly bound or unbound to the nucleus of the atom.
Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 83

8.3.1 Cold Collisionless Plasma Medium

We can first start with a simple electron driven by an electric field E in the absence of a
magnetic field B.2 If the electron is free to move, then the force acting on it, from the
Lorentz force law, is just −eE where q = −e is the charge of the electron (see Figure 8.1).
Then from Newton’s law, assuming a one dimensional case, it follows that

d2 x
me = −eE (8.3.3)
dt2
where the left-hand side is due to the inertial force of the mass of the electron, and the
right-hand side is the electric force acting on a charge of −e coulomb. Here, we assume that
E points in the x-direction, and we neglect the vector nature of the electric field or that we
assume that both x and E are in the same direction. Writing the above in the frequency
domain for time-harmonic fields, and using phasor technique, one gets

−ω 2 me x = −eE (8.3.4)

The above implies that the inertial force of the electron, given by −ω 2 me x, is of the same
polarity as the electric field force on the electron which is −eE. From this, we have
e
x= E (8.3.5)
ω 2 me

implying that the displacement x is linearly proportional to the electric field amplitude E, or
they are in phase. This, for instance, can happen in a plasma medium where the atoms are
ionized, and the electrons are free to roam [67]. Hence, we assume that the positive ions are
more massive, sluggish, and move very little compared to the electrons when an electric field
is applied.
The dipole moment formed by the displaced electron away from the ion due to the electric
field is then

e2
p = −ex = − E (8.3.6)
ω 2 me

for one electron. When there are N electrons per unit volume, the dipole moment density is
then given by

N e2
P = Np = − E (8.3.7)
ω 2 me

In general, P and E point in the opposite directions, and we can write

N e2 ωp 2
P=− 2
E = − 2 ε0 E (8.3.8)
ω me ω
2 Even if B 6= 0, the v × B force is small if the velocity of the electron is much smaller than the speed of

light.
84 Electromagnetic Field Theory

where we have defined ωp 2 = N e2 /(me ε0 ). Then,

ωp 2
 
D = ε0 E + P = ε0 1 − 2 E (8.3.9)
ω

In this manner, we see that the effective permittivity is

ωp 2
 
ε(ω) = ε0 1 − 2 (8.3.10)
ω

What the above math is saying is that the electric field E induces a dipole moment density
P that is negative to ε0 E, or the vacuum part of the contribution to D. This negative dipole
density cancels the contribution to the electric flux from the vacuum 0 E. For low frequency,
the effective permittivity is negative, disallowing the propagation of a wave as we shall see.
Hence, ε < 0 if p
ω < ωp = N/(me ε0 )e
√
Here, ωp is the plasma frequency. Since k = ω µε, if ε is negative, k = −jα becomes pure
imaginary, and a wave such as e−jkz decays exponentially as e−αz . This is also known as an
evanescent wave. In other words, the wave cannot propagate through such a medium: Our
ionosphere is such a medium. The plasma shields out electromagnetic waves that are below
the plasma frequency ωp .
Therefore, it was extremely fortuitous that Marconi, in 1901, was able to send a radio
signal from Cornwall, England, to Newfoundland, Canada. Nay sayers thought his experiment
would never succeed as the radio signal would propagate to outer space and never to return.
Fortunately so, it is the presence of the ionosphere that bounces the radio wave back to Earth,
making his experiment a resounding success and a very historic one! Serendipity occurs in
science and technology development more than once: the experiment also heralds in the age
of wireless communications.
This experiment also stirred interests into research on the ionosphere. It was an area again
where Oliver Heaviside made contributions; as a result, a layer of the ionosphere is named
Heaviside layer or Kennelly-Heaviside layer [68]. If you listen carefully to the broadway
musical “Cats” by Andrew Lloyd Weber, there is a mention about the Heaviside layer in one
of the verses!

Figure 8.2: The electron is bound to the ion by an attractive force. This can be approximately
modeled by a spring providing a restoring force to the electron.
Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 85

8.3.2 Bound Electron Case

The above model can be generalized to the case where the electron is bound to the ion, but
the ion now provides a restoring force similar to that of a spring (see Figure 8.2), namely,

d2 x
me + κx = −eE (8.3.11)
dt2
We assume that the ion provides a restoring force just like Hooke’s law. Again, for a time-
harmonic field, (8.3.11) can be solved easily in the frequency domain to yield
e e
x= E= 2 E (8.3.12)
(ω 2 me − κ) (ω − ω0 2 )me

where we have defined ω0 2 me = κ. The above is the typical solution of a lossless harmonic
oscillator (pendulum) driven by an external force, in this case the electric field. The dipole
moment due to an electric field then is
e2
p = −ex = − E (8.3.13)
(ω 2 − ω0 2 )me

Therefore, when the frequency is low or ω = 0, this dipole moment is of the same polarity as
the applied electric field E, contributing to a positive dipole moment. It contributes positively
to the displacement flux D via P.

8.3.3 Damping or Dissipation Case

Equation (8.3.11) can be generalized to the case when frictional, damping, or dissipation
forces are present, or that

d2 x dx
me 2
+ me Γ + κx = −eE (8.3.14)
dt dt
The second term on the left-hand side is a force that is proportional to the velocity dx/dt of
the electron. This is the hall-mark of frictional force. Frictional force is due to the collision
of the electrons with the background ions or lattice. It is proportional to the destruction (or
change) of momentum (me dx dt ) of an electron. In the average sense, the destruction of the
momentum is given by the product of the collision frequency and the momentum. In the
above, Γ has the unit of frequency, and for plasma, and conductor, it can be regarded as a
collision frequency.
Solving the above in the frequency domain, one gets
e
x= E (8.3.15)
(ω 2 − jωΓ − ω0 2 )me

Following the same procedure in arriving at (8.3.7), we get

−N e2
P = E (8.3.16)
(ω 2 − jωΓ − ω0 2 )me
86 Electromagnetic Field Theory

In this, one can identify that

−N e2
χ(ω) =
(ω 2 − jωΓ − ω0 2 )me ε0
ωp 2
=− 2 (8.3.17)
ω − jωΓ − ω0 2

where ωp is as defined before. A function with the above frequency dependence is also called
a Lorentzian function. It is the hallmark of a damped harmonic oscillator.

If Γ = 0, then when ω = ω0 , one sees an infinite resonance peak exhibited by the DLS
model. But in the real world, Γ 6= 0, and when Γ is small, but ω ≈ ω0 , then the peak value
of χ is

ωp 2 ωp 2
χ≈+ = −j (8.3.18)
jωΓ ωΓ

χ exhibits a large negative imaginary part, the hallmark of a dissipative medium, as in the
conducting medium we have previously studied.

8.3.4 Broad Applicability of Drude-Lorentz-Sommerfeld Model

The DLS model is a wonderful model because it can capture phenomenologically the essence
of the physics of many electromagnetic media, even though it is a purely classical model.3 It
captures the resonance behavior of an atom absorbing energy from light excitation. When
the light wave comes in at the correct frequency, it will excite electronic transition within an
atom which can be approximately modeled as a resonator with behavior similar to that of a
pendulum oscillator. This electronic resonances will be radiationally damped [34],4 and the
damped oscillation can be modeled by Γ 6= 0. By picking a mixture of multi-species DLS
oscillators, almost any shape of absorption spectra can be curve-fitted [69] (see Figure 8.3).

3 Whatwe mean here is that only Newton’s law has been used, and no quantum theory as yet.
4 The
oscillator radiates as it oscillates, and hence, loses energy to its environment. This causes the decay
of the oscillation, just as a damped LC tank circuit losing energy to the resistor.
Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 87

Figure 8.3: A Lorentzian has almost a bell-shape curve. By assuming multi-species of DLS
oscillators in a medium, one can fit absorption spectra of almost any shape (courtesy of
Wikipedia [69]).

Moreover, the above model can also be used to model molecular vibrations. In this case,
the mass of the electron will be replaced by the mass of the atom involved. The damping of
the molecular vibration is caused by the hindered vibration of the molecule due to interaction
with other molecules [70]. The hindered rotation or vibration of water molecules when excited
by microwave is the source of heat in microwave heating.
In the case of plasma, Γ 6= 0 represents the collision frequency between the free electrons
and the ions, giving rise to loss. In the case of a conductor, Γ represents the collision frequency
between the conduction electrons in the conduction band with the lattice of the material.5
Also, if there is no restoring force, then ω0 = 0. This is true for sea of electron moving in the
conduction band of a medium. Besides, for sufficiently low frequency, the inertial force can
be ignored. Thus, from (8.3.17), again we have6

ωp 2
χ ≈ −j (8.3.19)
ωΓ
and

ωp 2
 
ε = ε0 (1 + χ) = ε0 1 − j (8.3.20)
ωΓ

We recall that for a conductive medium, we define a complex permittivity to be

 
σ
ε = ε0 1 − j (8.3.21)
ωε0
5 It is to be noted that electron has a different effective mass in a crystal lattice [71, 72], and hence, the

electron mass has to be changed accordingly in the DLS model.

6 This equation is similar to (8.3.18). In both cases, collision force dominates in the equation of motion

(8.3.14).
88 Electromagnetic Field Theory

Comparing (8.3.20) and (8.3.21), we see that

ωp 2
σ = ε0 (8.3.22)
Γ
The above formula for conductivity can be arrived at using collision frequency argument as
is done in some textbooks [73].
As such, the DLS model is quite powerful: it can be used to explain a wide range of
phenomena from very low frequency to optical frequency. The fact that ε < 0 can be used to
explain many phenomena. The ionosphere is essentially a plasma medium described by

ωp 2
 
ε = ε0 1 − 2 (8.3.23)
ω

with ω0 = Γ = 0 called a cold collisionless plasma. Radio wave or microwave can only
penetrate through this ionosphere when ω > ωp , so that ε > 0. The electrons in many
conductive materials can be modeled as a sea of free electrons moving about quite freely with
an effective mass . As such, they behave like a plasma medium as shall be seen.

8.3.5 Frequency Dispersive Media

The DLS model shows that, except for vacuum, all media are frequency dispersive. It is
prudent to digress to discuss more on the physical meaning of a frequency dispersive medium.
The relationship between electric flux and electric field, in the frequency domain, still follows
the formula

D(ω) = ε(ω)E(ω) (8.3.24)

When the effective permittivity, ε(ω), is a function of frequency, it implies that the above
relationship in the time domain is via convolution, viz.,

D(t) = ε(t) ~ E(t) (8.3.25)

Since the above represents a linear time-invariant (LTI) system [52], it implies that an input
is not followed by an instantaneous output. In other words, there is a delay between the
input and the output. The reason is because an electron has a mass, and it cannot respond
immediately to an applied force: or it has inertial. (In other words, the system has memory
of what it was before when you try to move it.)
Even though the effective permittivity  is a function of frequency, the frequency domain
analysis we have done for a plane wave propagating in a dispersive medium still applies.
For a mono-chromatic signal, it will have a velocity, called the phase velocity, given by
√
v = 1/ µ0 ε. Here, it also implies that different frequency components will propagate with
different phase velocities through such a medium. Hence, a narrow pulse will spread in its
width because different frequency components are not in phase after a short distance of travel.
Also, the Lorentzian function is great for data fitting, as many experimentally observed
resonances have finite Q and a line width. The Lorentzian function models that well. If
multiple resonances occur in a medium or an atom, then multi-species DLS model can be
Lossy Media, Lorentz Force Law, Drude-Lorentz-Sommerfeld Model 89

used. It is now clear that all media have to be frequency dispersive because of the finite mass
of the electron and the inertial it has. In other words, there is no instantaneous response in
a dielectric medium due to the finiteness of the electron mass.
Even at optical frequency, many metals, which has a sea of freely moving electrons in the
conduction band, can be modeled approximately as a plasma. A metal consists of a sea of
electrons in the conduction band which are not tightly bound to the ions or the lattice. Also,
in optics, the inertial force due to the finiteness of the electron mass (in this case effective
mass , see Figure 8.4) can be sizeable compared to other forces. Then, ω0  ω or that the
restoring force is much smaller than the inertial force, in (8.3.17), and if Γ is small, χ(ω)
resembles that of a plasma, and ε of a metal can be negative.

Figure 8.4: Effective masses of electron in different metals.

8.3.6 Plasmonic Nanoparticles

When a plasmonic nanoparticle made of gold is excited by light, its response is given by (see
homework assignment)

a3 cos θ εs − ε0
ΦR = E0 (8.3.26)
r2 εs + 2ε0

In a plasma, εs can be negative, and thus, at certain frequency, if εs = −2ε0 , then ΦR → ∞.

Gold or silver with a sea of electrons, behaves like a plasma at optical frequencies, since the
inertial force in the DLS model is quite large.7 Therefore, when light interacts with such a
particle, it can sparkle brighter than normal. This reminds us of the saying “All that glitters
is not gold!” even though this saying has a different intended meaning.
Ancient Romans apparently knew about the potent effect of using gold and silver nanopar-
ticles to enhance the reflection of light. These nanoparticles were impregnated in the glass
or lacquer ware. By impregnating these particles in different media, the color of light will
sparkle at different frequencies, and hence, the color of the glass emulsion can be changed
(see website [74]).
7 In this case, ω 2  ω 2 , and ω 2  ωΓ; the binding force and the collision force can be ignored similar to a
0
cold plasma.
90 Electromagnetic Field Theory

Figure 8.5: Ancient Roman goblets whose laquer coating glisten better due to the presence
of gold nanoparticles. Gold or silver at optical frequencies behaves like plasma (courtesy of
Smithsonian.com).
Lecture 9

Waves in Gyrotropic Media,

Polarization

Gyrotropy is an important concept in electromagnetics. When a wave propagates through

a gyrotropic medium, the electric field rotates changing the polarization of the wave. Our
ionosphere is such a medium, and it affects radio and microwave communications between the
Earth and the satellite by affecting the polarization of the wave. We will study this important
topic in this lecture.

9.1 Gyrotropic Media and Faraday Rotation

This section derives the effective permittivity tensor of a gyrotropic medium in the ionsphere.
Our ionosphere is always biased by a static magnetic field due to the Earth’s magnetic field
[75]. But in this derivation, to capture the salient feature of the physics with a simple model,
we assume that the ionosphere has a static magnetic field polarized in the z direction, namely
that B = ẑB0 . Now, the equation of motion from the Lorentz force law for an electron with
q = −e, (in accordance with Newton’s second law that F = ma or force equals mass times
accelration) becomes
dv
me = −e(E + v × B) (9.1.1)
dt
Next, let us assume that the electric field is polarized in the xy plane. The derivative of v is
the acceleration of the electron, and also, v = dr/dt where r = x̂x + ŷy + ẑz. Again, assuming
linearity, we use frequency domain technique for the analysis. And in the frequency domain,
the above equation in the cartesian coordinates becomes
me ω 2 x = e(Ex + jωB0 y) (9.1.2)
2
me ω y = e(Ey − jωB0 x) (9.1.3)
The above constitutes two equations with two unknowns x and y. They cannot be solved easily
for x and y in terms of the electric field because they correspond to a two-by-two matrix system

91
92 Electromagnetic Field Theory

with cross coupling between the unknowns x and y. But they can be simplified as follows:
We can multiply (9.1.3) by ±j and add it to (9.1.2) to get two decoupled equations [76]:

me ω 2 (x + jy) = e[(Ex + jEy ) + ωB0 (x + jy)] (9.1.4)

2
me ω (x − jy) = e[(Ex − jEy ) − ωB0 (x − jy)] (9.1.5)

In the above, if we take the new unknowns to be x ± jy, the two equations are decoupled
with respect to to these two unknowns. Defining new variables such that

s± = x ± jy (9.1.6)
E± = Ex ± jEy (9.1.7)

then (9.1.4) and (9.1.5) become

me ω 2 s± = e(E± ± ωB0 s± ) (9.1.8)

Thus, solving the above yields

e
s± = E± = C± E± (9.1.9)
me ω 2 ∓ eB0 ω
where
e
C± = (9.1.10)
me ω 2 ∓ eB0 ω
(By this manipulation, the above equations (9.1.2) and (9.1.3) transform to new equations
where there is no cross coupling between s± and E± . The mathematical parlance for this is
the diagonalization of a matrix equation [77]. Thus, the new equation can be solved easily.)
Next, one can define Px = −N ex, Py = −N ey, and that P± = Px ± jPy = −N es± . Then
it can be shown that

P± = ε0 χ± E± (9.1.11)

The expression for χ± can be derived, and they are given as

N eC± Ne e ωp 2
χ± = − =− = − (9.1.12)
ε0 ε0 me ω 2 ∓ eBo ω ω 2 ∓ Ωω

where Ω and ωp are the cyclotron frequency1 and plasma frequency, respectively.

eB0 N e2
Ω= , ωp 2 = (9.1.13)
me me ε0
At the cyclotron frequency, |χ± | → ∞. In other words, P± is finite even when E± = 0, or a
solution exists to the equation of motion (9.1.1) without a forcing term, which in this case is
the electric field. Thus, at this frequency, the solution blows up if the forcing term, E± is not
1 This is also called the gyrofrequency.
Waves in Gyrotropic Media, Polarization 93

zero. This is like what happens to an LC tank circuit at resonance whose current or voltage
tends to infinity when the forcing term, like the voltage or current is nonzero.
Now, one can express the original variables Px , Py , Ex , Ey in terms of P± and E± . With
the help of (9.1.11), we arrive at
P+ + P− ε0 ε0
Px = = (χ+ E+ + χ− E− ) = [χ+ (Ex + jEy ) + χ− (Ex − jEy )]
2 2 2
ε0
= [(χ+ + χ− )Ex + j(χ+ − χ− )Ey ] (9.1.14)
2
P+ − P− ε0 ε0
Py = = (χ+ E+ − χ− E− ) = [χ+ (Ex + jEy ) − χ− (Ex − jEy )]
2j 2j 2j
ε0
= [(χ+ − χ− )Ex + j(χ+ + χ− )Ey ] (9.1.15)
2j
The above relationship in cartesian coordinates can be expressed using a tensor where

P = ε0 χ · E (9.1.16)

where P = [Px , Py ], and E = [Ex , Ey ]. From the above, χ is of the form

2 2
ω ω Ω
!
p
− ω2 −Ω −j ω(ω2p−Ω2 )
 
1 (χ+ + χ− ) j(χ+ − χ− ) 2
χ= = 2 2 (9.1.17)
2 −j(χ+ − χ− ) (χ+ + χ− ) ω
j ω(ω2p−Ω2 )
Ω ω p
− ω2 −Ω2

Notice that in the above, when the B field is turned off or Ω = 0, the above resembles
the solution of a collisionless, cold plasma again. For the B = 0 case with electric field in the
x (or y) direction, it will drive a motion of the electron to be in the x (or y) direction. In
this case, v × B term is zero, and the electron motion is unaffected by the magnetic field as
can be seen from the Lorentz force law or (9.1.1). Hence, it behaves like a simple collisionless
plasma without a biasing magnetic field.
Consequently, for the B 6= 0 case, the above can be generalized to 3D to give
 
χ0 jχ1 0
χ = −jχ1 χ0 0 (9.1.18)
0 0 χp

where χp = −ωp2 /ω 2 . Notice that since we assume that B = ẑB0 , the z component of (9.1.1)
is unaffected by the v × B force. Hence, the electron moving in the z is like that of a cold
collisionless plasma.
Using the fact that D = ε0 E + P = ε0 (I + χ) · E = ε · E, the above implies that
 
1 + χ0 jχ1 0
ε = ε0  −jχ1 1 + χ0 0  (9.1.19)
0 0 1 + χp

Now, ε is that of an anisotropic medium, of which a gyrotropic medium belongs. Please notice
that the above tensor is a hermitian tensor. We shall learn later that this is the hallmark of
a lossless medium.
94 Electromagnetic Field Theory

Another characteristic of a gyrotropic medium is that a linearly polarized wave will rotate
when passing through it. This is the Faraday rotation effect [76], which we shall learn more
later. This phenomenon poses a severe problem for Earth-to-satellite communication, using
linearly polarized wave as it requires the alignment of the Earth-to-satellite antennas. This
can be avoided using a rotatingly polarized wave, called a circularly polarized wave that we
shall learn in the next section.
As we have learnt, the ionosphere affects out communication systems two ways: It acts as
a mirror for low-frequency electromagnetic or radio waves (making the experiment of Marconi
a rousing success). It also affects the polarization of the wave. But the ionosphere of the
Earth and the density of electrons that are ionized is highly dependent on temperature, and
the effect of the Sun. The fluctuation of particles in the ionosphere gives rise to scintillation
effects due to electron motion and collision that affect radio wave communication systems [78].

9.2 Wave Polarization

Studying wave polarization is very important for communication purposes [32]. A wave whose
electric field is pointing in the x direction while propagating in the z direction is a linearly
polarized (LP) wave. The same can be said of one with electric field polarized in the y
direction. It turns out that a linearly polarized wave suffers from Faraday rotation when
it propagates through the ionosphere. For instance, an x polarized wave can become a y
polarized wave due to Faraday rotation. So its polarization becomes ambiguous as the wave
propagates through the ionosphere: to overcome this, Earth to satellite communication is
done with circularly polarized (CP) waves [79]. So even if the electric field vector is rotated
by Faraday’s rotation, it remains to be a CP wave. We will study these polarized waves next.
We can write a general uniform plane wave propagating in the z direction in the time
domain as

E = x̂Ex (z, t) + ŷEy (z, t) (9.2.1)

Clearly, ∇ · E = 0, and Ex (z, t) and Ey (z, t), by the principle of linear superposition, are so-
lutions to the one-dimensional wave equation. For a time harmonic field, the two components
may not be in phase, and we have in general

Ex (z, t) = E1 cos(ωt − βz) (9.2.2)

Ey (z, t) = E2 cos(ωt − βz + α) (9.2.3)

where α denotes the phase difference between these two wave components. We shall study
how the linear superposition of these two components behaves for different α’s. First, we set
z = 0 to observe this field. Then

E = x̂E1 cos(ωt) + ŷE2 cos(ωt + α) (9.2.4)

π
For α = 2

Ex = E1 cos(ωt), Ey = E2 cos(ωt + π/2) (9.2.5)

Waves in Gyrotropic Media, Polarization 95

Next, we evaluate the above for different ωt’s

ωt = 0, Ex = E1 , Ey = 0 (9.2.6)
√ √
ωt = π/4, Ex = E1 / 2, Ey = −E2 / 2 (9.2.7)
ωt = π/2, Ex = 0, Ey = −E2 (9.2.8)
√ √
ωt = 3π/4, Ex = −E1 / 2, Ey = −E2 / 2 (9.2.9)
ωt = π, Ex = −E1 , Ey =0 (9.2.10)

The tip of the vector field E traces out an ellipse as show in Figure 9.1. With the left-hand
thumb pointing in the z direction, and the wave rotating in the direction of the fingers, such
a wave is called left-hand elliptically polarized (LHEP) wave.

Figure 9.1: If one follows the tip of the electric field vector, it traces out an ellipse as a
function of time t.

When E1 = E2 , the ellipse becomes a circle, and we have a left-hand circularly polarized
(LHCP) wave. When α = −π/2, the wave rotates in the counter-clockwise direction, and
the wave is either right-hand elliptically polarized (RHEP), or right-hand circularly polarized
(RHCP) wave depending on the ratio of E1 /E2 . Figure 9.2 shows the different polarizations
of the wave wave for different phase differences and amplitude ratio. Figure 9.3 shows a
graphic picture of a CP wave propagating through space.
96 Electromagnetic Field Theory

Figure 9.2: Due to different phase difference between the Ex and Ey components of the field,
and their relative amplitudes E2 /E1 , different polarizations will ensure. The arrow indicates
the direction of rotation of the field vector.

Figure 9.3: The rotation of the field vector of a right-hand circular polarization wave as it
propagates in the right direction [80] (courtesy of Wikipedia).

9.2.1 Arbitrary Polarization Case and Axial Ratio2

As seen before, the tip of the field vector traces out an ellipse in space as it propagates.
The axial ratio (AR) is the ratio of the major axis to the minor axis of this ellipse. It is an
important figure of merit for designing CP (circularly polarized) antennas (antennas that will
radiate circularly polarized waves). The closer is this ratio to 1, the better is the antenna
design. We will discuss the general polarization and the axial ratio of a wave.
For the general case for arbitrary α, we let

Ex = E1 cos ωt, Ey = E2 cos(ωt + α) = E2 (cos ωt cos α − sin ωt sin α) (9.2.11)

2 This section is mathematically complicated. It can be skipped on first reading.
Waves in Gyrotropic Media, Polarization 97

Then from the above, expressing Ey in terms of Ex , one gets

"  2 #1/2
E2 Ex
Ey = Ex cos α − E2 1 − sin α (9.2.12)
E1 E1

Rearranging and squaring, we get

aEx 2 − bEx Ey + cEy 2 = 1 (9.2.13)

where
1 2 cos α 1
a= 2 , b= , c= (9.2.14)
E1 sin2 α E1 E2 sin2 α E2 2
sin2 α

After letting Ex → x, and Ey → y, equation (9.2.13) is of the form,

ax2 − bxy + cy 2 = 1 (9.2.15)

The equation of an ellipse in its self coordinates is

2 2
x0 y0
 
+ =1 (9.2.16)
A B

where A and B are axes of the ellipse as shown in Figure 9.4. We can transform the above
back to the (x, y) coordinates by letting

x0 = x cos θ − y sin θ (9.2.17)

y 0 = x sin θ + y cos θ (9.2.18)

to get

cos2 θ sin2 θ sin2 θ cos2 θ

     
2 1 1 2
x + − xy sin 2θ − 2 +y + =1 (9.2.19)
A2 B2 A2 B A2 B2

Comparing (9.2.13) and (9.2.19), one gets

 
1 2 cos αE1 E2
θ= tan−1 (9.2.20)
2 E2 2 − E1 2
 1/2
1+∆
AR = >1 (9.2.21)
1−∆

where AR is the axial ratio and

!1/2
4E1 2 E2 2 sin2 α
∆= 1−
2 (9.2.22)
E1 2 + E2 2
98 Electromagnetic Field Theory

Figure 9.4: This figure shows the parameters used to derive the axial ratio (AR) of an
elliptically polarized wave.

9.3 Polarization and Power Flow

For a linearly polarized wave in the time domain,
E0
E = x̂E0 cos(ωt − βz), H = ŷ cos(ωt − βz) (9.3.1)
η
Hence, the instantaneous power we have learnt previously in Section 5.3 becomes
E0 2
S(t) = E(t) × H(t) = ẑ cos2 (ωt − βz) (9.3.2)
η
indicating that for a linearly polarized wave, the instantaneous power is function of both time
and space. It travels as lumps of energy through space. In the above E0 is the amplitude of
the linearly polarized wave.
Next, we look at power flow for for elliptically and circularly polarized waves. It is to be
noted that in the phasor world or frequency domain, (9.2.1) becomes

E(z, ω) = x̂E1 e−jβz + ŷE2 e−jβz+jα (9.3.3)

For LHEP wave,

E(z, ω) = e−jβz (x̂E1 + j ŷE2 ) (9.3.4)

whereas for LHCP wave,

E(z, ω) = e−jβz E1 (x̂ + j ŷ) (9.3.5)

Waves in Gyrotropic Media, Polarization 99

For RHEP wave, the above becomes

E(z, ω) = e−jβz (x̂E1 − j ŷE2 ) (9.3.6)

whereas for RHCP wave, it is

E(z, ω) = e−jβz E1 (x̂ − j ŷ) (9.3.7)

Focussing on the circularly polarized wave,

E = (x̂ ± j ŷ)E0 e−jβz (9.3.8)

Using that
β×E
H= ,
ωµ
where β = ẑβ, then
E0 −jβz
H = (∓x̂ − j ŷ)j e (9.3.9)
η
p
where η = µ/ε is the intrinsic impedance of the medium. Therefore,

E(t) = x̂E0 cos(ωt − βz) ± ŷE0 sin(ωt − βz) (9.3.10)

E0 E0
H(t) = ∓x̂ sin(ωt − βz) + ŷ cos(ωt − βz) (9.3.11)
η η
Then the instantaneous power becomes

E0 2 E0 2 E0 2
S(t) = E(t) × H(t) = ẑ cos2 (ωt − βz) + ẑ sin2 (ωt − βz) = ẑ (9.3.12)
η η η
In other words, a CP wave delivers constant instantaneous power independent of space and
time, as opposed to a linearly polarized wave which delivers a non-constant instantaneous
power as shown in (9.3.2).
It is to be noted that the complex Poynting’s vector for a lossless medium

S = E × H∗ (9.3.13)
e
is real and constant independent of space both for linearly, circularly, and elliptically polarized
waves. This is because there is no reactive power in a plane wave of any polarization: the
stored energy in the plane wave cannot be returned to the source!
100 Electromagnetic Field Theory
Lecture 10

Spin Angular Momentum,

Complex Poynting’s Theorem,
Lossless Condition, Energy
Density

Figure 10.1: The local coordinates used to describe a circularly polarized wave: In cartesian
and polar coordinates.

In the last lecture, we study circularly polarized waves as well as linearly polarized waves. In
addition, these waves can carry power giving rise to power flow. But in addition to carrying
power, a travelling wave also has a momentum: for a linearly polarized wave, it carries linear
momention in the direction of the propagation of the traveling wave. But for a circularly
polarized wave, it carries angular momentum as well.
We have studied the complex Poynting’s theorem in the frequency domain with phasors.

101
102 Electromagnetic Field Theory

Here, we will derive the lossless conditions for the permittivity and permeability tensors.
Energy density is well defined for a lossless dispersionless medium, but it assumes a different
formula when the medium is dispersive.

10.1 Spin Angular Momentum and Cylindrical Vector

Beam

In this section, we will study the spin angular momentum of a circularly polarized (CP) wave.
It is to be noted that in cylindrical coordinates, as shown in Figure 10.1, x̂ = ρ̂ cos φ − φ̂ sin φ,
ŷ = ρ̂ sin φ + φ̂ cos φ, then a CP field is proportional to

(x̂ ± j ŷ) = ρ̂e±jφ ± j φ̂e±jφ = e±jφ (ρ̂ ± φ̂) (10.1.1)

Therefore, the ρ̂ and φ̂ of a CP is also an azimuthal traveling wave in the φ̂ direction in

addition to being a traveling wave e−jβz in the ẑ direction. This is obviated by writing

e−jφ = e−jkφ ρφ (10.1.2)

where kφ = 1/ρ is the azimuthal wave number, and ρφ is the arc length traversed by the
azimuthal wave. Notice that the wavenumber kφ is dependent on ρ: the larger the ρ, the
smaller is kφ , and hence, the larger the azimuthal wavelength. Thus, the wave possesses
angular momentum called the spin angular momentum (SAM), just as a traveling wave e−jβz
possesses linear angular momentum in the ẑ direction.

In optics research, the generation of cylindrical vector beam is in vogue. Figure 10.2 shows
a method to generate such a beam. A CP light passes through a radial analyzer that will only
allow the radial component of (10.1.1) to be transmitted. Then a spiral phase element (SPE)
compensates for the exp(±jφ) phase shift in the azimuthal direction. Finally, the light is a
cylindrical vector beam which is radially polarized without spin angular momentum. Such
a beam has been found to have nice focussing property, and hence, has aroused researchers’
interest in the optics community [81].
Spin Angular Momentum, Complex Poynting’s Theorem, Lossless Condition, Energy Density103

Figure 10.2: A cylindrical vector beam can be generated experimentally. The spiral phase
element (SPE) compensates for the exp(±jφ) phase shift (courtesy of Zhan, Q. [81]). The
half-wave plate rotates the polarization of a wave by 90 degrees.

10.2 Momentum Density of Electromagnetic Field

We have seen that a traveling wave carries power and has energy density associated with it. In
other words, the moving or traveling energy density gives rise to power flow. It turns out that
a traveling wave also carries a momentum with it. The momentum density of electromagnetic
field is given by

G=D×B (10.2.1)

also called the momentun density vector. With it, one can derive momentum conservation
theorem [32, p. 59] [48]. The derivation is rather long, but we will justify the above formula
and simplify the derivation using the particle or corpuscular nature of light or electromagnetic
field. The following derivation is only valid for plane waves.
It has been long known that electromagnetic energy is carried by photon, associated with
a packet of energy given by ~ω. It is also well known that a photon has momentum given by
~k. Assuming that there are photons, with density of N photons per unit volume, streaming
through space at the velocity of light c. Then the power flow associated with these streaming
photons is given by

E × H = ~ωN cẑ (10.2.2)

104 Electromagnetic Field Theory

Assuming that the plane wave is propagating in the z direction. Using k = ω/c, we can
rewrite the above more suggestively as

E × H = ~ωN cẑ = ~kN c2 ẑ (10.2.3)

where k = ω/c. Defining the momentum density vector to be

G = ~kN ẑ (10.2.4)

From the above, we deduce that

1
E × H = Gc2 = G (10.2.5)
µε
Or the above can be rewritten as

G=D×B (10.2.6)

where D = εE, and B = µH.1

10.3 Complex Poynting’s Theorem and Lossless Condi-

tions
10.3.1 Complex Poynting’s Theorem
It has been previously shown that the vector E(r, t) × H(r, t) has a dimension of watts/m2
which is that of power density. Therefore, it is associated with the direction of power flow
[32, 48]. As has been shown for time-harmonic field, a time average of this vector can be
defined as
T
1
hE(r, t) × H(r, t)i = lim E(r, t) × H(r, t) dt. (10.3.1)
T →∞ T
0

Given the phasors of time harmonic fields E(r, t) and H(r, t), namely, E(r, ω) and H(r, ω)
respectively, we can show that
1
hE(r, t) × H(r, t)i = <e{E(r, ω) × H∗ (r, ω)}. (10.3.2)
2
Here, the vector E(r, ω) × H∗ (r, ω), as previously discussed, is also known as the complex
Poynting vector. If we define the instantaneous Poynting’s vector to be

S(r, t) = E(r, t) × H(r, t) (10.3.3)

and the complex Poynting’s vector to be

S(r, ω) = E(r, ω) × H∗ (r, ω) (10.3.4)

^
1 The author is indebted to Wei SHA for this simple derivation.
Spin Angular Momentum, Complex Poynting’s Theorem, Lossless Condition, Energy Density105

then for time-harmonic fields,

1
hS(r, t)i =
<eS(r, ω) (10.3.5)
2
^
The above is often the source of confusion in the definition of Poynting’s vector.
In the above definition of complex Poynting’s vector and its aforementioned property, and
its dimension of power density, we will study its conservative property. To do so, we take its
divergence and use the appropriate vector identity to obtain2
∇ · (E × H∗ ) = H∗ · ∇ × E − E · ∇ × H∗ . (10.3.6)
Next, using Maxwell’s equations for ∇ × E and ∇ × H∗ , namely
∇ × E = −jωB (10.3.7)
∇ × H∗ = −jωD∗ + J∗ (10.3.8)
and the constitutive relations for anisotropic media that
B = µ · H, D∗ = ε∗ · E∗ (10.3.9)
we have
∇ · (E × H∗ ) = −jω H∗ · B + jω E · D∗ − E · J∗ (10.3.10)
= −jω H∗ · µ · H + jω E · ε∗ · E∗ − E · J∗ . (10.3.11)
The above is also known as the complex Poynting’s theorem. It can also be written in an
integral form using Gauss’ divergence theorem, namely,
  
dS · (E × H∗ ) = −jω dV (H∗ · µ · H − E · ε∗ · E∗ ) − dV E · J∗ . (10.3.12)
S
V V

where S is the surface bounding the volume V .

10.3.2 Lossless Conditions

For a region V that is lossless and source-free, J = 0. There should be no net time-averaged
power-flow out of or into this region V . Therefore,

<e dS · (E × H∗ ) = 0, (10.3.13)
S

Because of energy conservation, the real part of the right-hand side of (10.3.11), without the
E · J∗ term, must be zero. In other words, the right-hand side of (10.3.11) should be purely
imaginary. Thus

dV (H∗ · µ · H − E · ε∗ · E∗ ) (10.3.14)
V
2 We will drop the argument r, ω for the phasors in our discussion next as they will be implied.
106 Electromagnetic Field Theory

must be a real quantity.

Other than the possibility that the above is zero, the general requirement for (10.3.14) to
be real for arbitrary E and H, is that H∗ · µ · H and E · ε∗ · E∗ are real quantities. This is
only possible if µ is hermitian.3 Therefore, the conditions for anisotropic media to be lossless
are

µ = µ† , ε = ε† , (10.3.15)

requiring the permittivity and permeability tensors to be hermitian. If this is the case,
(10.3.14) is always real for arbitraty E and H, and (10.3.13) is true, implying a lossless
region V . Notice that for an isotropic medium, this lossless conditions reduce simply to
that =m(µ) = 0 and =m(ε) = 0, or that µ and ε are pure real quantities. Looking back,
many of the effective permittivities or dielectric constants that we have derived using the
Drude-Lorentz-Sommerfeld model cannot be lossless when the friction term is involved.
For a lossy medium which is conductive, we may define J = σ · E where σ is a general
conductivity tensor. In this case, equation (10.3.12), after combining the last two terms, may
be written as
 
jσ ∗
   
∗ ∗ ∗ ∗
dS · (E × H ) = −jω dV H · µ · H − E · ε + ·E (10.3.16)
ω
S V

∗
= −jω dV [H∗ · µ · H − E · ε̃ · E∗ ], (10.3.17)

where ε̃ = ε − jσ
ω which is the general complex permittivity tensor. In this manner, (10.3.17)
has the same structure as the source-free Poynting’s theorem. Notice here that the complex
permittivity tensor ε̃ is clearly non-hermitian corresponding to a lossy medium.
For a lossless medium without the source term, by taking the imaginary part of (10.3.12),
we arrive at
 
=m dS · (E × H∗ ) = −ω dV (H∗ · µ · H − E · ε∗ · E∗ ), (10.3.18)
S V

The left-hand side of the above is the reactive power coming out of the volume V , and hence,
the right-hand side can be interpreted as reactive power as well. It is to be noted that
H∗ · µ · H and E · ε∗ · E∗ are not to be interpreted as stored energy density when the medium
is dispersive. The correct expressions for stored energy density will be derived in the next
section.
But, the quantity H∗ · µ · H for lossless, dispersionless media is associated with the time-
averaged energy density stored in the magnetic field, while the quantity E · ε∗ · E∗ for lossless
dispersionless media is associated with the time-averaged energy density stored in the electric
field. Then, for lossless, dispersionless, source-free media, then the right-hand side of the
3 H∗ · µ · H is real only if its complex conjugate, or conjugate transpose is itself. Using some details from
t t t
matrix algebra that (A · B · C)t = C · B · A , implies that (in physics notation, the transpose of a vector is
implied in a dot product) (H · µ · H) = (H · µ∗ · H∗ )t = H∗ · µ† · H = H∗ · µ · H. The last equality in the
∗ †

above is possible only if µ = µ† or that µ is hermitian.

Spin Angular Momentum, Complex Poynting’s Theorem, Lossless Condition, Energy Density107

above can be interpreted as stored energy density. Hence, the reactive power is proportional
to the time rate of change of the difference of the time-averaged energy stored in the magnetic
field and the electric field.

10.4 Energy Density in Dispersive Media

A dispersive medium alters our concept of what energy density is.4 To this end, we assume
that the field has complex ω dependence in ejωt , where ω = ω 0 − jω 00 , rather than real ω de-
pendence. We take the divergence of the complex power for fields with such time dependence,
and let ejωt be attached to the field. So E(t) and H(t) are complex field but not exactly like
phasors since they are not truly time harmonic. In other words, we let

E(r, t) = E(r, ω)ejωt , H(r, t) = H(r, ω)ejωt (10.4.1)

e e
The above, just like phasors, can be made to satisfy Maxwell’s equations where the time
derivative becomes jω but with complex ω. We can study the quantity E(r, t) × H∗ (r, t)
which has the unit of power density. In the real ω case, their time dependence will exactly
cancel each other and this quantity becomes complex power again, but not in the complex ω
case. Hence,

∇ · [E(t) × H∗ (t)] = H∗ (t) · ∇ × E(t) − E(t) · ∇ × H∗ (t)

= −H∗ (t) · jωµH(t) + E(t) · jω ∗ ε∗ E∗ (t) (10.4.2)

where Maxwell’s equations have been used to substitute for ∇ × E(t) and ∇ × H∗ (t). The
space dependence of the field is implied, and we assume a source-free medium so that J = 0.
∗
If E(t) ∼ ejωt , then, due to ω being complex, now H∗ (t) ∼ e−jω t , and the term like
E(t) × H∗ (t) is not truly time independent but becomes
∗ 00
E(t) × H∗ (t) ∼ ej(ω−ω )t
= e2ω t
(10.4.3)

And each of the term above will have similar time dependence. Writing (10.4.2) more explic-
itly, by letting ω = ω 0 − jω 00 , we have

∇ · [E(t) × H∗ (t)] = −j(ω 0 − jω 00 )µ(ω)|H(t)|2 + j(ω 0 + jω 00 )ε∗ (ω)|E(t)|2 (10.4.4)

Assuming that ω 00  ω 0 , or that the field is quasi-time-harmonic, we can let, after using
Taylor series approximation, that

∂µ(ω 0 ) 0
∂ε(ω )
µ(ω 0 − jω 00 ) ∼
= µ(ω 0 ) − jω 00 , ε(ω 0 − jω 00 ) ∼
= ε(ω 0 ) − jω 00 (10.4.5)
∂ω 0 ∂ω 0
4 The derivation here is inspired by H.A. Haus, Electromagnetic Noise and Quantum Optical Measurements

[82]. Generalization to anisotropic media is given by W.C. Chew, Lectures on Theory of Microwave and Optical
Waveguides [83].
108 Electromagnetic Field Theory

Using (10.4.5) in (10.4.4), and collecting terms of the same order, and ignoring (ω 00 )2 terms,
gives5

∇ · [E(t) × H∗ (t)] ∼
= − jω 0 µ(ω 0 )|H(t)|2 + jω 0 ε∗ (ω 0 )|E(t)|2
∂µ
− ω 00 µ(ω 0 )|H(t)|2 − ω 0 ω 00 0 |H(t)|2
∂ω
∂ε∗
− ω 00 ε∗ (ω 0 )|E(t)|2 − ω 0 ω 00 0 |E(t)|2 (10.4.6)
∂ω
The above can be rewritten as

∇ · [E(t) × H∗ (t)] ∼
= −jω 0 µ(ω 0 )|H(t)|2 − ε∗ (ω 0 )|E(t)|2
 
 0 0
∂ω 0 ε∗ (ω 0 )

00 ∂ω µ(ω ) 2 2
−ω |H(t)| + |E(t)| (10.4.7)
∂ω 0 ∂ω 0

The above approximation is extremely good when ω 00  ω 0 . For a lossless medium, ε(ω 0 ) and
µ(ω 0 ) are purely real, and the first term of the right-hand side is purely imaginary while the
second term is purely real. In the limit when ω 00 → 0, when half the imaginary part of the
above equation is taken, we have
 
1 ∗ 0 1 2 1 2
∇ · =m [E × H ] = −ω µ|H| − ε|E| (10.4.8)
2 2 2
which has the physical interpretation of reactive power as has been previously discussed.
When half the real part of (10.4.7) is taken, we obtain

ω 00 ∂ω 0 µ ∂ω 0 ε 2
 
1
∇ · <e[E × H∗ ] = − |H|2
+ |E| (10.4.9)
2 2 ∂ω 0 ∂ω 0
00
Since the right-hand side has time dependence of e2ω t , it can be written as

∂ 1 ∂ω 0 µ ∂ω 0 ε 2
 
1 ∂
∇ · <e[E × H∗ ] = − |H|2
+ |E| = − hWT i (10.4.10)
2 ∂t 4 ∂ω 0 ∂ω 0 ∂t
Therefore, the time-average stored energy density can be identified as

1 ∂ω 0 µ ∂ω 0 ε 2
 
2
hWT i = |H| + |E| (10.4.11)
4 ∂ω 0 ∂ω 0
For a non-dispersive medium, the above reverts back to a familiar expression,
1
µ|H|2 + ε|E|2

hWT i = (10.4.12)
4
which is what we have derived before.
In the above analysis, we have used a quasi-time-harmonic signal with exp(jωt) depen-
dence. In the limit when ω 00 → 0, this signal reverts back to a time-harmonic signal, and to
5 This is the general technique of perturbation expansion [45].
Spin Angular Momentum, Complex Poynting’s Theorem, Lossless Condition, Energy Density109

our usual interpretation of complex power. However, by assuming the frequency ω to have a
small imaginary part ω 00 , it forces the stored energy to grow very slightly, and hence, power
has to be supplied to maintain the growth of this stored energy. By so doing, it allows us
to identify the expression for energy density for a dispersive medium. These expressions for
energy density were not discovered until 1960 by Brillouin [84], as energy density times group
velocity should be power flow. More discussion on this topic can be found in Jackson [48].
It is to be noted that if the same analysis is used to study the energy storage in a capacitor
or an inductor, the energy storage formulas have to be accordingly modified if the capacitor
or inductor is frequency dependent.
110 Electromagnetic Field Theory
Lecture 11

Transmission Lines

Transmission lines represent one of the most important electromagnetic technologies. The
reason being that they can be described by simple theory, similar to circuit theory. As such,
the theory is within the grasp of most practicing electrical engineers. Moreover, transmission
line theory fills the gap in the physics of circuit theory: Circuit theory alone cannot describe
wave phenomena, but when circuit theory is augmented with transmission line theory, wave
phenomena with its corresponding wave physics start to emerge.

Even though circuit theory has played an indispensable role in the development of the
computer chip industry, it has to be embellished by transmission line theory, so that high-
speed circuits can be designed. Retardation effect, which causes time delay, clock skew, and
phase shift, can be modeled simply using transmission lines. Nowadays, commercial circuit
solver software have the capability of including transmission line as an element in modeling.

111
112 Electromagnetic Field Theory

11.1 Transmission Line Theory

Figure 11.1: Various kinds of transmission lines. Schematically, all of them can be modeled
by two parallel wires.

Transmission lines were the first electromagnetic waveguides ever invented. The were driven
by the needs in telegraphy technology. It is best to introduce transmission line theory from
the viewpoint of circuit theory, which is elegant and one of the simplest theories of electrical
engineering. This theory is also discussed in many textbooks and lecture notes. Transmission
lines are so important in modern day electromagnetic engineering, that most engineering
electromagnetics textbooks would be incomplete without introducing the topics related to
them [31, 32, 44, 50, 54, 65, 79, 83, 85, 86].
Circuit theory is robust and is not sensitive to the detail shapes of the components in-
volved such as capacitors or inductors. Moreover, many transmission line problems cannot
be analyzed simply when the full form of Maxwell’s equations is used,1 but approximate
solutions can be obtained using circuit theory in the long-wavelength limit. We shall show
that circuit theory is an approximation of electromagnetic field theory when the wavelength
is very long: the longer the wavelength, the better is the approximation [50].
Examples of transmission lines are shown in Figure 11.1. The symbol for a transmission
line is usually represented by two pieces of parallel wires, but in practice, these wires need
not be parallel as showin in Figure 11.2.
1 Usually called full-wave analysis.
Transmission Lines 113

Figure 11.2: A twisted pair transmission line where the two wires are not parallel to each
other (courtesy of slides by A. Wadhwa, A.L. Dal, N. Malhotra [87].)

Circuit theory also explains why waveguides can be made sloppily when wavelength is long
or the frequency low. For instance, in the long-wavelength limit, we can make twisted-pair
waveguides with abandon, and they still work well (see Figure 11.2). Hence, it is simplest
to first explain the propagation of electromagnetic signal on a transmission line using circuit
analysis.

11.1.1 Time-Domain Analysis

We will start with performing the time-domain analysis of a simple, infinitely long transmis-
sion line. Remember that two pieces of metal can accumulate attractive positive and negative
charges between them, giving rise to capacitive coupling, with electric fields that start with
positive charges and end with negative charges, and hence stored energy in the electric field.
Moreover, a piece of wire carrying a current generates a magnetic field, and hence, yields
stored energy in the magnetic field. These stored energies are the sources of the capacitive
and inductive effects. But these capacitive and inductive effects are distributed over the spa-
tial dimension of the transmission line. Therefore, it is helpful to think of the two pieces of
metal as consisting of small segments of metal concatenated together. Each of these segments
will have a small inductance, as well as a small capacitive coupling between them. Hence, we
can model two pieces of metal with a distributed lumped element model as shown in Figure
11.3. For simplicity, we assume the other conductor to be a ground plane, so that it need not
be approximated with lumped elements.

In the transmission line, the voltage V (z, t) and the current I(z, t) are functions of both
space z and time t, but we will model the space variation of the voltage and current with
discrete step approximations. The voltage varies from node to node while the current varies
from branch to branch of the lump-element model.
114 Electromagnetic Field Theory

Figure 11.3: A long transmission line can be replaced by a concatenation of many short
transmission lines. For each pair of short wires, there are capacitive coupling between them.
Furtheremore, when current flows in the wire, magnetic field is generated making them behave
like an inductor. Therefore, it can be replaced by a lumped-element approximation as shown.

Telegrapher’s Equations
First, we recall that the V-I relation of an inductor is
dI0
V0 = L0 (11.1.1)
dt
where L0 is the inductor, V0 is the time-varying voltage drop across the inductor, and I0 is
the current through the inductor. Then using this relation between nodes 1 and 2, we have
∂I
V − (V + ∆V ) = L∆z (11.1.2)
∂t
The left-hand side is the voltage drop across the inductor, while the right-hand side follows
from the aforementioned V-I relation of an inductor, but we have replaced L0 = L∆z. Here,
L is the inductance per unit length (line inductance) of the transmission line. And L∆z is
the incremental inductance due to the small segment of metal of length ∆z. In the above, we
assume that V = V (z, t) and I = I(z, t), so that time derivative is replaced by partial time
derivative. Then the above can be simplified to
∂I
∆V = −L∆z (11.1.3)
∂t
Next, we make use of the V-I relation for a capacitor, which is
dV0
I0 = C 0 (11.1.4)
dt
where C0 is the capacitor, I0 is the current through the capacitor, and V0 is a time-varying
voltage drop across the capacitor. Thus, applying this relation at node 2 gives
∂ ∂V
−∆I = C∆z (V + ∆V ) ≈ C∆z (11.1.5)
∂t ∂t
Transmission Lines 115

where C is the capacitance per unit length, and C∆z is the incremental capacitance between
the small piece of metal and the ground plane. In the above, we have used Kirchhoff current
law to surmise that the current through the capacitor is −∆I, where ∆I = I(z+∆z, t)−I(z, t).
In the last approximation in (11.1.5), we have dropped a term involving the product of ∆z
and ∆V , since it will be very small or second order in magnitude.
In the limit when ∆z → 0, one gets from (11.1.3) and (11.1.5) that
∂V (z, t) ∂I(z, t)
= −L (11.1.6)
∂z ∂t
∂I(z, t) ∂V (z, t)
= −C (11.1.7)
∂z ∂t
The above are the telegrapher’s equations.2 They are two coupled first-order equations,
and can be converted into second-order equations easily. Therefore,
∂2V ∂2V
2
− LC 2 = 0 (11.1.8)
∂z ∂t
∂2I ∂2I
− LC =0 (11.1.9)
∂z 2 ∂t2
The above are wave equations that we have previously studied, where the velocity of the wave
is given by
1
v=√ (11.1.10)
LC
Furthermore, if we assume that

V (z, t) = V0 f+ (z − vt), I(z, t) = I0 f+ (z − vt) (11.1.11)

corresponding to a right-traveling wave, they can be verified to satisfy (11.1.6) and (11.1.7)
by back substitution.
Consequently, we can easily show that
r
V (z, t) L
= = Z0 (11.1.12)
I(z, t) C
where Z0 is the characteristic impedance of the transmission line. The above ratio is only
true for one-way traveling wave, in this case, one that propagates in the +z direction.
For a wave that travels in the negative z direction, we can let,

V (z, t) = V0 f− (z + vt), I(z, t) = I0 f− (z + vt) (11.1.13)

one can easily show that

r
V (z, t) L
=− = −Z0 (11.1.14)
I(z, t) C
2 They can be thought of as the distillation of the Faraday’s law and Ampere’s law from Maxwell’s equations

without the source term. Their simplicity gives them an important role in engineering electromagnetics.
116 Electromagnetic Field Theory

Time-domain analysis is very useful for transient analysis of transmission lines, especially
when nonlinear elements are coupled to the transmission line.3 Another major strength
of transmission line model is that it is a simple way to introduce time-delay (also called
retardation) in a simple circuit model.4
Time delay is a wave propagation effect, and it is harder to incorporate into circuit theory
or a pure circuit model consisting of R, L, and C only. In circuit theory, where the wavelength
is assumed very long, Laplace’s equation is usually solved, which is equivalent to Helmholtz
equation with infinite wave velocity, namely,
ω2
lim ∇2 Φ(r) + Φ(r) = 0 =⇒ ∇2 Φ(r) = 0 (11.1.15)
c→∞ c2
Hence, events in Laplace’s equation happen instantaneously. In other words, circuit theory
assumes that the velocity of the wave is infinite, and there is no retardation effect. This
is only true or a good approximation when the size of the structure is small compared to
wavelength.

11.1.2 Frequency-Domain Analysis–the Power of Phasor Technique

Again!
Frequency domain analysis is very popular as it makes the transmission line equations very
simple–one just replace ∂/∂t → jω. Moreover, generalization to a lossy system is quite
straight forward. Furthermore, for linear time invariant systems, the time-domain signals can
be obtained from the frequency-domain data by performing a Fourier inverse transform since
phasors and Fourier transforms of a time variable are just related to each other by a constant.
The telegrapher’s equations (11.1.6) and (11.1.7) then in frequency domain become
d
V (z, ω) = −jωLI(z, ω) (11.1.16)
dz
d
I(z, ω) = −jωCV (z, ω) (11.1.17)
dz
The corresponding Helmholtz equations are then
d2 V
+ ω 2 LCV = 0 (11.1.18)
dz 2
d2 I
+ ω 2 LCI = 0 (11.1.19)
dz 2
The above are second ordinary differential equations, and the general solutions to the above
are
V (z) = V+ e−jβz + V− ejβz (11.1.20)
−jβz jβz
I(z) = I+ e + I− e (11.1.21)
3 Remember that we can only use frequency domain technique or Fourier transform for linear time-invariant

systems.
4 By a simple circuit model, we mean a model that has lumped elements such as R, L, and C as well as a

transmission line element.

Transmission Lines 117
√
where β = ω LC. This is similar to what we have seen previously for plane waves in the
one-dimensional wave equation in free space, where

Ex (z) = E0+ e−jk0 z + E0− ejk0 z (11.1.22)

√
where k0 = ω µ0 0 . We see much similarity between (11.1.20), (11.1.21), and (11.1.22).
To see the solution in the time domain, we let the phasor V± = |V± |ejφ± in (11.1.20),
and the voltage signal above can then be converted back to the time domain using the key
formula in phasor technique as

V (z, t) = <e{V (z, ω)ejωt } (11.1.23)

= |V+ | cos(ωt − βz + φ+ ) + |V− | cos(ωt + βz + φ− ) (11.1.24)

As can be seen, the first term corresponds to a right-traveling wave, while the second term is
a left-traveling wave.
Furthermore, if we assume only a one-way traveling wave to the right by letting V− =
I− = 0, then it can be shown that, for a right-traveling wave, using (11.1.16) or (11.1.17)
r
V (z) V+ L
= = = Z0 (11.1.25)
I(z) I+ C

where Z0 is the characteristic impedance. Since Z0 is real, it implies that V (z) and I(z) are
in phase.
Similarly, applying the same process for a left-traveling wave only, by letting V+ = I+ = 0,
then
r
V (z) V− L
= =− = −Z0 (11.1.26)
I(z) I− C
◦
In other words, for the left-traveling waves, the voltage and current are 180 out of phase.

11.2 Lossy Transmission Line

Figure 11.4: In a lossy transmission line, series resistance can be added to the series induc-
tance, and the shunt conductance can be added to the shun susceptance of the capacitor.
However, this problem is homomorphic to the lossless case in the frequency domain.
118 Electromagnetic Field Theory

The power of frequency domain analysis is revealed in the study of lossy transmission lines.
The previous analysis, which is valid for lossless transmission line, can be easily generalized
to the lossy case in the frequency domain. In using frequency domain and phasor technique,
impedances will become complex numbers as shall be shown.
To include loss, we use the lumped-element model as shown in Figure 11.4. One thing to
note is that jωL is actually the series line impedance of the lossless transmission line, while
jωC is the shunt line admittance of the same line. First, we can rewrite the expressions for
the telegrapher’s equations in (11.1.16) and (11.1.17) in terms of series line impedance and
shunt line admittance to arrive at
d
V = −ZI (11.2.1)
dz

d
I = −Y V (11.2.2)
dz
where Z = jωL and Y = jωC. The above can be easily generalized to the lossy case as shall
be shown.
The geometry in Figure 11.4 is topologically similar to, or homomorphic5 to the lossless
case in Figure 11.3. Hence, when lossy elements are added in the geometry, we can surmise
that the corresponding telegrapher’s equations are similar to those above. But to include loss,
we need only to generalize the series line impedance and shunt admittance from the lossless
case to lossy case as follows:

Z = jωL → Z = jωL + R (11.2.3)

Y = jωC → Y = jωC + G (11.2.4)

where R is the series line resistance, and G is the shunt line conductance, and now Z and
Y are the series impedance and shunt admittance, (which are complex number rather than
being pure imaginary), respectively. Then, the corresponding Helmholtz equations are

d2 V
− ZY V = 0 (11.2.5)
dz 2
d2 I
− ZY I = 0 (11.2.6)
dz 2
or
d2 V
− γ2V = 0 (11.2.7)
dz 2
d2 I
− γ2I = 0 (11.2.8)
dz 2
where γ 2 = ZY , or that one can also think of γ 2 = −β 2 by comparing with (11.1.18) and
(11.1.19). Then the above is homomorphic to the lossless case except that now, β is a complex
5 A math term for “similar in math structure”. The term is even used in computer science describing a

emerging field of homomorphic computing.

Transmission Lines 119

number, indicating that the field is decaying as it propagates. As before, the above are second
order one-dimensional Helmholtz equations where the general solutions are

V (z) = V+ e−γz + V− eγz (11.2.9)

−γz γz
I(z) = I+ e + I− e (11.2.10)

and
√ p
γ= ZY = (jωL + R)(jωC + G) = jβ (11.2.11)

Here, β = β 0 − jβ 00 is now a complex number. In other words,

0
z−β 00 z
e−γz = e−jβ

is an oscillatory and decaying wave. Or focusing on the voltage case,

00
z−jβ 0 z 00
z+jβ 0 z
V (z) = V+ e−β + V − eβ (11.2.12)

Again, letting V± = |V± |ejφ± , the above can be converted back to the time domain as

V (z, t) = <e{V (z, ω)ejωt } (11.2.13)

−β 00 z 0 β 00 z 0
= |V+ |e cos(ωt − β z + φ+ ) + |V− |e cos(ωt + β z + φ− ) (11.2.14)

The first term corresponds to a decaying wave moving to the right while the second term is
also a decaying wave but moving to the left. When there is no loss, or√R = G = 0, and from
(11.2.11), we retrieve the lossless case where β 00 = 0 and γ = jβ = jω LC.
Notice that for the lossy case, the characteristic impedance, which is the ratio of the
voltage to the current for a one-way wave, can similarly be derived using homomorphism:
r s r s
V+ V− L jωL Z jωL + R
Z0 = =− = = → Z0 = = (11.2.15)
I+ I− C jωC Y jωC + G

The above Z0 is manifestly a complex number. Here, Z0 is the ratio of the phasors of the
one-way traveling waves, and apparently, their current phasor and the voltage phasor will not
be in phase for lossy transmission line.
In the absence of loss, the above again becomes
r
L
Z0 = (11.2.16)
C
the characteristic impedance for the lossless case previously derived.
120 Electromagnetic Field Theory
Lecture 12

More on Transmission Lines

As mentioned before, transmission line theory is indispensable in microwave engineering these

days. The theory is the necessary augmentation of circuit theory for higher frequency analysis,
and it is also indispensable to integrated circuit designers as computer clock rate becomes
faster. Over the years, engineers have developed some very useful tools and measurement
techniques to expand the design space of circuit designers. We will learn some of these tools
in this lecture.1
As seen in the previous lecture, the telegrapher’s equations are similar to the one-dimensional
form of Maxwell’s equations, and can be thought of as Maxwell’s equations in their simplest
form. Therefore, they entail a subset of the physics seen in the full Maxwell’s equations.

12.1 Terminated Transmission Lines

Figure 12.1: A schematic for a transmission line terminated with an impedance load ZL at
z = 0.

For an infinitely long transmission line, the solution consists of the linear superposition of a
wave traveling to the right plus a wave traveling to the left. If transmission line is terminated
1 Some of you may have studied this topic in your undergraduate electromagnetics course. However, this

topic is important, and you will have to muster your energy to master this knowledge again:)

121
122 Electromagnetic Field Theory

by a load as shown in Figure 12.1, a right-traveling wave will be reflected by the load, and
in general, the wave on the transmission line will be a linear superposition of the left and
right traveling waves. To simplify the analysis, we will assume that the line is lossless. The
generalization to the lossy case is quite straightforward. Thus, we assume that

V (z) = a+ e−jβz + a− ejβz = V+ (z) + V− (z) (12.1.1)

√
where β = ω LC. In the above, in general, a+ 6= a− . Besides, this is a linear system; hence,
we can define the right-going wave V+ (z) to be the input, and that the left-going wave V− (z)
to be the output as due to the reflection of the right-going wave V+ (z). Or we can define the
amplitude of the left-going reflected wave a− to be linearly related to the amplitude of the
right-going or incident wave a+ . In other words, at z = 0, we can let

V− (z = 0) = ΓL V+ (z = 0) (12.1.2)

thus, using the definition of V+ (z) and V− (z) as implied in (12.1.1), we have

a− = ΓL a+ (12.1.3)

where ΓL is the termed the reflection coefficient. Hence, (12.1.1) becomes

V (z) = a+ e−jβz + ΓL a+ ejβz = a+ e−jβz + ΓL ejβz

(12.1.4)

The corresponding current I(z) on the transmission line is given by using the telegrapher’s
equations as previously defined. By recalling that

dV
= −jωLI
dz
then for the general case,
a+ −jβz
− ΓL ejβz


I(z) = e (12.1.5)
Z0

Notice the sign change in the second term of the above expression.
Similar to ΓL , a general reflection coefficient can be defined (which is a function of z)
relating the left-traveling and right-traveling wave at location z such that

V− (z) = a− ejβz a− ejβz

Γ(z) = = = ΓL e2jβz (12.1.6)
V+ (z) = a+ e−jβz a+ e−jβz

Of course, Γ(z = 0) = ΓL . Furthermore, due to the V-I relation at an impedance load, we

must have2
V (z = 0)
= ZL (12.1.7)
I(z = 0)
2 One can also look at this from a differential equation viewpoint that this is a boundary condition.
More on Transmission Lines 123

or that using (12.1.4) and (12.1.5) with z = 0, the left-hand side of the above can be rewritten,
and we have

1 + ΓL 1 + ΓL ZL
Z0 = ZL , or = = ZnL (12.1.8)
1 − ΓL 1 − ΓL Z0

From the above, we can solve for ΓL in terms of ZnL to get

ZnL − 1 ZL − Z0
ΓL = = (12.1.9)
ZnL + 1 ZL + Z0

Thus, given the termination load ZL and the characteristic impednace Z0 , the reflection
coefficient ΓL can be found, or vice versa. Or given ΓL , the normalized load impedance,
ZnL = ZL /Z0 , can be found. It is seen that ΓL = 0 if ZL = Z0 . Thus a right-traveling wave
will not be reflected and the left-traveling is absent. This is the case of a matched load.
When there is no reflection, all energy of the right-traveling wave will be totally absorbed by
the load.
In general, we can define a generalized impedance at z 6= 0 to be

V (z) a+ (e−jβz + ΓL ejβz )

Z(z) = = 1 −jβz − Γ ejβz )
I(z) Z0 a+ (e L

1 + ΓL e2jβz 1 + Γ(z)
= Z0 = Z0 (12.1.10)
1 − ΓL e2jβz 1 − Γ(z)

where Γ(z) defined in (12.1.6) is used. The above can also be written as

1 + Γ(z)
Zn (z) = Z(z)/Z0 = (12.1.11)
1 − Γ(z)

where Zn (z) is the normalized generalized impedance. Conversely, one can write the above
as

Zn (z) − 1 Z(z) − Z0
Γ(z) = = (12.1.12)
Zn (z) + 1 Z(z) + Z0

From (12.1.10) above, one gets

1 + ΓL e2jβz
Z(z) = Z0 (12.1.13)
1 − ΓL e2jβz

One can show that by setting z = −l, using (12.1.9), and after some algebra,

ZL + jZ0 tan βl
Z(−l) = Z0 (12.1.14)
Z0 + jZL tan βl
124 Electromagnetic Field Theory

12.1.1 Shorted Terminations

Figure 12.2: The input reactance (X) of a shorted transmission line as a function of its length
l.

From (12.1.14) above, when we have a short such that ZL = 0, then

Z(−l) = jZ0 tan(βl) = jX (12.1.15)

When βl  1, then tan βl ≈ βl, and (12.1.15) becomes

Z(−l) ∼
= jZ0 βl (12.1.16)

p √
After using that Z0 = L/C and that β = ω LC, (12.1.16) becomes

Z(−l) ∼
= jωLl (12.1.17)

The above implies that a short length of transmission line connected to a short as a
load looks like an inductor with Leff = Ll, since much current will pass through this short
producing a strong magnetic field with stored magnetic energy. Remember here that L is the
line inductance, or inductance per unit length.
On the other hand, when the length of the shorted line increases, due to the standing wave
on the transmission line, certainly parts of the line will have charge accumulation giving rise
to strong electric field, while other parts have current flow giving rise to strong magnetic field.
Depending on this standing wave pattern, the line can become either capacitive or inductive.
More on Transmission Lines 125

12.1.2 Open Terminations

Figure 12.3: The input reactance (X) of an open transmission line as a function of its length
l.

When we have an open circuit such that ZL = ∞, then from (12.1.14) above

Z(−l) = −jZ0 cot(βl) = jX (12.1.18)

Then, when βl  1, cot(βl) ≈ 1/βl

Z0
Z(−l) ≈ −j (12.1.19)
βl
√ p
And then, again using β = ω LC, Z0 = L/C

1
Z(−l) ≈ (12.1.20)
jωCl
Hence, an open-circuited terminated short length of transmission line appears like an effective
capacitor with Ceff = Cl. Again, remember here that C is line capacitance or capacitance
per unit length.
Again, as shown in Figure 12.3, the impedance at z = −l is purely reactive, and goes
through positive and negative values due to the standing wave set up on the transmission
line. Therefore, by changing the length of l, one can make a shorted or an open terminated
line look like an inductor or a capacitor depending on its length l. This effect is shown in
Figures 12.2 and 12.3. Moreover, the reactance X becomes infinite or zero with the proper
choice of the length l. These are resonances or anti-resonances of the transmission line, very
much like an LC tank circuit. An LC circuit can look like an open or a short circuit at
resonances and depending on if they are connected in parallel or in series.
126 Electromagnetic Field Theory

12.2 Smith Chart

In general, from (12.1.13) and (12.1.14), a length of transmission line can transform a load ZL
to a range of possible complex values Z(−l). To understand this range of values better, we
can use the Smith chart (invented by P.H. Smith 1939 before the advent of the computer) [88].
The Smith chart is essentially a graphical calculator for solving transmission line problems. It
has been used so much by microwave engineers during the early days that its use has imposed
a strong impression on these engineers: it also has become an indispensable visual aid for
understanding and solving microwave engineering problems.

Equation (12.1.12) indicates that there is a unique map between the normalized impedance
Zn (z) = Z(z)/Z0 and reflection coefficient Γ(z). In the normalized impedance form where
Zn = Z/Z0 , from (12.1.10) and (12.1.12)

Zn − 1 1+Γ
Γ= , Zn = (12.2.1)
Zn + 1 1−Γ

Equations in (12.2.1) are related to a bilinear transform in complex variables [89]. It is a

conformal map that maps circles to circles. Such a map is shown in Figure 12.4, where lines
on the right-half of the complex Zn plane are mapped to the circles on the complex Γ plane.
Since straight lines on the complex Zn plane are circles with infinite radii, they are mapped
to circles on the complex Γ plane. The Smith chart shown on Figure 12.5 allows one to obtain
the corresponding Γ given Zn and vice versa as indicated in (12.2.1), but using a graphical
calculator or the Smith chart.

Notice that the imaginary axis on the complex Zn plane maps to the circle of unit radius on
the complex Γ plane. All points on the right-half plane are mapped to within the unit circle.
The reason being that the right-half plane of the complex Zn plane corresponds to passive
impedances that will absorb energy. Hence, such an impedance load will have reflection
coefficient with amplitude less than one, which are points within the unit circle.

On the other hand, the left-half of the complex Zn plane corresponds to impedances with
negative resistances. These will be active elements that can generate energy, and hence,
yielding |Γ| > 1, and will be outside the unit circle on the complex Γ plane.

Another point to note is that points at infinity on the complex Zn plane map to the point
at Γ = 1 on the complex Γ plane, while the point zero on the complex Zn plane maps to
Γ = −1 on the complex Γ plane. These are the reflection coefficients of an open-circuit load
and a short-circuit load, respectively. For a matched load, Zn = 1, and it maps to the zero
point on the complex Γ plane implying no reflection.
More on Transmission Lines 127

n −1
Figure 12.4: Bilinear map of the formulae Γ = Z 1+Γ
Zn +1 , and Zn = 1−Γ . The chart on the right,
called the Smith chart, allows the values of Zn to be determined quickly given Γ, and vice
versa.

The Smith chart also allows one to quickly evaluate the expression

Γ(−l) = ΓL e−2jβl (12.2.2)

and its corresponding Zn not by using (12.2.1) via a calculator, but by using a graphical
calculator—the Smith chart. Since β = 2π/λ, it is more convenient to write βl = 2πl/λ, and
measure the length of the transmission line in terms of wavelength. To this end, the above
becomes

Γ(−l) = ΓL e−4jπl/λ (12.2.3)

For increasing l, one moves away from the load to the generator (or source), l increases,
and the phase is decreasing because of the negative sign. So given a point for ΓL on the Smith
chart, one has negative phase or decreasing phase by rotating the point clockwise. Also, due
to the exp(−4jπl/λ) dependence of the phase, when l = λ/4, the reflection coefficient rotates
a half circle around the chart. And when l = λ/2, the reflection coefficient will rotate a full
circle, or back to the original point. Therefore, on the edge of the Smith chart, there are
indication as to which direction one should rotate if one were to move toward the generator
or toward the load.
Also, for two points diametrically opposite to each other on the Smith chart, Γ changes
sign, and it can be shown easily from (12.2.1) that the normalized impedances are reciprocal
of each other. Hence, the Smith chart can also be used to find the reciprocal of a complex
number quickly. A full blown Smith chart is shown in Figure 12.5.
128 Electromagnetic Field Theory

Figure 12.5: The Smith chart in its full glory. It was invented in 1939 before the age of digital
computers, but it still allows engineers to do mental estimations and rough calculations with
it, because of its simplicity.

12.3 VSWR (Voltage Standing Wave Ratio)

From the previous section, one sees that the voltage and current are not constant in a trans-
mission line. Therefore, one surmises that measuring the impedance of a device at microwave
frequency is a tricky business. At low frequency, one can use an ohm meter with two wire
probes to do such a measurement. But at microwave frequency, two pieces of wire become
More on Transmission Lines 129

inductors, and two pieces of metal become capacitors. More sophisticated ways to measure
the impedance need to be designed as described below.
Due to the interference between that forward traveling wave and the backward traveling
wave, V (z) is a function of position z on a terminated transmission line and it is given as
V (z) = V0 e−jβz + V0 ejβz ΓL
= V0 e−jβz 1 + ΓL e2jβz

= V0 e−jβz (1 + Γ(z)) (12.3.1)

where we have used (12.1.6) for Γ(z) with γ = jβ. Hence, V (z) is not a constant but
dependent on z, or
|V (z)| = |V0 ||1 + Γ(z)| (12.3.2)
For lack of a better name, this is called the standing wave, even though it is not truly a
standing wave.
In Figure 12.6, the relationship between variation of 1 + Γ(z) as z varies is shown.

Figure 12.6: The voltage amplitude on a transmission line depends on |V (z)|, which is pro-
portional to |1 + Γ(z)| per equation (12.3.2). This figure shows how |1 + Γ(z)| varies as z
varies on a transmission line.

Using the triangular inequality, one gets the lower and upper bounds or that
|V0 |(1 − |Γ(z)|) ≤ |V (z)| ≤ |V0 |(1 + |Γ(z)|) (12.3.3)
But from (12.1.6) and that β is pure real for a lossless line, then |Γ(z)| = |ΓL |; hence
Vmin = |V0 |(1 − |ΓL |) ≤ |V (z)| ≤ |V0 |(1 + |ΓL |) = Vmax (12.3.4)
The voltage standing wave ratio, VSWR is defined to be
Vmax 1 + |ΓL |
VSWR = = (12.3.5)
Vmin 1 − |ΓL |
130 Electromagnetic Field Theory

Conversely, one can invert the above to get

VSWR − 1
|ΓL | = (12.3.6)
VSWR + 1

Hence, the knowledge of voltage standing wave pattern (VSWP), as shown in Figure 12.7,
yields the knowledge of |ΓL | the amplitude of ΓL . Notice that the relations between VSWR
and |ΓL | are homomorphic to those between Zn and Γ. Therefore, the Smith chart can also
be used to evaluate the above equations.

Figure 12.7: The voltage standing wave pattern (VSWP) as a function of z on a load-
terminated transmission line.

The phase of ΓL can also be determined from the measurement of the voltage standing
wave pattern. The location of ΓL in Figure 12.6 is determined by the phase of ΓL . Hence,
the value of d1 in Figure 12.6 is determined by the phase of ΓL as well. The length of the
transmission line waveguide needed to null the original phase of ΓL to bring the voltage
standing wave pattern to a maximum value at z = −d1 is shown in Figure 12.7. Thus, d1 is
the value where the following equation is satisfied:

|ΓL |ejφL e−4πj(d1 /λ) = |ΓL | (12.3.7)

Therefore, by measuring the voltage standing wave pattern, one deduces both the amplitude
and phase of ΓL . From the complex value ΓL , one can determine ZL , the load impedance
from the Smith chart.
More on Transmission Lines 131

Figure 12.8: A slotted-line equipment which consists of a coaxial waveguide with a slot
opening at the top to allow the measurement of the field strength and hence, the voltage
standing wave pattern in the waveguide (courtesy of Microwave101.com).

In the old days, the voltage standing wave pattern was measured by a slotted-line equip-
ment which consists of a coaxial waveguide with a slot opening as shown in Figure 12.8. A
field probe can be inserted into the slotted line to determine the strength of the electric field
inside the coax waveguide. A typical experimental setup for a slotted line measurement is
shown in Figure 12.9. A generator source, with low frequency modulation, feeds microwave
energy into the coaxial waveguide. The isolator, allowing only the unidirectional propagation
of microwave energy, protects the generator. The attenuator protects the slotted line equip-
ment. The wavemeter is an adjustable resonant cavity. When the wavemeter is tuned to the
frequency of the microwave, it siphons off some energy from the source, giving rise to a dip
in the signal of the SWR meter (a short for voltage-standing-wave-ratio meter). Hence, the
wavemeter measures the frequency of the microwave.
The slotted line probe is usually connected to a square law detector with a rectifier that
converts the microwave signal to a low-frequency signal. In this manner, the amplitude of
the voltage in the slotted line can be measured with some low-frequency equipment, such as
the SWR meter. Low-frequency equipment is a lot cheaper to make and maintain. That is
also the reason why the source is modulated with a low-frequency signal. At low frequencies,
circuit theory prevails, engineering and design are a lot simpler.
The above describes how the impedance of the device-under-test (DUT) can be measured
at microwave frequencies. Nowadays, automated network analyzers make these measurements
a lot simpler in a microwave laboratory. More resource on microwave measurements can be
found on the web, such as in [90].
Notice that the above is based on the interference of the two traveling wave on a ter-
minated transmission line. Such interference experiments are increasingly difficult in optical
frequencies because of the much shorter wavelengths. Hence, many experiments are easier to
perform at microwave frequencies rather than at optical frequencies.
Many technologies are first developed at microwave frequency, and later developed at
optical frequency. Examples are phase imaging, optical coherence tomography, and beam
steering with phase array sources. Another example is that quantum information and quan-
tum computing can be done at optical frequency, but the recent trend is to use artificial atoms
132 Electromagnetic Field Theory

working at microwave frequencies. Engineering with longer wavelength and larger component
is easier; and hence, microwave engineering.
Another new frontier in the electromagnetic spectrum is in the terahertz range. Due to
the dearth of sources in the terahertz range, and the added difficulty in having to engineer
smaller components, this is an exciting and a largely untapped frontier in electromagnetic
technology.

Figure 12.9: An experimental setup for a slotted line measurement (courtesy of Pozar and
Knapp, U. Mass [91]).
Lecture 13

Multi-Junction Transmission
Lines, Duality Principle

A simple extension of the transmission line theory of the previous lecture is to the case when
tranmission lines of different characteristic impedances and wave velocities are concatenated
together. Myriads of devices can be designed using such admixture of transmission lines,
transistors, and diodes.

Also, due to the symmetry of Maxwell’s equations between the electric field and the
magnetic field, once a set of solutions have been found for Maxwell’s equations, new solutions
can be found by symmetry arguments. This is known as duality principle in electromagnetics.

13.1 Multi-Junction Transmission Lines

The real world is usually more complex than the world of our textbooks. However, we need to
distill problems in the real world into simpler problems that we can explain with our textbook
examples. Figure 13.1 shows many real world technologies, but they can be approximated
with transmission line models as shall be seen.

133
134 Electromagnetic Field Theory

Figure 13.1: Different kinds of waveguides operating in different frequencies in power lines,
RF circuits, microwave circuits, and optical fibers. Their salient physics or features can be
captured or approximated by transmission lines (courtesy of Owen Casha).

An area where multi-junction transmission lines play an important role is in the microwave
integrated circuit (MIC) area and the monolithic microwave integrated circuit (MMIC) area.
An MMIC circuit is shown in Figure 13.2. Many of the components can be approximated by
multi-junction transmission lines. They are clear motivation for studying this topic.

Figure 13.2: A generic GaAs monolithic microwave integrated circuit (MMIC). They are the
motivation for studying multi-junction transmission lines (courtesy of Wikipedia).
Multi-Junction Transmission Lines, Duality Principle 135

By concatenating sections of transmission lines of different characteristic impedances, a

large variety of devices such as resonators, filters, radiators, and matching networks can be
formed. We will start with a single junction transmission line first. A good reference for
such problem is the book by Collin [92], but much of the treatment here is not found in any
textbooks.

13.1.1 Single-Junction Transmission Lines

Consider two transmission lines connected at a single junction as shown in Figure 13.3. For
simplicity, we assume that the transmission line to the right is infinitely long so that a right-
traveling wave is not reflected; namely, there is no reflected wave. And we assume that the
two transmission lines have different characteristic impedances, Z01 and Z02 .

Figure 13.3: A single junction transmission line can be modeled by a equivalent transmission
line terminated in a load Zin2 .

The impedance of the transmission line at junction 1 looking to the right,using the formula
from previously derived,1 is
1 + ΓL,∞ e−2jβ2 l2
Zin2 = Z02 = Z02 (13.1.1)
1 − ΓL,∞ e−2jβ2 l2
Since no reflected wave exists, ΓL,∞ = 0, the above is just Z02 . As a result, transmission line
1 sees a load of ZL = Zin2 = Z02 hooked to its end. The equivalent circuit is shown in Figure
1 We should always remember that the relations between the reflection coefficient Γ and the normalized
n −1
impedance Zn are Γ = ZZ +1
1+Γ
and Zn = 1−Γ .
n
136 Electromagnetic Field Theory

13.3 as well. Hence, we deduce that the reflection coefficient at junction 1 between line 1 and
line 2, using the knowledge from the previous lecture, is Γ12 , and is given by

ZL − Z01 Zin2 − Z01 Z02 − Z01

Γ12 = = = (13.1.2)
ZL + Z01 Zin2 + Z01 Z02 + Z01

where we have assumed that ZL = Zin2 = Z02 .

13.1.2 Two-Junction Transmission Lines

Figure 13.4: A single-junction transmission line with a load ZL at the far (load) end of the
second line. But it can be reduced to the equivalent circuit shown in the bottom of Figure
13.3.

Now, we look at the two-junction case. To this end, we first look at when line 2 is terminated
by a load ZL at its load end as shown in Figure 13.4. Then, using the formula derived in the
previous lecture,

1 + Γ(−l2 ) 1 + ΓL2 e−2jβ2 l2

Zin2 = Z02 = Z02 (13.1.3)
1 − Γ(−l2 ) 1 − ΓL2 e−2jβ2 l2

where we have used the fact that Γ(−l2 ) = ΓL2 e−2jβ2 l2 . It is to be noted that here, using
knowledge from the previous lecture, that the reflection coefficient at the load end of line 2 is

ZL − Z02
ΓL2 = (13.1.4)
ZL + Z02

Now, line 1 sees a load of Zin2 hooked at its end. The equivalent circuit is the same as
that shown in Figure 13.3. The generalized reflection coefficient at junction 1, which includes
all the reflection of waves from its right, is now

Zin2 − Z01
Γ̃12 = (13.1.5)
Zin2 + Z01
Multi-Junction Transmission Lines, Duality Principle 137

Substituting (13.1.3) into (13.1.5), we have

1+Γ
Z02 ( 1−Γ ) − Z01
Γ̃12 = 1+Γ
(13.1.6)
Z02 ( 1−Γ ) + Z01

where Γ = ΓL2 e−2jβ2 l2 . The above can be rearranged to give

Z02 (1 + Γ) − Z01 (1 − Γ)
Γ̃12 = (13.1.7)
Z02 (1 + Γ) + Z01 (1 − Γ)

Finally, by further rearranging terms, after some somewhat tedious algebra, it can be shown
that the above becomes
Γ12 + Γ Γ12 + ΓL2 e−2jβ2 l2
Γ̃12 = = (13.1.8)
1 + Γ12 Γ 1 + Γ12 ΓL2 e−2jβ2 l2
where Γ12 , the local reflection coefficient at the junction between line 1 and line 2, is given
by (13.1.2), and Γ = ΓL2 e−2jβ2 l2 is the general reflection coefficient2 at z = −l2 due to the
load ZL . In other words,
ZL − Z02
ΓL2 = (13.1.9)
ZL + Z02

Figure 13.5: A two-junction transmission line with a load ZL at the far end. The input
impedance looking in from the far left can be found recursively using the formula (13.1.12)
and (13.1.13).

Equation (13.1.8) is a powerful formula for multi-junction transmission lines. Imagine

now that we add another section of transmission line as shown in Figure 13.5. We can use
2 We will use the term “general reflection coefficient” at location z to mean the ratio between the amplitudes

of the left-traveling wave and the right-traveling wave on a transmission line.

138 Electromagnetic Field Theory

the aforementioned method to first find Γ̃23 , the generalized reflection coefficient at junction
2. Using formula (13.1.8), it is given by

Γ23 + ΓL3 e−2jβ3 l3

Γ̃23 = (13.1.10)
1 + Γ23 ΓL3 e−2jβ3 l3

where ΓL3 is the load reflection coefficient due to the load ZL hooked to the end of transmission
line 3 as shown in Figure 13.5. Here, it is given as

ZL − Z03
ΓL3 = (13.1.11)
ZL + Z03

Given the knowledge of Γ̃23 , we can use (13.1.8) again to find the new Γ̃12 at junction 1.
It is now

Γ12 + Γ̃23 e−2jβ2 l2

Γ̃12 = (13.1.12)
1 + Γ12 Γ̃23 e−2jβ2 l2

The equivalent circuit is again that shown in Figure 13.3. Therefore, we can use (13.1.12)
recursively to find the generalized reflection coefficient for a multi-junction transmission line.
Once the reflection coefficient is known, the impedance at that location can also be found.
For instance, at junction 1, the impedance is now given by

1 + Γ̃12
Zin2 = Z01 (13.1.13)
1 − Γ̃12

instead of (13.1.3). In the above, Z01 is used because the generalized reflection coefficient
Γ̃12 is the total reflection coefficient for an incident wave from transmission line 1 that is sent
toward the junction 1. Previously, Z02 was used in (13.1.3) because the reflection coefficients
in that equation was for an incident wave sent from transmission line 2.
If the incident wave were to have come from line 2, then one can write Zin2 as

1 + Γ̃23 e−2jβ2 l2
Zin2 = Z02 (13.1.14)
1 − Γ̃23 e−2jβ2 l2

With some algebraic manipulation, it can be shown that (13.1.13) and (13.1.14) are identical.
But (13.1.13) is closer to an experimental scenario where one measures the reflection coefficient
by sending a wave from line 1 with no knowledge of what is to the right of junction 1.
Transmission lines can be made easily in microwave integrated circuit (MIC) by etching
or milling. A picture of a microstrip line waveguide or transmission line is shown in Figure
13.6.
Multi-Junction Transmission Lines, Duality Principle 139

Figure 13.6: Schematic of a microstrip line with the signal line above, and a ground plane
below (left). A strip line with each strip carrying currents of opposite polarity (right). A
ground plane is not needed in the second case.

13.1.3 Stray Capacitance and Inductance

Figure 13.7: A general microwave integrated circuit with different kinds of elements.

The junction between two transmission lines is not as simple as we have assumed. In the real
world, or in MIC, the waveguide junction has discontinuities in line width, or shape. This can
give rise to excess charge cumulation or constricted current flow. Excess charge gives rise to
excess electric field which corresponds to excess electric stored energy. This can be modeled
by stray or parasitic capacitances.
Alternatively, there could be excess or constricted current flow that give rise to excess
magnetic field. Excess magnetic field gives rise to excess magnetic stored energy. This can
be modeled by stray or parasitic inductances. Hence, a junction can be approximated by
a circuit model as shown in Figure 13.8 to account for these effects. The Smith chart or
the method we have outlined above can still be used to solve for the input impedances of a
140 Electromagnetic Field Theory

transmission circuit when these parasitic circuit elements are added.

Notice that when the frequency is zero or low, these stray capacitances and inductances are
negligible, we retrieve the simple junction model. But since their impedance and admittance
are jωLs and jωCs , respectively, they are non-negligible and are instrumental in modeling
high frequency circuits.

Figure 13.8: A junction between two microstrip lines can be modeled with stray junction
capacitance and stray inductances. The capacitance is used to account for excess charges
at the junction, while the inductances model the excess current at the junction. They are
important as the frequency increases.

13.2 Duality Principle

Duality principle exploits the inherent symmetry of Maxwell’s equations. Once a set of E
and H fields has been found to solve Maxwell’s equations for a certain geometry, another set
for a similar geometry can be found by invoking this principle. Maxwell’s equations in the
frequency domain, including the fictitious magnetic sources, are

∇ × E(r, ω) = −jωB(r, ω) − M(r, ω) (13.2.1)

∇ × H(r, ω) = jωD(r, ω) + J(r, ω) (13.2.2)
∇ · B(r, ω) = %m (r, ω) (13.2.3)
∇ · D(r, ω) = %(r, ω) (13.2.4)

One way to make Maxwell’s equations invariant is to do the following substitutions.

E → H, H → −E, D → B, B → −D (13.2.5)

M → −J, J → M, %m → −%, % → %m (13.2.6)

The above swaps retain the right-hand rule for plane waves. When material media is included,
such that D = ε · E, B = µ · H, for anisotropic media, Maxwell’s equations become

∇ × E = −jωµ · H − M (13.2.7)

∇ × H = jωε · E + J (13.2.8)
Multi-Junction Transmission Lines, Duality Principle 141

∇ · µ · H = %m (13.2.9)

∇·ε·E=% (13.2.10)

In addition to the above swaps, one need further to swap the material parameters, namely,

µ → ε, ε→µ (13.2.11)

13.2.1 Unusual Swaps3

There are other swaps where seemingly the right-hand rule is not preserved, e.g.,

E → H, H → E, M → −J, J → −M, (13.2.12)

%m → −%, % → −%m , µ → −ε, ε → −µ (13.2.13)

The above swaps will leave Maxwell’s equations invariant, but when applied to a plane wave,
the right-hand rule seems violated.

The deeper reason is that solutions to Maxwell’s equations are not unique, since there is
a time-forward as well as a time-reverse solution. In the frequency domain, this shows up in
√
the choice of the sign of the k vector where in a plane wave k = ±ω µε. When one does
a swap of µ → −ε and ε → −µ, k is still indeterminate, and one can always choose a root
where the right-hand rule is retained.

3 This section can be skipped on first reading.

142 Electromagnetic Field Theory

13.3 Fictitious Magnetic Currents

Figure 13.9: Sketches of the electric field due to an electric dipole and the magnetic field due
to a electric current loop. The E and H fields have the same pattern, and can be described
by the same formula.

Even though magnetic charges or monopoles do not exist, magnetic dipoles do. For instance,
a magnet can be regarded as a magnetic dipole. Also, it is believed that electrons have spins,
and these spins make electrons behave like tiny magnetic dipoles in the presence of a magnetic
field.
Also if we form electric current into a loop, it produces a magnetic field that looks like the
electric field of an electric dipole. This resembles a magnetic dipole field. Hence, a magnetic
dipole can be made using a small electric current loop (see Figure 13.9). Because of these
similarities, it is common to introduce fictitious magnetic charges and magnetic currents into
Maxwell’s equations. One can think that these magnetic charges always occur in pair and
together. Thus, they do not contradict the absence of magnetic monopole.
The electric current loops can be connected in series to make a toroidal antenna as shown
in Figure 13.10. The toroidal antenna is used to drive a current in an electric dipole. Notice
that the toroidal antenna ressembles the primary winding of a transformer circuit. In essence,
Multi-Junction Transmission Lines, Duality Principle 143

the toroidal loops, which mimic a magnetic current loop, produces an electric field that will
drive current through the electric dipole. This is dual to the fact that an electric current loop
produces a magnetic field.

Figure 13.10: A toroidal antenna used to drive an electric current through a conducting
cylinder of a dipole. It works similarly to a transformer: one can think of them as the
primary and secondary turns (windings) of a transformer (courtesy of Q. S. Liu [93]).
144 Electromagnetic Field Theory
Lecture 14

Reflection, Transmission, and

Interesting Physical Phenomena

We have seen the derivation of a reflection coefficient in a transmission line that relates the
amplitude of the reflected wave to that of the incident wave. By doing so, we have used a
simplified form of Maxwell’s equations, the telegrapher’s equations which are equations in
one dimension. Here, we will solve Maxwell’s equations in its full glory, but in order to do so,
we will look at a very simple problem of plane wave reflection and transmission at a single
plane interface.
This will give rise to the Fresnel reflection and transmission coefficients, and embedded
in them are interesting physical phenomena. We will study these interesting phenomena as
well.
(Much of the contents of this lecture can be found in Kong, and also the ECE 350X
lecture notes. They can be found in many textbooks, even though the notations can be
slightly different [31, 32, 44, 50, 54, 65, 79, 83, 85, 86].)

14.1 Reflection and Transmission—Single Interface Case

We will derive the plane-wave reflection coefficients for the single interface case between two
dielectric media. These reflection coefficients are also called the Fresnel reflection coefficients
because they were first derived by Austin-Jean Fresnel (1788-1827). Note that he lived before
the completion of Maxwell’s equations in 1865. But when Fresnel derived the reflection
coefficients in 1823, they were based on the elastic theory of light; and hence, the formulas
are not exactly the same as what we are going to derive (see Born and Wolf, Principles of
Optics, p. 40 [58]).
The single plane interface, plane wave reflection and transmission problem, with com-
plicated mathematics, is homomorphic to the transmission line problem. The complexity
comes because we have to keep track of the 3D polarizations of the electromagnetic fields in
this case. We shall learn later that the mathematical homomorphism can be used to exploit

145
146 Electromagnetic Field Theory

the simplicity of transmission line theory in seeking the solutions to the multiple dielectric
interface problems.

14.1.1 TE Polarization (Perpendicular or E Polarization)1

Figure 14.1: A schematic showing the reflection of the TE polarization wave impinging on a
dielectric interface.

To set up the above problem, the wave in Region 1 can be written as the superposition or sum
of the incident plane wave and reflected plane wave. Here, Ei (r) and Er (r) are the incident
and reflected plane waves, respectively. The total field is Ei (r) + Er (r) which are the phasor
representations of the fields. We assume plane wave polarized in the y direction where the
wave vectors are βi = x̂βix + ẑβiz , βr = x̂βrx − ẑβrz , βt = x̂βtx + ẑβtz , respectively for the
incident, reflected, and transmitted waves. Then, from Section 7.3, we have

Ei = ŷE0 e−jβi ·r = ŷE0 e−jβix x−jβiz z (14.1.1)

which represents a uniform incident plane wave, and

Er = ŷRT E E0 e−jβr ·r = ŷRT E E0 e−jβrx x+jβrz z (14.1.2)

which is a uniform reflected wave. In Region 2, we only have transmitted plane wave; hence

Et = ŷT T E E0 e−jβt ·r = ŷT T E E0 e−jβtx x−jβtz z (14.1.3)

1 These polarizations are also variously know as TE , or the s and p polarizations, a descendent from the
z
notations for acoustic waves where s and p stand for shear and pressure waves respectively.
Reflection, Transmission, and Interesting Physical Phenomena 147

In the above, the incident wave is known and hence, E0 is known. From (14.1.2) and (14.1.3),
RT E and T T E are unknowns yet to be sought. To find them, we need two boundary conditions
to yield two equations.2 These boundary conditions are tangential E field continuous and
tangential H field continuous, which are n̂ × E continuous and n̂ × H continuous conditions
at the interface.
Imposing n̂ × E continuous at z = 0, we get

E0 e−jβix x + RT E E0 e−jβrx x = T T E E0 e−jβtx x , ∀x (14.1.4)

where ∀ implies “for all”. In order for the above to be valid for all x, it is necessary that
βix = βrx = βtx , which is also known as the phase matching condition.3 From the above, by
letting βix = βrx = β1 sin θi = β1 sin θr , we obtain that θr = θi or that the law of reflection
that the angle of reflection is equal to the angle of incidence. By letting βix = β1 sin θi =
βtx = β2 sin θt , we obtain Snell’s law that β1 sin θi = β2 sin θt . (This law of refraction that
was also known in the Islamic world in the 900 AD. [94]).
The exponential terms or the phase terms on both sides of (14.1.4) are the same. Now,
canceling common terms on both sides of the equation (14.1.4), the above simplifies to

1 + RT E = T T E (14.1.5)

To impose n̂ × H continuous, one needs to find the H field using ∇ × E = −jωµH. By

letting ∇ with −jβ, then we have H = −jβ × E/(−jωµ) = β × E/(ωµ). By so doing4

βi × Ei βi × ŷ ẑβix − x̂βiz
Hi = = E0 e−jβi ·r = E0 e−jβi ·r (14.1.6)
ωµ1 ωµ1 ωµ1
βr × Er βr × ŷ T E ẑβrx + x̂βrz T E
Hr = = R E0 e−jβr ·r = R E0 e−jβr ·r (14.1.7)
ωµ1 ωµ1 ωµ1
βt × Et βt × ŷ T E ẑβtx − x̂βtz T E
Ht = = T E0 e−jβt ·r = T E0 e−jβt ·r (14.1.8)
ωµ2 ωµ2 ωµ2
Imposing n̂ × H continuous or Hx continuous at z = 0, we have
βiz βrz T E βtz T E
− E0 e−jβix x + R E0 e−jβrx x = − T E0 e−jβtx x (14.1.9)
ωµ1 ωµ1 ωµ2
As mentioned before, the phase-matching condition requires that βix = βrx = βtx . The
dispersion relation for plane waves requires that in their respective media,
2 2 2 2
βix + βiz = βrx + βrz = ω 2 µ1 ε1 = β12 , medium 1 (14.1.10)
2 2 2
βtx + βtz = ω µ2 ε2 = β22 , medium 2 (14.1.11)
2 Here, we will treat this problem as a boundary value problem where the unknowns are sought from

equations obtained from boundary conditions.

3 The phase-matching condition can also be proved by taking the Fourier transform of the equation with

respect to x. Among the physics community, this is also known as momentum matching, as the wavenumber
of a wave is related to the momentum of the particle.
4 We note here that field theory is a lot more complicated than transmission line theory. That is the

triumph of transmission line theory as well.

148 Electromagnetic Field Theory

Since
βix = βrx = βtx = βx (14.1.12)
the above implies that
βiz = βrz = β1z (14.1.13)
Moreover, βtz = β2z 6= β1z usually since β1 6= β2 . Then (14.1.9) simplifies to
β1z β2z T E
(1 − RT E ) = T (14.1.14)
µ1 µ2
p p
where β1z = β12 − βx2 , and β2z = β22 − βx2 .
Solving (14.1.5) and (14.1.14) for RT E and T T E yields
 , 
TE β 1z β 2z β1z β2z
R = − + (14.1.15)
µ1 µ2 µ1 µ2
  ,  
β1z β1z β2z
TTE = 2 + (14.1.16)
µ1 µ1 µ2

14.1.2 TM Polarization (Parallel or H Polarization)5

Figure 14.2: A similar schematic showing the reflection of the TM polarization wave impinging
on a dielectric interface. The solution to this problem can be easily obtained by invoking
duality principle.
5 Also known as TMz polarization.
Reflection, Transmission, and Interesting Physical Phenomena 149

The solution to the TM polarization case can be obtained by invoking duality principle where
we do the substitution E → H, H → −E, and µ
ε as shown in Figure 14.2. The reflection
coefficient for the TM magnetic field is then

  
β1z β2z β1z β2z
RT M = − + (14.1.17)
ε1 ε2 ε1 ε2

  
TM β1z β1z β2z
T =2 + (14.1.18)
ε1 ε1 ε2

Please remember that RT M and T T M are reflection and transmission coefficients for the
magnetic fields, whereas RT E and T T E are those for the electric fields. Some textbooks may
define these reflection coefficients based on electric field only, and they will look different, and
duality principle cannot be applied.

14.2 Interesting Physical Phenomena

Three interesting physical phenomena emerge from the solutions of the single-interface prob-
lem. They are total internal reflection, Brewster angle effect, and surface plasmonic
resonance. We will look at them next.
150 Electromagnetic Field Theory

14.2.1 Total Internal Reflection

Figure 14.3: Wave-number surfaces in two regions showing the phase matching condition.
The wave number in medium t is larger than the wave number in medium 0. The wave
vectors for the incident wave, reflected wave, and transmitted wave have to be aligned in
such a way that their components parallel to the interface are equal in order to satisfy the
phase-matching condition. One can see that Snell’s law is satisfied when the phase-matching
condition is satisfied.

Total internal reflection comes about because of phase matching (also called momentum
matching). This phase-matching condition can be illustrated using β-surfaces (same as k-
surfaces in some literature), as shown in Figure 14.3. It turns out that because of phase
matching, for certain interfaces, β2z becomes pure imaginary.
Reflection, Transmission, and Interesting Physical Phenomena 151

Figure 14.4: Wave-number surfaces in two regions showing the phase matching condition.
The wave number in medium t is smaller than the wave number in medium 0. The figure
shows in incident wave vector coming in at the critical angle. Then the transmitted wave
vector is parallel to the interface as shown. When the incident angle is larger than the critical
angle, βz becomes an imaginary number the wave vector in Region t is complex and cannot
be drawn.
2 2
As shown in Figures 14.3 and 14.4, because of the dispersion relation that βrx + βrz =
2 2 2 2 2 2
βix + βiz = β1 , βtx + βtz = β2 , they are equations of two circles in 2D whose radii are β1
and β2 , respectively. (The tips of the β vectors for Regions 1 and 2 have to be on a spherical
surface in the βx , βy , and βz space in the general 3D case, but in this figure, we only show a
cross section of the sphere assuming that βy = 0.)
Phase matching implies that the x-component of the β vectors are equal to each other
as shown. One sees that θi = θr in Figure 14.4, and also as θi increases, θt increases. For
an optically less dense medium where β2 < β1 , according to the Snell’s law of refraction, the
transmitted β will refract away from the normal, as seen in the figure (where k is synonymous
with our β). Therefore, eventually the vector βt becomes parallel to the x axis when βix =
√
βrx = β2 = ω µ2 ε2 and θt = π/2. The incident angle at which this happens is termed the
critical angle θc (see Figure 14.4).
Since βix = β1 sin θi = βrx = β1 sin θr = β2 , or
√
β2 µ2 ε2 n2
sin θr = sin θi = sin θc = =√ = (14.2.1)
β1 µ1 ε1 n1
√ √
where n1 is the reflective index defined as c0 /vi = µi εi / µ0 ε0 where vi is the phase velocity
152 Electromagnetic Field Theory

of the wave in Region i. Hence,

θc = sin−1 (n2 /n1 ) (14.2.2)
p
When θi > θc . βx > β2 and β2z = β2 2 − βx 2 becomes pure imaginary. When β2z
becomes pure imaginary, the wave cannot propagate in Region 2, or β2z = −jα2z , and the
wave becomes evanescent. The reflection coefficient (14.1.15) becomes of the form
RT E = (A − jB)/(A + jB) (14.2.3)
Since the numerator is the complex conjugate of the denominator. It is clear that |RT E | = 1
and that RT E = ejθT E . Therefore, a total internally reflected wave suffers a phase shift. A
phase shift in the frequency domain corresponds to a time delay in the time domain. Such
a time delay is achieved by the wave traveling laterally in Region 2 before being refracted
back to Region 1. Such a lateral shift is called the Goos-Hanschen shift as shown in Figure
14.5 [58]. A wave that travels laterally along the surface of two media is also known as lateral
waves [95, 96].
(Please be reminded that total internal reflection comes about entirely due to the phase-
matching condition when Region 2 is a faster medium than Region 1. Hence, it will occur
with all manner of waves, such as elastic waves, sound waves, seismic waves, quantum waves
etc.)

Figure 14.5: Goos-Hanschen Shift. A phase delay is equivalent to a time delay (courtesy of
Paul R. Berman (2012), Scholarpedia, 7(3):11584 [97]).

The guidance of a wave in a dielectric slab is due to total internal reflection at the dielectric-
to-air interface. The wave bounces between the two interfaces of the slab, and creates evanes-
cent waves outside, as shown in Figure 14.6. The guidance of waves in an optical fiber works
by similar mechanism of total internal reflection, as shown in Figure 14.7. Due to the tremen-
dous impact the optical fiber has on modern-day communications, Charles Kao, the father of
the optical fiber, was awarded the Nobel Prize in 2009. His work was first published in [98].
Reflection, Transmission, and Interesting Physical Phenomena 153

Figure 14.6: The total internal reflections at the two interfaces of a thin-film waveguide can
be used to guide an optical wave (courtesy of E.N. Glytsis, NTUA, Greece [99]).

Figure 14.7: An optical fiber consists of a core and a cladding region. Total internal reflections
occur at the core-cladding interface. They can guide an optical wave in the fiber (courtesy of
Wikepedia [100]).

Waveguides have affected international communications for over a hundred year now.
Since telegraphy was in place before the full advent of Maxwell’s equations, submarine cables
for global communications were laid as early as 1850’s. Figure 14.8 shows a submarine cable
from 1869 using coaxial cable, and one used in the modern world using optical fiber.
154 Electromagnetic Field Theory

Figure 14.8: The picture of an old 1869 submarine cable made of coaxial cables (left), and
modern submarine cable made of optical fibers (right) (courtesy of Atlantic-Cable [101], and
Wikipedia [102]).
Lecture 15

More on Interesting Physical

Phenomena, Homomorphism,
Plane Waves, and Transmission
Lines

Though simple that it looks, embedded in the TM Fresnel reflection coefficient are a few more
interesting physical phenomena. We have looked at the physics of total internal reflection,
which has inspired many interesting technologies such as waveguides, the most important of
which is the optical fiber. In this lecture, we will look at other physical phenomena. These
are the phenomena of Brewster’s angle [103, 104] and that of surface plasmon resonance, or
polariton [105, 106].
Even though transmission line theory and the theory of plane wave reflection and trans-
mission look quite different, they are very similar in their underlying mathematical structures.
For lack of a better name, we call this mathematical homomorphism (math analogy).1 Later,
to simplify the mathematics of waves in layered media, we will draw upon this mathemati-
cal homomorphism between multi-section transmission line theory and plane-wave theory in
layered media.

15.1 Brewster’s Angle

We will continue with understanding some interesting phenomena associated with the single-
interface problem starting with the Brewster’s angle.

1 The use of this term could be to the chagrin of a math person, but it has also being used in a subject

called homomorphic encryption or computing [107].

155
156 Electromagnetic Field Theory

Figure 15.1: A figure showing a plane wave being reflected and transmitted at the Brewster’s
angle. In Region t, the polarization current or dipoles are all pointing in the βr direction,
and hence, there is no radiation in that direction.

Brewster angle was discovered in 1815 [103, 104]. Furthermore, most materials at optical
frequencies have ε2 6= ε1 , but µ2 ≈ µ1 . In other words, it is hard to obtain magnetic materials
at optical frequencies. Therefore, the TM polarization for light behaves differently from TE
polarization. Hence, we shall focus on the reflection and transmission of the TM polarization
of light, and we reproduce the previously derived TM reflection coefficient here:
  
β1z β2z β1z β2z
RT M = − + (15.1.1)
ε1 ε2 ε1 ε2

The transmission coefficient is easily gotten by the formula T T M = 1 + RT M . Observe that

for RT M , it is possible that RT M = 0 if

ε2 β1z = ε1 β2z (15.1.2)

p
Squaring the above, making the note that βiz = βi2 − βx2 , one gets

ε2 2 (β1 2 − βx 2 ) = ε1 2 (β2 2 − βx 2 ) (15.1.3)

Solving the above, assuming µ1 = µ2 = µ, gives

r
√ ε1 ε2
βx = ω µ = β1 sin θ1 = β2 sin θ2 (15.1.4)
ε1 + ε2
More on Interesting Physical Phenomena, Homomorphism, Plane Waves, and Transmission Lines157

The latter two equalities come from phase matching at the interface or Snell’s law. Therefore,

r r
ε2 ε1
sin θ1 = , sin θ2 = (15.1.5)
ε1 + ε2 ε1 + ε2

or squaring the above and adding them,

sin2 θ1 + sin2 θ2 = 1, (15.1.6)

Then, assuming that θ1 and θ2 are less than π/2, and using the identity that cos2 θ1 +sin2 θ1 =
1, we infer that cos2 θ1 = sin2 θ2 . Then it follows that

sin θ2 = cos θ1 (15.1.7)

In other words,

θ1 + θ2 = π/2 (15.1.8)

The above formula can be used to explain why at Brewster angle, no light is reflected back
to Region 1. Figure 15.1 shows that the induced polarization dipoles in Region 2 always
have their axes aligned in the direction of reflected wave. A dipole does not radiate along its
axis, which can be verified heuristically by field sketch and looking at the Poynting vector.
Therefore, these induced dipoles in Region 2 do not radiate in the direction of the reflected
wave. Notice that when the contrast is very weak meaning that ε1 ∼ = ε2 , then θ1 ∼
= θ2 ∼
= π/4,
and (15.1.8) is satisfied.
Because of the Brewster angle effect for TM polarization when ε2 6= ε1 , |RT M | has to go
through a null when θi = θb . Therefore, |RT M | ≤ |RT E | as shown in the plots in Figure 15.2.
Then when a randomly (or arbitrarily) polarized light is incident on a surface, the polarization
where the electric field is parallel to the surface (TE polarization) is reflected more than the
polarization where the magnetic field is parallel to the surface (TM polarization). This
phenomenon is used to design sun glasses to reduce road surface glare for drivers. For light
reflected off a road surface, they are predominantly horizontally polarized with respect to the
surface of the road. When sun glasses are made with vertical polarizers,2 they will filter out
and mitigate the reflected rays from the road surface to reduce road glare. This phenomenon
can also be used to improve the quality of photography by using a polarizer filter as shown
in Figure 15.3.

2 Defined as one that will allow vertical polarization to pass through.

158 Electromagnetic Field Theory

Figure 15.2: Because |RT M | has to through a null when θi = θb , therefore, plots of |RT M | ≤
|RT E | for all θi as shown above.

Figure 15.3: Because that TM and TE lights will be reflected differently, polarizer filter can
produce remarkable effects on the quality of the photograph by reducing glare [104].

15.1.1 Surface Plasmon Polariton

Surface plasmon polariton occurs for the same mathematical reason for the Brewster angle
effect but the physical mechanism is quite different. Many papers and textbooks will introduce
this phenomenon from a different angle. But here, we will see it from the Fresnel reflection
coefficient for the TM waves. When the denominator of the reflection coefficient RT M is zero,
it can become infinite. This is possible if ε2 < 0, which is possible if medium 2 is a plasma
medium. In this case, the criterion for the denominator to be zero is

−ε2 β1z = ε1 β2z (15.1.9)

When RT M becomes infinite, it implies that a reflected wave exists when there is no incident
wave. Or when Href = Hinc RT M , and RT M = ∞, Hinc can be zero, and then Href can assume
More on Interesting Physical Phenomena, Homomorphism, Plane Waves, and Transmission Lines159

any value.3 Hence, there is a plasmonic resonance or guided mode existing at the interface
without the presence of an incident wave. It is a self-sustaining wave propagating in the x
direction, and hence, is a guided mode propagating in the x direction.
Solving (15.1.9) after squaring it, as in the Brewster angle case, yields

r
√ ε1 ε2
βx = ω µ (15.1.10)
ε1 + ε2

This is the same equation for the Brewster angle except now that ε2 is negative. Even if
ε2 < 0, but ε1 + ε2 < 0 is still possible so that the expression under the square root sign
(15.1.10) is positive. Thus, βx can be pure real. The corresponding β1z and β2z in (15.1.9)
can be pure imaginary as explained below, and (15.1.9) can still be satisfied.
This corresponds to a guided wave propagating in the x direction. When this happens,

1/2 1/2
ε21
  
√ ε2 √
q
2 2
β1z = β1 − βx = ω µ ε1 1 − =ω µ (15.1.11)
ε1 + ε2 ε1 + ε2

p
Since ε2 < 0, ε2 /(ε1 +ε2 ) > 1, then β1z becomes pure imaginary. Moreover, β2z = β2 2 − βx 2
and β2 2 < 0 making β2z becomes even a larger imaginary number. This corresponds to a
trapped wave (or a bound state) at the interface. The wave decays exponentially in both
directions away from the interface and they are evanescent waves. This mode is shown in
Figure 15.4, and is the only case in electromagnetics where a single interface can guide a
surface wave, while such phenomenon abounds for elastic waves.
When one operates close to the resonance of the mode so that the denominator in (15.1.10)
is almost zero, then βx can be veryp large. The wavelength in the x direction becomes very
short in this case, and since βiz = βi2 − βx2 , then β1z and β2z become even larger imaginary
numbers. Hence, the mode becomes tightly confined or bound to the interface, making the
confinement of the mode very tight. This evanescent wave is much p more rapidly decaying
than that offered by the total internal reflection, which is βz = βt2 − βx2 where βx is no
larger than β1 . It portends use in tightly packed optical components, and has caused some
excitement in the optics community.

3 In other words, infinity times zero is undefined. This is often encountered in a resonance system like an

LC tank circuit. Current flows in the tank circuit despite the absence of an exciting or driving voltage. In
an ordinary differential equation or partial differential equation without a driving term (source term), such
solutions are known as homogeneous solutions (to clarify the potpouri of math terms, homogeneous solutions
here refer to a solution with zero source term). In a matrix equation A · x = b without a right-hand side or
that b = 0, it is known as a null-space solution.
160 Electromagnetic Field Theory

Figure 15.4: Figure showing a surface plasmonic mode propagating at an air-plasma interface.
As in all resonant systems, a resonant mode entails the exchange of energies. In the case
of surface plasmonic resonance, the energy is exchanged between the kinetic energy of the
electrons and the energy store in the electic field (courtesy of Wikipedia [108]).

15.2 Homomorphism of Uniform Plane Waves and Trans-

mission Lines Equations

Transmission line theory is very simple due to its one-dimensional nature. But the problem
of reflection and transmission of plane waves at a planar interface is actually homormophic
to that of the transmission line problem. Therefore, the plane waves through layered medium
can be mapped into the multi-section transmission line problem due to mathematical homo-
morphism between the two problems. Hence, we can kill two birds with one stone: apply all
the transmission line techniques and equations that we have learnt to solve for the solutions
of waves through layered medium problems.4
For uniform plane waves, since they are proportional to exp(−jβ · r), we know that with
∇ → −jβ, Maxwell’s equations become

β × E = ωµH (15.2.1)
β × H = −ωεE (15.2.2)

for a general isotropic homogeneous medium. We will specialize these equations for different
polarizations.

4 This treatment is not found elsewhere, and is peculiar to these lecture notes.
More on Interesting Physical Phenomena, Homomorphism, Plane Waves, and Transmission Lines161

15.2.1 TE or TEz Waves

For this, one assumes a TE wave traveling in the z direction with electric field polarized in
the y direction, or E = ŷEy , H = x̂Hx + ẑHz . Then we have from (15.2.1)

βz Ey = −ωµHx (15.2.3)
βx Ey = ωµHz (15.2.4)

From (15.2.2), we have

βz Hx − βx Hz = −ωεEy (15.2.5)

The above equations involve three variables, Ey , Hx , and Hz . But there are only two variables
in the telegrapher’s equations which are V and I. To this end, we will eliminate one of the
variables from the above three equations. Then, expressing Hz in terms of Ey from (15.2.4),
we can show from (15.2.5) that

βx2
βz Hx = −ωεEy + βx Hz = −ωεEy + Ey
ωµ
= −ωε(1 − βx2 /β 2 )Ey = −ωε cos2 θEy (15.2.6)

where βx = β sin θ has been used.

Eqns. (15.2.3) and (15.2.6) can be written to look like the telegrapher’s equations by
letting −jβz → d/dz to get

d
Ey = jωµHx (15.2.7)
dz
d
Hx = jωε cos2 θEy (15.2.8)
dz
If we let Ey → V , Hx → −I, µ → L, ε cos2 θ → C, the above is exactly analogous to the
telegrapher’s equations. The equivalent characteristic impedance of these equations above is
then
r r r
L µ 1 µ β ωµ
Z0 = = = = (15.2.9)
C ε cos θ ε βz βz

The above ωµ/βz is the wave impedance for a propagating plane wave with propagation
direction or the β inclined with an angle θ respect to the z axis. It is analogous to the
characteristic impedance Z0 of a transmission line. When θ = 0, the wave impedance becomes
the intrinsic impedance of space.
A two region, single-interface reflection problem can then be mathematically mapped to
a single-junction connecting two-transmission-lines problem discussed in Section 13.1.1. The
equivalent characteristic impedances of these two regions are then
ωµ1 ωµ2
Z01 = , Z02 = (15.2.10)
β1z β2z
162 Electromagnetic Field Theory

We can use the above to find Γ12 as given by

Z02 − Z01 (µ2 /β2z ) − (µ1 /β1z )

Γ12 = = (15.2.11)
Z02 + Z01 (µ2 /β2z ) + (µ1 /β1z )

The above is the same as the Fresnel reflection coefficient found earlier for TE waves or RT E
after some simple re-arrangement.
Assuming that we have a single junction transmission line, one can define a transmission
coefficient given by

2Z02 2(µ2 /β2z )

T12 = 1 + Γ12 = = (15.2.12)
Z02 + Z01 (µ2 /β2z ) + (µ1 /β1z )

The above is similar to the continuity of the voltage across the junction, which is the same
as the continuity of the tangential electric field across the interface. It is also the same as the
Fresnel transmission coefficient T T E .

15.2.2 TM or TMz Waves

For the TM polarization, by invoking duality principle, the corresponding equations are, from
(15.2.7) and (15.2.8),

d
Hy = −jωεEx (15.2.13)
dz
d
Ex = −jωµ cos2 θHy (15.2.14)
dz
Just for consistency of units, since electric field is in V m−1 , and magnetic field is in A m−1
we may chose the following map to convert the above into the telegrapher’s equations, viz;

Ey → V, Hy → I, µ cos2 θ → L, ε→C (15.2.15)

Then, the equivalent characteristic impedance is now

r r r
L µ µ βz βz
Z0 = = cos θ = = (15.2.16)
C ε ε β ωε
The above is also termed the wave impedance of a TM propagating wave making an inclined
angle θ with respect to the z axis. Notice again that this wave impedance becomes the
intrinsic impedance of space when θ = 0.
Now, using the reflection coefficient for a single-junction transmission line, and the appro-
priate characteristic impedances for the two lines as given in (15.2.16), we arrive at

(β2z /ε2 ) − (β1z /ε1 )

Γ12 = (15.2.17)
(β2z /ε2 ) + (β1z /ε1 )

Notice that (15.2.17) has a sign difference from the definition of RT M derived earlier in the
last lecture. The reason is that RT M is for the reflection coefficient of magnetic field while
More on Interesting Physical Phenomena, Homomorphism, Plane Waves, and Transmission Lines163

Γ12 above is for the reflection coefficient of the voltage or the electric field. This difference
is also seen in the definition for transmission coefficients.5 A voltage transmission coefficient
can be defined to be
2(β2z /ε2 )
T12 = 1 + Γ12 = (15.2.18)
(β2z /ε2 ) + (β1z /ε1 )

But this will be the transmission coefficient for the voltage, which is not the same as T T M
which is the transmission coefficient for the magnetic field or the current. Different textbooks
may define different transmission coefficients for this polarization.

5 This is often the source of confusion for these reflection and transmission coefficients.
164 Electromagnetic Field Theory
Lecture 16

Waves in Layered Media

Waves in layered media is an important topic in electromagnetics. Many media can be

approximated by planarly layered media. For instance, the propagation of radio wave on the
earth surface was of interest and first tackled by Sommerfeld in 1909 [109]. The earth can be
approximated by planarly layered media to capture the important physics behind the wave
propagation. Many microwave components are made by planarly layered structures such as
microstrip and coplanar waveguides. Layered media are also important in optics: they can
be used to make optical filters such as Fabry-Perot filters. As technologies and fabrication
techniques become better, there is an increasing need to understand the interaction of waves
with layered structures or laminated materials.

16.1 Waves in Layered Media

Figure 16.1: Waves in layered media. A wave entering the medium from above can be multiply
reflected before emerging from the top again.

Because of the homomorphism between the transmission line problem and the plane-wave
reflection by interfaces, we will exploit the simplicity of the transmission line theory to arrive

165
166 Electromagnetic Field Theory

at formulas for plane wave reflection by layered media. This treatment is not found in any
other textbooks.

16.1.1 Generalized Reflection Coefficient for Layered Media

Figure 16.2: The equivalence of a layered medium problem to a transmission line prob-
lem. This equivalence is possible even for oblique incidence. For normal incidence, the wave
impedance becomes intrinsic impedances (courtesy of J.A. Kong, Electromagnetic Wave The-
ory).

Because of the homomorphism between transmission line problems and plane waves in layered
medium problems, one can capitalize on using the multi-section transmission line formulas
for generalized reflection coefficient, which is

Γ12 + Γ̃23 e−2jβ2 l2

Γ̃12 = (16.1.1)
1 + Γ12 Γ̃23 e−2jβ2 l2

In the above, Γ12 is the local reflection at the 1,2 junction, whereas Γ̃ij are the generalized
reflection coefficient at the i, j interface. For instance, Γ̃12 includes multiple reflections from
behind the 1,2 junction. It can be used to study electromagnetic waves in layered media
shown in Figures 16.1 and 16.2.
Using the result from the multi-junction transmission line, by analogy we can write down
the generalized reflection coefficient for a layered medium with an incident wave at the 1,2
interface, including multiple reflections from behind the interface. It is given by

R12 + R̃23 e−2jβ2z l2

R̃12 = (16.1.2)
1 + R12 R̃23 e−2jβ2z l2
Waves in Layered Media 167

where R12 is the local Fresnel reflection coefficient and R̃ij is the generalized reflection coef-
ficient at the i, j interface. Here, l2 is now the thickness of the region 2. In the above, we
assume that the wave is incident from medium (region) 1 which is semi-infinite, the general-
ized reflection coefficient R̃12 above is defined at the media 1 and 2 interface. It is assumed
that there are multiple reflections coming from the right of the 2,3 interface, so that the 2,3
reflection coefficient is the generalized reflection coefficient R̃23 .
Figure 16.2 shows the case of a normally incident wave into a layered media. For this
case, the wave impedance becomes the intrinsic impedance of homogeneous space.

16.1.2 Ray Series Interpretation of Generalized Reflection Coeffi-

cient

Figure 16.3: The expression of the generalized reflection coefficient into a ray series. Here,
l2 = d2 − d1 is the thickness of the slab (courtesy of [110]).

For simplicity, we will assume that R̃23 = R23 in this section. By manipulation, one can con-
vert the generalized reflection coefficient R̃12 into a form that has a ray physics interpretation.
By adding the term
2
R12 R23 e−2jβ2z l2
to the numerator of (16.1.2), it can be shown to become

R23 e−2jβ2z l2 (1 − R12

2
)
R̃12 = R12 + −2jβ l
(16.1.3)
1 + R12 R23 e 2z 2

By using the fact that R12 = −R21 and that Tij = 1 + Rij , the above can be rewritten as

T12 T21 R23 e−2jβ2z l2

R̃12 = R12 + (16.1.4)
1 + R12 R23 e−2jβ2z l2

Then using the fact that (1 − x)−1 = 1 + x + x2 + . . . +, the above can be rewritten as

R̃12 = R12 + T12 R23 T21 e−2jβ2z l2 + T12 R23

2
R21 T21 e−4jβ2z l2 + · · · . (16.1.5)

The above allows us to elucidate the physics of each of the terms. The first term in the
above is just the result of a single reflection off the first interface. The n-th term above
168 Electromagnetic Field Theory

is the consequence of the n-th reflection from the three-layer medium (see Figure 16.3).
Hence, the expansion of (16.1.2) into 16.1.5 renders a lucid interpretation for the generalized
reflection coefficient. Consequently, the series in 16.1.5 can be thought of as a ray series or a
geometrical optics series. It is the consequence of multiple reflections and transmissions in
region 2 of the three-layer medium. It is also the consequence of expanding the denominator of
the second term in (21). Hence, the denominator of the second term in (21) can be physically
interpreted as a consequence of multiple reflections within region 2.

16.1.3 Guided Modes from Generalized Reflection Coefficients

We shall discuss finding guided waves in a layered medium next using the generalized reflection
coefficient. For a general guided wave along the longitudinal direction parallel to the interfaces
(x direction in our notation), the wave will propagate in the manner of
e−jβx x
For instance, the surface plasmon mode that we found previously can be thought of as a
wave propagating in the x direction. The wave number in the x direction, βx for the surface
plasmon case can be found in closed form, but it is a rather complicated function of frequency.
Hence, it has very interesting phase velocity, which is frequency dependent. From Section
15.1.1, we have derived that for a surface plasmon mode,
r
√ ε1 ε2
βx = ω µ = β1 sin θ1 = β2 sin θ2 (16.1.6)
ε1 + ε2
Because ε2 is a plasma medium with complex frequency dependence, βx is in general, a
complicated function of ω or frequency. Therefore, this wave has very interesting phase and
group velocities, which is typical of waveguide modes. Hence, it is prudent to understand
what phase and group velocities are before studying waveguide modes in greater detail.

16.2 Phase Velocity and Group Velocity

Now that we know how a medium can be frequency dispersive in a complicated fashion as
in the Drude-Lorentz-Sommerfeld (DLS) model, we are ready to investigate the difference
between the phase velocity and the group velocity

16.2.1 Phase Velocity

The phase velocity is the velocity of the phase of a wave. It is only defined for a mono-
chromatic signal (also called time-harmonic, CW (constant wave), or sinusoidal signal) at one
given frequency. Given a sinusoidal wave signal, e.g., the voltage signal on a transmission
line, using phasor technique, its representation in the time domain can be easily found and
take the form
V (z, t) = V0 cos(ωt − kz + α)
h ω  i
= V0 cos k t−z +α (16.2.1)
k
Waves in Layered Media 169

This sinusoidal signal moves with a velocity

ω
vph = (16.2.2)
k

√
where, for example, k = ω µε, inside a simple coax. Hence,

√
vph = 1/ µε (16.2.3)

But a dielectric medium can be frequency dispersive, or ε(ω) is not a constant but a function
of ω as has been shown with the Drude-Lorentz-Sommerfeld model. Therefore, signals with
different ω’s will travel with different phase velocities.

More bizarre still, what if the coax is filled with a plasma medium where

ωp 2
 
ε = ε0 1− 2 (16.2.4)
ω

Then, ε < ε0 always meaning that the phase velocity given by (16.2.3) can be larger than
the velocity of light in vacuum (assuming µ = µ0 ). Also, ε = 0 when ω = ωp , implying that
k = 0; then in accordance to (16.2.2), vph = ∞. These ludicrous observations can be justified
or understood only if we can show that information can only be sent by using a wave packet.1
The same goes for energy which can only be sent by wave packets, but not by CW signal;
only in this manner can a finite amount of energy be sent. Therefore, it is prudent for us
to study the velocity of a wave packet which is not a mono-chromatic signal. These wave
packets can only travel at the group velocity as shall be shown, which is always less than the
velocity of light.

1 In information theory, according to Shannon, the basic unit of information is a bit, which can only be

sent by a digital signal, or a wave packet.

170 Electromagnetic Field Theory

16.2.2 Group Velocity

Figure 16.4: A Gaussian wave packet can be thought of as a linear superposition of monochro-
matic waves of slightly different frequencies. If one Fourier transforms the above signal, it
will be a narrow-band signal centered about certain ω0 (courtesy of Wikimedia [111]).

Now, consider a narrow band wave packet as shown in Figure 16.4. It cannot be mono-
chromatic, but can be written as a linear superposition of many frequencies. One way to
express this is to write this wave packet as an integral in terms of Fourier transform, or a
summation over many frequencies, namely2
∞
V (z, t) = dωV (z, ω)ejωt (16.2.5)
−∞ e

To make V (z, t) be related to a traveling wave, we assume that V (z, ω) is the solution to the
one-dimensional Helmholtz equation3 e

d2
V (z, ω) + k 2 (ω)V (z, ω) = 0 (16.2.6)
dz 2 e e
To derive this equation, one can easily extend the derivation in Section 7.2 to a dispersive
medium where V (z, ω) = Ex (z, ω). Alternatively, one can generalize the derivation in Section
11.2 to the case of dispersive transmission lines. For instance, when the co-axial transmission
2 The Fourier transform technique is akin to the phasor technique, but different. For simplicity, we will use

V (z, ω) to represent the Fourier transform of V (z, t).

e 3 In this notes, we will use k and β interchangeably for wavenumber. The transmission line community
tends to use β while the optics community uses k.
Waves in Layered Media 171

line is filled with a dispersive material, then k 2 = ω 2 µ0 ε(ω). Thus, upon solving the above
equation, one obtains that V (z, ω) = V0 (ω)e−jkz , and

∞
V (z, t) = dωV0 (ω)ej(ωt−kz) (16.2.7)
−∞

In the general case, k is a complicated function of ω as shown in Figure 16.5.

Figure 16.5: A typical frequency dependent k(ω) albeit the frequency dependence can be
more complicated than shown here.

Since this is a wave packet, we assume that V0 (ω) is narrow band centered about a
frequency ω0 , the carrier frequency as shown in Figure 16.6. Therefore, when the integral in
(16.2.7) is performed, we need only sum over a narrow range of frequencies in the vicinity of
ω0 .
172 Electromagnetic Field Theory

Figure 16.6: The frequency spectrum of V0 (ω) which is the Fourier transform of V0 (t).

Thus, we can approximate the integrand in the vicinity of ω = ω0 , in particular, k(ω) by

Taylor series expansion, and let
dk(ω0 ) 1 d2 k(ω0 )
k(ω) ∼
= k(ω0 ) + (ω − ω0 ) + (ω − ω0 )2 + ··· (16.2.8)
dω 2 dω 2
To ensure the real-valuedness of (16.2.5), one ensures that −ω part of the integrand is exactly
the complex conjugate of the +ω part, or V (z, −ω) = V ∗ (z, ω). Thus the integral can be
folded to integrate only over the positive frequency
e components.
e Another way is to sum over
only the +ω part of the integral and take twice the real part of the integral. So, for simplicity,
we rewrite (16.2.7) as
∞
V (z, t) = 2<e dωV0 (ω)ej(ωt−kz) (16.2.9)
0

Thus the above follows from the reality condition of V (z, t).
Since we need to integrate over ω ≈ ω0 , we can substitute (16.2.8) into (16.2.9) and rewrite
it as
 

∞
 
 
dk 
∼ j(ω−ω0 )t −j(ω−ω0 ) dω z 
 j[ω t−k(ω )z]
V (z, t) = 2<e e
 0 0
dωV0 (ω)e e  (16.2.10)
 
 0
| {z }
F (t− dω
dk
z)

where more specifically,

  ∞
dk dk
F t− z = dωV0 (ω)ej(ω−ω0 )t e−j(ω−ω0 ) dω z (16.2.11)
dω
0
Waves in Layered Media 173

It can be seen that the above integral now involves the integral summation over a small range
of ω in the vicinity of ω0 . By a change of variable by letting Ω = ω − ω0 , it becomes
   +∆
dk
dΩV0 (Ω + ω0 )ejΩ(t− dω z)
dk
F t− z = (16.2.12)
dω −∆

When Ω ranges from −∆ to +∆ in the above integral, the value of ω ranges from ω0 − ∆ to
ω0 + ∆. It is assumed that outside this range of ω, V0 (ω) is sufficiently small so that its value
can be ignored.
The above itself is a Fourier transform integral that involves only the low frequencies of
the Fourier spectrum where ejΩ(t− dω z) is evaluated over small Ω values. Hence, F is a slowly
dk

varying function. Moreover, this function F moves with a velocity

dω
vg = (16.2.13)
dk
Here, F (t− vzg ) in fact is the velocity of the envelope in Figure 16.4. In (16.2.10), the envelope
function F (t − vzg ) is multiplied by the rapidly varying function

ej[ω0 t−k(ω0 )z] (16.2.14)

before one takes the real part of the entire function. Hence, this rapidly varying part represents
the rapidly varying carrier frequency shown in Figure 16.4. More importantly, this carrier,
the rapidly varying part of the signal, moves with the velocity
ω0
vph = (16.2.15)
k(ω0 )
which is the phase velocity.

16.3 Wave Guidance in a Layered Media

Now that we have understood phase and group velocity, we are at ease with studying the
propagation of a guided wave in a layered medium. We have seen that in the case of a
surface plasmonic resonance, the wave is guided by an interface because the Fresnel reflection
coefficient becomes infinite. This physically means that a reflected wave exists even if an
incident wave is absent or vanishingly small. This condition can be used to find a guided
mode in a layered medium, namely, to find the condition under which the generalized reflection
coefficient (16.1.2) will become infinite.4

16.3.1 Transverse Resonance Condition

Therefore, to have a guided mode exist in a layered medium due to multiple bounces, the
generalized reflection coefficient becomes infinite, the denominator of (16.1.2) is zero, or that
1 + R12 R̃23 e−2jβ2z l2 = 0 (16.3.1)
4 As mentioned previously in Section 15.1.1, this is equivalent to finding a solution to a problem with no

driving term (forcing function), or finding the homogeneous solution to an ordinary differential equation or
partial differential equation. It is also equivalent to finding the null space solution of a matrix equation.
174 Electromagnetic Field Theory

where t is the thickness of the dielectric slab. Since R12 = −R21 , the above can be written as

1 = R21 R̃23 e−2jβ2z l2 (16.3.2)

The above has the physical meaning that the wave, after going through two reflections at
the two interfaces, 21, and 23 interfaces, which are R21 and R̃23 , plus a phase delay given
by e−2jβ2z l2 , becomes itself again. This is also known as the transverse resonance condition.
When specialized to the case of a dielectric slab with two interfaces and three regions, the
above becomes

1 = R21 R23 e−2jβ2z l2 (16.3.3)

The above can be generalized to finding the guided mode in a general layered medium. It
can also be specialized to finding the guided mode of a dielectric slab.
Lecture 17

Dielectric Waveguides

As mentioned before, the dielectric waveguide shares many salient features with the optical
fiber waveguide, one of the most important waveguides of this century. Before we embark
on the study of dielectric waveguides, we will revisit the transverse resonance again. The
transverse resonance condition allows one to derive the guidance conditions for a dielectric
waveguide easily without having to match the boundary conditions at the interface again: The
boundary conditions are already used when deriving the Fresnel reflection coefficients, and
hence they are embedded in these reflection coefficients and generalized reflection coefficients.
Much of the materials in this lecture can be found in [32, 44, 83].

17.1 Generalized Transverse Resonance Condition

The guidance conditions, the transverse resonance condition given previously, can also be
derived for the more general case. The generalized transverse resonance condition is a powerful
condition that can be used to derive the guidance condition of a mode in a layered medium.
To derive this condition, we first have to realize that a guided mode in a waveguide is due
to the coherent or constructive interference of the waves. This implies that if a plane wave
starts at position 1 (see Figure 17.1)1 and is multiply reflected as shown, it will regain its
original phase in the x direction at position 5. Since this mode progresses in the z direction,
all these waves (also known as partial waves) are in phase in the z direction by the phase
matching condition. Otherwise, the boundary conditions cannot be satisfied. That is, waves
at 1 and 5 will gain the same phase in the z direction. But, for them to add coherently or
interfere coherently in the x direction, the transverse phase at 5 must be the same as 1.
Assuming that the wave starts with amplitude 1 at position 1, it will gain a transverse
phase of e−jβ0x t when it reaches position 2. Upon reflection at x = x2 , at position 3, the wave
becomes R̃+ e−jβ0x t where R̃+ is the generalized reflection coefficient at the right interface
of Region 0. Finally, at position 5, it becomes R̃− R̃+ e−2jβ0x t where R̃− is the generalized
1 The waveguide convention is to assume the direction of propagation to be z. Since we are analyzing a

guided mode in a layered medium, z axis is as shown in this figure, which is parallel to the interfaces. This
is different from before.

175
176 Electromagnetic Field Theory

Figure 17.1: The transverse resonance condition for a layered medium. The phase of the wave
at position 5 should be equal to the transverse phase at position 1 for constructive interference
to occur.

reflection coefficient at the left interface of Region 0. For constructive interference to occur
or for the mode to exist, we require that

R̃− R̃+ e−2jβ0x t = 1 (17.1.1)

The above is the generalized transverse resonance condition for the guidance condition for a
plane wave mode traveling in a layered medium.
In (17.1.1), a metallic wall has a reflection coefficient of 1 for a TM wave; hence if R̃+ is
1, Equation (17.1.1) becomes
1 − R̃− e−2jβ0x t = 0. (17.1.2)

On the other hand, in (17.1.1), a metallic wall has a reflection coefficient of −1, for TE wave,
and Equation (17.1.1) becomes
1 + R̃− e−2jβ0x t = 0. (17.1.3)

17.2 Dielectric Waveguide

The most important dielectric waveguide of the modern world is the optical fiber, whose
invention was credited to Charles Kao [98]. He was awarded the Nobel prize in 2009 [112].
However, the analysis of the optical fiber requires the use of cylindrical coordinates and special
functions such as Bessel functions. In order to capture the essence of dielectric waveguides,
one can study the slab dielectric waveguide, which shares many salient features with the
optical fiber. This waveguide is also used as thin-film optical waveguides (see Figure 17.2).
We start with analyzing the TE modes in this waveguide.
Dielectric Waveguides 177

Figure 17.2: An optical thin-film waveguide is made by coating a thin dielectric film or sheet
on a metallic surface. The wave is guided by total internal reflection at the top interface, and
by metallic reflection at the bottom interface.

17.2.1 TE Case

Figure 17.3: Schematic for the analysis of a guided mode in the dielectric waveguide. Total
internal reflections occur at the top and bottom interfaces. If the waves add coherently, the
wave is guided along the dielectric slab.

We shall look at the application of the transverse resonance condition to a TE wave guided
in a dielectric waveguide. Again, we assume the direction of propagation of the guided mode
to be in the z direciton in accordance with convention. Specializing the above equation to
the dielectric waveguide shown in Figure 17.3, we have the guidance condition as

1 = R10 R12 e−2jβ1x d (17.2.1)

where d is the thickness of the dielectric slab. Guidance of a mode is due to total internal
reflection, and hence, we expect Region 1 to be optically more dense (in terms of optical
refractive indices)2 than region 0 and 2.
To simplify the analysis further, we assume Region 2 to be the same as Region 0 so that
R12 = R10 . The new guidance condition is then
2 −2jβ1x d
1 = R10 e (17.2.2)
2 Optically more dense means higher optical refractive index, or higher dielectric constant.
178 Electromagnetic Field Theory

By phase-matching, βz is the same in all the three regions of Figure 17.3. By expressing all
the βix in terms of the variable βz , the above is an implicit equation for βz . Also, we assume
that ε1 > ε0 so that total internal reflection occurs at both interfaces as the wave bounces
around so that β0x = −jα0x . Therefore, for TE polarization, the local, single-interface, or
Fresnel reflection coefficient is
µ0 β1x − µ1 β0x µ0 β1x + jµ1 α0x
R10 = = = ejθT E (17.2.3)
µ0 β1x + µ1 β0x µ0 β1x − jµ1 α0x
where θT E is the Goos-Hanschen shift for total internal reflection. It is given by
 
−1 µ1 α0x
θT E = 2 tan (17.2.4)
µ0 β1x

The guidance condition for constructive interference according to (17.2.1) is such that

2θT E − 2β1x d = 2nπ (17.2.5)

From the above, dividing it by four, and taking its tangent, we get
   
θT E nπ β1x d
tan = tan + (17.2.6)
2 2 2

or using (17.2.4) for the left-hand side,

 
µ1 α0x nπ β1x d
= tan + (17.2.7)
µ0 β1x 2 2

The above gives rise to

 
β1x d
µ1 α0x = µ0 β1x tan , n even (17.2.8)
2
 
β1x d
−µ1 α0x = µ0 β1x cot , n odd (17.2.9)
2

It can be shown that when n is even, the mode profile is even, whereas when n is odd, the
mode profile is odd. The above can also be rewritten as
 
µ0 β1x d β1x d α0x d
tan = , even modes (17.2.10)
µ1 2 2 2
 
µ0 β1x d β1x d α0x d
− cot = , odd modes (17.2.11)
µ1 2 2 2

Again, the above equations can be expressed in the βz variable, but they do not have closed
form solutions, save for graphical solutions (or numerical solutions). We shall discuss their
graphical solutions.3
3 This technique has been put together by the community of scholars in the optical waveguide area.
Dielectric Waveguides 179

To solve the above graphically, it is best to plot them in terms of one common variable.
It turns out the β1x is the simplest common variable to use for graphical solutions. To this
2
end, using the fact that −α0x = β02 − βz2 , and that β1x
2
= β12 − βz2 , eliminating βz from these
two equations, one can show that

1
α0x = [ω 2 (µ1 1 − µ0 0 ) − β1x
2 2
] (17.2.12)

Thus (17.2.10) and (17.2.11) become

 
µ0 β1x d β1x d α0x d
tan =
µ1 2 2 2
s 2
d2

β1x d
= ω 2 (µ1 1 − µ0 0 ) − , even modes (17.2.13)
4 2
 
µ0 β1x d β1x d α0x d
− cot =
µ1 2 2 2
s 2
d2

β1x d
= ω 2 (µ1 1 − µ0 0 ) − , odd modes (17.2.14)
4 2

We can solve the above graphically by plotting

 
µ0 β1x d β1x d
y1 = tan even modes (17.2.15)
µ1 2 2
 
µ0 β1x d d
y2 = − cot β1x odd modes (17.2.16)
µ1 2 2
" 2 # 12
d2

2 β1x d α0x d
y3 = ω (µ1 1 − µ0 0 ) − = (17.2.17)
4 2 2
180 Electromagnetic Field Theory

Figure 17.4: A way to solve (17.2.13) and (17.2.14) is via a graphical method. In this method,
both the right-hand side and the left-hand side of the equations are plotted on the same plot.
The solutions are at the intersection points of these plots.

In the above, y3 is the equation of a circle; the radius of the circle is given by

1 d
ω(µ1 1 − µ0 0 ) 2 . (17.2.18)
2

The solutions to (17.2.13) and (17.2.14) are given by the intersections of y3 with y1 and
y2 . We note from (17.2.1) that the radius of the circle can be increased in three ways: (i)
by increasing the frequency, (ii) by increasing the contrast µµ10 10 , and (iii) by increasing the
thickness d of the slab.4 The mode profiles of the first two modes are shown in Figure 17.5.

4 These area important salient features of a dielectric waveguide. These features are also shared by the

optical fiber.
Dielectric Waveguides 181

Figure 17.5: Mode profiles of the TE0 and TE1 modes of a dielectric slab waveguide (courtesy
of J.A. Kong [32]).

When β0x = −jα0x , the reflection coefficient for total internal reflection is

  
TE µ0 β1x + jµ1 α0x −1 µ1 α0x
R10 = = exp +2j tan (17.2.19)
µ0 β1x − jµ1 α0x µ0 β1x

 TE
and R10  = 1. Hence, the wave is guided by total internal reflections.
Cut-off occurs when the total internal reflection ceases to occur, i.e. when the frequency
decreases such that α0x = 0. From Figure 17.4, we see that α0x = 0 when

1 d mπ
ω(µ1 1 − µ0 0 ) 2 = , m = 0, 1, 2, 3, . . . (17.2.20)
2 2

or

mπ
ωmc = 1 , m = 0, 1, 2, 3, . . . (17.2.21)
d(µ1 1 − µ0 0 ) 2

The mode that corresponds to the m-th cut-off frequency above is labeled the TEm mode.
Thus TE0 mode is the mode that has no cut-off or propagates at all frequencies. This is
shown in Figure 17.6 where the TE mode profiles are similar since they are dual to each
other. The boundary conditions at the dielectric interface is that the field and its normal
derivative have to be continuous. The TE0 or TM0 mode can satisfy this boundary condition
at all frequencies, but not the TE1 or TM1 mode. At the cut-off frequency, the field outside
the slab has to become flat implying the α0x = 0 implying no guidance.
182 Electromagnetic Field Theory

Figure 17.6: Mode profiles of the TM modes of a dielectric slab. The TE modes are dual to
the TM modes and have similar mode profiles.

Next, we will elucidate more physics of the dielectric slab guided mode. At cut-off, α0x = 0,
2
and from the dispersion relation that α0x = βz2 − β02 ,
√
βz = ω µ0 0 ,

for all the modes. Hence, the phase velocity, ω/βz , and the group velocity, dω/dβz are that
of the outer region. This is because when α0x = 0, the wave is not evanescent outside, and
the energy of the mode is predominantly carried by the exterior field.
When ω → ∞, the radius of the circle in the plot of y3 becomes increasingly larger. As
seen from Figure 17.4, the solution for β1x → nπ d for all the modes. From the dispersion
relation for Region 1,

2 ≈ ω √µ  ,
q
βz = ω 2 µ1 1 − β1x 1 1 ω→∞ (17.2.22)

since ω 2 µ1 1  β1x
2
= (nπ/d)2 . Hence the group and phase velocities approach that of the
dielectric slab. This is because when ω → ∞, α0x → ∞, implying the rapid exponential
decay of the fields outside the waveguide. Therefore, the fields are trapped or confined in the
slab and propagating within it. Because of this, the dispersion diagram of the different modes
appear as shown in Figure 17.7. In this figure,5 kc1 , kc2 , and kc3 are the cut-off wave number
or frequency of the first three modes. Close to cut-off, the field is traveling mostly outside the
√
waveguide, and kz ≈ ω µ0 ε0 . Hence, both the phase and group velocities approach that of
5 Please note again that in this course, we will use β and k interchangeably for wavenumbers.
Dielectric Waveguides 183

the outer medium as shown in the figure. When the frequency increases, the mode is tightly
√
confined in the dielectric slab, and hence, kz ≈ ω µ1 ε1 . Both the phase and group velocities
approach that of Region 1 as shown.

Figure 17.7: Here, we have kz versus k1 plots for dielectric slab waveguide. Near its cut-off,
the energy of the mode is in the outer region, and hence, its group velocity is close to that of
the outer region. At high frequencies, the mode is tightly bound to the slab, and its group
velocity approaches that of the dielectric slab (courtesy of J.A. Kong [32]).

17.2.2 TM Case
For the TM case, a similar guidance condition analogous to (17.2.1) can be derived but with
the understanding that the reflection coefficients in (17.2.1) are now TM reflection coefficients.
Similar derivations show that the above guidance conditions, for 2 = 0 , µ2 = µ0 , reduce to
s 2
d2

0 d d 2
d
β1x tan β1x = ω (µ1 1 − µ0 0 ) − β1x , even modes (17.2.23)
1 2 2 4 2
s 2
d2

0 d d d
− β1x cot β1x = ω 2 (µ1 1 − µ0 0 ) − β1x , odd modes (17.2.24)
1 2 2 4 2
184 Electromagnetic Field Theory

Note that for equation (17.2.1), when we have two parallel metallic plates, RT M = 1, and
RT E = −1, and the guidance condition becomes
mπ
1 = e−2jβ1x d ⇒ β1x = , m = 0, 1, 2, . . . , (17.2.25)
d
These are just the guidance conditions for parallel plate waveguides.

17.2.3 A Note on Cut-Off of Dielectric Waveguides

The concept of cut-off in dielectric waveguides is quite different from that of hollow waveguides
that we shall learn next. A mode is guided in a dielectric waveguide if the wave is trapped
inside, in this case, the dielectric slab. The trapping is due to the total internal reflections at
the top and the bottom interfaces of the waveguide. When total internal reflection ceases to
occur at any of the two interfaces, the wave is not guided or trapped inside the dielectric slab
anymore. This happens when αix = 0 where i can indicate the top-most or the bottom-most
region. In other words, the wave ceases to be evanescent in one of the Region i’s.
Lecture 18

Hollow Waveguides

Hollow waveguides are useful for high-power microwaves. Air has a higher breakdown voltage
compared to most materials, and hence, it could be a good medium for propagating high
electromagnetic energy. Also, hollow metallic waveguides are sufficiently shielded from the
rest of the world so that interference from other sources is minimized. Furthermore, for radio
astronomy, they can provide a low-noise system immune to interference. Air generally has
less loss than materials, and loss is often the source of thermal noise. A low loss waveguide
is also a low noise waveguide.1

18.1 General Information on Hollow Waveguides

Many waveguide problems can be solved in closed form. An example is the coaxial waveguide
previously discussed. In addition, there are many other waveguide problems that have closed
form solutions. Closed form solutions to Laplace and Helmholtz equations are obtained by the
separation of variables method. The separation of variables method works only for separable
coordinate systems. (There are 11 separable coordinates for Helmholtz equation, but 13 for
Laplace equation.) Some examples of separable coordinate systems are cartesian, cylindrical,
and spherical coordinates. But these three coordinates are about all we need to know for
solving many engineering problems. For other than these three coordinates, complex special
functions need to be defined for their solutions, which are hard to compute. Therefore, more
complicated cases are now handled with numerical methods using computers.
When a waveguide has a center conductor or two conductors like a coaxial cable, it can
support a TEM wave where both the electric field and the magnetic field are orthogonal to
the direction of propagation. The uniform plane wave is an example of a TEM wave, for
instance. However, when the waveguide is hollow or is filled completely with a homogeneous
medium, without a center conductor, it cannot support a TEM mode as we shall prove next.
1 The fluctuation dissipation theorem [113, 114] says that when a system loses energy to the environment,

it also receives the same amount of energy from the environment for energy conservation. In a word, a lossy
system loses energy to its environment, but it receives energy back from the environment in terms of thermal
noise.

185
186 Electromagnetic Field Theory

Much of the materials of this lecture can be found in [32, 83, 92].

18.1.1 Absence of TEM Mode in a Hollow Waveguide

Figure 18.1: Absence of TEM mode in a hollow waveguide enclosed by a PEC wall. The
magnetic field lines form a closed loop due to the absence of magnetic charges.

We would like to prove by contradiction (reductio ad absurdum) that a hollow waveg-

uide as shown in Figure 18.1 (i.e. without a center conductor) cannot support a TEM mode
as follows. If we assume that TEM mode does exist, then the magneticfield has to end on
itself due to the absence of magnetic charges. In this case, it is clear that C Hs · dl 6= 0 about
any closed contour following the magnetic field lines. But Ampere’s law states that the above
is equal to
  
Hs · dl = jωε E · dS + J · dS (18.1.1)
C S S

The left-hand side of the above equation is clearly nonzero by the above argument. But for
a hollow waveguide, J = 0 and the above becomes
 
Hs · dl = jωε E · dS (18.1.2)
C S

Hence, this equation cannot be satisfied unless on the right-hand side there are Ez 6= 0
component. This implies that a TEM mode where both Ez and Hz are zero in a hollow
waveguide without a center conductor cannot exist.
By the above argument, in a hollow waveguide filled with homogeneous medium, only
TEz (TE to z) or TMz (TM to z) modes can exist like the case of a layered medium. For
a TEz wave (or TE wave), Ez = 0, Hz 6= 0 while for a TMz wave (or TM wave), Hz = 0,
Ez 6= 0. These classes of problems can be decomposed into two scalar problems like the layered
medium case, by using the pilot potential method. However, when the hollow waveguide is
filled with a center conductor, the TEM mode can exist in addition to TE and TM modes.
Hollow Waveguides 187

We begin by studying some simple closed form solutions to hollow waveguides, such as
the rectangular waveguides. These closed form solutions offer physical insight into the prop-
agation of waves in a hollow waveguide. Another waveguide with slightly more complicated
closed form solutions is the circular hollow waveguide. The solutions need to be sought in
terms of Bessel functions. Another waveguide with very complicated closed form solutions
is the elliptical waveguide. However, the solutions are too complicated to be considered; the
preferred method of solving these complicated problems is via numerical method.

18.1.2 TE Case (Ez = 0, Hz 6= 0, TEz case)

In this case, the field inside the waveguide is TE to z or TEz . To ensure such a TE field, we
can write the E field as

E(r) = ∇ × ẑΨh (r) (18.1.3)

By construction, equation (18.1.3) will guarantee that Ez = 0. Here, Ψh (r) is a scalar

potential and ẑ is called the pilot vector. 2
The waveguide is assumed source free and filled with a lossless, homogeneous material.
Eq. (18.1.3) also satisfies the source-free condition since, clearly, ∇ · E = 0. And hence, from
Maxwell’s equations that

∇ × E = jωµH (18.1.4)
∇ × H = −jωεE (18.1.5)

it can be shown that

∇ × ∇ × E − ω 2 µεE = 0 (18.1.6)

Furthermore, using the appropriate vector identiy, such as the back-of-the-cab formula, it
can be shown that the electric field E(r) satisfies the following Helmholtz wave equation (or
partial differential equation) that

(∇2 + β 2 )E(r) = 0 (18.1.7)

where β 2 = ω 2 µε. Substituting (18.1.3) into (18.1.7), we get

(∇2 + β 2 )∇ × ẑΨh (r) = 0 (18.1.8)

In the above, we assume that ∇2 ∇ × ẑΨ = ∇ × ẑ(∇2 Ψ), or that these operators commute.3
Then it follows that

∇ × ẑ(∇2 + β 2 )Ψh (r) = 0 (18.1.9)

2 It“pilots” the field so that it is transverse to z.
3 Thisis a mathematical parlance, and a commutator is defined to be [A, B] = AB − BA for two operators
A and B. If these two operators commute, then [A, B] = 0.
188 Electromagnetic Field Theory

Thus, if Ψh (r) satisfies the following Helmholtz wave equation or partial differential equa-
tion

(∇2 + β 2 )Ψh (r) = 0 (18.1.10)

then (18.1.9) is satisfied, and so is (18.1.7). Hence, the E field constructed with (18.1.3)
satisfies Maxwell’s equations, if Ψh (r) satisfies (18.1.10).

Figure 18.2: A hollow metallic waveguide with a center conductor (left), and without a center
conductor (right).

Next, we look at the boundary condition for Ψh (r) which is derivable from the boundary
condition for E. The boundary condition for E is that n̂ × E = 0 on C, the PEC wall of the
waveguide. But from (18.1.3), using the back-of-the-cab (BOTC) formula,

n̂ × E = n̂ × (∇ × ẑΨh ) = −n̂ · ∇Ψh = 0 (18.1.11)

In applying the BOTC formula, one has to be mindful that ∇ operates on a function to its
right, and the function Ψh should be placed to the right of the ∇ operator.
∂ ∂
In the above n̂ · ∇ = n̂ · ∇s where ∇s = x̂ ∂x + ŷ ∂y (a 2D gradient operator) since n̂ has
no z component. The boundary condition (18.1.11) then becomes

n̂ · ∇s Ψh = ∂n Ψh = 0 on C (18.1.12)

where C is the waveguide wall where ∂n is a shorthand notation for n̂ · ∇s operator which is a
scalar operator. The above is also known as the homogeneous Neumann boundary condition.

Furthermore, in a waveguide, just as in a transmission line case, we are looking for traveling
wave solutions of the form exp(∓jβz z) for (18.1.10), or that

Ψh (r) = Ψhs (rs )e∓jβz z (18.1.13)

Hollow Waveguides 189

where rs = x̂x + ŷy, or in short, Ψhs (rs ) = Ψhs (x, y) is a 2D function. Thus, ∂n Ψh = 0
∂2 2
implies that ∂n Ψhs = 0. With this assumption, ∂z 2 → −βz , and (18.1.10) becomes even

simpler, namely that,

(∇s 2 + β 2 − βz 2 )Ψhs (rs ) = (∇s 2 + βs2 )Ψhs (rs ) = 0 , ∂n Ψhs (rs ) = 0, on C (18.1.14)
where ∇2s 2 2 2
= ∂ /∂x + ∂ /∂y and 2
=β − βs2 2
βz2 .
The above is a boundary value problem
(BVP) for a 2D waveguide problem. The above 2D wave equation is also called the reduced
wave equation.

18.1.3 TM Case (Ez 6= 0, Hz = 0, TMz Case)

Repeating similar treatment for TM waves, the TM magnetic field is then
H = ∇ × ẑΨe (r) (18.1.15)
where
(∇2 + β 2 )Ψe (r) = 0 (18.1.16)
We need to derive the boundary condition for Ψe (r) from the fundamental boundary condition
that n̂ × E = 0 on the waveguide wall. To this end, we find the corresponding E field by
taking the curl of the magnetic field in (18.1.15), and thus the E field is proportional to
∂
Ψe + ẑβ 2 Ψe
E ∼ ∇ × ∇ × ẑΨe (r) = ∇∇ · (ẑΨe ) − ∇2 ẑΨe = ∇ (18.1.17)
∂z
where we have used the BOTC formula to simplify the above. The tangential component of
the above is n̂ × E which is proportional to
∂
Ψe + n̂ × ẑβ 2 Ψe
n̂ × ∇
∂z
In the above, n̂ × ∇ is a tangential derivative, and it is clear that the above will be zero if
Ψe = 0 on the waveguide wall. Therefore, if
Ψe (r) = 0 on C, (18.1.18)
where C is the waveguide wall, then,
n̂ × E(r) = 0 on C (18.1.19)
Equation (18.1.18) is also called the homogeneous Dirichlet boundary condition.
Next, we assume that
z
Ψe (r) = Ψes (rs )e∓jβz (18.1.20)
This will allow us to replace ∂ 2 /∂z 2 = −βz2 . Thus, with some manipulation, the boundary
value problem (BVP) related to equation (18.1.16) reduces to a simpler 2D problem, i.e.,
(∇s 2 + βs2 )Ψes (rs ) = 0 (18.1.21)
with the homogeneous Dirichlet boundary condition that
Ψes (rs ) = 0, rs on C (18.1.22)
To illustrate the above theory, we can solve some simple waveguides problems.
190 Electromagnetic Field Theory

18.2 Rectangular Waveguides

Rectangular waveguides are among the simplest waveguides to analyze because closed form
solutions exist in cartesian coordinates. One can imagine traveling waves in the xy directions
bouncing off the walls of the waveguide causing standing waves to exist inside the waveguide.
As shall be shown, it turns out that not all electromagnetic waves can be guided by
a hollow waveguide. Only when the wavelength is short enough, or the frequency is high
enough that an electromagnetic wave can be guided by a waveguide. When a waveguide
mode cannot propagate in a waveguide, that mode is known to be cut-off. The concept of
cut-off for hollow waveguide is quite different from that of a dielectric waveguide we have
studied previously.

18.2.1 TE Modes (Hz 6= 0, H Modes or TEz Modes)

For this mode, the scalar potential Ψhs (rs ) satisfies

∂
(∇s 2 + βs 2 )Ψhs (rs ) = 0, Ψhs (rs ) = 0 on C (18.2.1)
∂n

where βs 2 = β 2 − βz 2 . A viable solution using separation of variables4 for Ψhs (x, y) is then

Ψhs (x, y) = A cos(βx x) cos(βy y) (18.2.2)

where βx 2 + βy2 = βs2 . One can see that the above is the representation of standing waves
in the xy directions. It is quite clear that Ψhs (x, y) satisfies the BVP and boundary condi-
tions defined by equation (18.2.1). Furthermore, cosine functions, rather than sine functions
are chosen with the hindsight that the above satisfies the homogenous Neumann boundary
condition at x = 0, and y = 0 surfaces.

Figure 18.3: The schematic of a rectangular waveguide. By convention, the length of the
longer side is usually named a.
4 For those who are not familiar with this topic, please consult p. 385 of Kong [32].
Hollow Waveguides 191

To further satisfy the boundary condition at x = a, and y = b surfaces, it is necessary

that the boundary condition for eq. (18.1.12) is satisfied or that

∂x Ψhs (x, y)|x=a ∼ sin(βx a) cos(βy y) = 0, (18.2.3)

∂y Ψhs (x, y)|y=b ∼ cos(βx x) sin(βy b) = 0, (18.2.4)

The above puts constraints on βx and βy , implying that βx a = mπ, βy b = nπ where m and
n are integers. Hence, (18.2.2) becomes
 mπ   nπ 
Ψhs (x, y) = A cos x cos y (18.2.5)
a b
where
 mπ 2  nπ 2
βx2 + βy2 = + = βs2 = β 2 − βz 2 (18.2.6)
a b
Clearly, (18.2.5) satisfies the requisite homogeneous Neumann boundary condition at the
entire waveguide wall.
At this point, it is prudent to stop and ponder on what we have done. Equation (18.2.1)
is homomorphic to a matrix eigenvalue problem

A · xi = λi xi (18.2.7)

where xi is the eigenvector and λi is the eigenvalue. Therefore, βs2 is actually an eigenvalue,
and Ψhs (rs ) is an eigenfunction (or an eigenmode), which is analogous to an eigenvector. Here,
the eigenvalue βs2 is indexed by m, n, so is the eigenfunction in (18.2.5). The corresponding
eigenmode is also called the TEmn mode.
The above condition on βs2 expressed by (18.2.6) is also known as the guidance condition
for the modes in the waveguide. Furthermore, from (18.2.6),
r  mπ 2  nπ 2
p
βz = β − βs = β 2 −
2 2 − (18.2.8)
a b
And from (18.2.8), when the frequency is low enough, then
 mπ 2  nπ 2
βs2 = + > β 2 = ω 2 µε (18.2.9)
a b
and βz becomes pure imaginary and the mode cannot propagate or becomes evanescent in
the z direction.5 For fixed m and n, the frequency at which the above happens is called the
cutoff frequency of the TEmn mode of the waveguide. It is given by
r
1 mπ 2  nπ 2
ωmn,c = √ + (18.2.10)
µε a b
5 We have seen this happening in a plasma medium earlier and also in total internal reflection.
192 Electromagnetic Field Theory

When ω < ωmn,c , or the wavelength is longer than a certain value, the TEmn mode is
evanescent and cannot propagate inside the waveguide. A corresponding cutoff wavelength is
then
2
λmn,c = (18.2.11)
m 2 n 2 1/2



[ a + b ]

So when λ > λmn,c , the mode cannot propagate inside the waveguide.

Lowest Guided Mode in a Rectangular Waveguide

When m = n = 0, then Ψh (r) = Ψhs (x, y) exp(∓jβz z) is a function independent of x and y.
Then E(r) = ∇ × ẑΨh (r) = ∇s × ẑΨh (r) = 0. It turns out the only way for Hz 6= 0 is for
H(r) = ẑH0 which is a static field in the waveguide. This is not a very interesting mode, and
thus TE00 propagating mode is assumed not to exist and not useful. So the TEmn modes
cannot have both m = n = 0. As such, the TE10 mode, when a > b, is the mode with the
lowest cutoff frequency or longest cutoff wavelength. Only when the frequency is above this
cutoff frequency and the wavelength is shorter than this cutoff wavelength, the TE10 mode
can propagate.
For the TE10 mode, for the mode to propagate, from (18.2.11), it is needed that

λ < λ10,c = 2a (18.2.12)

The above has the nice physical meaning that the wavelength has to be smaller than 2a in
order for the mode to fit into the waveguide. As a mnemonic, we can think that photons have
“sizes”, corresponding to its wavelength. Only when its wavelength is small enough can the
photons go into (or be guided by) the waveguide. The TE10 mode, when a > b, is also the
mode with the lowest cutoff frequency or longest cutoff wavelength.
It is seen with the above analysis, when the wavelength is short enough, or frequency is
high enough, many modes can be guided. Each of these modes has a different group and
phase velocity. But for most applications, a single guided mode only is desirable. Hence,
the knowledge of the cutoff frequencies of the fundamental mode (the mode with the lowest
cutoff frequency) and the next higher mode is important. This allows one to pick a frequency
window within which only a single mode can propagate in the waveguide.
It is to be noted that when a mode is cutoff, the field is evanescent, and there is no real
power flow down the waveguide: Only reactive power is carried by such a mode.
Lecture 19

More on Hollow Waveguides

We have seen that the hollow waveguide is the simplest of waveguides. Closed form solutions
exist for such waveguides as seen in the rectangular waveguide case. The solution is elegantly
simple and beautiful requiring only trigonometric functions. So we will continue with the
study of the rectangular waveguide, and then address another waveguide, the circular waveg-
uide where closed form solutions exist also. However, the solution has to be expressed in
terms of “Bessel functions”, called special functions. As the name implies, these functions are
seldom used outside the context of studying wave phenomena. Bessel functions in cylindrical
coordinates are the close cousin of the sinusoidal functions in cartesian coordinates. Whether
Bessel functions are more complex or esoteric compared to sinusoidal functions is in the eye
of the beholder. Once one is familiarized with them, they are simple. They are also the
function that describes the concentric ripple wave that you see in your tea cup every morning
(see Figure 19.1)!

Figure 19.1: The ripple wave in your tea cup is describable by a Bessel function (courtesy of
dreamstime.com).

193
194 Electromagnetic Field Theory

19.1 Rectangular Waveguides, Contd.

We have seen the mathematics for the TE modes of a rectangular waveguide. We shall study
the TM modes and the modes of a circular waveguide in this lecture.

19.1.1 TM Modes (Ez 6= 0, E Modes or TMz Modes)

These modes are not the exact dual of the TE modes because of the boundary conditions.
The dual of a PEC (perfect electric conducting) wall is a PMC (perfect magnetic conducting)
wall. However, the previous exercise for TE modes can be repeated for the TM modes with
caution on the boundary conditions. The scalar wave function (or eigenfunction/eigenmode)
for the TM modes, satisfying the homogeneous Dirichlet (instead of Neumann)1 boundary
condition with (Ψes (rs ) = 0) on the entire waveguide wall is

 mπ   nπ 
Ψes (x, y) = A sin x sin y (19.1.1)
a b

Here, sine functions are chosen for the standing waves, and the chosen values of βx and βy
ensure that the boundary condition is satisfied on the x = a and y = b walls. Neither of the
m and n can be zero, lest Ψes (x, y) = 0, or the field is zero. Hence, both m > 0, and n > 0
are needed. Thus, the lowest TM mode is the TM11 mode. Thinking of this as an eigenvalue
problem, then the eigenvalue is

 mπ 2  nπ 2
βs2 = βx2 + βy2 = + (19.1.2)
a b

which is the same as the TE case. Therefore, the corresponding cutoff frequencies and cutoff
wavelengths for the TMmn modes are the same as the TEmn modes. Also, these TE and TM
modes are degenerate when they share the same eigevalues. Furthermore, the lowest modes,
TE11 and TM11 modes have the same cutoff frequency. Figure 19.2 shows the dispersion
curves for different modes of a rectangular waveguide. Notice that the group velocities of all
the modes are zero at cutoff, and then the group velocities approach that of the waveguide
medium as frequency becomes large. These observations can be explained physically.

1 Again, “homogeneous” here means “zero”.

More on Hollow Waveguides 195

Figure 19.2: Dispersion curves for a rectangular waveguide (courtesy of J.A. Kong [32]).
Notice that the lowest TM mode is the TM11 mode, and k is equivalent to β in this course.
At cutoff, the guided mode does not propagate in the z direction, and here, the group velocity
is zero. But when ω → ∞, the mode propagates in direction almost parallel to the axis of
the waveguide, and hence, the group velocity approaches that of the waveguide medium.

19.1.2 Bouncing Wave Picture

We have seen that the transverse variation of a mode in a rectangular waveguide can be
expanded in terms of sine and cosine functions which represent standing waves which are
superposition of two traveling waves, or that they are

[exp(−jβx x) ± exp(jβx x)] [exp(−jβy y) ± exp(jβy y)]

When the above is expanded and together with the exp(−jβz z) the mode propagating in the
z direction in addition to the standing waves in the tranverse direction. Or we see four waves
bouncing around in the xy directions and propagating in the z direction. The picture of this
bouncing wave can be depicted in Figure 19.3.
196 Electromagnetic Field Theory

Figure 19.3: The waves in a rectangular waveguide can be thought of as bouncing waves off
the four walls as they propagate in the z direction.

19.1.3 Field Plots

Given the knowledge of the scalar piloting potential of a waveguide, one can derive all the
field components. For example, for the TE modes, if we know Ψh (r), then

E = ∇ × ẑΨh (r), H = −∇ × E/(jωµ) (19.1.3)

Then all the electromagnetic field of a waveguide mode can be found, and similarly for TM
modes.

Plots of the fields of different rectangular waveguide modes are shown in Figure 19.4.
Notice that for higher m’s and n’s, with βx = mπ/a and βy = nπ/b, the corresponding βx
and βy are larger with higher spatialq frequencies. Thus, the transverse spatial wavelengths
are getting shorter. Also, since βz = β 2 − βx2 − βy2 , higher frequencies are needed to make
βz real in order to propagate the higher order modes or the high m and n modes.

Notice also how the electric field and magnetic field curl around each other. Since ∇×H =
jωεE and ∇ × E = −jωµH, they do not curl around each other “immediately” but with a
π/2 phase delay due to the jω factor. Therefore, the E and H fields do not curl around each
other at one location, but at a displaced location due to the π/2 phase difference. This is
shown in Figure 19.5.
More on Hollow Waveguides 197

Figure 19.4: Transverse field plots of different modes in a rectangular waveguide (courtesy
of Andy Greenwood. Original plots published in Lee, Lee, and Chuang, IEEE T-MTT, 33.3
(1985): pp. 271-274. [115]).

Figure 19.5: Field plot of a mode propagating in the z direction of a rectangular waveguide.
Notice that the E and H fields do not exactly curl around each other.
198 Electromagnetic Field Theory

19.2 Circular Waveguides

Another waveguide where closed-form solutions can be easily obtained is the circular hollow
waveguide as shown in Figure 19.6, but they involve the use of Bessel functions.

Figure 19.6: Schematic of a circular waveguide in cylindrical coordinates. It is one of the

separable coordinate systems.

19.2.1 TE Case
For a circular waveguide, it is best first to express the Laplacian operator, ∇s 2 = ∇s · ∇s , in
cylindrical coordinates. The first ∇s is a gradient operator while the second ∇s · is a divergence
operator: they have different physical meanings. Formulas for grad and div operators are
given in many text books [32, 116]. Doing a table lookup,

∂ 1 ∂
∇s Ψ = ρ̂ Ψ + φ̂
∂ρ ρ ∂φ

1 ∂ 1 ∂
∇s · A = ρAρ + Aφ
ρ ∂ρ ρ ∂φ
Then
1 ∂2
 
2 2

1 ∂ ∂ 2
∇s + βs Ψhs = ρ + + βs Ψhs (ρ, φ) = 0 (19.2.1)
ρ ∂ρ ∂ρ ρ2 ∂φ2

The above is the partial differential equation for field in a circular waveguide. It is an eigen-
value problem where βs2 is the eigenvalue, and Ψhs (rs ) is the eigenfunction (equivalence of an
eigenvector). Using separation of variables, we let

Ψhs (ρ, φ) = Bn (βs ρ)e±jnφ (19.2.2)

More on Hollow Waveguides 199

∂2
Then ∂φ2 → −n2 , and (19.2.1) simplifies to an ordinary differential equation which is

n2
 
1 d d
ρ − 2 + βs 2 Bn (βs ρ) = 0 (19.2.3)
ρ dρ dρ ρ

Here, dividing the above equation by βs2 , we can let βs ρ in (19.2.2) and (19.2.3) be x. Then
the above can be rewritten as
n2
 
1 d d
x − + 1 Bn (x) = 0 (19.2.4)
x dx dx x2
The above is known as the Bessel equation whose solutions are special functions denoted as
Bn (x).2
(1) (2)
These special functions are Jn (x), Nn (x), Hn (x), and Hn (x) which are called Bessel,
Neumann, Hankel function of the first kind, and Hankel function of the second kind, respec-
tively, where n is the order, and x is the argument.3 Since this is a second order ordinary
differential equation, it has only two independent solutions. Therefore, two of the four com-
monly encountered solutions of Bessel equation are independent. Thus, they can be expressed
in terms of each other. Their relationships are shown below:4
1
Bessel, Jn (x) = [Hn (1) (x) + Hn (2) (x)] (19.2.5)
2
1
Neumann, Nn (x) = [Hn (1) (x) − Hn (2) (x)] (19.2.6)
2j
Hankel–First kind, Hn (1) (x) = Jn (x) + jNn (x) (19.2.7)
(2)
Hankel–Second kind, Hn (x) = Jn (x) − jNn (x) (19.2.8)

It can be shown that

r
(1) 2 jx−j(n+ 1 ) π
Hn (x) ∼ e 2 2 , x→∞ (19.2.9)
πx
r
2 −jx+j(n+ 1 ) π
Hn (2) (x) ∼ e 2 2 , x→∞ (19.2.10)
πx
They correspond to traveling wave solutions when x = βs ρ → ∞. Since Jn (x) and Nn (x)
are linear superpositions of these traveling wave solutions, they correspond to standing wave
solutions. Moreover, Nn (x), Hn (1) (x), and Hn (2) (x) → ∞ when x → 0. Since the field
has to be regular when ρ → 0 at the center of the waveguide shown in Figure 19.6, the
only viable solution for the hollow waveguide, to be chosen from (19.2.5) to (19.2.9), is that
Bn (βs ρ) = AJn (βs ρ). Thus for a circular hollow waveguide, the eigenfunction or mode is of
the form

Ψhs (ρ, φ) = AJn (βs ρ)e±jnφ (19.2.11)

2 Studied by Friedrich Wilhelm Bessel, 1784-1846.
3 Some textbooks use Yn (x) for Neumann functions.
4 Their relations with each other are similar to those between exp(−jx), sin(x), and cos(x).
200 Electromagnetic Field Theory

To ensure that the eigenfunction and the eigenvalue are unique, boundary condition for the
partial differential equation is needed. The homogeneous Neumann boundary condition,5 . or
that ∂n Ψhs = 0, on the PEC waveguide wall then translates to
d
Jn (βs ρ) = 0, ρ=a (19.2.12)
dρ

Defining Jn 0 (x) = d
dx Jn (x),
6
the above is the same as

Jn 0 (βs a) = 0 (19.2.13)

The above are the zeros of the derivative of Bessel function and they are tabulated in many
textbooks and handbooks. The m-th zero of Jn 0 (x) is denoted to be βnm in many books.7
Plots of Bessel functions and their derivatives are shown in Figure 19.8, and some zeros of
Bessel function and its derivative are also shown in Figure 19.9. With this knowledge, the
guidance condition for a waveguide mode is then

βs = βnm /a (19.2.14)

for the TEnm mode. From the above, βs2 can be obtained which is the eigenvalue of (19.2.1)
and (19.2.3). It is a constant independent
p of frequency.
Using the fact that βz = β 2 − βs2 , then βz will become pure imaginary if β 2 is small
enough (or the frequency low enough) so that β 2 < βs2 or β < βs . From this, the corresponding
cutoff frequency (the frequency below which βz becomes pure imaginary) of the TEnm mode
is
1 βnm
ωnm,c = √ (19.2.15)
µε a

When ω < ωnm,c , the corresponding mode cannot propagate in the waveguide as βz becomes
pure imaginary. The corresponding cutoff wavelength is
2π
λnm,c = a (19.2.16)
βnm
By the same token, when λ > λnm,c , the corresponding mode cannot be guided by the
waveguide. It is not exactly precise to say this, but this gives us the heuristic notion that if
wavelength or “size” of the wave or photon is too big, it cannot fit inside the waveguide.

19.2.2 TM Case
The corresponding partial differential equation and boundary value problem for this case is
1 ∂2
 
1 ∂ ∂ 2
ρ + + βs Ψes (ρ, φ) = 0 (19.2.17)
ρ ∂ρ ∂ρ ρ2 ∂φ2
5 Note that “homogeneous” here means “zero” in math.
6 Note that this is a standard math notation, which has a different meaning in some engineering texts.
7 Notably, Abramowitz and Stegun, Handbook of Mathematical Functions [117]. An online version is

available at [118].
More on Hollow Waveguides 201

with the homogeneous Dirichlet boundary condition, Ψes (a, φ) = 0, on the waveguide wall.
The eigenfunction solution is
Ψes (ρ, φ) = AJn (βs ρ)e±jnφ (19.2.18)
with the boundary condition that Jn (βs a) = 0. The zeros of Jn (x) are labeled as αnm is
many textbooks, as well as in Figure 19.9; and hence, the guidance condition is that for the
TMnm mode is that
αnm
βs = (19.2.19)
a
wherepthe eigenvalue for (19.2.17) is βs2 which is a constant independent of frequency. With
βz = β 2 − βs2 , the corresponding cutoff frequency is
1 αnm
ωnm,c = √ (19.2.20)
µε a
or when ω < ωnm,c , the mode cannot be guided. The cutoff wavelength is
2π
λnm,c = a (19.2.21)
αnm
with the notion that when λ > λnm,c , the mode cannot be guided.
It turns out that the lowest mode in a circular waveguide is the TE11 mode. It is actually a
close cousin of the TE10 mode of a rectangular waveguide. This can be gathered by comparing
their field plots: these modes morph into each other as we deform the shape of a rectangular
waveguide into a circular waveguide.

Figure 19.7: Side-by-side comparison of the field plots of the TE10 mode of a rectangular
waveguide versus that of the TE11 mode of a circular waveguide. If one is imaginative enough,
one can see that the field plot of one mode morphs into that of the other mode. Electric fields
are those that have to end on the waveguide walls.

Figure 19.8 shows the plots of Bessel function Jn (x) and its derivative Jn0 (x). Tables in
Figure 19.9 show the roots of Jn0 (x) and Jn (x) which are important for determining the cutoff
frequencies of the TE and TM modes of circular waveguides.
202 Electromagnetic Field Theory

Figure 19.8: Plots of the Bessel function, Jn (x), and its derivatives Jn0 (x). The zeros of
these functions are used to find the eigenvalue βs2 of the problem, and hence, the guidance
condition. The left figure is for TM modes, while the right figure is for TE modes. Here,
Jn0 (x) = dJn (x)/dx.

Figure 19.9: Table 2.3.1 shows the zeros of Jn0 (x), which are useful for determining the
guidance conditions of the TEmn mode of a circular waveguide. On the other hand, Table
2.3.2 shows the zeros of Jn (x), which are useful for determining the guidance conditions of
the TMmn mode of a circular waveguide.
More on Hollow Waveguides 203

Figure 19.10: Transverse field plots of different modes in a circular waveguide (courtesy of
Andy Greenwood. Original plots published in Lee, Lee, and Chuang [115]). The axially
symmetric TE01 mode has the lowest loss, and finds a number of real-world applications as
in radio astronomy.
204 Electromagnetic Field Theory
Lecture 20

More on Waveguides and

Transmission Lines

Waveguide is a fundamental component of microwave circuits and systems. The study of

closed form solutions offers us physical insight. One can use such insight to design more
complex engineering systems. We will use heuristics to understand some systems whose
designs follow from physical insight of simpler systems.
Also, we will show that the waveguide problem is homomorphic to the transmission line
problem. Here again, many transmission line techniques can be used to solve some complex
waveguide problems encountered in microwave and optical engineering by adding junction
capacitances and inductances.

20.1 Circular Waveguides, Contd.

As in the rectangular waveguide case, the guidance of the wave in a circular waveguide can
be viewed as bouncing waves in the radial direction. But these bouncing waves give rise
to standing waves expressible in terms of Bessel functions. The scalar potential (or pilot
potential) for the modes in the waveguide is expressible as

Ψαs (ρ, φ) = AJn (βs ρ)e±jnφ (20.1.1)

where α = h for TE waves and α = e for TM waves. The Bessel function or wave is expressible
in terms of Hankel functions as in (19.2.5). Since Hankel functions are traveling waves, Bessel
functions represent standing waves. Therefore, the Bessel waves can be thought of as bouncing
traveling waves as in the rectangular waveguide case. In the azimuthal direction, one can
express e±jnφ as traveling waves in the φ direction, or they can be expressed as cos(nφ) and
sin(nφ) which are standing waves in the φ direction.

205
206 Electromagnetic Field Theory

20.1.1 An Application of Circular Waveguide

Figure 20.1: Bouncing wave picture of the Bessel wave inside a circular waveguide for the
TE01 mode.

When a real-world waveguide is made, the wall of the metal waveguide is not made of perfect
electric conductor, but with some metal of finite conductivity. Hence, tangential E field is
not zero on the wall implying that n̂ · (E × H∗ ) 6= 0. Thus energy can dissipate into the
waveguide wall. It turns out that due to symmetry, the TE01 mode of a circular waveguide
has the lowest loss of all the waveguide modes including rectangular waveguide modes. Hence,
this waveguide mode is of interest to astronomers who are interested in building low-loss and
low-noise systems.1
The TE01 mode has electric field given by E = φ̂Eφ . Furthermore, looking at the magnetic
field, the current is mainly circumferential flowing in the φ direction. Moreover, by looking at
a bouncing wave picture of the guided waveguide mode, this mode has a small component of
tangential magnetic field on a waveguide wall: It becomes increasingly smaller as the frequency
increases (see Figure 20.1). The reason is that the wave vector for the waveguide becomes
increasingly parallel to the axis of the waveguide with a large βz component compared to the
βs component.2 The wave becomes paraxial in the high-frequency limit.
The tangential magnetic field needs to be supported by a surface current on the waveguide
wall. This implies that the surface current on the waveguide wall becomes smaller as the
frequency increases. The wall loss (or copper loss or eddy current loss) of the waveguide,
hence, becomes smaller for higher frequencies. In fact, for high frequencies, the TE01 mode
has the smallest copper loss of the waveguide modes: It becomes the mode of choice (see
Figure 20.2). Waveguides supporting the TE01 modes are used to connect the antennas of
the very large array (VLA) for detecting extra-terrestrial signals in radio astronomy [120] as
shown in Figure 20.3.

1 Low-loss systems are also low-noise due to energy conservation and the fluctuation dissipation theorem

[113, 114, 119].

2 Recall that for a fixed mode, β is independent of frequency.
s
More on Waveguides and Transmission Lines 207

Figure 20.2: Losses of different modes in a circular waveguide or radius 1.5 cm . It is seen
that at high frequencies, the TE01 mode has the lowest loss (courtesy of [121]).

Figure 20.3: Picture of the Very Large Array in New Mexico, USA (courtesy of [120]).
208 Electromagnetic Field Theory

Figure 20.4 shows two ways of engineering a circular waveguide so that the TE01 mode
is enhanced: (i) by using a mode filter that discourages the guidance of other modes but
not the TE01 mode, and (ii), by designing corrugated waveguide wall to discourage the flow
of axial current and hence, the propagation of the non-TE01 mode. More details of circular
waveguides can be found in [121]. Typical loss of a circular waveguide can be as low as 2
dB/km.3

As shall be shown, an open circular waveguide can be made into an aperture antenna
quite easily, because the fields of the aperture are axially symmetric. Such antenna is called a
horn antenna. Because of this, the radiation pattern of such an antenna is axially symmetric,
which can be used to produce axially symmetric circularly polarized (CP) waves. Ways to
enhance the TE01 mode are also desirable [122] as shown in Figure 20.5.

Figure 20.4: Ways to enhance the TE01 mode in a circular waveguide: (a) Using mode
filter that only allows the mode to go through. (b) Use corrugated waveguide to discourage
axial current flow. The field plot of the mode is shown in (c). Such waveguide is used in
radio astronomy to design the communication links between antennas in a very large array
(VLA [120]), or it is used in a circular horn antenna [122].

3 For optical fiber, this figure of merit (FOM) can be lower than 1 dB/km making the optical fiber a darling

for long-distance communication.

More on Waveguides and Transmission Lines 209

Figure 20.5: Picture of a circular horn antenna where corrugated wall is used to enhance the
TE01 mode (courtesy of [123]).

20.2 Remarks on Quasi-TEM Modes, Hybrid Modes,

and Surface Plasmonic Modes
We have analyzed some simple structures where closed form solutions are available. These
simple elegant solutions offer physical insight into how waves are guided, and how they are
cutoff from guidance. As has been shown, for some simple waveguides, the modes can be
divided into TEM, TE, and TM modes. However, most waveguides are not simple. We will
remark on various complexities that arise in real world applications.

20.2.1 Quasi-TEM Modes

Figure 20.6: Some examples of practical coaxial-like waveguides are microstrip line and co-
planar waveguide (left), and the optical fiber (right). The environments of these waveguides
are inhomogeneous media, and hence, a pure TEM mode cannot propagate on these waveg-
uides.
210 Electromagnetic Field Theory

Many waveguides cannot support a pure TEM mode even when two conductors are present.
For example, two pieces of metal make a transmission line, and in the case of a circular coax,
a TEM mode can propagate in the waveguide. But most two-metal transmission lines do not
support a pure TEM mode: Instead, they support a quasi-TEM mode. In the optical fiber
case, when the index contrast of the fiber is very small, the mode is quasi-TEM as it has to
degenerate to the TEM case when the contrast is absent.

Absence of TEM Modes in Inhomogeneously-Filled Waveguides

In the following, we will give physical arguments as to why a pure TEM mode cannot exists
in a microstrip line, a coplanar waveguide, and the optical fiber. When a wave is TEM, it
is necessary that the wave propagates with the phase velocity of the medium. But when a
uniform waveguide has inhomogeneity in between, as shown in Figure 20.6, this is not possible
anymore. We can prove this assertion by reductio ad absurdum.
Assume only TE wave in a piecewise homogeneous region, then the E field is
1
E= ∇ × ∇ × (ẑΨe ) (20.2.1)
jωεi
where εi is the permittivity of the region. By doing some algebra, and assume that the field
is a waveguide mode such that Ψe has e−jβz z dependence, then one can show that Ez is given
by
1
Ez = (β 2 − βz2 )Ψe (20.2.2)
jωεi i
The above derivation is certainly valid in a piecewise homogeneous region. But each of the
piecewise homogeneous media can be made arbitrary small, and hence, it is also valid for
inhomogeneous media. If this mode becomes TEM, then Ez = 0 and this is possible only if
βz = βi . In other words, the phase velocity of the waveguide mode is the same as a plane
TEM wave in the same medium.
Now assume that a TEM wave exists in both inhomogeneous regions of the microstrip line
or all three dielectric regions of the optical fiber in Figure 20.6. Then the phase velocities
in the z direction, determined by ω/βz of each region will be ω/βi of the respective region
where βi is the wavenumber of the i-th region. Hence, phase matching is not possible, and
the boundary condition cannot be satisfied at the dielectric interfaces.
Nevertheless, the lumped element circuit model of the transmission line is still a very good
model for such a waveguide. If the line capacitance and line inductances of such lines can be
estimated, βz can still be estimated. As shall be shown later, circuit theory is valid when the
frequency is low, or the wavelength is large compared to the size of the structures.

20.2.2 Hybrid Modes–Inhomogeneously-Filled Waveguides

For most inhomogeneously filled waveguides, the modes (eigenmodes or eigenfunctions) inside
are not cleanly classed into TE and TM modes, but with some modes that are the hybrid of
TE and TM modes. If the inhomogeneity is piecewise constant, some of the equations we have
derived before are still valid: In other words, in the homogeneous part (or constant part) of the
More on Waveguides and Transmission Lines 211

waveguide filled with piecewise constant inhomogeneity, the fields can still be decomposed into
TE and TM fields. But these fields are coupled to each other by the presence of inhomogeneity,
i.e., by the boundary conditions requisite at the interface between the piecewise homogeneous
regions. Or both TE and TM waves are coupled together and are present simultaneously, and
both Ez 6= 0 and Hz 6= 0. Some examples of inhomogeneously-filled waveguides where hybrid
modes exist are shown in Figure 20.7.
Sometimes, the hybrid modes are called EH or HE modes, as in an optical fiber. Never-
theless, the guidance is via a bouncing wave picture, where the bouncing waves are reflected
off the boundaries of the waveguides. In the case of an optical fiber or a dielectric waveguide,
the reflection is due to total internal reflection. But in the case of metalic waveguides, the
reflection is due to the metal walls.

Figure 20.7: Some examples of inhomogeneously filled waveguides where hybrid modes exist:
(top-left) A general inhomogeneously filled waveguide, (top-right) slab-loaded rectangular
waveguides, and (bottom) an optical fiber with core and cladding.

20.2.3 Guidance of Modes

Propagation of a plane wave in free space is by the exchange of electric stored energy and
magnetic stored energy. So the same thing happens in a waveguide. For example. in the
transmission line, the guidance is by the exchange of electric and magnetic stored energy via
the coupling between the line capacitance and the line inductance of the line. In this case,
the waveguide size, like the cross-section of a coaxial cable, can be made much smaller than
the wavelength.
In the case of hollow waveguides, the E and H fields are coupled through their space and
time variations. Namely,

∇ × E = −jωµH, ∇ × H = jωεE (20.2.3)

212 Electromagnetic Field Theory

Hence, the exchange of the energies stored is via the space that stores these energies, like
that of a plane wave. These waveguides work only when these plane waves can “enter” the
waveguide. Hence, the size of these waveguides has to be about half a wavelength.
The surface plasmonic waveguide is an exception in that the exchange is between the
electric field energy stored with the kinetic energy stored in the moving electrons in the
plasma instead of magnetic energy stored. This form of energy stored is sometimes referred
to as coming from kinetic inductance. Therefore, the dimension of the waveguide can be very
small compared to wavelength, and yet the surface plasmonic mode can be guided.

20.3 Homomorphism of Waveguides and Transmission

Lines
Previously, we have demonstrated mathematical homomorphism between plane waves in lay-
ered medium and transmission lines. Such homomorphism can be further extended to waveg-
uides and transmission lines. But unlike the plane wave in layered medium case, we cannot
replace the ∇ operator with −jβ in a waveguide. Hence, the mathematics is slightly more
elaborate. We can show this first for TE modes in a hallow waveguide, and the case for TM
modes can be established by invoking duality principle.4

20.3.1 TE Case
For this case, Ez = 0, and from Maxwell’s equations

∇ × H = jωεE (20.3.1)
∂
By letting ∇ = ∇s + ∇z , H = Hs + Hz where ∇z = ẑ ∂z , and Hz = ẑHz , and the subscript
s implies transverse to z components, then

(∇s + ∇z ) × (Hs + Hz ) = ∇s × Hs + ∇z × Hs + ∇s × Hz (20.3.2)

where it is understood that ∇z × Hz = 0. Notice that the first term on the right-hand side
of the above is pointing in the z direction. Therefore, by letting E = Es + Ez , and equating
transverse components in (20.3.1), we have5

∇z × Hs + ∇s × Hz = jωεEs (20.3.3)

To simplify the above equation, we shall remove Hz from above. Next, from Faraday’s law,
we have

∇ × E = −jωµH (20.3.4)

Again, by letting E = Es + Ez , we can let (20.3.4) be written as

∇s × Es + ∇z × Es + ∇s × Ez = −jωµ(Hs + Hz ) (20.3.5)
4I have not seen exposition of such mathematical homomorphism elsewhere except in very simple cases [32].
5 And from the above, it is obvious that ∇s × Hs = jωεEz , but this equation will not be used in the
subsequent derivation.
More on Waveguides and Transmission Lines 213

Equating z components of the above, we have

∇s × Es = −jωµHz (20.3.6)

The above allows us to express Hz in terms of Es . Using (20.3.6), Eq.(20.3.3) can be rewritten
as
1
∇z × Hs + ∇s × ∇s × Es = +jωεEs (20.3.7)
−jωµ

The above can be further simplified by noting that

∇s × ∇s × Es = ∇s (∇s · Es ) − ∇s · ∇s Es (20.3.8)

But since ∇ · E = 0, and Ez = 0 for TE modes, it also implies that ∇s · Es = 0. Also, from
Maxwell’s equations, we have previously shown that for a homogeneous source-free medium,

(∇2 + β 2 )E = 0 (20.3.9)

with Ez = 0 for TE mode, or that

(∇2 + β 2 )Es = 0 (20.3.10)

Assuming that we have a guided mode, then

∂2
Es ∼ e∓jβz z , Es = −βz 2 Es (20.3.11)
∂z 2
Therefore, (20.3.10) becomes

(∇s 2 + β 2 − βz 2 )Es = 0 (20.3.12)

or that

(∇s 2 + βs 2 )Es = 0 (20.3.13)

where βs2 = β 2 − βz2 is the transverse wave number. Consequently, from (20.3.8)

∇s × ∇s × Es = −∇2s Es = βs2 Es (20.3.14)

As such, (20.3.7) becomes

1
∇z × Hs = jωεEs + ∇s × ∇s × Es
jωµ
1
= jωεEs + βs 2 Es
jωµ
βs 2 βz 2
 
= jωε 1 − 2 = jωε 2 Es (20.3.15)
β β
214 Electromagnetic Field Theory

Letting βz = β cos θ, then the above can further be rewritten as

∇z × Hs = jωε cos2 θEs (20.3.16)

The above now resembles one of the two telegrapher’s equations that we seek. Now looking
at (20.3.4) again, assuming Ez = 0, equating transverse components, we have

∇z × Es = −jωµHs (20.3.17)

More explicitly, we can rewrite (20.3.16) and (20.3.17) in the above as

∂
ẑ × Hs = jωε cos2 θEs (20.3.18)
∂z

∂
ẑ × Es = −jωµHs (20.3.19)
∂z
The above now resembles the telegrapher’s equations. We can multiply (20.3.19) by ẑ× to
get

∂
Es = jωµẑ × Hs (20.3.20)
∂z
Now (20.3.18) and (20.3.20) are a set of coupled equations that look even more like the
telegrapher’s equations. We can have Es → V , ẑ × Hs → −I. µ → L, ε cos2 θ → C, and the
above resembles the telegrapher’s equations, or that the waveguide problem is homomorphic
to the transmission line problem. The characteristic impedance of this line is then
r r r
L µ µ 1 ωµ
Z0 = = 2
= = (20.3.21)
C ε cos θ ε cos θ βz

Therefore, the TE modes of a waveguide can be mapped into a transmission problem. This
can be done, for instance, for the TEmn mode of a rectangular waveguide. Then, in the above
r  mπ 2  nπ 2
βz = β 2 − − (20.3.22)
a b
Therefore, each TEmn mode will be represented by a different characteristic impedance Z0 ,
since βz is different for different TEmn modes.

20.3.2 TM Case
This case can be derived using duality principle. Invoking duality, and after some algebra,
then the equivalence of (20.3.18) and (20.3.20) become

∂
Es = jωµ cos2 θẑ × Hs (20.3.23)
∂z
More on Waveguides and Transmission Lines 215

∂
ẑ × Hs = jωεEs (20.3.24)
∂z
To keep the dimensions commensurate, we can let Es → V , ẑ × Hs → −I, µ cos2 θ → L,
ε → C, then the above resembles the telegrapher’s equations. We can thus let
r r r
L µ cos2 θ µ βz
Z0 = = = cos θ = (20.3.25)
C ε ε ωε
Please note that (20.3.21) and (20.3.25) are very similar to that for the plane wave case, which
are the wave impedance for the TE and TM modes, respectively.

Figure 20.8: A waveguide filled with layered medium is mathematically homomorphic to

a multi-section transmission line problem. Hence, transmission-line method can be used to
solve this problem, but junction capacitance and inductance are needed to model the junctions
correctly.

The above implies that if we have a waveguide of arbitrary cross section filled with layered
media, the problem can be mapped to a multi-section transmission line problem, and solved
with transmission line methods. When V and I are continuous at a transmission line junction,
Es and Hs will also be continuous. Hence, the transmission line solution would also imply
continuous E and H field solutions.

Figure 20.9: A multi-section waveguide is not exactly homormorphic to a multi-section trans-

mission line problem, when the cross section of the waveguides are not equal to each other.
Circuit elements are needed at the junctions to capture the physics at the waveguide junctions
as shown in the next figure.
216 Electromagnetic Field Theory

20.3.3 Mode Conversion

In the waveguide shown in Figure 20.8, there is no mode conversion at the junction interface.
Assuming a rectangular waveguide as an example, what this means is that if we send at TE10
into the waveguide, this same mode will propagate throughout the length of the waveguide.
The reason is that only this mode alone is sufficient to satisfy the boundary condition at the
junction interface. The mode profile does not change throughout the length of the waveguide.

To elaborate further, from our prior knowledge, the transverse fields of the waveguide,
e.g., for the TM mode, can be derived to be

Hs = ∇ × ẑΨes (rs )e∓jβz z (20.3.26)

∓βz
Es = ∇s Ψes (rs )e∓jβz z (20.3.27)
ωε

In the above, βs2 and Ψes (rs ) are eigenvalue and eigenfunction, respectively, that depend
only on the geometrical shape of the waveguide, but not the materials filling the waveguide.
These eigenfunctions are the same throughout different sections of the waveguide. Therefore,
boundary conditions can be easily satisfied at the junctions.

However, for a multi-junction waveguide show in Figure 20.9, tangential E and H con-
tinuous condition cannot be satisfied by a single mode in each waveguide alone: V and I
continuous at a transmission line junction will not guarantee the continuity of tangential E
and tangential H fields at the waveguide junction.

Multi-modes have to be assumed on both sides of the junction at each section in order
to match boundary conditions at the junction [83]. Moreover, mode matching method for
multiple modes has to be used at each junction. Typically, a single mode incident at a
junction will give rise to multiple modes reflected and multiple modes transmitted. The
multiple modes give rise to the phenomenon of mode conversion at a junction. Hence,
the waveguide may need to be modeled with multiple transmission lines where each mode is
modeled by a different transmission line with different characteristic impedances.

However, the operating frequency can be chosen so that only one mode is propagating at
each section of the waveguide, and the other modes are cutoff or evanescent. In this case, the
multiple modes at a junction give rise to localized energy storage at a junction. These energies
can be either inductive or capacitive. The junction effect may be modeled by a simple circuit
model as shown in Figure 20.10. These junction elements also account for the physics that
the currents and voltages are not continuous anymore across the junction. Moreover, these
junction lumped circuit elements account for the stored electric and magnetic energies at the
junction.
More on Waveguides and Transmission Lines 217

Figure 20.10: Junction circuit elements are used to account for stored electric and magnetic
energies at the junction. They also account for that the currents and voltages are not con-
tinuous across the junctions anymore as the fields of the dominant modes in each section as
shown in Figure 20.9 are not continuous anymore.
218 Electromagnetic Field Theory
Lecture 21

Cavity Resonators

Cavity resonators are important components of microwave and optical systems. They work
by constructive and destructive interference of waves in an enclosed region. They can be
used as filters, or as devices to enhance certain physical interactions. These can be radiation
antennas or electromagnetic sources such as magnetrons or lasers. They can also be used to
enhance the sensitivity of sensors. We will study a number of them, and some of them, only
heuristically in this lecture.

21.1 Transmission Line Model of a Resonator

The simplest cavity resonator is formed by using a transmission line. The source end can be
terminated by ZS and the load end can be terminated by ZL . When ZS and ZL are non-
dissipative, such as when they are reactive loads (capacitive or inductive), then no energy is
dissipitated as a wave is reflected off them. Therefore, if the wave can bounce and interfere
constructively between the two ends, a coherent solution or a resonant solution can exist due
to constructive inference.

The resonant solution exists even when the source is turned off. In mathematical par-
lance, this is a homogeneous solution to a partial differential equation or ordinary differential
equation, since the right-hand side of the pertinent equation is zero. The right-hand side of
these equations usually corresponds to a source term or a driving term. In physics parlance,
this is a natural solution since it exists naturally without the need for a driving or exciting
source.

219
220 Electromagnetic Field Theory

Figure 21.1: A simple resonator can be made by terminating a transmission line with two
reactive loads at its two ends, the source end with ZS and the load end with ZL .

The transverse resonance condition for 1D problem can be used to derive the resonance
condition, namely that

1 = ΓS ΓL e−2jβz d (21.1.1)

where ΓS and ΓL are the reflection coefficients at the source and the load ends, respectively,
βz the the wave number of the wave traveling in the z direction, and d is the length of
the transmission line. For a TEM mode in the transmission line, as in a coax filled with
homogeneous medium, then βz = β, where β is the wavenumber for the homogeneous medium.
Otherwise, for a quasi-TEM mode, βz = βe where βe is some effective wavenumber for a z-
propagating wave in a mixed medium. In general,

βe = ω/ve (21.1.2)

where ve is the effective phase velocity of the wave in the heterogeneous structure.
When the source and load impedances are replaced by short or open circuits, then the
reflection coefficients are −1 for a short, and +1 for an open circuit. The (21.1.1) above then
becomes

±1 = e−2jβe d (21.1.3)

The ± sign corresponds to different combinations of open and short circuits at the two ends
of the transmission lines. When a “+” sign is chosen, which corresponds to either both ends
are short circuit, or are open circuit, the resonance condition is such that

βe d = pπ, p = 0, 1, 2, . . . , or integer (21.1.4)

For a TEM or a quasi-TEM mode in a transmission line, p = 0 is not allowed as the voltage
will be uniformly zero on the transmisson line or V (z) = 0 for all z implying a trivial solution.
The lowest mode then is when p = 1 corresponding to a half wavelength on the transmission
line.
When the line is open at one end, and shorted at the other end in (21.1.1), the resonance
condition corresponds to the “−” sign in (21.1.3), which gives rise to

βe d = pπ/2, p odd (21.1.5)

Cavity Resonators 221

The lowest mode is when p = 1 corresponding to a quarter wavelength on the transmission

line, which is smaller than that of a transmission line terminated with short or open at both
ends. Designing a small resonator is a prerogative in modern day electronic design. For
example, miniaturization in cell phones calls for smaller components that can be packed into
smaller spaces.
A quarter wavelength resonator made with a coax is shown in Figure 21.2. It is easier to
make a short indicated at the left end, but it is hard to make a true open circuit as shown at
the right end. A true open circuit means that the current has to be zero. But when a coax is
terminated with an open, the electric current does end abruptly. The fringing field at the right
end gives rise to stray capacitance through which displacement current can flow in accordance
to the generalized Ampere’s law. Hence, we have to model the right end termination with
a small stray or fringing field capacitance as shown in Figure 21.2. This indicates that the
current does not abruptly go to zero at the right-hand side due to the presence of fringing
field and hence, displacement current.

Figure 21.2: A short and open circuited transmission line can be a resonator, but the open
end has to be modeled with a fringing field capacitance Cf since there is no exact open circuit.

21.2 Cylindrical Waveguide Resonators

Since a cylindrical waveguide1 is homomorphic to a transmission line, we can model a mode

in this waveguide as a transmission line. Then the termination of the waveguide with either
a short or an open circuit at its end makes it into a resonator.
Again, there is no true open circuit in an open ended waveguide, as there will be fringing
fields at its open ends. If the aperture is large enough, the open end of the waveguide radiates
and may be used as an antenna as shown in Figure 21.3.

1 Both rectangular and circular waveguides are cylindrical waveguides.

222 Electromagnetic Field Theory

Figure 21.3: A rectangular waveguide terminated with a short at one end, and an open circuit
at the other end. The open end can also act as an antenna as it also radiates (courtesy of
RFcurrent.com).

As previously shown, single-section waveguide resonators can be modeled p

with a transmis-
sion line using homomorphism with the appropriately chosen βz . Then, βz = β 2 − βs2 where
βs can be found by first solving a 2D waveguide problem corresponding to the reduced-wave
equation.
For a rectangular waveguide, for example, from previous lecture,
r  mπ 2  nπ 2
βz = β2 − − (21.2.1)
a b

for both TEmn and TMmn modes.2 If the waveguide is terminated with two shorts (which is
easy to make) at its ends, then the resonance condition is that

βz = pπ/d, p integer (21.2.2)

Together, using (21.2.1), we have the condition that

ω2  mπ 2  nπ 2  pπ 2
β2 = = + + (21.2.3)
c2 a b d
The above can only be satisfied by certain select frequencies, and these frequencies are the
resonant frequencies of the rectangular cavity. The corresponding mode is called the TEmnp
mode or the TMmnp mode depending on if these modes are TE to z or TM to z.
2 It is noted that for a certain mn mode, with a choice of frequency, β = 0 which does not happen in a
z
transmission line.
Cavity Resonators 223

The entire electromagnetic fields of the cavity can be found from the scalar potentials
previously defined, namely that
E = ∇ × ẑΨh , H = ∇ × E/(−jω) (21.2.4)
H = ∇ × ẑΨe , E = ∇ × H/(jωε) (21.2.5)

Figure 21.4: A waveguide filled with layered dielectrics can also become a resonator. The
transverse resonance condition can be used to find the resonant modes.

Since the layered medium problem in a waveguide is the same as the layered medium
problem in open space, we can use the generalized transverse resonance condition to find the
resonant modes of a waveguide cavity loaded with layered medium as shown in Figure 21.4.
This condition is repeated below as:
R̃− R̃+ e−2jβz d = 1 (21.2.6)
where d is the length of the waveguide section where the above is applied, and R̃− and R̃+
are the generalized reflection coefficient to the left and right of the center waveguide section.
The above is similar to the resonant condition using the transmission line model in (21.1.1),
except that now, we have replaced the transmission line reflection coefficient with TE or TM
generalized reflection coefficients.

21.2.1 βz = 0 Case
In this case, we can still look at the TE and the TM modes in the waveguide. This corresponds
to a waveguide mode that bounces off the waveguide wall, but make no progress in the z
direction. The modes are independent of z since βz = 0. It is quite easy to show that for the
TE case, a z-independent H = ẑH0 , and E = Es exist inside the waveguide, and for the TM
case, a z-independent E = ẑE0 , and H = Hs being the only components in the waveguide.
Consider now a single section waveguide. For the TE mode, if either one of the ends of
the waveguide is terminated with a PEC wall, then n̂ · H = 0 at the end. This will force
the z-independent H field to be zero in the entire waveguide. Thus for the TE mode, it can
only exist if both ends are terminated with open, but this mode is not trapped inside since it
easily leaks energy to the outside via the ends of the waveguide.
For the TM mode, since E = ẑE0 , it easily satisfy the boundary condition if both ends are
terminated with PEC walls since the boundary condition is that n̂ × E = 0. The wonderful
part about this mode is that the length or d of the cavity can be as short as possible.
224 Electromagnetic Field Theory

21.2.2 Lowest Mode of a Rectangular Cavity

The lowest TM mode in a rectanglar waveguide is the TM11 mode. At the cutoff of this mode,
the βz = 0 or p = 0, implying no variation of the field in the z direction. When the two ends
are terminated with metallic shorts, the tangential magnetic field is not shorted out. But the
tangential electric field is shorted to zero in the entire cavity, or that the TE mode cannot
exist. Therefore, the longitudinal electric field of the TM mode still exists (see Figures 21.5
and 21.6). As such, for the TM mode, m = 1, n = 1 and p = 0 is possible giving a non-zero
field in the cavity. This is the TM110 mode of the resonant cavity, which is the lowest mode in
the cavity if a > b > d. The top and side views of the fields of this mode is shown in Figures
21.5 and 21.6. The corresponding resonant frequency of this mode satisfies the equation
2  π 2  π 2
ω110
= + (21.2.7)
c2 a b

Figure 21.5: The top view of the E and H fields of a rectangular resonant cavity.

Figure 21.6: The side view of the E and H fields of a rectangular resonant cavity (courtesy
of J.A. Kong [32]).

For the TE modes, it is required that p 6= 0, otherwise, the field is zero in the cavity. For
example, it is possible to have the TE101 mode with nonzero E field. The resonant frequency
Cavity Resonators 225

of this mode is
2  π 2  π 2
ω101
= + (21.2.8)
c2 a d
Clearly, this mode has a higher resonant frequency compared to the TM110 mode if d < b.
The above analysis can be applied to circular and other cylindrical waveguides with βs
determined differently. For instance, for a circular waveguide, βs is determined differently
using Bessel functions, and for a general arbitrarily shaped waveguide, βs may have to be
determined numerically.

Figure 21.7: A circular resonant cavity made by terminating a circular waveguide (courtesy
of Kong [32]).

For a spherical cavity, one would have to analyze the problem in spherical coordinates.
The equations will have to be solved by the separation of variables using spherical harmonics.
Details are given on p. 468 of Kong [32].

21.3 Some Applications of Resonators

Resonators in microwaves and optics can be used for designing filters, energy trapping devices,
and antennas. As filters, they are used like LC resonators in circuit theory. A concatenation
of them can be used to narrow or broaden the bandwidth of a filter. As an energy trapping
device, a resonator can build up a strong field inside the cavity if it is excited with energy
close to its resonance frequency. They can be used in klystrons and magnetrons as microwave
sources, a laser cavity for optical sources, or as a wavemeter to measure the frequency of
226 Electromagnetic Field Theory

the electromagnetic field at microwave frequencies. An antenna is a radiator that we will

discuss more fully later. The use of a resonator can help in resonance tunneling to enhance
the radiation efficiency of an antenna.

21.3.1 Filters

Microstrip line resonators are often used to make filters. Transmission lines are often used to
model microstrip lines in a microwave integrated circuits (MIC)or monolithic MIC (MMIC).
In these circuits, due to the etching process, it is a lot easier to make an open circuit rather
than a short circuit. But a true open circuit is hard to make as an open ended microstrip line
has fringing field at its end as shown in Figure 21.8 [124, 125]. The fringing field gives rise to
fringing field capacitance as shown in Figure 21.2. Then the appropriate ΓS and ΓL can be
used to model the effect of fringing field capacitance. Figure 21.9 shows a concatenation of
several microstrip resonators to make a microstrip filter. This is like using a concatenation
of LC tank circuits to design filters in circuit theory.
Optical filters can be made with optical etalon as in a Fabry-Perot resonator, or concate-
nation of them. This is shown in Figure 21.10.

Figure 21.8: End effects and junction effects in a microwave integrated circuit [124, 125]
(courtesy of Microwave Journal).
Cavity Resonators 227

Figure 21.9: A microstrip filter designed using concatenated resonators. The connectors to
the coax cable are the SMA (sub-miniature type A) connectors (courtesy of aginas.fe.up.pt).

Figure 21.10: Design of a Fabry-Perot resonator [56, 83, 126, 127].

21.3.2 Electromagnetic Sources

Microwave sources are often made by transferring kinetic energy from an electron beam
to microwave energy. Klystrons, magnetrons, and traveling wave tubes are such devices.
228 Electromagnetic Field Theory

However, the cavity resonator in a klystron enhances the interaction of the electrons with the
microwave field allowing for such energy transfer, causing the field to grow in amplitude as
shown in Figure 21.11.

Magnetron cavity works also by transferring the kinetic energy of the electron into the
microwave energy. By injecting hot electrons into the magnetron cavity, the electromagnetic
cavity resonance is magnified by the absorbing kinetic energy from the hot electrons, giving
rise to microwave energy.

Figure 21.13 shows laser cavity resonator to enhance of light wave interaction with material
media. By using stimulated emission of electronic transition, light energy can be produced.

Figure 21.11: A klystron works by converting the kinetic energy of an electron beam into the
energy of a traveling microwave next to the beam (courtesy of Wiki [128]).
Cavity Resonators 229

Figure 21.12: A magnetron works by having a high-Q microwave cavity resonator. When the
cavity is injected with energetic electrons from the cathode to the anode, the kinetic energy
of the electron feeds into the energy of the microwave (courtesy of Wiki [129]).

Figure 21.13: A simple view of the physical principle behind the working of the laser (courtesy
of www.optique-ingenieur.org).

Energy trapping of a waveguide or a resonator can be used to enhance the efficiency of

a semiconductor laser as shown in Figure 21.14. The trapping of the light energy by the
heterojunctions as well as the index profile allows the light to interact more strongly with
230 Electromagnetic Field Theory

the lasing medium or the active medium of the laser. This enables a semiconductor laser
to work at room temperature. In 2000, Z. I. Alferov and H. Kroemer, together with J.S.
Kilby, were awarded the Nobel Prize for information and communication technology. Alferov
and Kroemer for the invention of room-temperature semiconductor laser, and Kilby for the
invention of electronic integrated circuit (IC) or the chip.

Figure 21.14: A semiconductor laser at work. Room temperature lasing is possible due to
both the tight confinement of light and carriers (courtesy of Photonics.com).

21.3.3 Frequency Sensor

A cavity resonator can be used as a frequency sensor. It acts as an energy trap, because it
will siphon off energy from a microwave when the microwave frequency hits the resonance
frequency of the cavity resonator. This can be used to determine the frequency of the passing
wave. Wavemeters are shown in Figures 21.15 and 21.16. As seen in the picture, there is
an entry microwave port for injecting microwave into the cavity, and another exit port for
the microwave to leave the cavity sensor. The passing microwave, when it hits the resonance
frequency of the cavity, will create a large field inside it. The larger field will dissipate more
energy on the cavity metallic wall, and gives rise to less energy leaving the cavity. This dip
in energy transmission reveals the frequency of the microwave.
Cavity Resonators 231

Figure 21.15: An absorption wave meter can be used to measure the frequency of microwave
(courtesy of Wiki [130]).

Figure 21.16: The innards of a wavemeter (courtesy of eeeguide.com).

232 Electromagnetic Field Theory
Lecture 22

Quality Factor of Cavities, Mode

Orthogonality

Cavity resonators are important for making narrow band filters. The bandwidth of a filter
is related to the Q or the quality factor to the cavity. A concatenation of cavity resonators
can be used to engineer different filter designs. Resonators can also be used to design various
sensing systems, as well as measurement systems. We will study this concept of Q in this
lecture.

Also, before we leave the lectures on waveguides and resonators, it will be prudent to
discuss mode orthogonality. Since this concept is very similar to eigenvector orthogonality
found in matrix or linear algebra, we will relate mode orthogonality in waveguides and cavities
to eigenvector orthogonality.

22.1 The Quality Factor of a Cavity–General Concept

The quality factor of a cavity or its Q measures how ideal or lossless a cavity resonator is.
An ideal lossless cavity resonator will sustain free oscillations forever, while most resonators
sustain free oscillations for a finite time. This is because of losses coming from radiation,
dissipation in the dielectric material filling the cavity, or resistive loss of the metallic part of
the cavity.

233
234 Electromagnetic Field Theory

22.1.1 Analogue with an LC Tank Circuit

Figure 22.1: For the circuit on the left, it will resonate forever even if the source is turned
off. But for the circuit on the right, the current in the circuit will decay with time due to
dissipation in the resistor.

One of the simplest resonators imaginable is the LC tank circuit. By using it as an analogue,
we can better understand the resonance of a cavity. When there is no loss in an LC tank
circuit, it can oscillate forever. Moreover, if we turn off the source, a free oscillation solution
exists.1 One can write the voltage-current relation in the circuit as
V
I= =VY (22.1.1)
jωL + 1/(jωC)
where
1
Y (ω) = (22.1.2)
jωL + 1/(jωC)
The above Y (ω) can be thought of as the transfer function of the linear system where the in
put is V (ω) and the output is I(ω). When the voltage is zero or turned off, a non-zero current
exist or persist at the resonance frequency of the oscillator. The resonant frequency is when
the denominator in the above equation is zero, so that I is finite despite V = 0. This √ resonant
frequency, obtained by setting the denominator of Y to zero, is given by ωR = 1/ LC.
When a small resistor is added in the circuit to give rise to loss, the voltage-current relation
becomes
V
I= =VY (22.1.3)
jωL + R + 1/(jωC)
1
Y = (22.1.4)
jωL + R + 1/(jωC)
1 This is analogous to the homogeneous solution of an ordinary differential equation.
Quality Factor of Cavities, Mode Orthogonality 235

Now, the denominator of the above functions can never go to zero for real ω. But there exists
complex ω that will make Y become infinite. These are the complex resonant frequencies of
the circuit. The homogeneous solution (also called the natural solution, or free oscillation)
can only exists at the complex resonant frequencies. With complex resonances, the voltage
and the current are decaying sinusoids.
By the same token, because of losses, the free oscillation in a cavity has electromagnetic
field with time dependence as follows:

E ∝ e−αt cos(ωt + φ1 ), H ∝ e−αt cos(ωt + φ2 ) (22.1.5)

That is, they are decaying sinusoids. The total time-average stored energy, which is propor-
2 2
tional to 14  |E| + 14 µ |H| is of the form
.
hWT i = hWE i + hWH i = W0 e−2αt (22.1.6)

If there is no loss, hWT i will remain constant. However, with loss, the average stored energy
1
will decrease to 1/e of its original value at t = τ = 2α . The Q of a cavity is defined as
the number of free oscillations in radians (rather than cycles) that the field undergoes before
the energy stored decreases to 1/e of its original value (see Figure 22.2). In a time interval
1 ω
τ = 2α , the number of free oscillations in radians is ωτ or 2α ; hence, the Q is defined to
be [32]
. ω
Q= (22.1.7)
2α
Q is an approximate concept, and makes sense only if the system has low loss.

Figure 22.2: A typical time domain response of a high Q system (courtesy of Wikipedia).
236 Electromagnetic Field Theory

Furthermore, by energy conservation, the decrease in stored energy per unit time must be
equal to the total power dissipated in the losses of a cavity, in other words,
dhWT i
hPD i = − (22.1.8)
dt
By further assuming that WT has to be of the form in (22.1.6), then

dhWT i .
− = 2αW0 e−2αt = 2αhWt i (22.1.9)
dt
From the above, we can estimate the decay constant

. hPD i
α= (22.1.10)
2hWT i

Hence, we can write equation (22.1.7) for the Q as

. ωhWT i
Q= (22.1.11)
hPD i

By further letting ω = 2π/T , we lent further physical interpretation to express Q as

. hWT i total energy stored

Q = 2π = 2π (22.1.12)
hPD iT Energy dissipated/cycle
In a cavity, the energy can dissipate in either the dielectric loss or the wall loss of the cavity
due to the finiteness of the conductivity. It has to be re-emphasized the Q is a low-loss
concept, and hence, the above formulas are only approximately true.

22.1.2 Relation to Bandwidth and Pole Location

The resonance of a system is related to the pole of the transfer function . For instance, in our
previous examples of the RLC tank circuit, the admittance Y can be thought of as a transfer
function in linear system theory: The input is the voltage, while the output is the current.
If we encounter the resonance of the system at a particular frequency, the transfer function
becomes infinite. This infinite value can be modeled by a pole of the transfer function in the
complex ω plane. In other words, in the vicinity of the pole in the frequency domain, the
transfer function of the system can be approximated by a single pole which is
A A
Y (ω) ∼ = (22.1.13)
ω − ωR ω − ω0 − jα
where we have assumed that ωR = ω0 + jα, the resonant frequency is complex. In principle,
when ω = ωR , the reflection coefficient becomes infinite, but this does not happen in practice
because ωR is complex, and ω, the operating frequency is real. In other words, the pole is
displaced slightly off the real axis to account for loss. Using a single pole approximation, it
√ that |Y (ω)| would peak at ω = ω0 . At ω = ω0 ± α, the magnitude of |Y (ω)|
is quite clear
will be 1/ 2 smaller, or that the power which is proportional to |Y (ω)|2 will be half as small.
Quality Factor of Cavities, Mode Orthogonality 237

Therefore, the half-power points compared to the peak are at ω = ω0 ± α. Therefore, the
full-width half maximum (FWHM) bandwidth is ∆ω = 2α. Thus the Q can be written as in
terms of the half-power bandwidth ∆ω of the system, viz.,
.
Q = ω0 /∆ω (22.1.14)
The above implies that the narrower the bandwidth, the higher the Q of the system. Typical
plots of transfer function versus frequency for a system with different Q’s are shown in Figure
22.3.

Figure 22.3: A typical system response versus frequency for different Q’s using (22.1.4). The
Q is altered by changing the resistor R in the circuit.

22.1.3 Wall Loss and Q for a Metallic Cavity

If the cavity is filled with air, then the loss comes mainly from the metallic loss or copper-loss
from the cavity wall. In this case, the power dissipated on the wall is given by [32]
 
1 ∗ 1
hPD i = <e (E × H ) · n̂ dS = <e (n̂ × E) · H∗ dS (22.1.15)
2 S 2 S

where S is the surface of the cavity wall.2 Here, (n̂ × E) is the tangential component of
the electric field which would have been zero if the cavity wall is made of ideal PEC. Also,
2 We have used the cyclic identity that a · (b × c) = b · (c × a) = c · (a × b) in the above (see Some Useful

Mathematical Formulas).
238 Electromagnetic Field Theory

n̂ is taken to be the outward pointing normal at the surface S. The β (or k) vector in
the transmitted medium is very large due to the high-conductivity of the wall. Due to the
phase-matching condition, the transmitted wave β vector is almost normal to the interface.
Therefore, we can approximate the transmitted wave as a wave propagating normal to the
interface. In the metal, it decays predominantly in the direction of propagation which is
normal to the surface.

Figure 22.4: Due to high σ in the metal, and large |βt |, phase matching condition requires
the transmitted wave vector to be almost normal to the surface.

For such a wave, we can approximate n̂ × E = Ht Zm where Zmqis the intrinsic

q impedance
for the metallic conductor, as shown in Section 8.1, which is Zm = εm ≈ −j σ = ωµ
µ µ
p
2σ (1+ ω

j),3 where we have assumed that εm ≈ − jσ ω , and Ht is the tangential magnetic field. This
relation between E and H will ensure that power is flowing into the metallic surface. Hence,
 r r 
1 ωµ 2 1 ωµ 2
hPD i = <e (1 + j) |Ht | dS = |Ht | dS (22.1.16)
2 S 2σ 2 2σ S

By further assuming that the stored electric and magnetic energies of a cavity are equal
to each other at resonance, the stored energy can be obtained by

1
hWT i = µ |H|2 dV (22.1.17)
2 V

Written explicitly, the Q becomes

2 2

p |H| dV 2 |H| dV
Q = 2ωµσ  V 2 = V 2 (22.1.18)
|Ht | dS δ |Ht | dS
S S

In the above, δ is the skin depth of the metallic wall. Hence, the more energy stored there
is with respect to the power dissipated, the higher the Q of a resonating system. Also, the
lower the metal loss, or the smaller the skin depth, the higher the Q would be.
Notice that in (22.1.18), the numerator is a volume integral and hence, is proportional to
volume, while the denominator is a surface integral and is proportional to surface. Thus, the
3 When an electromagnetic wave enters a conductive region with a large β, it can be shown that the wave

is refracted to propagate normally to the surface as shown in Figure 22.4, and hence, this formula can be
applied.
Quality Factor of Cavities, Mode Orthogonality 239

Q, a dimensionless quantity, is roughly proportional to

V
Q∼ (22.1.19)
Sδ
where V is the volume of the cavity, while S is its surface area. From the above, it is noted
that a large cavity compared to its skin depth has a larger Q than an small cavity.

22.1.4 Example: The Q of TM110 Mode

For the TM110 mode, as can be seen from the previous lecture, the only electric field is
E = ẑEz , where
 πx   πy 
Ez = E0 sin sin (22.1.20)
a b
The magnetic field can be derived from the electric field using Maxwell’s equation or Faraday’s
law, and
j πb


jω ∂  πx   πy 
Hx = 2 EZ = E0 sin cos (22.1.21)
ω µ ∂y ωµ a b
j π


−jω ∂  πx   πy 
Hy = 2 EZ = − a E0 cos sin (22.1.22)
ω µ ∂x ωµ a b
Therefore
  0  b  a h i
2 2 2
|H| dV = dx dy dz |Hx | + |Hy |
V −d 0 0
a2  0  b 
|E0 |
= 2 2 dx dy dz
ω µ −d 0 0
   
π 2 2  πx     π 2
2 πy 2 πx
 
2 πy
sin cos + cos sin
b a b a a b
2 2  
|E0 | π a b
= 2 2 + d (22.1.23)
ω µ 4 b a
A cavity has six faces, finding the tangential exponent at each face and integrate
  b  a h i
2 2
|Ht | dS = 2 dx dy |Hx | + |Hy |
0 0
S
 0  a  0  b
2 2
+2 dx dz |Hx (y = 0)| + 2 dy dz |Hy (x = 0)|
−d 0 −d 0
2

2
2
2 πb
2 2 πa
 
2 |E0 | π ab 1 1 2 ad 2 bd
= 2 2 + 2 + 2 2 |E0 | + 2 2 |E0 |
ω µ 4 a2 b ω µ 2 ω µ 2
2 2  
π |E0 | b a ad bd
= + + 2 + 2 (22.1.24)
ω 2 µ2 2a 2b b a
240 Electromagnetic Field Theory

Hence the Q is
ad bd


2 b + a
Q= b a ad bd

(22.1.25)
δ 2a + 2b + b2 + a2

The result shows that the larger the cavity, the higher the Q. This is because the Q, as
mentioned before, is the ratio of the energy stored in a volume to the energy dissipated over
the surface of the cavity.

22.2 Mode Orthogonality and Matrix Eigenvalue Prob-

lem
It turns out that the modes of a waveguide or a resonator are orthogonal to each other. This
is intimately related to the orthogonality of eigenvectors of a matrix operator.4 Thus, it is
best to understand this by the homomorphism between the electromagnetic mode problem
and the matrix eigenvalue problem. Because of this similarity, electromagnetic modes are also
called eigenmodes. Thus it is prudent that we revisit the matrix eigenvalue problem (EVP)
here.

22.2.1 Matrix Eigenvalue Problem (EVP)

It is known in matrix theory that if a matrix is hermitian, then its eigenvalues are all real.
Furthermore, their eigenvectors with distinct eigenvalues are orthogonal to each other [77].
Assume that an eigenvalue and an eigenvector exists for the hermitian matrix A. Then
A · vi = λi vi (22.2.1)

Dot multiplying the above from the left by vi† where † indicates conjugate transpose, then
the above becomes
vi† · A · vi = λi vi† · vi (22.2.2)
†
Since A is hermitian, or A = A, then the quantity vi† · A · vi is purely real. Moreover, the
quantity vi† · vi is positive real. So in order for the above to be satisfied, λi has to be real.
To prove orthogonality of eigenvectors, now, assume that A has two eigenvectors with
distinct eigenvalues such that
A · vi = λi vi (22.2.3)
A · vj = λj vj (22.2.4)

Left dot multiply the first equation with vj† and do the same to the second equation with vi† ,
one gets
vj† · A · vi = λi vj† · vi (22.2.5)
vi† · A · vj = λj vi† · vj (22.2.6)
4 This mathematical homomorphism is not discussed in any other electromagnetic textbooks.
Quality Factor of Cavities, Mode Orthogonality 241

Taking the conjugate transpose of (22.2.5) in the above, and since A is hermitian, their left-
hand sides (22.2.5) and (22.2.6) become the same. Subtracting the two equations, we arrive
at

0 = (λi − λj )vj† · vi (22.2.7)

For distinct eigenvalues, λi 6= λj , the only way for the above to be satisfied is that

vj† · vi = Ci δij (22.2.8)

Hence, eigenvectors of a hermitian matrix with distinct eigenvalues are orthogonal to each
other. The eigenvalues are also real.

22.2.2 Homomorphism with the Waveguide Mode Problem

We shall next show that the problem for finding the waveguide modes or eigenmodes is
analogous to the matrix eigenvalue problem. The governing equation for a waveguide mode
is a BVP involving the reduced wave equation previously derived, or

∇2s ψi (rs ) + βis

2
ψi (rs ) = 0, ⇒ −∇2s ψi (rs ) = βis
2
ψi (rs ) (22.2.9)

with the pertinent homogeneous Dirichlet or Neumann boundary condition, depending on if

TE or TM modes are considered. In the above, the differential operator ∇2s is analogous to
2
the matrix operator A, the eigenfunction ψi (rs ) is analogous to the eigenvector vi , and βis
is analogous to the eigenvalue λi .
In the above, we can think of ∇2s as a linear operator that maps a function to another
function, viz.,

∇2s ψi (rs ) = fi (rs ) (22.2.10)

Hence it is analogous to a matrix operator A, that maps a vector to another vector, viz.,

A·x=f (22.2.11)

We shall elaborate this further next.

Discussion on Functional Space

To think of a function ψ(rs ) as a vector, one has to think in the discrete world.5 If one needs
to display the function ψ(rs ), on a computer, one will evaluate the function ψ(rs ) at discrete
N locations rls , where l = 1, 2, 3, . . . , N . For every rls or every l, there is a scalar number
ψ(rls ). These scalar numbers can be stored in a column vector in a computer indexed by l.
The larger N is, the better is the discrete approximation of ψ(rs ). In theory, N needs to be
infinite to describe this function exactly.6
5 Some of these concepts are discussed in [35, 131].
6 In mathematical parlance, the index for ψ(rs ) is uncountably infinite or nondenumerable.
242 Electromagnetic Field Theory

From the above discussion, a function is analogous to a vector and a functional space is
analogous to a vector space in linear or matrix algebra. However, a functional space is infinite
dimensional. But in order to compute on a computer with finite resource, such functions are
approximated with finite length vectors. Or infinite dimensional vector spaces are replaced
with finite dimensional ones to make the problem computable. Such an infinite dimensional
functional space is also called a Hilbert space.
It is also necessary to define the inner product between two vectors in a functional space
just as inner product between two vectors in an matrix vector space. The inner product (or
dot product) between two vectors in matrix vector space is
N
X
vit · vj = vi,l vj,l (22.2.12)
l=1

The analogous inner product between two vectors in function space is7

hψi , ψj i = drs ψi (rs )ψj (rs ) (22.2.13)
S

where S denotes the cross-sectional area of the waveguide over which the integration is per-
formed. The left-hand side is the shorthand notation for inner product in functional space or
the infinite dimensional functional Hilbert space.
Another requirement for a vector in a functional Hilbert space is that it contains finite
energy or that

Ef = drs |ψi (rs )|2 (22.2.14)
S

is finite. The above is analogous to that for a matrix vector v as

N
X
Em = |vl |2 (22.2.15)
l=1

The square root of the above is often used to denote the “length”, the “metric”, or the “norm”
of the vector. Finite energy also implies that the vectors are of finite length.

22.2.3 Proof of Orthogonality of Waveguide Modes8

Because of the aforementioned discussion, we see the similarity between a functional Hilbert
space, and the matrix vector space. In order to use the result of the matrix EVP, one key
step is to prove that the operator ∇2s is hermitian. In matrix algebra, if the elements of a
matrix is explicitly available, a matrix A is hermitian if

Aij = A∗ji (22.2.16)

7 In many math books, the conjugation of the first function ψ is implied, but here, we follow the electro-
i
magnetic convention that the conjugation of ψi is not implied unless explicitly stated.
8 This may be skipped on first reading.
Quality Factor of Cavities, Mode Orthogonality 243

But if a matrix element is not explicitly available, a different way to define hermiticity is
required: we will use inner products to define it. A matrix operator is hermitian if
 †
†  †  ∗
x†i · A · xj = x†j · A · xi = x†j · A · xi = x†j · A · xi (22.2.17)

The first equality follows from standard matrix algebra,9 the second equality follows if A =
†
A , or that A is hermitian. The last equality follows because the quantity in the parenthesis
is a scalar, and hence, its conjugate transpose is just its conjugate. Therefore, if a matrix A
is hermitian, then
 ∗
x†i · A · xj = x†j · A · xi (22.2.18)

The above inner product can be used to define the hermiticity of the matrix operator A.
It can be easily extended to define the hermiticity of an operator in an infinite dimensional
Hilbert space where the matrix elements are not explicitly available.
Hence, using the inner product definition in (22.2.13) for infinite dimensional functional
Hilbert space, a functional operator ∇2s is hermitian if
hψi∗ , ∇2s ψj i = (hψj∗ , ∇2s ψi i)∗ (22.2.19)
We can rewrite the left-hand side of the above more explicitly as

hψi∗ , ∇2s ψj i = drs ψi∗ (rs )∇2s ψj (rs ) (22.2.20)
S

and then the right-hand side of (22.2.19) can be rewritten more explicitly as

(hψj∗ , ∇2s ψi i)∗ = drs ψj (rs )∇2s ψi∗ (rs ) (22.2.21)
S

To prove the above equality in (22.2.19), one uses the identity that
∇s · [ψi∗ (rs )∇s ψj (rs )] = ψi∗ (rs )∇2s ψj (rs ) + ∇s ψi∗ (rs ) · ∇s ψj (rs ) (22.2.22)
Integrating the above over the cross sectional area S, and invoking Gauss divergence theorem
in 2D, one gets that
 
dln̂ · (ψi∗ (rs )∇s ψj (rs )) = drs ψi∗ (rs )∇2s ψj (rs )


C S

+ drs (∇ψi∗ (rs ) · ∇s ψj (rs )) (22.2.23)
S

where C the the contour bounding S or the waveguide wall. By applying the boundary
condition that ψi (rs ) = 0 or that n̂ · ∇s ψj (rs ) = 0, or a mixture thereof, then the left-hand
side of the above is zero. This will be the case be it TE or TM modes.
 
∗ 2
drs (∇ψi∗ (rs ) · ∇s ψj (rs ))


0= drs ψi (rs )∇s ψj (rs ) + (22.2.24)
S S
† † †
9 (A ·B· C)† = C · B · A [77].
244 Electromagnetic Field Theory

Applying the same treatment to (22.2.21), we get

 
drs ψj (rs )∇2s ψi∗ (rs ) drs (∇ψi∗ (rs ) · ∇s ψj (rs ))


0= + (22.2.25)
S S

The above two equations indicate that

hψi∗ , ∇2s ψj i = (hψj∗ , ∇2s ψi i)∗ (22.2.26)

proving that the operator ∇2s is hermitian. One can then use the above property to prove
the orthogonality of the eigenmodes when they have distinct eigenvalues, the same way we
have proved the orthogonality of eigenvectors. The above proof can be extended to the case
of a resonant cavity. The orthogonality of resonant cavity modes is also analogous to the
orthogonality of eigenvectors of a hermitian operator.
Lecture 23

Scalar and Vector Potentials

23.1 Scalar and Vector Potentials for Time-Harmonic

Fields

Now that we have studied the guidance of waves by waveguides, and the trapping of elec-
tromagnetic waves by cavity resonator, it will be interesting to consider how electromagnetic
waves radiate from sources. This is best done via the scalar and vector potential formulation.

Previously, we have studied the use of scalar potential Φ for electrostatic problems. Then
we learnt the use of vector potential A for magnetostatic problems. Now, we will study
the combined use of scalar and vector potential for solving time-harmonic (electrodynamic)
problems.

This is important for bridging the gap between static regime where the frequency is zero
or low, and dynamic regime where the frequency is not low. For the dynamic regime, it
is important to understand the radiation of electromagnetic fields. Electrodynamic regime
is important for studying antennas, communications, sensing, wireless power transfer ap-
plications, and many more. Hence, it is imperative that we understand how time-varying
electromagnetic fields radiate from sources.

It is also important to understand when static or circuit (quasi-static) regimes are im-
portant. The circuit regime solves problems that have fueled the microchip and integrated
circuit design (ICD) industry, and it is hence imperative to understand when electromagnetic
problems can be approximated with simple circuit problems and solved using simple laws
such as Kirchhoff current law (KCL) an Kirchhoff voltage law (KVL).

245
246 Electromagnetic Field Theory

Figure 23.1: Plot of the dynamic electric field around a dipole source, where only the field
in the upper half-space is shown. Close to the source, the field ressembles that of a static
electric dipole, but far away from the source, the electromagnetic field is detached from the
source: the source starts to shed energy to the far region.

23.2 Scalar and Vector Potentials for Statics, A Review

Previously, we have studied scalar and vector potentials for electrostatics and magnetostatics
where the frequency ω is identically zero. The four Maxwell’s equations for a homogeneous
medium are then

∇×E=0 (23.2.1)
∇×H=J (23.2.2)
∇ · εE = % (23.2.3)
∇ · µH = 0 (23.2.4)

Looking at the first equation above, and using the knowledge that ∇ × (∇Φ) = 0, we can
construct a solution to (23.2.1) easily. Thus, in order to satisfy the first of Maxwell’s equations
or Faraday’s law above, we let

E = −∇Φ (23.2.5)

Using the above in (23.2.3), we get,

∇ · ε∇Φ = −% (23.2.6)

Then for a homogeneous medium where ε is a constant, ∇ · ε∇Φ = ε∇ · ∇Φ = ε∇2 Φ, and we

have
%
∇2 Φ = − (23.2.7)
ε
which is the Poisson’s equation for electrostatics.
Scalar and Vector Potentials 247

Now looking at (23.2.4) where ∇ · µH = 0, we let

µH = ∇ × A (23.2.8)

Since ∇ · (∇ × A) = 0, the last of Maxwell’s equations (23.2.4) is automatically satisfied.

Next, using the above in the second of Maxwell’s equations above, we get

∇ × ∇ × A = µJ (23.2.9)

Now, using the fact that ∇ × ∇ × A = ∇(∇ · A) − ∇2 A, and Coulomb gauge that ∇ · A = 0,
we arrive at

∇2 A = −µJ (23.2.10)

which is the vector Poisson’s equation. Next, we will repeat the above derivation when ω 6= 0.

23.2.1 Scalar and Vector Potentials for Electrodynamics

To this end, assuming linearity, we will start with frequency domain Maxwell’s equations with
sources J and % included, and later see how these sources J and % can radiate electromagnetic
fields. Maxwell’s equations in the frequency domain are

∇ × E = −jωµH (23.2.11)
∇ × H = jωεE + J (23.2.12)
∇ · εE = % (23.2.13)
∇ · µH = 0 (23.2.14)

In order to satisfy the last Maxwell’s equation, as before, we let

µH = ∇ × A (23.2.15)

Now, using (23.2.15) in (23.2.11), we have

∇ × (E + jωA) = 0 (23.2.16)

Since ∇ × (∇Φ) = 0, the above implies that E + jωA = −∇Φ, or

E = −jωA − ∇Φ (23.2.17)

The above implies that the electrostatic theory of letting E = −∇Φ we have learnt previously
in Section 3.3.1 is not exactly correct when ω 6= 0. The second term above, in accordance to
Faraday’s law, is the contribution to the electric field from the time-varying magnetic field,
and hence, is termed the induction term.1
Furthermore, the above shows that once A and Φ are known, one can determine the fields
H and E assuming that J and % are given. To this end, we will derive equations for A and
Φ in terms of the sources J and %. Substituting (23.2.15) and (23.2.17) into (23.2.12) gives

∇ × ∇ × A = −jωµε(−jωA − ∇Φ) + µJ (23.2.18)

1 Notice that in electrical engineering, most concepts related to magnetic fields are inductive!
248 Electromagnetic Field Theory

Or upon rearrangement, after using that ∇ × ∇ × A = ∇∇ · A − ∇ · ∇A, we have

∇2 A + ω 2 µεA = −µJ + jωµε∇Φ + ∇∇ · A (23.2.19)
Moreover, using (23.2.17) in (23.2.14), we have
%
∇ · (jωA + ∇Φ) = − (23.2.20)
ε
In the above, (23.2.19) and (23.2.20) represent two equations for the two unknowns A and
Φ, expressed in terms of the known quantities, the sources J and %. But these equations are
coupled to each other. They look complicated and are rather unwieldy to solve at this point.
Fortunately, the above can be simplified! As in the magnetostatic case, the vector potential
A is not unique. To show this, one can always construct a new A0 = A + ∇Ψ that produces
the same magnetic field µH via (23.2.8), since ∇ × (∇Ψ) = 0. It is quite clear that µH =
∇ × A = ∇ × A0 . Moreover, one can further show that Φ is also non-unique [48]. Namely,
with
A0 = A + ∇Ψ (23.2.21)
Φ0 = Φ − jωΨ (23.2.22)
it can be shown that the new A0 and Φ0 produce the same E and H field. The above is known
as gauge transformation, clearly showing the non-uniqueness of A and Φ.
To make them unique, in addition to specifying what ∇ × A should be in (23.2.15), we
need to specify its divergence or ∇ · A as in the electrostatic case.2 A clever way to specify
the divergence of A is to choose it to simplify the complicated equations above in (23.2.19).
We choose a gauge so that the last two terms in the equation (23.2.19) cancel each other. In
other words, we let
∇ · A = −jωµεΦ (23.2.23)
The above is judiciously chosen so that the pertinent equations (23.2.19) and (23.2.20) will
be simplified and decoupled. With the use of (23.2.23) in (23.2.19) and (23.2.20), they now
become
∇2 A + ω 2 µεA = −µJ (23.2.24)
%
∇2 Φ + ω 2 µεΦ = − (23.2.25)
ε
Equation (23.2.23) is known as the Lorenz gauge3 and the above equations are Helmholtz
equations with source terms. Not only are these equations simplified, they can be solved
independently of each other since they are decoupled from each other.
Equations (23.2.24) and (23.2.25) can be solved using the Green’s function method we
have learnt previously. Equation (23.2.24) actually constitutes three scalar equations for the
three x, y, z components, namely that
∇2 Ai + ω 2 µεAi = −µJi (23.2.26)
2 This is akin to that given a vector A, and an arbitrary vector k, in addition to specifying what k × A is,

it is also necessary to specify what k · A is to uniquely specify A.

3 Please note that this Lorenz is not the same as Lorentz.
Scalar and Vector Potentials 249

where i above can be x, y, or z. Therefore, (23.2.24) and (23.2.25) together constitute

four scalar equations similar to each other. Hence, we need only to solve their point-source
response, or the Green’s function of these equations by solving

∇2 g(r, r0 ) + β 2 g(r, r0 ) = −δ(r − r0 ) (23.2.27)

where β 2 = ω 2 µε.
Previously, we have shown that when β = 0,
1
g(r, r0 ) = g(|r − r0 |) =
4π|r − r0 |

When β 6= 0, the correct solution is

0
0 0 e−jβ|r−r |
g(r, r ) = g(|r − r |) = (23.2.28)
4π|r − r0 |

which can be verified by back substitution.

By using the principle of linear superposition, or convolution, the solutions to (23.2.24)
and (23.2.25) are then
  0
e−jβ|r−r |
A(r) = µ dr0 J(r0 )g(|r − r0 |) = µ dr0 J(r0 ) (23.2.29)
4π|r − r0 |
  0
1 0 0 0 1 0 0e−jβ|r−r |
Φ(r) = dr %(r )g(|r − r |) = dr %(r ) (23.2.30)
ε ε 4π|r − r0 |

In the above dr0 is the shorthand notation for dx dy dz and hence, they are still volume
integrals. The above are three-dimensional convolutional integrals in space.

23.2.2 More on Scalar and Vector Potentials

It is to be noted that Maxwell’s equations are symmetrical and this is especially so when we
add a magnetic current M to Maxwell’s equations and magnetic charge %m to Gauss’s law.4
Thus the equations then become

∇ × E = −jωµH − M (23.2.31)
∇ × H = jωεE + J (23.2.32)
∇ · µH = %m (23.2.33)
∇ · εE = % (23.2.34)

The above can be solved in two stages, using the principle of linear superposition. First, we
can set M = 0, %m = 0, and J 6= 0, % 6= 0, and solve for the fields as we have done. Second,
we can set J = 0, % = 0 and M 6= 0, %m 6= 0 and solve for the fields next. Then the total
general solution, by linearity, is just the linear superposition of these two solutions.
4 In fact, Maxwell exploited this symmetry [38].
250 Electromagnetic Field Theory

For the second case, we can define an electric vector potential F such that

D = −∇ × F (23.2.35)

and a magnetic scalar potential Φm such that

H = −∇Φm − jωF (23.2.36)

By invoking duality principle, one gather that [47]

  0
e−jβ|r−r |
F(r) = ε dr0 M(r0 )g(|r − r0 |) = ε dr0 M(r0 ) (23.2.37)
4π|r − r0 |
  0
1 0 0 0 1 0 0 e−jβ|r−r |
Φm (r) = dr %m (r )g(|r − r |) = dr %m (r ) (23.2.38)
µ µ 4π|r − r0 |

As mentioned before, even though magnetic sources do not exist, they can be engineered.
In many engineering designs, one can use fictitious magnetic sources to enrich the diversity
of electromagnetic technologies.

23.3 When is Static Electromagnetic Theory Valid?

We have learnt in the previous section that for electrodynamics,

E = −∇Φ − jωA (23.3.1)

where the second term above on the right-hand side is due to induction, or the contribution
to the electric field from the time-varying magnetic filed. Hence, much things we learn in
potential theory that E = −∇Φ is not exactly valid. But simple potential theory that
E = −∇Φ is very useful because of its simplicity. We will study when static electromagnetic
theory can be used to model electromagnetic systems.
Since the third and the fourth Maxwell’s equations are derivable from the first two, let
us first study when we can ignore the time derivative terms in the first two of Maxwell’s
equations, which, in the frequency domain, are

∇ × E = −jωµH (23.3.2)
∇ × H = jωεE + J (23.3.3)

When the terms multiplied by jω above, which are associated with time derivatives, can
be ignored, then electrodynamics can be replaced with static electromagnetics, which are
much simpler.5 Quasi-static electromagnetic theory eventually gave rise to circuit theory
and telegraphy technology. Circuit theory consists of elements like resistors, capacitors, and
inductors. Given that we have now seen electromagnetic theory in its full form, we would
like to ponder when we can use simple static electromagnetics to describe electromagnetic
phenomena.
5 That is why Ampere’s law, Coulomb’s law, and Gauss’ law were discovered first.
Scalar and Vector Potentials 251

Figure 23.2: The electric and magnetic fields are great contortionists around a perfectly
conducting particle. They deform themselves to satisfy the boundary conditions, n̂ × E = 0,
and n̂ · H = 0 on the PEC surface, even when the particle is very small. In other words, the
fields vary on the length-scale of 1/L. Hence, ∇ ∼ 1/L which is large when L is small.

To see this lucidly, it is best to write Maxwell’s equations in dimensionless units or the
same units. Say if we want to solve Maxwell’s equations for the fields close to an object of
size L as shown in Figure 23.2. This object can be a small particle like the sphere, or it could
be a capacitor, or an inductor, which are small; but how small should it be before we can
apply static electromagnetics?
It is clear that these E and H fields will have to satisfy boundary conditions, n̂ × E = 0,
and n̂ · H = 0 on the PEC surface, which is de rigueur in the vicinity of the object as shown
in Figure 23.2 even when the frequency is low or the wavelength long. The fields become
great contortionists in order to do so. Hence, we do not expect a constant field around the
object but that the field will vary on the length scale of L. So we renormalize our length scale
by this length L by defining a new dimensionless coordinate system such that6
x y z
x0 = , y0 = , z0 = (23.3.4)
L L L
In other words, by so doing, then Ldx0 = dx, Ldy 0 = dy, and Ldz 0 = dz, and
∂ 1 ∂ ∂ 1 ∂ ∂ 1 ∂
= , = , = (23.3.5)
∂x L ∂x0 ∂y L ∂y 0 ∂z L ∂z 0
Take for example a function f (x) = sin(πx/L) which is a periodic function varying on the
length scale of L having a period of L. Then it is quite clear that df (x)/dx = (π/L) cos(πx/L).
When L is small representing a rapidly varying function, df (x)/dx ∼ O(1/L).7 But in the
new variable, f (Lx0 ) = sin(πx0 ) and df /dx0 = cos(πx0 ) which is O(1).
6 This dimensional analysis is often used by fluid dynamicists to study fluid flow problems [132].
7 Reads order of 1/L of order 1/L.
252 Electromagnetic Field Theory

In this manner, ∇ = L1 ∇0 where ∇0 is dimensionless; or ∇ will be very large when it

operates on fields that vary on the length scale of very small L, where ∇0 , which is an O(1)
operator, will not be large because it is in coordinates normalized with respect to L.
Then, the first two of Maxwell’s equations become
1 0
∇ × E = −jωµ0 H (23.3.6)
L
1 0
∇ × H = jωε0 E + J (23.3.7)
L
Here, we still have apples and oranges to compare with since E and H have different units;
we cannot compare quantities if they have different units. For instance, the ratio of E to the
H field has a dimension of impedance. To bring them to the same unit, we define a new E0
such that

η0 E0 = E (23.3.8)

where η0 = µ0 /ε0 ∼
p
= 377 ohms in vacuum. In this manner, the new E0 has the same unit
as the H field. Then, (23.3.6) and (23.3.7) become
η0 0
∇ × E0 = −jωµ0 H (23.3.9)
L
1 0
∇ × H = jωε0 η0 E0 + J (23.3.10)
L
With this change, the above can be rearranged to become
L
∇0 × E0 = −jωµ0 H (23.3.11)
η0
∇0 × H = jωε0 η0 LE0 + LJ (23.3.12)
p
By letting η0 = µ0 /ε0 , the above can be further simplified to become
ω
∇0 × E0 = −j LH (23.3.13)
c0
ω
∇0 × H = j LE0 + LJ (23.3.14)
c0
Notice now that in the above, H, E0 , and LJ have the same unit, and ∇0 is dimensionless
and is of order one, and ωL/c0 is also dimensionless.
Therefore, in the above, one can compare terms, and ignore the frequency dependent jω
term when
ω
L1 (23.3.15)
c0
The above is the same as kL  1 where k = ω/c0 = 2π/λ0 . Thus, when
L
2π 1 (23.3.16)
λ0
Scalar and Vector Potentials 253

the jω terms can be ignored and the first two Maxwell’s equations become static equations.
Consequently, the above criteria are for the validity of the static approximation when the time-
derivative terms in Maxwell’s equations can be ignored. When these criteria are satisfied, then
Maxwell’s equations can be simplified to and approximated by the following equations

∇0 × E0 = 0 (23.3.17)
∇0 × H = LJ (23.3.18)

which are the static equations, Faraday’s law and Ampere’s law of electromagnetic theory.
They can be solved together with Gauss’ laws, or the third or the fourth Maxwell’s equations.
In other words, one can solve, even in optics, where ω is humongous or the wavelength
very short, using static analysis if the size of the object L is much smaller than the optical
wavelength which is about 400 nm for blue light. Nowadays, plasmonic nano-particles of
about 10 nm can be made. If the particle is small enough compared to wavelength of the
light, electrostatic analysis can be used to study their interaction with light. And hence,
static electromagnetic theory can be used to analyze the wave-particle interaction. This was
done in one of the homeworks. Figure 23.3 shows an incident light whose wavelength is much
longer than the size of the particle. The incident field induces an electric dipole moment on
the particle, whose external field can be written as
 a 3
Es = (r̂2 cos θ + θ̂ sin θ) Es (23.3.19)
r

while the incident field E0 and the interior field Ei to the particle can be expressed as

E0 = ẑE0 = (r̂ cos θ − θ̂ sin θ)E0 (23.3.20)

Ei = ẑEi = (r̂ cos θ − θ̂ sin θ)Ei (23.3.21)

By matching boundary conditions, as was done in the homework, it can be shown that

εs − ε
Es = E0 (23.3.22)
εs + 2ε
3ε
Ei = E0 (23.3.23)
εs + 2ε

For a plasmonic nano-particle, the particle medium behaves like a plasma (see Lecture 8), and
εs in the above can be negative, making the denominators of the above expression very close
to zero. Therefore, the amplitude of the internal and scattered fields can be very large when
this happens, and the nano-particles will glitter in the presence of light. Even the ancient
Romans realized this!
Figure 23.4 shows a nano-particle induced in plasmonic oscillation by a light wave. Figure
23.5 shows that different color fluids can be obtained by immersing nano-particles in fluids
with different background permittivity (ε in (23.3.22) and (23.3.23)) causing the plasmonic
particles to resonate at different frequencies.
254 Electromagnetic Field Theory

Figure 23.3: A plane electromagnetic wave incident on a particle. When the particle size is
small compared to wavelength, static analysis can be used to solve this problem (courtesy of
Kong [32]).

Figure 23.4: A nano-particle undergoing electromagnetic oscillation when an electromagnetic

wave impinges on it (courtesy of sigmaaldric.com). The oscillation is inordinately large when
the incident wave’s frequency coincides with the resonance frequency of the plasmonic particle.
Scalar and Vector Potentials 255

Figure 23.5: Different color fluids containing nano-particles can be obtained by changing the
permittivity of the background fluids (courtesy of nanocomposix.com).

In (23.3.16), this criterion has been expressed in terms of the dimension of the object
L compared to the wavelength λ0 . Alternatively, we can express this criterion in terms of
transit time. The transit time for an electromagnetic wave to traverse an object of size L is
τ = L/c0 and ω = 2π/T where T is the period of one time-harmonic oscillation. Therefore,
(23.3.15) can be re-expressed as
2πτ
1 (23.3.24)
T
The above implies that if the transit time τ needed for the electromagnetic field to traverse
the object of length L is much small than the period of oscillation of the electromagnetic
field, then static theory can be used.
The finite speed of light gives rise to delay or retardation of electromagnetic signal when
it propagates through space. When this retardation effect can be ignored, then static elec-
tromagnetic theory can be used. In other words, if the speed of light had been infinite, then
there would be no retardation effect, and static electromagnetic theory could always be used.
Alternatively, the infinite speed of light will give rise to infinite wavelength, and criterion
(23.3.16) will always be satisfied, and static theory prevails always.

23.3.1 Quasi-Static Electromagnetic Theory

In closing, we would like to make one more remark. The right-hand side of (23.3.11), which
is Faraday’s law, is essential for capturing the physical mechanism of an inductor and flux
linkage. And yet, if we drop it, there will be no inductor in this world. To understand this
dilemma, let us rewrite (23.3.11) in integral form, namely,
 
0 L
E · dl = −jωµ0 dS · H (23.3.25)
C η 0 S

In the inductor, the right-hand side has been amplified by multiple turns, effectively increasing
S, the flux linkage area. Or one can think of an inductor as having a much longer effective
length Leff when untwined so as to compensate for decreasing frequency ω. Hence, the
importance of flux linkage or the inductor in circuit theory is not diminished unless ω = 0.
256 Electromagnetic Field Theory

By the same token, displacement current in (23.3.11) can be enlarged by using capacitors.
In this case, even when no electric current J flows through the capacitor, displacement current
flows and the generalized Ampere’s law becomes
 
H · dl = jωεη0 L dS · E0 (23.3.26)
C S

The displacement in a capacitor cannot be ignored unless ω = 0. Therefore, when ω 6= 0, or

in quasi-static case, inductors and capacitors in circuit theory are important, because they
amplify the flux linkage and the displacement current effects, as we shall study next. In
summary, the full physics of Maxwell’s equations is not lost in circuit theory: the induction
term in Faraday’s law, and the displacement current in Ampere’s law are still retained. That
explains the success of circuit theory in electromagnetic engineering!
Lecture 24

Circuit Theory Revisited

Circuit theory is one of the most successful and often used theories in electrical engineering.
Its success is mainly due to its simplicity: it can capture the physics of highly complex circuits
and structures, which is very important in the computer and micro-chip industry (or the IC
design industry). Simplicity rules! Now, having understood electromagnetic theory in its
full glory, it is prudent to revisit circuit theory and study its relationship to electromagnetic
theory [31, 32, 54, 65].
The two most important laws in circuit theory are Kirchoff current law (KCL) and Kirch-
hoff voltage law (KVL) [14, 52]. These two laws are derivable from the current continuity
equation and from Faraday’s law.

24.1 Kirchhoff Current Law

Figure 24.1: Schematic showing the derivation of Kirchhoff current law. All currents flowing
into a node must add up to zero.

257
258 Electromagnetic Field Theory

Kirchhoff current law (KCL) is a consequence of the current continuity equation, or that

∇ · J = −jω% (24.1.1)

It is a consequence of charge conservation. But it is also derivable from generalized Ampere’s

law and Gauss’ law for charge.1
First, we assume that all currents are flowing into a node as shown in Figure 24.1, and
that the node is non-charge accumulating with ω → 0. Then the charge continuity equation
becomes2

∇·J=0 (24.1.2)

By integrating the above current continuity equation over a volume containing the node, it
is easy to show that
N
X
Ii = 0 (24.1.3)
i

which is the statement of KCL. This is shown for the schematic of Figure 24.1.

24.2 Kirchhoff Voltage Law

Kirchhoff voltage law is the consequence of Faraday’s law. For the truly static case when
ω = 0, it is

∇×E=0 (24.2.1)

The above implies that E = −∇Φ, from which we can deduce that

− E · dl = 0 (24.2.2)
C

For statics, the statement that E = −∇Φ also implies that we can define a voltage drop
between two points, a and b to be
 b  b
Vba = − E · dl = ∇Φ · dl = Φ(rb ) − Φ(ra ) = Vb − Va (24.2.3)
a a

b
The equality a ∇Φ · dl = Φ(rb ) − Φ(ra ) can be understood by expressing this integral in one
dimension along a straight line segment, or that
 b
d
Φ · dx = Φ(rb ) − Φ(ra ) (24.2.4)
a dx
1 Some authors will say that charge conservation is more fundamental, and that Gauss’ law and Ampere’s

law are consistent with charge conservation and the current continuity equation.
2 One can also say that this is a consequenc of static Ampere’s law, ∇ × H = J. By taking the divergence

of this equation yields (24.1.2) directly.

Circuit Theory Revisited 259

A curved line can be thought of as a concatenation of many small straight line segments.

As has been shown before, to be exact, E = −∇Φ − ∂/∂tA, but we have ignored the
induction effect. Therefore, this concept is only valid in the low frequency or long wavelength
limit, or that the dimension over which the above is applied is very small so that retardation
effect can be ignored.

A good way to remember the above formula is that if Vb > Va . Since E = −∇Φ, then
the electric field points from point a to point b: Electric field always points from the point
of higher potential to point of lower potential. Faraday’s law when applied to the static case
for a closed loop of resistors shown in Figure 24.2 gives Kirchhoff voltage law (KVL), or that

N
X
Vj = 0 (24.2.5)
i

Notice that the voltage drop across a resistor is always positive, since the voltages to the left
of the resistors in Figure 24.2 are always higher than the voltages to the right of the resistors.
This implies that internal to the resistor, there is always an electric field that points from the
left to the right.

If one of the voltage drops is due to a voltage source, it can be modeled by a negative
resistor as shown in Figure 24.3. The voltage drop across a negative resistor is opposite to
that of a positive resistor. As we have learn from the Poynting’s theorem, negative resistor
gives out energy instead of dissipates energy.

Figure 24.2: Kichhoff voltage law where the sum of all voltages around a loop is zero, which
is the consequence of static Faraday’s law.
260 Electromagnetic Field Theory

Figure 24.3: A voltage source can also be modeled by a negative resistor.

Faraday’s law for the time-varying B flux case is

∂B
∇×E=− (24.2.6)
∂t

Writing the above in integral form, one gets

 
d
− E · dl = B · dS (24.2.7)
C dt s

We can apply the above to a loop shown in Figure 24.4, or a loop C that goes from a to b to
c to d to a. We can further assume that this loop is very small compared to wavelength so
that potential theory that E = −∇Φ can be applied. Furthermore, we assume that this loop
C does not have any magnetic flux through it so that the right-hand side of the above can be
set to zero, or


− E · dl = 0 (24.2.8)
C
Circuit Theory Revisited 261

Figure 24.4: The Kirchhoff voltage law for a circuit loop consisting of resistor, inductor, and
capacitor can also be derived from Faraday’s law at low frequency (courtesy of Ramo et al).

Figure 24.5: The voltage-current relation of an inductor can be obtained by unwrapping an

inductor coil, and then calculating its flux linkage.

Notice that this loop does not go through the inductor, but goes directly from c to d.
Then there is no flux linkage in this loop and thus
 b  c  d  a
− E · dl − E · dl − E · dl − E · dl = 0 (24.2.9)
a b c d

Inside the source or the battery, it is assumed that the electric field points opposite to the
direction of integration dl, and hence the first term on the left-hand side of the above is
positive and equal to V0 (t), while the other terms are negative. Writing out the above more
explicitly, after using (24.2.3), we have
V0 (t) + Vcb + Vdc + Vad = 0 (24.2.10)
Notice that in the above, in accordance to (24.2.3), Vb > Vc , Vc > Vd , and Va > Va . Therefore,
Vcb , Vdc , and Vad are all negative quantities but V0 (t) > 0. We will study the contributions
to each of the terms, the inductor, the capacitor, and the resistor more carefully next.
262 Electromagnetic Field Theory

24.3 Inductor
To find the voltage current relation of an inductor, we apply Faraday’s law to a closed loop
C 0 formed by dc and the inductor coil shown in the Figure 24.5 where we have unwrapped
the solenoid into a larger loop. Assume that the inductor is made of a perfect conductor,
so that the electric field E in the wire is zero. Then the only contribution to the left-hand
side of Faraday’s law is the integration from point d to point c, the only place in the loop
C 0 where E is not zero. We assume that outside the loop in the region between c and d,
potential theory applies, and hence, E = −∇Φ. Now, we can connect Vdc in the previous
equation to the flux linkage to the inductor. When the voltage source attempts to drive an
electric current into the loop, Lenz’s law (1834)3 comes into effect, essentially, generating an
opposing voltage. The opposing voltage gives rise to charge accumulation at d and c, and
therefore, a low frequency electric field at the gap at dc.
To this end, we form a new C 0 that goes from d to c, and then continue onto the wire that
leads to the inductor. But this new loop will contain the flux B generated by the inductor
current. Thus
  c 
d
E · dl = E · dl = −Vdc = − B · dS (24.3.1)
C0 d dt S0
0
As mentioned before, since the wire
is a PEC, the integration around the loop C is only
nonzero from d to c. In the above, S 0 B · dS is the flux linkage. The inductance L is defined
as the flux linkage per unit current, or
 
L= B · dS /I (24.3.2)
S0

So the voltage in (24.3.1) is then

d dI
Vdc = (LI) = L (24.3.3)
dt dt
since L is time independent.
Had there been a finite resistance in the wire of the inductor, then the electric field is
non-zero inside the wire. Taking this into account, we have
 
d
E · dl = RL I − Vdc = − B · dS (24.3.4)
dt S

Consequently,
dI
Vdc = RL I + L (24.3.5)
dt
Thus, to account for the loss of the coil, we add a resistor in the equation. The above becomes
simpler in the frequency domain, namely

Vdc = RL I + jωLI (24.3.6)

3 Lenz’s law can also be explained from Faraday’s law (1831).
Circuit Theory Revisited 263

24.4 Capacitance
The capacitance is the proportionality constant between the charge Q stored in the capacitor,
and the voltage V applied across the capacitor, or Q = CV . Then
Q
C= (24.4.1)
V
From the current continuity equation, one can easily show that in Figure 24.6,
dQ d dVda
I= = (CVda ) = C (24.4.2)
dt dt dt
where C is time independent. Integrating the above equation, one gets
 t
1
Vda (t) = Idt0 (24.4.3)
C −∞

The above looks quite cumbersome in the time domain, but in the frequency domain, it
becomes

I = jωCVda (24.4.4)

Figure 24.6: Schematic showing the calculation of the capacitance of a capacitor.

24.5 Resistor
The electric field is not zero inside the resistor as electric field is needed to push electrons
through it. As discussed in Section 8.3, a resistor is a medium where collision dominates. As
is well known,

J = σE (24.5.1)
264 Electromagnetic Field Theory

where σ is the conductivity of the medium. From this, we deduce that Vcb = Vc − Vb is a
negative number given by
 c  c
J
Vcb = − E · dl = − · dl (24.5.2)
b b σ

where we assume a uniform current J = ˆlI/A in the resistor where ˆl is a unit vector pointing
in the direction of current flow in the resistor. We can assumed that I is a constant along the
length of the resistor, and thus, J · dl = Idl/A, implying that
 c  c
Idl dl
Vcb = − = −I = −IR (24.5.3)
b σA b σA

where4
 c  c
dl ρdl
R= = (24.5.4)
b σA b A

Again, for simplicity, we assume long wavelength or low frequency in the above derivation.

24.6 Some Remarks

In this course, we have learnt that given the sources % and J of an electromagnetic system,
one can find Φ and A, from which we can find E and H. This is even true at DC or statics.
We have also looked at the definition of inductor L and capacitor C. But clever engineering
is driven by heuristics: it is better, at times, to look at inductors and capacitors as energy
storage devices, rather than flux linkage and charge storage devices.
Another important remark is that even though circuit theory is simpler that Maxwell’s
equations in its full glory, not all the physics is lost in it. The physics of the induction term
in Faraday’s law and the displacement current term in generalized Ampere’s law are still
retained and amplified by capacitor and inductor, respectively. In fact, wave physics is still
retained in circuit theory: one can make slow wave structure out a series of inductors and
capacitors. The lumped-element model of a transmission line is an example of a slow-wave
structure design. Since the wave is slow, it has a smaller wavelength, and resonators can
be made smaller: We see this in the LC tank circuit which is a much smaller resonator in
wavelength with L/λ  1 compared to a microwave cavity resonator for instance. Therefore,
circuit design is great for miniaturization. The short coming is that inductors and capacitors
generally have higher losses than air or vacuum.

24.7 Energy Storage Method for Inductor and Capacitor

Often time, it is more expedient to think of inductors and capacitors as energy storage devices.
This enables us to identify stray (also called parasitic) inductances and capacitances more
4 The resisitivity ρ = 1/σ where ρ has the unit of ohm-m, while σ has the unit of siemen/m.
Circuit Theory Revisited 265

easily. This manner of thinking allows for an alternative way of calculating inductances and
capacitances as well [31].
The energy stored in an inductor is due to its energy storage in the magnetic field, and it
is alternatively written, according to circuit theory, as

1 2
Wm = LI (24.7.1)
2

Therefore, it is simpler to think that an inductance exists whenever there is stray magnetic
field to store magnetic energy. A piece of wire carries a current that produces a magnetic
field enabling energy storage in the magnetic field. Hence, a piece of wire in fact behaves
like a small inductor, and it is non-negligible at high frequencies: Stray inductances occur
whenever there are stray magnetic fields.
By the same token, a capacitor can be thought of as an electric energy storage device
rather than a charge storage device. The energy stored in a capacitor, from circuit theory, is

1
We = CV 2 (24.7.2)
2

Therefore, whenever stray electric field exists, one can think of stray capacitances as we have
seen in the case of fringing field capacitances in a microstrip line.

24.8 Finding Closed-Form Formulas for Inductance and

Capacitance
Finding closed form solutions for inductors and capacitors is a difficult endeavor. As in
solving Maxwell’s equations or the waveguide proble, only certain geometries are amenable
to closed form solutions. Even a simple circular loop does not have a closed form solution for
its inductance L. If we assume a uniform current on a circular loop, in theory, the magnetic
field can be calculated using Bio-Savart law that we have learnt before, namely that


I(r0 )dl0 × R̂
H(r) = (24.8.1)
4πR2

But the above cannot be evaluated in closed form save in terms of complicate elliptic integrals
[117, 133]. Thus it is simpler to just measure the inductance.
However, if we have a solenoid as shown in Figure 24.7, an approximate formula for the
inductance L can be found if the fringing field at the end of the solenoid can be ignored.
The inductance can be found using the flux linkage method [29, 31]. Figure 24.8 shows the
schematic used to find the approximate inductance of this inductor.
266 Electromagnetic Field Theory

Figure 24.7: The flux-linkage method is used to estimate the inductor of a solenoid (courtesy
of SolenoidSupplier.Com).

Figure 24.8: Finding the inductor flux linkage approximately by assuming the magnetic field
is uniform inside a long solenoid.

The capacitance of a parallel plate capacitor can be found by solving a boundary value
problem (BVP) for electrostatics as shown in Section 3.3.4. The electrostatic BVP for ca-
pacitor involves Poisson’s equation and Laplace equation which are scalar equations [48].
Alternatively, variational expressions can be used to find the lower and upper bounds of
capacitors using, for example, Thomson’s theorem [48].
Circuit Theory Revisited 267

Figure 24.9: The capacitance between two charged conductors can be found by solving a
boundary value problem (BVP) involving Laplace equation as discussed in 3.3.4.

Assume a geometry of two conductors charged to +V and −V volts as shown in Figure

24.9. Surface charges will accumulate on the surfaces of the conductors. Using Poisson’s
equations, and Green’s function for Poisson’s equation, one can express the potential in
between the two conductors as due to the surface charges density σ(r). It can be expressed
as

1 σ(r0 )
Φ(r) = dS 0 (24.8.2)
ε S 4π|r − r0 |

where S = S1 + S2 is the union of two surfaces S1 and S2 . Since Φ has values of +V and −V
on the two conductors, we require that
 0
(
1 σ(r ) +V, r ∈ S1
Φ(r) = dS 0 = (24.8.3)
ε S 4π|r − r0 | −V, r ∈ S2

In the above, σ(r0 ), the surface charge density, is the unknown yet to be sought and it is
embedded in an integral. But the right-hand side of the equation is known. Hence, this
equation is also known as an integral equation. The integral equation can be solved by
numerical methods.
Having found σ(r), then it can be integrated to find Q, the total charge on one of the
conductors. Since the voltage difference between the two conductors is known, the capacitance
can be found as C = Q/(2V ).

24.9 Importance of Circuit Theory in IC Design

The clock rate of computer circuits has peaked at about 3 GHz due to the resistive loss, or
the I 2 R loss. At this frequency, the wavelength is about 10 cm. Since transistors and circuit
components are shrinking due to the compounding effect of Moore’s law, most components,
which are of nanometer dimensions, are much smaller than the wavelength. Thus, most of the
physics of electromagnetic signal in a microchip circuit can be captured using circuit theory.
268 Electromagnetic Field Theory

Figure 24.10 shows the schematic and the cross section of a computer chip at different
levels: with the transistor level at the bottom-most. The signals are taken out of a transistor
by XY lines at the middle level that are linked to the ball-grid array at the top-most level
of the chip. And then, the signal leaves the chip via a package. Since these nanometer-size
structures are much smaller than the wavelength, they are usually modeled by lumped R,
L, and C elements when retardation effect can be ignored. If retardation effect is needed, it
is usually modeled by a transmission line. This is important at the package level where the
dimensions of the components are larger.
A process of parameter extraction where computer software or field solvers (software that
solve Maxwell’s equations numerically) are used to extract these lumped-element parameters.
Finally, a computer chip is modeled as a network involving a large number of transistors,
diodes, and R, L, and C elements. Subsequently, a very useful commercial software called
SPICE (Simulation Program with Integrated-Circuit Emphasis) [134], which is a computer-
aided software, solves for the voltages and currents in this network.

Figure 24.10: Cross section of a chip (top left) and the XY lines in the chip (top right), and
the interconnects in the package needed to take the signal out of the chip (bottom right)
(courtesy of Wikipedia and Intel).

Initially, SPICE software was written primarily to solve circuit problems. But the SPICE
software now has many capabilities, including modeling of transmission lines for microwave
engineering, which are important for modeling retardation effects. Figure 24.11 shows an
interface of an RF-SPICE that allows the modeling of transmission line with a Smith chart
interface.
Circuit Theory Revisited 269

Figure 24.11: SPICE is also used to solve RF problems. A transmission line is used in com-
bination with circuit theory to account for retardation effects in a computer circuit (courtesy
of EMAG Technologies Inc.).

24.10 Decoupling Capacitors and Spiral Inductors

Decoupling capacitor is an important part of modern computer chip design. They can regulate
voltage supply on the power delivery network of the chip as they can remove high-frequency
noise and voltage fluctuation from a circuit as shown in Figure 24.12. Figure 24.13 shows a
3D IC computer chip where decoupling capacitors are integrated into its design.

Figure 24.12: A decoupling capacitor is essentially a low-pass filter allowing low-frequency

signal to pass through, while high-frequency signal is short-circuited (courtesy learningabout-
electronics.com).
270 Electromagnetic Field Theory

Figure 24.13: Modern computer chip design is 3D and is like a jungle. There are different
levels in the chip and they are connected by through silicon vias (TSV). IMD stands for inter-
metal dielectrics. One can see different XY lines serving as power and ground lines (courtesy
of Semantic Scholars).

Inductors are also indispensable in IC design, as they can be used as a high frequency
choke. However, designing compact inductor is still a challenge. Spiral inductors are used
because of their planar structure and ease of fabrication. However, miniaturizing inductor is
a difficult frontier research topic [135].

Figure 24.14: Spiral inductors are difficult to build on a chip, but by using laminal structure,
it can be integrated into the IC fabrication process (courtesy of Quan Yuan, Research Gate).
Lecture 25

Radiation by a Hertzian Dipole

Radiation of electromagnetic field is of ultimate importance for wireless communication sys-

tems. The first demonstration of the wave nature of electromagnetic field was by Heinrich
Hertz in 1888 [18], some 23 years after Maxwell’s equations were fully established. Guglielmo
Marconi, after much perserverence with a series of experiments, successfully transmitted
wireless radio signal from Cornwall, England to Newfoundland, Canada in 1901 [136]. The
experiment was serendipitous since he did not know that the ionosphere was on his side:
The ionosphere helped to bounce the radio wave back to earth from outer space. Marconi’s
success ushered in the age of wireless communication, which is omni-present in our daily
lives. Hence, radiation by arbitrary sources is an important topic for antennas and wireless
communications. We will start with studying the Hertzian dipole which is the simplest of
radiation sources we can think of.

25.1 History

The original historic Hertzian dipole experiment is shown in Figure 25.1. It was done in 1887
by Heinrich Hertz [18]. The schematic for the original experiment is also shown in Figure
25.2.
A metallic sphere has a capacitance in closed form with respect to infinity or a ground
plane.1 Hertz could use those knowledge to estimate the capacitance of the sphere, and also,
he could estimate the inductance of the leads that are attached to the dipole, and hence, the
resonance frequency of his antenna. The large sphere is needed to have a large capacitance,
so that current can be driven through the wires. As we shall see, the radiation strength of
the dipole is proportional to p = ql the dipole moment.

1 We shall learn later that this problem can be solved in closed form using image theorem.

271
272 Electromagnetic Field Theory

Figure 25.1: Hertz’s original experiment on a small dipole (courtesy of Wikipedia [18]).

Figure 25.2: More on Hertz’s original experiment on a small dipole (courtesy of Wikipedia
[18]). The antenna was powered by a transformer. The radiated electromagnetic field was
picked up by a loop receiver antenna that generates a spark at its gap M .
Radiation by a Hertzian Dipole 273

25.2 Approximation by a Point Source

Figure 25.3: Schematic of a small Hertzian dipole which is a close approximation of that first
proposed by Hertz.

Figure 25.3 is the schematic of a small Hertzian dipole resembling the original dipole that
Hertz made. Assuming that the spheres at the ends store charges of value q, and l is the
effective length of the dipole, then the dipole moment p = ql. The charge q is varying in time
harmonically because it is driven by the generator. Since

dq
= I,
dt

we have the current moment

dq
Il = l = jωql = jωp (25.2.1)
dt

for this Hertzian dipole.

A Hertzian dipole is a dipole which is much smaller than the wavelength under consid-
eration so that we can approximate it by a point current distribution, or a current density.
Mathematically, it is given by [32, 44]

J(r) = ẑIlδ(x)δ(y)δ(z) = ẑIlδ(r) (25.2.2)

The dipole is as shown in Figure 25.3 schematically. As long as we are not too close to the
dipole so that it does not look like a point source anymore, the above is a good mathematical
model and approximation for describing a Hertzian dipole.
274 Electromagnetic Field Theory

Figure 25.4: The field of a point dipole field versus that of a dipole field. When one is far
away from the dipole sources, their fields are similar to each other (courtesy of Wikepedia).

We have learnt previously that the vector potential is related to the current as follows:
 0
e−jβ|r−r |
A(r) = µ dr0 J(r0 ) (25.2.3)
4π|r − r0 |
Since the current is a 3D delta function in space, using the sifting property of a delta function,
the corresponding vector potential is given by
µIl −jβr
A(r) = ẑ e (25.2.4)
4πr
Since the vector potential A(r) is cylindrically symmetric, the corresponding magnetic field
is obtained, using cylindrical coordinates, as
 
1 1 1 ∂ ∂
H= ∇×A= ρ̂ Az − φ̂ Az (25.2.5)
µ µ ρ ∂φ ∂ρ
∂
p
where ∂φ = 0, r = ρ2 + z 2 . In the above, we have used the chain rule that

∂ ∂r ∂ ρ ∂ ρ ∂
= =p = .
∂ρ ∂ρ ∂r ρ2 + z 2 ∂r r ∂r

As a result,
 
ρ Il 1 1
H = −φ̂ − − jβ e−jβr (25.2.6)
r 4π r2 r
ρ
In spherical coordinates, r = sin θ, and (25.2.6) becomes [32]
Il
H = φ̂ (1 + jβr)e−jβr sin θ (25.2.7)
4πr2
Radiation by a Hertzian Dipole 275

The electric field can be derived using Maxwell’s equations.

 
1 1 1 ∂ 1 ∂
E= ∇×H= r̂ sin θHφ − θ̂ rHφ
jω jω r sin θ ∂θ r ∂r
Ile−jβr h 2 2
i
= r̂2 cos θ(1 + jβr) + θ̂ sin θ(1 + jβr − β r ) (25.2.8)
jω4πr3

Figure 25.5: Spherical coordinates are used to calculate the fields of a Hertzian dipole.

25.2.1 Case I. Near Field, βr  1

Since βr  1, retardation effect within this short distance from the point dipole can be
ignored. Also, we let βr → 0, and keeping the largest terms (or leading order terms in math
parlance), then from (25.2.8), with Il = jωp
p
E∼
= (r̂2 cos θ + θ̂ sin θ), βr  1 (25.2.9)
4πr3
For the H field, from (25.2.7), with βr  1, then
jωp
H = φ̂ sin θ (25.2.10)
4πr2
or
jβrp
η0 H = φ̂ sin θ (25.2.11)
4πεr3
Thus, it is seen that

η0 H  E, when βr  1 (25.2.12)
276 Electromagnetic Field Theory

where p = ql is the dipole moment.2 The above implies that in the near field, the electric
field dominates over the magnetic field.
In the above, βr could be made very small by making λr small or by making ω → 0. The
above is like the static field of a dipole.
Another viewpoint is that in the near field, the field varies rapidly, and space derivatives
are much larger than the time derivative.3
For instance,
∂ ∂

∂x c∂t
Alternatively, we can say that the above is equivalent to
∂ ω

∂x c
or that
1 ∂2
∇2 − ≈ ∇2
c2 ∂t2
In other words, static theory prevails over dynamic theory when βr  1. The above approx-
imations are consistent with that the retardation effect is negligible over this lengthscale.

25.2.2 Case II. Far Field (Radiation Field), βr  1

This is also known as the far zone. In this case, retardation effect is important. In other
words, phase delay cannot be ignored.
Il −jβr
E∼
= θ̂jωµ e sin θ (25.2.13)
4πr
and
Il −jβr
H∼
= φ̂jβ e sin θ (25.2.14)
4πr
ωµ
Eθ
pµ
Note that H φ
= β =  = η0 . Here, E and H are orthogonal to each other and they are
both orthogonal to the direction of propagation, as in the case of a plane wave. Or in a word,
a spherical wave resembles a plane wave in the far field approximation.

25.3 Radiation, Power, and Directive Gain Patterns

The time average power flow in the far field is given by
 2
1 1 2 η0 βIl
hSi = <e[E × H∗ ] = r̂ η0 |Hφ | = r̂ sin2 θ (25.3.1)
2 2 2 4πr
p
2 Here, η0 = µ/ε. We multiply H by η0 so that the quantities we are comparing have the same unit.
3 This is in agreement with our observation that electromagnetic fields are great contortionists: They will
deform themselves to match the boundary first before satisfying Maxwell’s equations. Since the source point
is very small, the fields will deform themselves so as to satisfy the boundary conditions near to the source
region. If this region is small compared to wavelength, the fields will vary rapidly over a small lengthscale
compared to wavelength.
Radiation by a Hertzian Dipole 277

The radiation field pattern of a Hertzian dipole is the plot of |E| as a function of θ at a
constant r. Hence, it is proportional to sin θ, and it can be proved that it is a circle. The
radiation power pattern is the plot of hSr i at a constant r.

Figure 25.6: Radiation field pattern of a Hertzian dipole. It can be shown that the pattern
is a circle.

Figure 25.7: Radiation power pattern of a Hertzian dipole which is also the same as the
directive gain pattern.
278 Electromagnetic Field Theory

The total power radiated by a Hertzian dipole is thus given by

 2π  π  π  2
η0 βIl
P = dφ dθr2 sin θhSr i = 2π dθ sin3 θ (25.3.2)
0 0 0 2 4π

Since

 π  −1  1
3 2 4
dθ sin θ = − (d cos θ)[1 − cos θ] = dx(1 − x2 ) = (25.3.3)
0 1 −1 3

then

2
η0 (βIl)2

4 βIl
P = πη0 = (25.3.4)
3 4π 12π

The directive gain of an antenna, G(θ, φ), is defined as [32]

hSr i hSr i
G(θ, φ) = = P (25.3.5)
hSav i 4πr 2

where

P
hSav i = (25.3.6)
4πr2

is the power density if the power P were uniformly distributed over a sphere of radius r.
Notice that hSav i is independent of angle. Hence, the angular dependence of the directive
gain G(θ, φ) is coming from hSr i.
Substituting (25.3.1) and (25.3.4) into the above, we have

 2
η0 βIl
2 4πr sin2 θ 3 2
G(θ, φ) =  2 = sin θ (25.3.7)
1 4 βIl 2
4πr 2 3 η0 π 4π

The peak of G(θ, φ) is known as the directivity of an antenna. It is 1.5 in the case of
a Hertzian dipole. If an antenna is radiating isotropically, its directivity is 1, which is the
lowest possible value, whereas it can be over 100 for some antennas like reflector antennas
(see Figure 25.8). A directive gain pattern is a plot of the above function G(θ, φ) and it
resembles the radiation power pattern.
Radiation by a Hertzian Dipole 279

Figure 25.8: The gain of a reflector antenna can be increased by deflecting the power radiated
in the desired direction by the use of a reflector (courtesy of racom.eu).

If the total power fed into the antenna instead of the total radiated power is used in the
denominator of (25.3.5), the ratio is known as the power gain or just gain and the pattern
is the power gain pattern. The total power fed into the antenna is not equal to the total
radiated power because there could be some loss in the antenna system like metallic loss.

25.3.1 Radiation Resistance

Engineers love to replace complex systems with simpler systems. Simplicity rules again! This
will make interface with electronic driving circuits for the antenna easier to derive. A raw
Hertzian dipole, when driven by a voltage source, essentially looks like a capacitor due to the
preponderance of electric field energy stored in the dipole field. But at the same time, the
dipole radiates giving rise to radiation loss. Thus a simple circuit equivalence of a Hertzian
dipole is a capacitor in series with a resistor. The resistor accounts for radiation loss of the
dipole.
280 Electromagnetic Field Theory

Figure 25.9: (a) Equivalent circuit of a raw Hertzian dipole without matching. (b) Equivalent
circuit of a matched Hertzian dipole (using maximum power transfer theorem). (c) Equivalent
circuit of a matched dipole at the resonance frequency of the LC tank circuit.

Hence, the way to drive the Hertzian dipole effectively is to use matching network for
maximum power transfer. Or an inductor has to be added in series with the intrinsic ca-
pacitance of the Hertzian dipole to cancel it at the resonance frequency of the tank circuit.
Eventually, after matching, the Hertzian dipole can be modeled as just a resistor. Then the
power absorbed by the Hertzian dipole from the driving source is P = 12 I 2 Rr . Thus, the
radiation resistance Rr is the effective resistance that will dissipate the same power as the
radiation power P when a current I flows through the resistor. Hence, it is defined by [32]

2P (βl)2
Rr = = η0 ≈ 20(βl)2 , where η0 = 377 ≈ 120π Ω (25.3.8)
I2 6π

For example, for a Hertzian dipole with l = 0.1λ, Rr ≈ 8Ω.

The above assumes that the current is uniformly distributed over the length of the Hertzian
dipole. This is true if there are two charge reservoirs at its two ends. For a small dipole with
no charge reservoir at the two ends, the currents have to vanish at the tips of the dipole as
shown in Figure 25.10. The effective length of an equivalent Hertzian dipole for the dipole
with triangular distribution is half of its actual length due to the manner the currents are
distributed.4 Such a formula can be used to estimate the radiation resistance of a dipole.
For example, a half-wave dipole does not have a triangular current distribution a sinusoidal
one as shown in Figure 25.11. Nevertheless, we approximate the current distribution of a half-
wave dipole with a triangular distribution, and apply the above formula. We pick a = λ2 , and
let leff = λ4 in (25.3.8), we have

Rr ≈ 50Ω (25.3.9)

4 As shall be shown, when the dipole is short, the details of the current distribution is inessential in

determining the radiation field. It is the area under the current distribution that is important.
Radiation by a Hertzian Dipole 281

Figure 25.10: The current pattern on a short dipole can be approximated by a triangle since
the current has to vanish at the end points of the short dipole. Furthermore, this dipole can
be approximated by an effective Hertzian dipole half its length with uniform current.

The true current distribution on a half-wave dipole resembles that shown in Figure 25.11.
The current is zero at the end points, but the current has a more sinusoidal-like distribution
as in a transmission line. Hence, a half-wave dipole is not much smaller than a wavelength
and does not qualify to be a Hertzian dipole. Furthermore, the current distribution on the
half-wave dipole is not triangular in shape as above. A more precise calculation shows that
Rr = 73Ω for a half-wave dipole [54]. This also implies that a half-wave dipole with sinusoidal
current distribution is a better radiator than a dipole with triangular current distribution.

In fact, one can think of a half-wave dipole as a flared, open transmission line. In the
beginning, this flared open transmission line came in the form of biconical antennas which are
shown inp Figure 25.12 [137]. If we recall that the characteristic impedance of a transmission
line is L/C, then as the spacing of the two metal pieces becomes bigger, the equivalent
characteristic impedance gets bigger. Therefore, the impedance can gradually transform
from a small impedance like 50 Ω to that of free space, which is 377 Ω. This impedance
matching helps mitigate reflection from the ends of the flared transmission line, and enhances
radiation. Because of the matching nature of bicone antennas, they are better radiators with
higher radiaton loss and lower Q. Thus they have a broader bandwidth, and are important
in UWB (ultra-wide band) antennas [138].
282 Electromagnetic Field Theory

Figure 25.11: Approximate current distribution on a half-wave dipole (courtesy of electronics-

notes.co). The currents are zero at the two end tips due to the current continuity equation,
or KCL.

Figure 25.12: A bicone antenna can be thought of as a transmission line with gradually
changing characteristic impedance. This enhances impedance matching and the radiation of
the antenna (courtesy of antennasproduct.com).
Lecture 26

Radiation Fields

Figure 26.1: (a) Electric field around a time-oscillating dipole (courtesy of physics stack
exchange). (b) Equi-potential lines aroud a moving charge that gives rise Cherenkov radiation
(courtesy of J.D. Jackson [48]). We will not study Cherenkov (Cerenkov) radiation in this
course, but it is written up in [48] and [32]. Its was a Nobel Prize winning discovery.

The reason why charges radiate is because they move or accelerate. In the case of a dipole
antenna, the charges move back and forth between poles of the antenna. Near to the dipole
source, quasi-static physics prevails, and the field resembles that of a static dipole. If the
dipole is flipping sign constantly due to the change in the direction of the current flow, the
field would also have to flip sign constantly. But electromagnetic wave travels with a finite
velocity. The field from the source ultimately cannot keep up with the sign change of the
source field: it has to be ‘torn’ away from the source field and radiate. Another interesting
radiation is the Cherenkov radiation. It is due to a charge moving faster than the velocity of
light. As an electron cannot move faster than the speed of light in a vacuum, this can only

283
284 Electromagnetic Field Theory

happen in the material media or plasma, where the velocity of the electron can be faster than
the group velocity of wave in the medium. Ultimately, the electric field from the particle is
‘torn’ off from the charge and radiate. These two kinds of radiation are shown in the Figure
26.1.
We have shown how to connect the vector and scalar potentials to the sources J and %
of an electromagnetic system. This is a very important connection: it implies that once we
know the sources, we know how to find the fields. But the relation between the fields and
the sources are in general rather complex. In this lecture, we will simplify this relation by
making a radiation field or far-field approximation by assuming that the point where the field
is observed is very far from the source location in terms of wavelegth. This approximation
is very useful for understanding the physics of the radiation field from a source such as an
antenna. It is also important for understanding the far field of an optical system. As shall
be shown, this radiation field carries the energy generated by the sources to infinity.

26.1 Radiation Fields or Far-Field Approximation

Figure 26.2: The relation of the observation point located a r to the source location at r0 .
The distance of the observation point r to the source location r0 is |r − r0 |.

In the previous lecture, we have derived the relation of the vector and scalar potentials to the
sources J and % as shown in (23.2.29) and (23.2.30)1 They are given by
 0
0 0 e−jβ|r−r |
A(r) = µ dr J(r ) (26.1.1)
V 4π|r − r0 |
 0
1 e−jβ|r−r |
Φ(r) = dr0 %(r0 ) (26.1.2)
ε V 4π|r − r0 |
1 This topic is found in many standard textbooks in electromagnetics [32, 47, 54]. They are also found in

lecture notes [44, 139].

Radiation Fields 285

√
where β = ω µε = ω/c is the wavenumber. The integrals in (26.1.1) and (26.1.2) are
normally untenable, but when the observation point is far from the source, approximation to
the integrals can be made giving them a nice physical interpretation.

Figure 26.3: The relation between |r| and |r − r0 | using the parallax method, or that |r − r0 | ≈
|r| − r0 · r̂. It is assumed that r is almost parallel to r − r0 .

26.1.1 Far-Field Approximation

When |r|  |r0 |, then |r − r0 | ≈ r − r0 · r̂, where r = |r| . This approximation can be
shown algebraically or by geometrical argument as shown in Figure 26.3. Thus (26.1.1) above
becomes
 
µ 0 J(r0 ) −jβr+jβr0 ·r̂ µe−jβr 0
A(r) ≈ dr e ≈ dr0 J(r0 )ejβr ·r̂ (26.1.3)
4π V r − r0 · r̂ 4πr V

In the above, r0 · r̂ is small compared to r. Hence, we have made use of that 1/(1 − ∆) ≈ 1
when ∆ is small, so that 1/(r − r0 · r̂) can be approximate by 1/r. Also, we assume that the
0
frequency is sufficiently high such that βr0 · r̂ is not necessarily small. Thus, ejβr ·r̂ 6= 1, unless
0
βr · r̂  1. Hence, we keep the exponential term in (26.1.3) but simplify the denominator to
arrive at the last expression above.
If we let β = βr̂, which is the β vector (or k vector in optics), and r0 = x̂x0 + ŷy 0 + ẑz 0 ,
then
0 0 0 0 0
ejβr ·r̂ = ejββ ·r = ejβx x +jβy y +jβz z (26.1.4)

Therefore (26.1.3) resembles a 3D Fourier transform integral,2 namely, the above integral
2 Except that the vector β is of fixed length.
286 Electromagnetic Field Theory

becomes

µe−jβr 0
A(r) ≈ dr0 J(r0 )ejβ·r (26.1.5)
4πr V

and (26.1.5) can be rewritten as

µe−jβr
A(r) ∼
= β)
F(β (26.1.6)
4πr
where

0
β) =
F(β dr0 J(r0 )ejβ·r (26.1.7)
V

is the 3D Fourier transform of J(r0 ) with the Fourier transform variable β = r̂β.
It is to be noted that this is not a normal 3D Fourier transform because |β|2 = βx 2 +βy 2 +
2
βz = β 2 which is a constant for a fixed frequency. In other words, the length of the vector β
is fixed to be β, whereas in a normal 3D Fourier transform, βx , βy , and βz are independent
variables, each with values in the range [−∞, ∞]. Or the value of βx 2 + βy 2 + βz 2 ranges from
zero to infinity.
The above is the 3D “Fourier transform” of the current source J(r0 ) with Fourier variables,
βx , βy , βz restricted to lying on a sphere of radius β and β = βr̂. This spherical surface in
the Fourier space is also called the Ewald sphere.

26.1.2 Locally Plane Wave Approximation

We can write r̂ or β in terms of direction cosines in spherical coordinates or that

r̂ = x̂ cos φ sin θ + ŷ sin φ sin θ + ẑ cos θ (26.1.8)

Hence,

β ) = F(βr̂) = F(β, θ, φ)
F(β (26.1.9)

It is not truly a 3D function, since β, the length of the vector β, is fixed. It is a 3D Fourier
transform with data restricted on a spherical surface.
Also in (26.1.6), when r  r0 · r̂, and when the frequency is high or β is large, e−jβr is
now a rapidly varying function of r while, F(β β ) is only a slowly varying function of r̂ or of θ
and φ, the observation angles. In other words, the prefactor in (26.1.6), exp(−jβr)/r, can be
thought of as resembling a spherical wave. Hence, if one follows a ray of this spherical wave
and moves in the r direction, the predominant variation of the field is due to e−jβr , whereas
the direction of the vector β changes little, and hence, F(β) changes little. Furthermore, r0
in (26.1.7) are restricted to small or finite number, making F(β) a weak function of β (see
Figure 26.4).
The above shows that in the far field, the wave radiated by a finite source resembles a
spherical wave. Moreover, a spherical wave resembles a plane wave when one is sufficiently
far from the source such that βr  1, or 2πr/λ  1. Or r is many wavelengths away from
Radiation Fields 287

the source. Hence, we can write e−jβr = e−jββ ·r where β = r̂β and r = r̂r so that a spherical
wave resembles a plane wave locally. This phenomenon is shown in Figure 26.4 and Figure
26.5

Figure 26.4: A source radiates a field that resembles a spherical wave. In the vicinity of
the observation point r, when β is large, the field is strongly dependent on r via exp(−jβr)
but weakly dependent on β (beta hardly changes direction in the vicinity of the observation
point). Hence, the field becomes locally a plane wave in the far field.
288 Electromagnetic Field Theory

Figure 26.5: (a) A leaky hole in a waveguide leaks a spherical (courtesy of MEEP, MIT). (b)
A point source radiates a spherical wave (courtesy of ME513, Purdue Engineering). Most of
these simulations are done with FDTD (finite-difference time-domain) method that we will
learn later in the course. When the wavelength is short, or the frequency high, a spherical
wave front looks locally plane. This is similar to the notion that as humans, who are short,
think that the earth is flat around us. Up to this day, some people still believe that the earth
is flat:)

Then, it is clear that with the local plane-wave approximation, ∇ → −jβ

β = −jβr̂, we
have
1 β β
H= ∇ × A ≈ −j r̂ × (θ̂Aθ + φ̂Aφ ) = j (θ̂Aφ − φ̂Aθ ) (26.1.10)
µ µ µ
Similarly [44, 139],
1 β
E= ∇×H∼
= −j r̂ × H ∼
= −jω(θ̂Aθ + φ̂Aφ ) (26.1.11)
jωε ω
Notice that β = βr̂, the direction of propagation of the local plane wave, is orthogonal to E
and H in the far field, a property of a plane wave since the wave is locally a plane wave.
Moreover, there are more than one way to derive the electric field E. For instance, using
(26.1.10) for the magnetic field, the electric field can also be written as
1
E= ∇×∇×A (26.1.12)
jωµε
Using the formula for the double-curl operator, the above can be rewritten as
1 1
∇∇ · A − ∇2 A ∼ −ββ + β 2 I · A



E= = (26.1.13)
jωµε jωµε
Radiation Fields 289

where we have used that ∇ ∼ = −jβ and ∇2 A = −β 2 A.3 Alternatively, we can factor β 2 =
2
ω µ out of the parenthesis, and rewrite the above as
 
E∼


= −jω −β̂ β̂ + I · A = −jω −r̂r̂ + I · A (26.1.14)

Since I = r̂r̂ + θ̂θ̂ + φ̂φ̂, then the above becomes

 
E∼= −jω θ̂θ̂ + φ̂φ̂ · A = −jω(θ̂Aθ + φ̂Aφ ) (26.1.15)

which is the same as previously derived. It also shows that the electric field is transverse to
the β vector.4
Furthermore, it can be shown that in the far field, using the local plane-wave approxima-
tion,

|E|/|H| ≈ η (26.1.16)

where η is the intrinsic impedance of free space, which is a property of a plane wave. Moreover,
one can show that the time average Poynting’s vector, or the power density flow, in the far
field is
1 1
hSi = <e (E × H∗ ) ≈ |E|2 r̂ = hSr ir̂ (26.1.17)
2 2η

which resembles also the property of a plane wave.5 Since the radiated field is a spherical
wave, the Poynting’s vector is radial. Therefore,
1
hSi = r̂hSr (θ, φ)i, where hSr (θ, φ)i = |E|2 (26.1.18)
2η

and hSr i is the time-average radial power density. The plot of |E(θ, φ)| is termed the far-field
pattern or the radiation pattern of an antenna or the source, while the plot of |E(θ, φ)|2 is its
far-field power pattern.

26.1.3 Directive Gain Pattern Revisited

We have defined the directive gain pattern for a Hertzian dipole before in Section 25.3. But
this concept can be applied to a general radiating source or antenna. Once the far-field
radiation power pattern or the radial power density hSr i is known, the total power radiated
by the antenna in the far field can be found by integrating over all angles, viz.,
 π  2π
PT = r2 sin θdθdφhSr (θ, φ)i (26.1.19)
0 0
3 Note that ∇ · A 6= 0 here.
4 We can also arrive at the above by letting E = −jωA − ∇Φ, and using the appropriate formula for the
scalar potential. There is more than one road that lead to Rome!
5 To avoid confusion, we will use S to denote instantaneous Poynting’s vector and S to denote complex

Poynting’s vector (see 10.3.1). e

290 Electromagnetic Field Theory

The above evaluates to a constant independent of r due to energy conservation. Now assume
that this same antenna is radiating isotropically in all directions, then the average power
density of this fictitious isotropic radiator as r → ∞ is

PT
hSav i = (26.1.20)
4πr2
A dimensionless directive gain pattern can be defined as before in Section 25.3 such that
[32, 139]

hSr (θ, φ)i 4πr2 hSr (θ, φ)i

G(θ, φ) = = (26.1.21)
hSav i PT

This directive gain pattern is a measure of the radiation power pattern of the antenna or
source compared to when it radiates isotropically. The above function is independent of r in
the far field since Sr ∼ 1/r2 in the far field. As in the Hertzian dipole case, the directivity of
an antenna D = max(G(θ, φ)), is the maximum value of the directive gain. It is to be noted
that by its mere definition,

dΩG(θ, φ) = 4π (26.1.22)

2π
π
where dΩ = 0 0 sin θdθdφ. It is seen that since the directive gain pattern is normalized,
when the radiation power is directed to the main lobe of the antenna, the corresponding side
lobes and back lobes will be diminished.
An antenna also has an effective area or aperture Ae , such that if a plane wave carrying
power density denoted by hSinc i impinges on the antenna, then the power received by the
antenna, Preceived is given by

Preceived = hSinc iAe (26.1.23)

Here, the transmit antenna and the receive antenna are in the far field of each other. Hence,
we can approximate the field from the transmit antenna to be a plane wave when it reaches
the receive antenna. If the receive antenna is made of PEC, induced current will form on the
receive antenna so as to generated a field that will cancel the incident field on the PEC surface.
This induced current generates a voltage at the receiver load, and hence power received by
the antenna.
A wonderful relationship exists between the directive gain pattern G(θ, φ) and the effective
aperture, namely that6

λ2
Ae = G(θ, φ) (26.1.24)
4π
Therefore, the effective aperture of an antenna is also direction dependent. The above im-
plies that the radiation property of an antenna is related to its receiving property. This is a
6 The proof of this formula is beyond the scope of this lecture, but we will elaborate on it when we discuss

reciprocity theorem.
Radiation Fields 291

beautiful consequence of reciprocity theorem that we will study later! The constant of pro-
portionality, λ2 /(4π) is a universal constant that is valid for all antennas satisfying reciprocity
theorem. The derivation of this constant for a Hertzian dipole is given in Kong [32], or using
blackbody radiation law [139, 140].

The directivity and the effective aperture can be enhanced by designing antennas with
different gain patterns. When the radiative power of the antenna can be directed to be in a
certain direction, then the directive gain and the effective aperture (for that given direction)
of the antenna is improved. This is shown in Figure 26.6. Such focussing of the radiation fields
of the antenna can be achieved using reflector antennas or array antennas. Array antennas,
as shall be shown, work by constructive and destructive wave field of the antenna.

Being able to do point-to-point communications at high data rate is an important modern

application of antenna array. Figure 26.7 shows the gain pattern of a sophisticated antenna
array design for 5G applications.

Figure 26.6: The directive gain pattern of an array antenna. The directivity is increased by
constructive interference (courtesy of Wikepedia).
292 Electromagnetic Field Theory

Figure 26.7: The directive gain pattern of a sophisticated array antenna for 5G applications
(courtesy of Ozeninc.com).
Lecture 27

Array Antennas, Fresnel Zone,

Rayleigh Distance

Our world is beset with the dizzying impact of wireless communication. It has greatly im-
pact our lives.1 Wireless communication is impossible without using antennas. Hence it is
important to design these communication systems with utmost efficiency and sensitivity. We
have seen that a simple Hertzian dipole has low directivity in Section 25.3. The radiation
pattern looks like that of a donut, and the directivity of the antenna is 1.5. Hence, for point-
to-point communications, much power is wasted. However, the directivity of antennas can
be improved if a group or array of dipoles can work cooperatively together. They can be
made to constructively interfere in the desired direction, and destructively interfere in other
directions to enhance their directivity. Since the far-field approximation of the radiation field
can be made, and the relationship between the far field and the source is a Fourier transform
relationship, clever engineering can be done borrowing knowledge from the signal processing
area. After understanding the far-field physics, one can also understand many optical phe-
nomena, such as how a laser pointer works. Many textbooks have been written about array
antennas some of which are [141, 142].

27.1 Linear Array of Dipole Antennas

Antenna array can be designed so that the constructive and destructive interference in the far
field can be used to steer the direction of radiation of the antenna, or the far-field radiation
pattern of an antenna array. This is because the far field of a source is related to the source
by a Fourier transform relationship. The relative phases of the array elements can be changed
in time so that the beam of an array antenna can be steered in real time. This has important
applications in, for example, air-traffic control. It is to be noted that if the current sources are
impressed current sources, and they are the input to Maxwell’s equations, and the fields are
1 I cannot imagine a day that I do not receive messages from my relatives and friends in Malaysia and

Singapore. It has made the world look smaller, and more transparent.

293
294 Electromagnetic Field Theory

the output of the system, then we are dealing with a linear system here whereby linear system
theory can be used. For instance, we can use Fourier transform to analyze the problem in the
frequency domain. The time domain response can be obtained by inverse Fourier transform.
This is provided that the current sources are impressed and they are not affected by the fields
that they radiate.

Figure 27.1: Schematics of a dipole array. To simplify the math, the far-field approximation
can be used to find its far field or radiation field.

A simple linear dipole array is shown in Figure 27.1. First, without loss of generality and
for simplicity to elucidate the physics, we assume that this is a linear array of point Hertzian
dipoles aligned on the x axis. The current can then be described mathematically as follows:
J(r0 ) = ẑIl[A0 δ(x0 ) + A1 δ(x0 − d1 ) + A2 δ(x0 − d2 ) + · · ·
+ AN −1 δ(x0 − dN −1 )]δ(y 0 )δ(z 0 ) (27.1.1)
The far field can be found using the approximate formula derived in the previous lecture, viz.,
(26.1.3) reproduced below:

µe−jβr 0
A(r) ≈ dr0 J(r0 )ejβ·r (27.1.2)
4πr V

To reiterate, the above implies that the far field is related to the Fourier transform of the
current source J(r0 ).

27.1.1 Far-Field Approximation

The vector potential on the xy-plane in the far field, using the sifting property of delta
function, yield the following equation for A(r) using (27.1.2),

µIl −jβr 0
A(r) ∼
= ẑ e dr0 [A0 δ(x0 ) + A1 δ(x0 − d1 ) + · · · ]δ(y 0 )δ(z 0 )ejβr ·r̂ (27.1.3)
4πr
In the above, for simplicity, we will assume that the observation point is on the xy plane, or
that r = ρ = x̂x + ŷy. Thus, r̂ = x̂ cos φ + ŷ sin φ. Also, since the sources are aligned on the
Array Antennas, Fresnel Zone, Rayleigh Distance 295

0 0
x axis, then r0 = x̂x0 , and r0 · r̂ = x0 cos φ. Consequently, ejβr ·r̂ = ejβx cos φ
. By so doing, we
have
µIl −jβr
A(r) ∼
= ẑ e [A0 + A1 ejβd1 cos φ + A2 ejβd2 cos φ + · · · + AN −1 ejβdN −1 cos φ ] (27.1.4)
4πr
Next, for further simplification, we let dn = nd, implying an equally space array with dis-
tance d between adjacent elements. Then we let An = ejnψ , which assumes a progressively
increasing phase shift between different elements. Such an antenna array is called a linear
phase array. Thus, (27.1.3) in the above becomes
µIl −jβr
A(r) ∼
= ẑ e [1 + ej(βd cos φ+ψ) + ej2(βd cos φ+ψ) + · · ·
4πr
+ej(N −1)(βd cos φ+ψ) ] (27.1.5)

With the simplifying assumptions, the above series can be summed in closed form.

27.1.2 Radiation Pattern of an Array

The above(27.1.5) can be summed in closed form using the formula
N −1
X 1 − xN
xn = (27.1.6)
n=0
1−x

Then in the far field,

µIl −jβr 1 − ejN (βd cos φ+ψ)

A(r) ∼
= ẑ e (27.1.7)
4πr 1 − ej(βd cos φ+ψ)

Ordinarily, as shown previously in (26.1.11), E ≈ −jω(θ̂Aθ + φ̂Aφ ). But since A is ẑ directed,

Aφ = 0. Furthermore, on the xy plane, Eθ ≈ −jωAθ = jωAz . As a consequence,

 1 − ejN (βd cos φ+ψ) 

 
|Eθ | = |E0 | 
 , r → ∞
1 − ej(βd cos φ+ψ) 
 
 sin N (βd cos φ + ψ)

2
= |E0 | 
 , r → ∞ (27.1.8)

 sin 21 (βd cos φ + ψ) 

The factor multiplying |E0 | above is also called the array factor. The above can be used to
plot the far-field pattern of an antenna array.
Equation (27.1.8) has an array factor that is of the form

|sin (N x)|
|sin x|
This function appears in digital signal processing frequently, and is known as the digital sinc
function [143]. The reason why this is so is because the far field is proportional to the Fourier
transform of the current. The current in this case a finite array of Hertzian dipole, which is
296 Electromagnetic Field Theory

a product of a box function and infinite array of Hertzian dipole. The Fourier transform of
such a current, as is well known in digital signal processing, is the digital sinc.
Plots of |sin(3x)| and |sin x| are shown as an example and the resulting |sin(3x)|
|sin x| is also
shown in Figure 27.2. The function peaks when both the numerator and the denominator
of the digital sinc vanish. The value can be found by Taylor series expansion. This happens
when x = nπ for integer n. Such an apparent singularity is called a removable singularity.

|sin 3x|
Figure 27.2: Plot of the digital sinc, |sin x| .

In equation (27.1.8), x = 12 (βd cos φ+ψ). We notice that the maximum in (27.1.8) would
occur if x = nπ, or if

βd cos φ + ψ = 2nπ, n = 0, ±1, ±2, ±3, · · · (27.1.9)

The zeros or nulls will occur at N x = nπ, or

2nπ
βd cos φ + ψ = , n = ±1, ±2, ±3, · · · , n 6= mN (27.1.10)
N
For example,

Case I. ψ =
0, βd = π, principal maximum is at φ = ± π2 . If N = 5, nulls are at

φ = ± cos−1 2n ◦ ◦ ◦ ◦
5 , or φ = ±66.4 , ±36.9 , ±113.6 , ±143.1 . The radiation pattern is seen
to form lobes. The largest lobe is called the main lobe, while the smaller lobes are called
side lobes. Since ψ = 0, the radiated fields in the y direction are in phase and the peak of
the radiation lobe is in the y direction or the broadside direction. Hence, this is called a
broadside array. The radiation pattern of such an array is shown in Figure 27.3.

Case II. ψ =
π, βd = π, principal maximum is at φ = 0, π. If N = 4, nulls are at

φ = ± cos−1 n2 − 1 , or φ = ±120◦ , ±90◦ , ±60◦ . Since the sources are out of phase by 180◦ ,
and N = 4 is even, the radiation fields cancel each other in the broadside, but add in the x
Array Antennas, Fresnel Zone, Rayleigh Distance 297

direction or the end-fire direction. This is called the endfire array. Figure 27.4 shows the
radiation pattern of such an array.

Figure 27.3: The radiation pattern of a five-element broadside array. The broadside and
endfire directions of the array are also labeled.

Figure 27.4: By changing the phase of the linear array, the radiation pattern of the antenna
array can be changed to become an endfire array.

From the above examples, it is seen that the interference effects between the different
antenna elements of a linear array focus the power in a given direction. We can use linear
array to increase the directivity of antennas. Moreover, it is shown that the radiation patterns
can be changed by adjusting the spacings of the elements as well as the phase shift between
298 Electromagnetic Field Theory

them. The idea of antenna array design is to make the main lobe of the pattern to be much
higher than the side lobes so that the radiated power of the antenna is directed along the
main lobe or lobes rather than the side lobes. So side-lobe level suppression is an important
goal of designing a highly directive antenna. Also, by changing the phase of the antenna
elements in real time, the beam of the antenna can be steered in real time with no moving
parts.

27.2 When is Far-Field Approximation Valid?

In making the far-field approximation in (27.1.3), it will be interesting to ponder when the
far-field approximation is valid? That is, when we can approximate
0 0
e−jβ|r−r | ≈ e−jβr+jβr ·r̂ (27.2.1)

to arrive at (27.1.3). This is especially important because when we integrate over r0 , it can
range over large values especially for a large array. In this case, r0 can be as large as (N − 1)d.
The above approximation is important also because it tells when the field generated by an
array antenna become a spherical wave.
To answer this question, we need to study the approximation in (27.2.1) more carefully.
First, we have
2
|r − r0 |2 = (r − r0 ) · (r − r0 ) = r2 − 2r · r0 + r0 (27.2.2)

We can take the square root of the above to get

!1/2
2
2r · r0 r0
|r − r0 | = r 1 − + (27.2.3)
r2 r2

Next, we use the Taylor series expansion to get, for small x, that
n(n − 1) 2
(1 + x)n ≈ 1 + nx + x + ··· (27.2.4)
2!
or that
1 1
(1 + x)1/2 ≈ 1 + x − x2 + · · · (27.2.5)
2 8
We can apply this approximation by letting
2
2r · r0 r0
x∼
=− 2
+ 2
r r
To this end, we arrive at2
" 2 #
2
r · r0 1 r0 r · r0

0 1
|r − r | ≈ r 1 − 2 + − + ··· (27.2.6)
r 2 r2 2 r2
2 The art of making such approximation is called perturbation expansion [45].
Array Antennas, Fresnel Zone, Rayleigh Distance 299

In the above, we have not kept every term of the x2 terms by assuming that r02  r0 · r, and
terms much smaller than the last term in (27.2.6) can be neglected.
We can multiply out the right-hand side of the above to further arrive at
2
r · r0 1 r0 1 (r · r0 )2
|r − r0 | ≈ r − + − + ···
r 2 r 2 r3
2
1 r0 1
= r − r̂ · r0 + − (r̂ · r0 )2 + · · · (27.2.7)
2 r 2r
The last two terms in the last line of (27.2.7) are of the same order.3 Moreover, their sum is
2
bounded by r0 /(2r) since r̂ · r0 is always less than r0 . Hence, the far field approximation is
valid if
2
r0
β 1 (27.2.8)
2r
In the above, β is involved because the approximation has to be valid in the exponent, namely
exp(−jβ|r − r0 |) where β multiplies |r − r0 | or its approximation. If (27.2.8) is valid, then
r0 2
ejβ 2r ≈1

and thus, the first two terms on the right-hand side of (27.2.7) suffice to approximate |r − r0 |
on the left-hand side, which are the two terms we have kept in the far-field approximation.

27.2.1 Rayleigh Distance

Figure 27.5: The right half of a Gaussian beam [82] displays the physics of the near field, the
Fresnel zone, and the far zone. In the far zone, the field behaves like a spherical wave.
3 The math parlance for saying that these two terms are approximately of the same magnitude as each

other.
300 Electromagnetic Field Theory

If we have an infinite time-harmonic current sheet, it can be shown that by matching boundary
conditions, it will launch plane waves on both sides of the current sheet [32][p. 652]. Thus
if we have an aperture antenna like the opening of a waveguide that is much larger than the
wavelength, it will launch a wave that is almost like a plane wave from it. Thus, when a wave
field leaves an aperture antenna, it can be approximately described by a Gaussian beam [82]
(see Figure 27.5). Near to the antenna aperture, or in the near zone, it is approximately a
plane wave with wave fronts parallel to the aperture surface. Far from the antenna aperture,
or in the far zone, the field behaves like a spherical wave, with its typical wave front. In
between, we are in the Fresnel zone.
Consequently, after using that β = 2π/λ, for the far-field approximation to be valid, we
need (27.2.8) to be valid, or that
π 02
r r (27.2.9)
λ
If the aperture of the antenna is of radius W , then r0 < rmax ∼
= W and the far field approxi-
mation is valid if
π 2
r W = rR (27.2.10)
λ
If r is larger than this distance, then an antenna beam behaves like a spherical wave and starts
to diverge. This distance rR is also known as the Rayleigh distance. After this distance, the
wave from a finite size source resembles a spherical wave which is diverging in all directions
(see Figure 27.5). Also, notice that the shorter the wavelength λ, the larger is this distance.
This also explains why a laser pointer works. A laser pointer light can be thought of radiation
from a finite size source located at the aperture of the laser pointer as shall be shown using
equivalence theorem later. The laser pointer beam remains collimated for quite a distance,
before it becomes a divergent beam or a beam with a spherical wave front.
In some textbooks [32], it is common to define acceptable phase error to be π/8. The
Rayleigh distance is the distance beyond which the phase error is below this value. When the
phase error of π/8 is put on the right-hand side of (27.2.8), one gets
2
r0 π
β ≈ (27.2.11)
2r 8
Using the approximation, the Rayleigh distance is defined to be

2D2
rR = (27.2.12)
λ
where D = 2W is the diameter of the antenna aperture. This concept is important in both
optics and microwave.

27.2.2 Near Zone, Fresnel Zone, and Far Zone

Therefore, when a source radiates, the radiation field is divided into the near zone, the Fresnel
zone, and the far zone (also known as the radiation zone, or the Fraunhofer zone in optics).
Array Antennas, Fresnel Zone, Rayleigh Distance 301

The Rayleigh distance is the demarcation boundary between the Fresnel zone and the far
zone. The larger the aperture of an antenna array is, the further one has to be in order to
reach the far zone of an antenna. This distance becomes larger too when the wavelength is
short. In the far zone, the far field behaves like a spherical wave, and its radiation pattern is
proportional to the Fourier transform of the current.
In some sources, like the Hertzian dipole, in the near zone, much reactive energy is stored
in the electric field or the magnetic field near to the source. This near zone receives reactive
power from the source, which corresponds to instantaneous power that flows from the source,
but is return to the source after one time harmonic cycle. Thus, reactive power corresponds to
energy that sloshes back and forth between the source and the near field. Hence, a Hertzian
dipole has input impedance that looks like that of a capacitor, because much of the near field
of this dipole is the electric field.
The field in the far zone carries power that radiates to infinity. As a result, the field in the
near zone decays rapidly, but the field in the far zone decays as 1/r for energy conservation.
In other words, the far field convects energy to infinity.
302 Electromagnetic Field Theory
Lecture 28

Different Types of
Antennas—Heuristics

We have studied different closed form solutions and approximate solutions to Maxwell’s equa-
tions. Examples of closed form solutions are transmission lines, waveguides, resonators, and
dipoles solutions. Examples of approximate solutions are circuit theory and far field approx-
imations. These solutions offer us insights into the physical behaviour of electromagnetic
fields, and also the physical mechanisms as to how things work. These physical insights often
inspire us for new designs.

Fortunately for us, Maxwell’s equations are accurate from sub-atomic lengthscales to
galactic lengthscales. In vacuum, they have been validated to extremely high accuracy (see
Section 1.1). Furthermore, in the last few decades since the 1960s, very many numerical
solutions have been possible for Maxwell’s equations of complex structures. This field of
solving Maxwell’s equations numerically is known as computational electromagnetics. It
shall be discussed later in this course, and many commercial software are available to solve
Maxwell’s equations to high fidelity. Therefore, design engineers these days do not require
higher knowledge of math and physics, and the solutions of Maxwell’s equations can be
obtained by learning how to use these commercial software. This is a boon to many design
engineers: by running these software with cut-and-try engineering, wonderful systems can be
designed. It used to be said that if we lock 100 monkeys in a room and let them punch at the
100 keyboards, they will never type out Macbeth nor Hamlet. But with 100 engineers trained
with good physical insight, when locked up in a room with commercial software, with enough
time and patience, they can come up with wonderful designs of different electromagnetic
systems. In the parlance of the field, it is known as virtual proto-typing. It is mainly driven
by heuristics and cut-and-try engineering. Therefore, we will discuss the functions of different
antennas heuristically in this lecture.

303
304 Electromagnetic Field Theory

28.1 Resonance Tunneling in Antenna

We realize the power of resonance enhancement when we were young by playing on a swing
in the park. By pumping the swing at its resonance frequency, we can cause it to swing at a
large amplitude without a Herculean effort. A simple antenna like a short dipole behaves like
a Hertzian dipole with an effective length. A short dipole has an input impedance resembling
that of a capacitor. Hence, it is difficult to drive current into the antenna unless other elements
are added. Hertz was clever by using two metallic spheres to increase the current flow. A
large current flow on the stem of the antenna makes the stem resemble an inductor. Thus,
the end-cap capacitances and the stem inductance together can act like a resonator enhancing
the current flow on the antenna.

Some antennas are deliberately built to resonate with its structure to enhance its radiation.
A half-wave dipole is such an antenna as shown in Figure 28.1 [137]. These antennas are using
resonance tunneling to increase the currents on them to enhance their radiation efficiencies.
A half-wave dipole can also be thought of as a flared open transmission line in order to make
it radiate. It can be gradually morphed from a quarter-wavelength transmission line as shown
in Figure 28.1. A transmission line is a poor radiator, because the electromagnetic energy
is trapped between two pieces of metal. But a flared transmission line can radiate its field
to free space. The dipole antenna, though a simple device, has been extensively studied by
King [144].1

Figure 28.1: A half-wave dipole can be thought of as a resonator with radiation loss. It can
be thought of as a quarter-wavelength transmission line that is gradually opened up (courtesy
of electronics-notes.com).

1 He has reputed to have produced over 100 PhD students studying the dipole antenna.
Different Types of Antennas—Heuristics 305

Figure 28.2: The electromagnetic field around a Goubau line (courtesy of [145]). The field
resembles that of a coaxial cable with the outer conductor located at infinity.

One can also think of a piece of a wire as a waveguide. It is called a Goubau line as
shown in Figure 28.2, which can be thought of the limiting case of a coaxial cable where the
outer conductor is infinitely far away [145]. The wave is weakly guided since it now can shed
energy to infinity. The behavior of a wire as a Goubau line waveguide can be used to explain
heuristically why a half-wave dipole resonates when it is about half wavelength.

A folded dipole is often used to alter the input impedance of a dipole antenna [146]. Even
though it can have a resonant frequency lower than that of a normal dipole, the lowest resonant
mode does not radiate well. The mode that radiates well has the same resonant length as an
unfolded dipole. It has a radiation resistance four times that of a half-wave dipole of similar
length which is about 300 ohms. This is equal to the characteristic impedance of a twin-lead
transmission line [147]. Figure 28.3 shows a Yagi-Uda antenna driven by a folded dipole.
306 Electromagnetic Field Theory

Figure 28.3: A Yagi-Uda antenna was invented by heuristics in 1926. The principal element of
the antenna is the folded dipole. When a wire dipole antenna is less than half a wavelength,
it acts as a waveguide, or a director. When the wire antenna is slightly more than half a
wavelength, it acts as a reflector [148]. Therefore, the antenna radiates predominantly in one
direction (courtesy of Wikipedia [149]).

A Yagi-Uda antenna is also another interesting invention. It was invented in 1926 by Yagi
and Uda in Japan by plainly using physical intuition [148]. Physical intuition was a tool
of engineers of yesteryears while modern engineers tend to use sophisticated computer-aided
design (CAD) software. The principal driver element of the antenna is the folded dipole.
Surprisingly, the elements of dipoles in front of the driver element are acting like a waveguide
in space, while the element at the back acts like a reflector. Therefore, the field radiated by
the driver element will be directed toward the front of the antenna. Thus, this antenna has
higher directivity than just a stand alone dipole. Due to its simplicity, this antenna has been
made into nano-antennas which operate at optical frequencies [150].

Figure 28.4: A cavity-backed slot antenna radiates well because when the small dipole radiates
close to the resonant frequency of the cavity, the field strength is strongly enhanced inside
the cavity, and hence around the slot. This makes the slot into a good radiator (courtesy of
antenna-theory.com).

Slot antenna is a simple antenna to make [151]. To improve the radiation efficiency of slot
Different Types of Antennas—Heuristics 307

antenna, it is made to radiate via a cavity. A cavity-backed slot antenna that uses such a
concept is shown in Figure 28.4. A small dipole with poor radiation efficiency is placed inside
the cavity. When the operating frequency is close to the resonant frequency of the cavity, the
field strength inside the cavity becomes very strong, and much of the energy can leak from
the cavity via the slot on its side. This makes the antenna radiate more efficiently into free
space compared to just the small dipole alone.
Another antenna that resembles a cavity backed slot antenna is the microstrip patch an-
tenna, or patch antenna. This is shown in in Figure 28.5. This antenna also radiates efficiently
by resonant tunneling. Roughly, when L (see left of Figure 28.5) is half a wavelength, the
patch antenna resonates. This is similar to the resonant frequency of a transmission line with
open circuit at both ends. The current sloshes back and forth across the length of the patch
antenna along the L direction. The second design (right of Figure 28.5) has an inset feed.
This allows the antenna to resonate at a lower frequency because the current has a longer
path to slosh through when it is at resonance.

Figure 28.5: A microstrip patch antenna also radiates well when it resonates. The patch
antenna resembles a cavity resonator with magnetic wall (courtesy of emtalk.com).

28.2 Horn Antennas

The impedance of free space is 377 ohms while that of most transmission line is 50 ohms.
This impedance mismatch can be mitigated by using a flared horn (see Figure 28.6) [152].
One can thinkp that the characteristic impedance of a transmission line made of two pieces
of metal as Z0 = L/C. As the horn flares, C becomes smaller, increasing its characteristic
impedance to get close to that of free space. This allows for better impedance matching from
the source to free space. This is similar to the quarter wave transformer for matching the
characteristic impedance Z0 of a line to a load with impedance √ ZL . The requirement is that
the quarter wave transformer has an impedance given by ZT = Z0 ZL
A corrugated horn, as we have discussed previously in a circular waveguide in Section
20.1.1, discourages current flows in the non-axial symmetric mode. It encourages the prop-
agation of the TE01 mode in the circular waveguide and hence, the circular horn antenna.
This mode is axially symmetric, and thus, this antenna can radiate fields that are axially
symmetric [153, 154].
308 Electromagnetic Field Theory

Figure 28.6: A horn antenna works with the same principle as the biconical antenna. Its
flared horn changes the waveguide impedance so as to match the impedance of a waveguide
to the impedance of free space. The lower figure is that of a corrugated circular horn antenna.
The corrugation enhances the propagation of the TE01 mode in the circular waveguide, and
thus it enhances the cylindrical symmetry of the mode and the radiation field (courtesy of
tutorialpoints.com and comsol.com).

A Vivaldi antenna (invented by P. Gibson in 1978 [155]), is shown in Figure 28.7.2 It is

also called a notch antenna. It works by the same principle to gradually match the impedance
of the source to that of free space. But such a gradually flared horn has the element of a
frequency independent antenna. The low frequency component of the signal will radiate from
the wide end of the flared notch, while the high frequency component will radiate from the
narrow end of the notch. Thus, this antenna can radiate effectively over a broad range of
frequencies, and thus this gives the antenna a broad bandwidth performance. It is good for
transmitting a pulsed signal which has a broad frequency spectrum.
2 He must have loved the musician Vivaldi so much:)
Different Types of Antennas—Heuristics 309

Figure 28.7: A Vivaldi antenna, also called a notch antenna, works like a horn antenna, but
uses very little metal. Hence, it is cheap to build, and its flared notch makes it broadband
(courtesy of Wikipedia [156]).

28.3 Quasi-Optical Antennas

High-frequency or short wavelength electromagnetic field behaves like light ray as in optics.
Therefore, many high-frequency antennas are designed based on the principle of ray optics.
A reflector antenna is such an antenna as shown in Figure 28.8. The reflector antenna in this
case is a Cassegrain design [157]3 where a sub-reflector is present. This allows the antenna
to be fed from behind the parabolic dish where the electronics can be stored and isolated as
well. Reflector antennas [159] are prevalent in radio astronomy and space exploration due to
their high directivity and sensitivity.

Figure 28.8: The left picture of an NRAO radio telescope antenna of Cassegrain design in
Virginia, USA (courtesy of Britannica.com). The bottom is the detail of the Cassegrain design
(courtesy of rev.com).
3 The name came from an optical telescope of similar design [158]
310 Electromagnetic Field Theory

Another recent invention is the reflectarray antenna [160, 161] which is very popular. One
of them is as shown in Figure 28.9. Due to recent advent in simulation technology, complicated
structures can be simulated on a computer, including one with a complicated surface design.
Patch elements can be etched onto a flat surface as shown, giving it an effective impedance
that is spatially varying, making it reflect like a curved surface. Such a surface is known as a
meta-surface [162, 163]. It can greatly economize on the space usage compared to a reflector
antenna.

Figure 28.9: A reflectarray where the reflector is a flat surface. Patches are unequally spaced
to give the array the focussing effect (courtesy of antenna-theory.com).

Another quasi-optical antenna is the lens antenna as shown in Figure 28.10 [164]. The
design of this antenna follows lens optics, and is only valid when the wavelength is very short
compared to the curvature of the surfaces. In this case, reflection and transmission at a curve
surface is similar to that of a flat surface. This is called the tangent-plane approximation of
a curve surface, and is valid at high frequencies.
Different Types of Antennas—Heuristics 311

Figure 28.10: The left figure shows a lens antenna where the lens is made of artificial dielectrics
made from metallic strips (courtesy of electriciantutoring.tpub.com). The right figure shows
some dielectric lens at the aperture of an open waveguide to focus the microwave exiting from
the waveguide opening (courtesy of micro-radar.de).

28.4 Small Antennas

Small antennas are in vogue these days due to the advent of the cell phone, and the im-
portance of economizing on the antenna size due to miniaturization requirements. Also, the
antennas should have enough bandwidth to accommodate the signals from different cell phone
companies, which use different carrier frequencies. An interesting small antenna is the PIFA
(planar inverted F antenna) shown in Figure 28.11 [165]. Because it is shorted at one end
and open circuit at the other end, it acts like a quarter wavelength resonator, making it
substantially smaller. But the resonator has a low Q because of the “slots” or “openings”
around it from whom energy can leak. The low Q gives this antenna a broader bandwidth.

Figure 28.11: A PIFA (planar inverted F antenna) is compact, broadband, and easy to
fabricate. It is good for cell phone antennas due to its small size (courtesy of Mathworks).
312 Electromagnetic Field Theory

An interesting small antenna is the U-slot antenna shown in Figure 28.12 [166, 167]. Be-
cause the current is forced to follow a longer tortuous path by the U-slot, it can resonant with
a longer wavelength (lower frequency) and hence, can be made smaller compared to wave-
length. In order to give the antenna a larger bandwidth, its Q is made smaller by etching it on
a thick dielectric substrate (shown as the dielectric material region in the figure). But feeding
it with a longer probe will make the bandwidth of the antenna smaller, due to the larger
inductance of the probe.4 An ingenious invention is to use an L probe [168]. The L probe has
an inductive part as well as a capacitive part. Their reactance cancel each other, allowing
the electromagnetic energy to tunnel through the antenna, making it a better radiator.

Figure 28.12: The top figure shows a U slot patch antenna design. The bottom figure shows
a patch antenna fed by an L probe with significant increase in bandwidth (courtesy of K.M.
Luk) [168].

Another area where small antennas are needed is in RFID (radio frequency identification)
tag [169]. Since tags are placed outside the packages of products, e.g., in a warehouse, an
RFID tag has a transmit-receive antenna that can communicate with the external world. The
communication is done through an RFID reader. The RFID reader can talk to a small com-
4 Remember that larger inductance implies more store magnetic field energy, and hence, the higher Q of

the system.
Different Types of Antennas—Heuristics 313

puter chip embedded in the tag where data about the package can be stored. Thus, an RFID
reader can quickly and remotely communicate with the RFID tag to retrieve information
about the package. Such a small antenna design for RFID tag is shown in Figure 28.13. It
uses image theorem (that we shall learn later) so that the antenna can be made half as small.
Then slots are cut into the radiating patch, so that the current follows a longer path. This
lowers the resonant frequency of the antenna, allowing it to be made smaller. The take home
message here is that to make an antenna a few times smaller than a wavelength to resonate,
the current on the antenna has to flow through a tortuous path. In this manner, the antenna
can be made a few times smaller than the wavelength.

Figure 28.13: Some RFID antennas designed at The University of Hong Kong (courtesy of P.
Yang, Y. Li, J. Huang, L.J. Jiang, S.Q. He, T. Ye, and W.C. Chew).

An RFID reader can be designed to read the information from a batch of vials or test tubes
containing different chemicals. Hence, a large loop antenna is needed but at a sufficiently high
frequency (for large bandwidth). However, a loop antenna, if we look at a piece of wire as a
Goubau line [145], will have resonant frequencies. When a loop antenna resonates, the current
is non-uniform on it. This happens at higher frequencies. (Fundamentally, this comes from
the retardation effect of electromagnetic field.) This will result in a non-uniform field inside
the loop defeating the design of the RFID reader.
One way to view how the non-uniform current come about is that a piece of wire becomes
a tiny inductor. Across an inductor, V = jωLI, implying a 90◦ phase shift between the
voltage and the current. In other words, the voltage drop is always nonzero, and therefore,
the voltage cannot be constant around the loop. Since the voltage and current are locally
314 Electromagnetic Field Theory

related by the local inductance, the current cannot be constant also.

To solve the problem of the current and voltage being non-constant around the loop, a
local inductor is connected in series with a capacitor [170]. This causes the local LC tank
circuit to resonate. At resonance, the current-voltage relationship across the LC tank circuit
is such that there is no voltage drop across the tank circuit since it becomes a short circuit. In
this way, the voltage is equilized between two points and becomes uniform across the loop so
is the current. Therefore, one way to enable a uniform current in a large loop is to capacitively
load the loop. This will ensure a constant phase, or a more uniform current around the loop,
and hence, a more efficient reader. Such a design is shown in Figure 28.14.
Different Types of Antennas—Heuristics 315

Figure 28.14: The top figure shows a RFID reader designed by [171]. The bottom figure
shows simulation and measurement done at The University of Hong Kong (courtesy of Z.N.
Chen [171] and P. Yang, Y. Li, J. Huang, L.J. Jiang, S.Q. He, T. Ye, and W.C. Chew).
316 Electromagnetic Field Theory
Lecture 29

Uniqueness Theorem

The uniqueness of a solution to a linear system of equations is an important concept in

mathematics. Under certain conditions, ordinary differential equation partial differential
equation and matrix equations will have unique solutions. But uniqueness is not always
guaranteed as we shall see. This issue is discussed in many math books and linear algebra
books [77,89]. The proof of uniqueness for Laplace and Poisson equations are given in [31,54]
which is slightly different from electrodynamic problems.
Just imagine how bizzare it would be if there are more than one possible solutions. One
has to determine which is the real solution. To quote Star Trek, we need to know who the
real McCoy is!1

29.1 The Difference Solutions to Source-Free Maxwell’s

Equations
In this section, we will prove uniqueness theorem for electrodynamic problems [32, 35, 50, 65,
83]. First, let us assume that there exist two solutions in the presence of one set of common
impressed sources Ji and Mi .2 Namely, these two solutions are Ea , Ha , Eb , Hb . Both of them
satisfy Maxwell’s equations and the same boundary conditions. Are Ea = Eb , Ha = Hb ?
To study the uniqueness theorem, we consider general linear anisotropic inhomogeneous
media, where the tensors µ and ε can be complex so that lossy media can be included. In
the frequency domain, lets assume two possible solutions with one given set of sources Ji and

1 This phrase was made popular to the baby-boom generation, or the Trekkies by Star Trek. It actually

refers to an African American inventor.

2 It is not clear when the useful concept of impressed sources were first used in electromagnetics even though

it was used in [172] in 1936. These are immutable sources that cannot be changed by the environment in
which they are immersed.

317
318 Electromagnetic Field Theory

Mi , it follows that

∇ × Ea = −jωµ · Ha − Mi (29.1.1)
b b
∇ × E = −jωµ · H − Mi (29.1.2)
∇ × Ha = jωε · Ea + Ji (29.1.3)
∇ × Hb = jωε · Eb + Ji (29.1.4)

By taking the difference of these two solutions, we have

∇ × (Ea − Eb ) = −jωµ · (Ha − Hb ) (29.1.5)

a b a b
∇ × (H − H ) = jωε · (E − E ) (29.1.6)

Or alternatively, defining δE = Ea − Eb and δH = Ha − Hb , we have

∇ × δE = −jωµ · δH (29.1.7)
∇ × δH = jωε · δE (29.1.8)

The difference solutions, δE and δH, satisfy the original source-free Maxwell’s equations.
Source-free here implies that we are looking at the homogeneous solutions of the pertinent
partial differential equations constituted by (29.1.7) and (29.1.8).
To prove uniqueness, we would like to find a simplifying expression for ∇ · (δE × δH∗ ).
By using the product rule for divergence operator, it can be shown that

∇ · (δE × δH∗ ) = δH∗ · ∇ × δE − δE · ∇ × δH∗ (29.1.9)

Then by taking the left dot product of δH∗ with (29.1.7), and then the left dot product of
δE with the complex conjugation of (29.1.8), we obtain

δH∗ · ∇ × δE = −jωδH∗ · µ · δH
δE · ∇ × δH∗ = −jωδE · ε∗ · δE∗ (29.1.10)

Now, taking the difference of the above, we get

δH∗ · ∇ × δE − δE · ∇ × δH∗ = ∇ · (δE × δH∗ )

= −jωδH∗ · µ · δH + jωδE · ε∗ · δE∗ (29.1.11)
Uniqueness Theorem 319

Figure 29.1: Geometry for proving the uniqueness theorem. We like to know the requisite
boundary conditions on S plus the type of media inside V in order to guarantee the uniqueness
of the solution in V .

Next, we integrate the above equation (29.1.11) over a volume V bounded by a surface S
as shown in Figure 29.1. After making use of Gauss’ divergence theorem, we arrive at
 
∇ · (δE × δH∗ )dV = (δE × δH∗ ) · dS
V
 S

= [−jωδH∗ · µ · δH + jωδE · ε∗ · δE∗ ]dV (29.1.12)

V

And next, we would like to know the kind of boundary conditions that would make the
left-hand side equal to zero. It is seen that the surface integral on the left-hand side will be
zero if:3
1. If n̂ × E is specified over S for the two possible solutions, so that n̂ × Ea = n̂ × Eb on S.
Then n̂ × δE = 0, which is the PEC boundary condition for δE, and then4
S
(δE × δH ) · n̂dS = S (n̂ × δE) · δH∗ dS = 0.
∗

2. If n̂ × H is specified over S for the two possible solutions, so that n̂ × Ha = n̂ × Hb on S.

Then n̂ × δH = ∗0, which is the PMC boundary
∗
condition for δH, and then
S
(δE × δH ) · n̂dS = − S
(n̂ × δH ) · δEdS = 0.

3. Let the surface S be divided into two mutually exclusive surfaces S1 and S2 .5 If n̂ × E is
3 In the following, please be reminded that PEC stands for “perfect electric conductor”, while PMC stands

for “perfect magnetic conductor”. PMC is the dual of PEC. Also, a fourth case of impedance boundary
condition is possible, which is beyond the scope of this course. Interested readers may consult Chew, Theory
of Microwave and Optical Waveguides [83].
4 We use the vector identity that a · (b × c) = c · (a × b) = b · (c × a). Also, from Section 1.3.3, dS = n̂dS.
5 In math parlance, S ∪ S = S.
1 2
320 Electromagnetic Field Theory

specified over S1 , and n̂ × H is specified over S2 . Then n̂ × δE = 0 (PEC boundary condition)

on S1 , and n̂ × δH = 0 (PMC boundary condition) on S2 . Therefore, the left-hand side
becomes
  
(δE × δH∗ ) · n̂dS = S1 + S2 = S1 (n̂ × δE) · δH∗ dS
S 
− S2 (n̂ × δH∗ ) · δEdS = 0.

Thus, under the above three scenarios, the left-hand side of (29.1.12) is zero, and then
the right-hand side of (29.1.12) becomes

[−jωδH∗ · µ · δH + jωδE · ε∗ · δE∗ ]dV = 0 (29.1.13)
V

For lossless media, µ and ε are hermitian tensors (or matrices6 ), then it can be seen, using
the properties of hermitian matrices or tensors, that δH∗ · µ · δH and δE · ε∗ · δE∗ are purely
real. Taking the imaginary part of the above equation yields

[−δH∗ · µ · δH + δE · ε∗ · δE∗ ]dV = 0 (29.1.14)
V

The above two terms correspond to stored magnetic field energy and stored electric field
energy in the difference solutions δH and δE, respectively. The above being zero does not
imply that δH and δE are zero.
For resonant solutions, the stored electric energy can balance the stored magnetic energy.
The above resonant solutions are those of the difference solutions satisfying PEC or PMC
boundary condition or mixture thereof. Also, they are the resonant solutions of the source-free
Maxwell’s equations (29.1.7). Therefore, δH and δE need not be zero, even though (29.1.14)
is zero. This happens when we encounter solutions that are the resonant modes of the volume
V bounded by the surface S.

29.2 Conditions for Uniqueness

Uniqueness can only be guaranteed if the medium is lossy as shall be shown later. It is also
guaranteed if lossy impedance boundary conditions are imposed.7 First we begin with the
isotropic case.

29.2.1 Isotropic Case

It is easier to see this for lossy isotropic media. Then (29.1.13) simplifies to

[−jωµ|δH|2 + jωε∗ |δE|2 ]dV = 0 (29.2.1)
V

6 Tensors are a special kind of matrices.

7 See Chew, Theory of Microwave and Optical Waveguides.
Uniqueness Theorem 321

For isotropic lossy media, µ = µ0 − jµ00 and ε = ε0 − jε00 . Taking the real part of the above,
we have from (29.2.1) that

[−ωµ00 |δH|2 − ωε00 |δE|2 ]dV = 0 (29.2.2)
V

Since the integrand in the above is always negative definite, the integral can be zero only if

δE = 0, δH = 0 (29.2.3)

everywhere in V , implying that Ea = Eb , and that Ha = Hb . Hence, it is seen that uniqueness

is guaranteed only if the medium is lossy.
The physical reason is that when the medium is lossy, a homogeneous solution (also called
a natural solution) which is pure time-harmonic solution cannot exist due to loss. The modes,
which are the source-free solutions of Maxwell’s equations, are decaying sinusoids. But when
we express equations (29.1.1) to (29.1.4) in the frequency domain, we are seeking solutions
for which ω is real. Thus decaying sinusoids are not among the possible solutions, and hence,
they are not in the solution space.
Notice that the same conclusion can be drawn if we make µ00 and ε00 negative. This
corresponds to active media, and uniqueness can be guaranteed for a time-harmonic solution.
In this case, no time-harmonic solution exists, and the resonant solution is a growing sinusoid.

29.2.2 General Anisotropic Case

The proof for general anisotropic media is more complicated. For the lossless anisotropic
media, we see that (29.1.13) is purely imaginary. However, when the medium is lossy, this
same equation will have a real part. Hence, we need to find the real part of (29.1.13) for the
general lossy case.

About taking the Real and Imaginary Parts of a Complicated Expression

To this end, we digress on taking the real and imaginary parts of a complicated expression.
Here, we need to find the complex conjugate8 of (29.1.13), which is scalar, and add it to itself
to get its real part. To this end, we will find the conjugate of its integrand which is a scalar
number.
First, the complex conjugate of the first scalar term in the integrand of (29.1.13) is9

(−jωδH∗ · µ · δH)∗ = jωδH · µ∗ · δH∗ = jωδH∗ · µ† · δH (29.2.4)

Similarly, the complex conjugate of the second scalar term in the same integrand is

(jωδE · ε∗ · δE∗ )∗ = −jωδE∗ · ε† · δE (29.2.5)

8 Also called hermitian conjugate.
9 To arrive at these expressions, one makes use of the matrix algebra rule that if D = A · B · C, then
t t t t
D = C · B · A . This is true even for non-square matrices. But for our case here, A is a 1 × 3 row vector,
and C is a 3 × 1 column vector, and B is a 3 × 3 matrix. In vector algebra, the transpose of a vector is implied.
Also, in our case here, D is a scalar, and hence, its transpose is itself.
322 Electromagnetic Field Theory

But
jωδE · ε∗ · δE∗ = jωδE∗ · ε† · δE (29.2.6)
The above gives us the complex conjugate of the scalar quantity (29.1.13) and adding it to
itself, we have

[−jωδH∗ · (µ − µ† ) · δH − jωδE∗ · (ε − ε† ) · δE]dV = 0 (29.2.7)
V

For lossy media, −j(µ − µ† ) and −j(ε − ε† ) are hermitian positive matrices. Hence the
integrand is always positive definite, and the above equation cannot be satisfied unless δH =
δE = 0 everywhere in V . Thus, uniqueness is guaranteed in a lossy anisotropic medium.
Similar statement can be made for the isotropic case if the medium is active. Then the
integrand is positive definite, and the above equation cannot be satisfied unless δH = δE = 0
everywhere in V , thereby proving that uniqueness is satisfied.

29.3 Hind Sight Using Linear Algebra

The proof of uniqueness for Maxwell’s equations is very similar to the proof of uniqueness for
a matrix equation [77]. As you will see, the proof using linear algebra is a lot simpler due to
the simplicity of notations. To see this, consider a linear algebraic equation

A·x=b (29.3.1)

If a solution to a matrix equation exists without excitation, namely, when b = 0, then the
solution is the null space solution [77], namely, x = xN . In other words,

A · xN = 0 (29.3.2)

These null space solutions exist without a “driving term” b on the right-hand side. For
Maxwell’s equations, b corresponds to the source terms. The solution in (29.3.2) is like the
homogeneous solution of an ordinary differential equation or a partial differential equation
[89]. In an enclosed region of volume V bounded by a surface S, homogeneous solutions
are the resonant solutions (or the natural solutions) of this Maxwellian system. When these
solutions exist, they give rise to non-uniqueness. Note that these resonant solutions in the
time domain exist for all time if the cavity is lossless.
Also, notice that (29.1.7) and (29.1.8) are Maxwell’s equations without the source terms.
In a closed region V bounded by a surface S, only resonant solutions for δE and δH with the
relevant boundary conditions can exist when there are no source terms.
As previously mentioned, one way to ensure that these resonant solutions (or homoge-
neous solutions) are eliminated is to put in loss or gain. When loss or gain is present, then
the resonant solutions are decaying sinusoids or growing sinusoids (see Section 22.1.1 for an
analogue with LC tank circuit). Since we are looking for solutions in the frequency domain,
or time harmonic solutions, the solutions considered are on the real ω axis on the complex ω
plane. Thus the non-sinusoidal solutions are outside the solution space: They are not part of
the time-harmonic solutions (which are on the real axis) that we are looking for. Therefore,
Uniqueness Theorem 323

complex resonant solutions which are off the real axis, and are homogeneous solutions, are
not found on the real axis.
We see that the source of non-uniqueness is the homogeneous solutions or the resonant
solutions of the system that persist for all time. These solutions are non-causal, and they are
there in the system since the beginning of time to time ad infinitum. One way to remove these
resonant solutions is to set them to zero at the beginning by solving an initial value problem
(IVP). However, this has to be done in the time domain. One reason for non-uniqueness is
because we are seeking the solutions in the frequency domain.

29.4 Connection to Poles of a Linear System

The output to input of a linear system can be represented by a transfer function H(ω) [52,173].
If H(ω) has poles, and if the system is lossless, the poles are on the real axis. Therefore, when
ω = ωpole , the function H(ω) becomes undefined. In other words, one can add a constant
term to the output, and the ratio between output to input is still infinity. This also gives
rise to non-uniqueness of the output with respect to the input. Poles usually correspond to
resonant solutions, and hence, the non-uniqueness of the solution is intimately related to the
non-uniqueness of Maxwell’s equations at the resonant frequencies of a structure. This is
illustrated in the upper part of Figure 29.2.

Figure 29.2: The non-uniqueness problem is intimately related to the locations of the poles
of a transfer function being on the real axis, when one solves a linear system using Fourier
transform technique. For a lossless system, the poles are located on the real axis. When
performing a Fourier inverse transform to obtain the solution in the time domain, then the
Fourier inversion contour is undefined, and the solution cannot be uniquely determined.
324 Electromagnetic Field Theory

If the input function is f (t), with Fourier transform F (ω), then the output y(t) is given
by the following Fourier integral, viz.,
 ∞
1
y(t) = dωejωt H(ω)F (ω) (29.4.1)
2π −∞

where the Fourier inversion integral path is on the real axis on the complex ω plane. The
Fourier inversion integral is undefined or non-unique if poles exist on the real ω axis.
However, if loss is introduced, these poles will move away from the real axis as shown
in the lower part of Figure 29.2. Then the transfer function is uniquely determined for all
frequencies on the real axis. In this way, the Fourier inversion integral in (29.4.1) is well
defined, and uniqueness of the solution is guaranteed.
When the poles are located on the real axis yielding possibly non-unique solutions, a
remedy to this problem is to use Laplace transform technique [52]. The Laplace transform
technique allows the specification of initial values, which is similar to solving the problem as
an initial value problem (IVP).
If you have problem wrapping your head around this concept, it is good to connect back
to the LC tank circuit example. The transfer function H(ω) is similar to the Y (ω) of (22.1.4).
The transfer function has two poles. If there is no loss, then the poles are located on the
real axis, rendering the Fourier inversion contour undefined in (29.4.1). Hence, the solution
is non-unique. However, if infinitesimal loss is introduced by setting R 6= 0, then the poles
will migrate off the real axis making (29.4.1) well defined!

29.5 Radiation from Antenna Sources and Radiation Con-

dition
The above uniqueness theorem guarantees that if we have some antennas with prescribed
current sources on them, the radiated field from these antennas are unique. To see how this
can come about, we first study the radiation of sources into a region V bounded by a large
surface Sinf as shown in Figure 29.4 [35].
Even when n̂ × E or n̂ × H are specified on the surface at Sinf , the solution is nonunique
because the volume V bounded by Sinf , can have many resonant solutions. In fact, the region
will be replete with resonant solutions as one makes Sinf become very large.
To have more insight, we look at the resonant condition of a large rectangular cavity given
by (21.2.3) reproduced here as

ω2  mπ 2  nπ 2  pπ 2
β2 = = + + (29.5.1)
c2 a b d
The above is an equation of an Ewald sphere in a 3D mode space, but the values of βx = mπa ,
pπ
βy = nπb , and βz = d are discrete. We can continuously change the operating frequency
ω above until the above equation is satisfied. When this happens, we encounter a resonant
frequency of the cavity. At this operating frequency, the solution to Maxwell’s equations
inside the cavity is non-unique. As the dimensions of the cavity become large or a, b, and
d are large, then the number of ω’s or resonant frequencies that the above equation can be
Uniqueness Theorem 325

satisfied or approximately satisfied becomes very large. This is illustrated Figure 29.3 in 2D.
Hence, the chance of the operating frequency ω coinciding with a resonant mode of the cavity
is very high giving rise to nonuniqueness. This is even more so when the cavity becomes very
large. Hence, the chance of operating inside the cavity with unique solution is increasingly
difficult. This above argument applies to cavities of other shapes as well.

Figure 29.3: For very large cavity, the grid spacing in the mode space (or Fourier space)
becomes very small. Then the chance that the sphere surface encounters a resonant mode is
very high. When this happens, the solution to the cavity problem is non-unique.

The way to remove these resonant solutions is to introduce an infinitesimal amount of loss
in region V . Then these resonant solutions will disappear from the real ω axis, where we seek
a time-harmonic solution. Now we can take Sinf to infinity, and the solution will always be
unique even if the loss is infinitesimally small.
Notice that if Sinf → ∞, the waves that leave the sources will never be reflected back
because of the small amount of loss. The radiated field will just disappear into infinity.
This is just what radiation loss is: power that propagates to infinity, but never to return.
In fact, one way of guaranteeing the uniqueness of the solution in region V when Sinf is
infinitely large, or that V is infinitely large is to impose the radiation condition: the waves
that radiate to infinity are outgoing waves only, and never do they return. This is also called
the Sommerfeld radiation condition [174]. . Uniqueness of the field outside the sources
is always guaranteed if we assume that the field radiates to infinity and never to return. This
is equivalent to solving the cavity solutions with an infinitesimal loss, and then letting the
size of the cavity become infinitely large.
326 Electromagnetic Field Theory

Figure 29.4: The solution for antenna radiation is unique because we impose the Sommerfeld
radiation condition when seeking the solution. That is we assume that the radiation wave
travels to infinity but never to return. This is equivalent to assuming an infinitesimal loss
when seeking the solution in V .
Lecture 30

Reciprocity Theorem

Reciprocity theorem is one of the most important theorems in electromagnetics. With it we

can develop physical intuition to ascertain if a certain design or experiment is right or wrong.
It also tells us what is possible or impossible in the design of many systems. Reciprocity
theorem is like “tit-for-tat” relationship in humans: Good-will is reciprocated with good will
while ill-will is countered with ill-will. Both Confucius (551 BC–479 BC) and Jesus Christ
(4 BC–AD 30) epoused the concept that, “Don’t do unto others that you don’t like others to
do unto you.” But in electromagnetics, this beautiful relationship can be expressed precisely
and succinctly using mathematics. We shall see how this is done.

Figure 30.1: (Left) A depiction of Confucius from a stone fresco from the Western Han
dynasty (202 BC–9 AD). The emphasis of the importance of “reciprocity” by Confucius
Analects translated by D. Hinton [175]. (Right) A portrait of Jesus that is truer to its form.
Jesus teaching from the New Testament says, “Do unto others as you would have them do
unto you.” Luke 6:31 and Matthew 7:12 [176]. The subsequent portraits of these two sages
are more humanly urbane.

327
328 Electromagnetic Field Theory

30.1 Mathematical Derivation

Figure 30.2: The geometry for proving reciprocity theorem. We perform two experiments on
the same object or scatterer: (a) With sources J1 and M1 turned on, generating fields E1
and H1 , and (b) With sources J2 and M2 turned on, generating fields E2 and H2 . Magnetic
currents, by convention, are denoted by double arrows.

Consider a general anisotropic inhomogeneous medium in the frequency domain where both
µ(r) and ε(r) are described by permeability tensor and permittivity tensor over a finite part
of space as shown in Figure 30.2. This representation of the medium is quite general, and
it can include dispersive and conductive media as well. It can represent complex terrain, or
complicated electronic circuit structures in circuit boards or microchips, as well as complicated
antenna structures.
We can do a Gedanken experiment1 where a scatterer or an object is illuminated by fields
from two sets of sources which are turned on and off consecutively. This is illustrated in
Figure 30.2: When only J1 and M1 are turned on, they generate fields E1 and H1 in this
medium. On the other hand, when only J2 and M2 are turned on, they generate E2 and
H2 in this medium. Therefore, the pertinent equations in the frequency domain, for linear
time-invariant systems, for these two cases are2

∇ × E1 = −jωµ · H1 − M1 (30.1.1)
∇ × H1 = jωε · E1 + J1 (30.1.2)
∇ × E2 = −jωµ · H2 − M2 (30.1.3)
∇ × H2 = jωε · E2 + J2 (30.1.4)

We would like to find a simplifying expression for the divergence of the following quantity,

∇ · (E1 × H2 ) = H2 · ∇ × E1 − E1 · ∇ · H2 (30.1.5)

so that the divergence theorem can be invoked. To this end, and from the above, we can show
1 Thought experiment in German.
2 The current sources are impressed currents so that they are immutable, and not changed by the environ-
ment they are immersed in [50, 172].
Reciprocity Theorem 329

that (after left dot-multiply (30.1.1) with H2 and (30.1.4) with E1 ),

H2 · ∇ × E1 = −jωH2 · µ · H1 − H2 · M1 (30.1.6)
E1 · ∇ × H2 = jωE1 · ε · E2 + E1 · J2 (30.1.7)

Then, using the above, and the following identity, we get the second equality in the following
expression:

∇ · (E1 × H2 ) = H2 · ∇ × E1 − E1 · ∇ × H2
= −jωH2 · µ · H1 − jωE1 · ε · E2 − H2 · M1 − E1 · J2 (30.1.8)

By the same token,

∇ · (E2 × H1 ) = −jωH1 · µ · H2 − jωE2 · ε · E1 − H1 · M2 − E2 · J1 (30.1.9)

If one assumes that

µ = µt , ε = εt (30.1.10)

or when the tensors are symmetric, then H1 · µ · H2 = H2 · µ · H1 and E1 · ε · E2 = E2 · ε · E1 .3

Upon subtracting (30.1.8) and (30.1.9), one gets

∇ · (E1 × H2 − E2 × H1 ) = −H2 · M1 − E1 · J2 + H1 · M2 + E2 · J1 (30.1.11)

Figure 30.3: The geometry for proving reciprocity theorem when the surface S: (a) does not
enclose the sources, and (b) encloses the sources. In the figure, the sources are supposed to
be either (M1 , J1 ) producing fields (E1 , H1 ) or (M2 , J2 ) producing fields (E2 , H2 ).

Now, integrating (30.1.11) over a volume V bounded by a surface S, and invoking Gauss’
divergence theorem, we have the reciprocity theorem that

dS · (E1 × H2 − E2 × H1 )
S

=− dV [H2 · M1 + E1 · J2 − H1 · M2 − E2 · J1 ] (30.1.12)
V
3 It is to be noted that in matrix algebra, the dot product between two vectors are often written as at · b,

but in the physics literature, the transpose on a is implied. Therefore, the dot product between two vectors
is just written as a · b.
330 Electromagnetic Field Theory

When the volume V contains no sources (see Figure 30.3), the reciprocity theorem reduces to

dS · (E1 × H2 − E2 × H1 ) = 0 (30.1.13)
S

The above is also called Lorentz reciprocity theorem by some authors.4

Next, when the surface S contains all the sources (see Figure 30.3), then the right-hand
side of (30.1.12) will not be zero. On the other hand, when the surface S → ∞, E1 and
H2 becomes spherical waves which can be approximated by plane waves sharing the same β
vector. Moreover, under the plane-wave approximation, ωµ0 H2 = β × E2 , ωµ0 H1 = β × E1 ,
then

E1 × H2 ∼ E1 × (β × E2 ) = E1 (β · E2 ) − β(E1 · E2 ) (30.1.14)
E2 × H1 ∼ E2 × (β × E1 ) = E2 (β · E1 ) − β(E2 · E1 ) (30.1.15)

But β · E2 = β · E1 = 0 in the far field and the β vectors are parallel to each other. Therefore,
the two terms on the left-hand side of (30.1.12) cancel each other, and it vanishes when
S → ∞. (They cancel each other so that the remnant field vanishes faster than 1/r2 . This is
necessary as the surface area S is growing larger and proportional to r2 .)
As a result, when S → ∞, (30.1.12) can be rewritten simply as
 
dV [E2 · J1 − H2 · M1 ] = dV [E1 · J2 − H1 · M2 ] (30.1.16)
V V

The inner product symbol is often used to rewrite the above as

hE2 , J1 i − hH2 , M1 i = hE1 , J2 i − hH1 , M2 i (30.1.17)

where the inner product hA, Bi = V dV A(r) · B(r).
The above inner product is also called reaction, a concept introduced by Rumsey [177].
The above is also called the Rumsey reaction theorem. Sometimes, the above is rewritten
more succinctly and tersely as

h2, 1i = h1, 2i (30.1.18)

where

h2, 1i = hE2 , J1 i − hH2 , M1 i (30.1.19)

The concept of inner product or reaction can be thought of as a kind of “measurement”. The
reciprocity theorem can be stated as that the fields generated by sources 2 as “measured”
by sources 1 is equal to fields generated by sources 1 as “measured” by sources 2. This
measurement concept is more lucid if we think of these sources as Hertzian dipoles.
4 Harrington, Time-Harmonic Electric Field [50].
Reciprocity Theorem 331

30.2 Conditions for Reciprocity

It is seen that the above proof hinges on (30.1.10). In other words, the anisotropic medium
has to be described by symmetric tensors. They include conductive media, but not gyrotropic
media which is non-reciprocal. A ferrite biased by a magnetic field is often used in electronic
circuits, and it corresponds to a gyrotropic, non-reciprocal medium.5 Also, our starting
equations (30.1.1) to (30.1.4) assume that the medium and the equations are linear time
invariant so that Maxwell’s equations can be written down in the frequency domain easily.

30.3 Application to a Two-Port Network and Circuit

Theory

Figure 30.4: A geometry for proving the circuit relationship between two antennas using
reciprocity theorem. Circuit relationship is possible when the ports of the antennas are small
compared to wavelength.

The reciprocity theorem can be used to distill and condense the interaction between two
antennas over a complex terrain as long as the terrain comprises reciprocal media, namely,
if µ = µt and ε = εt for these media.6 In Figure 30.4, we assume that antenna 1 is driven
by impressed current J1 while antenna 2 is driven by impressed current J2 . It is assumed
that the antennas are made from reciprocal media, such as conductive media. Since the
system is linear time invariant, it can be written as the interaction between two ports as in
circuit theory as shown in Figure 30.5. Assuming that these two ports are small compared
to wavelengths, then we can apply circuit concepts like potential theory by letting E = −∇Φ
in the neighborhood of the ports. Thus, we can define voltages and currents at these ports,
and V-I relationships can be established in the manner of circuit theory.

5 Non-reciprocal media are important for making isolators in microwave. Microwave signals can travel from

Port 1 to Port 2, but not vice versa.

6 It is to be noted that a gyrotropic medium considred in Section 9.1 does not safisfy this reciprocity criteria,

but it does satisfy the lossless criteria of Section 10.3.2.

332 Electromagnetic Field Theory

Figure 30.5: The interaction between two antennas in the far field of each other can be
reduced to a circuit theory description since the input and output ports of the antennas are
small compared to wavelength.

Focusing on a two-port network as shown in Figure 30.5, we have from circuit theory
that [178]
    
V1 Z11 Z12 I1
= (30.3.1)
V2 Z21 Z22 I2

This form is permissible since we have a linear time-invariant system, and this is the most
general way to establish a linear relationship between the voltages and the currents. Further-
more, the matrix elements Zij can be obtained by performing a series of open-circuit and
short-circuit measurements as in circuit theory.
Then assuming that the port 2 is turned on with J2 6= 0, while port 1 is turned off with
J1 = 0. In other words, port 1 is open circuit, and the source J2 is an impressed current
source 7 that will produce an electric field E2 at port 1. Since the current at port 1 is turned
off, or that J1 = 0, the voltage measured at port 1 is the open-circuit voltage V1oc . Please
note here that J1 and J2 are impressed currents and are only defined in their respective port.
Consequently, the reaction
 
hE2 , J1 i = dV (E2 · J1 ) = I1 E2 · dl = −I1 V1oc (30.3.2)
V Port 1

Even though port 1 is assumed to be off, the J1 is the impressed current to be used above
is the J1 when port 1 is turned on. Given that the port is in the circuit physics regime, we
assume the currents in wire to be constant, then the current J1 is a constant current at the
port when it is turned on. Or the current I1 can be taken outside the integral. In slightly
more details, the current J1 = ˆlI1 /A where A is the cross-sectional area of the wire, and ˆl is
a unit vector aligned with the axis of the wire. The volume

integral dV = Adl, and hence the
second equality follows above, where dl = ˆldl. Since Port 1 E2 · dl = −V1oc , we have the last
equality above.
We can repeat the derivation with port 2 to arrive at the reaction

hE1 , J2 i = I2 E1 · dl = −I2 V2oc (30.3.3)
Port 2
7 This is the same as the current source concept in circuit theory.
Reciprocity Theorem 333

Reciprocity requires these two reactions to be equal, and hence,

I1 V1oc = I2 V2oc
But from (30.3.1), we can set the pertinent currents to zero to find these open circuit voltages.
Therefore, V1oc = Z12 I2 , V2oc = Z21 I1 . Since I1 V1oc = I2 V2oc by the reaction concept or by
reciprocity, then Z12 = Z21 . The above analysis can be easily generalized to an N -port
network.
The simplicity of the above belies its importance. The above shows that the reciprocity
concept in circuit theory is a special case of reciprocity theorem for electromagnetic theory.
The terrain can also be replaced by complex circuits as in a circuit board, as long as the
materials in the terrain or circuit board are reciprocal, linear and time invariant. For instance,
the complex terrain can also be replaced by complex antenna structures. It is to be noted
that even when the transmit and receive antennas area miles apart, as long as the transmit
and receive ports of the linear system can be characterized by a linear relation expounded by
(30.3.1), and the ports small enough compared to wavelength so that circuit theory prevails,
we can apply the above analysis! This relation that Z12 = Z21 is true as long as the medium
traversed by the fields is a reciprocal medium.
Before we conclude this section, it is to be mentioned that some researchers advocate the
use of circuit theory to describe electromagnetic theory. Such is the case in the transmission
line matrix (TLM) method [179], and the partial element equivalence circuit (PEEC) method
[180]. Circuit theory is so simple that many people fall in love with it!

30.4 Voltage Sources in Electromagnetics

Figure 30.6: Two ways to model voltage sources: (i) An impressed current source Ja driving
a very short antenna, and (ii) An impressed magnetic frill source (loop source) Ma driving a
very short antenna (courtesy of Kong, Electromagnetic Wave Theory [32]).

In the above discussions, we have used current sources in reciprocity theorem to derive certain
circuit concepts. Before we end this section, it is prudent to mention how voltage sources are
334 Electromagnetic Field Theory

modeled in electromagnetic theory. The use of the impressed currents so that circuit concepts
can be applied is shown in Figure 30.6. The antenna in (a) is driven by a current source.
But a magnetic current can be used as a voltage source in circuit theory as shown by Figure
30.6b. By using duality concept, an electric field has to curl around a magnetic current just
in Ampere’s law where magnetic field curls around an electric current. This electric field will
cause a voltage drop between the metal above and below the magnetic current loop making
it behave like a voltage source.8

30.5 Hind Sight

The proof of reciprocity theorem for Maxwell’s equations is very deeply related to the sym-
metry of the operator involved. We can see this from linear algebra. Given a matrix equation
driven by two different sources b1 and b2 with solutions x1 and x2 , they can be written
succinctly as

A · x1 = b1 (30.5.1)
A · x2 = b2 (30.5.2)

We can left dot multiply the first equation with x2 and do the same with the second equation
with x1 to arrive at

xt2 · A · x1 = xt2 · b1 (30.5.3)

xt1 · A · x2 = xt1 · b2 (30.5.4)

If A is symmetric, the left-hand side of both equations are equal to each other.9 Therefore,
we can equate their right-hand side to arrive at

xt2 · b1 = xt1 · b2 (30.5.5)

The above is analogous to the statement of the reciprocity theorem which is

hE2 , J1 i = hE1 , J2 i (30.5.6)

where the reaction inner product, as mentioned before, is hEi , Jj i = V Ei (r)·Jj (r). The inner
product in linear algebra is that of dot product in matrix theory, but the inner product for
reciprocity theorem is that for infinite dimensional spaces.10 So if the operators in Maxwell’s
equations are symmetrical, then reciprocity theorem applies.

8 More can be found in Jordain and Balmain, Electromagnetic Waves and Radiation Systems [54].
9 This can be easily proven by taking the transpose of a scalar, and taking the transpose of the product of
matrices.
10 Such spaces are called Hilbert space.
Reciprocity Theorem 335

30.6 Transmit and Receive Patterns of an Antennna

Figure 30.7: The schematic diagram for studying the transmit and receive properties of
antennas. The two antennas are assumed to be identical, and each switches between transmit
and receive modes in this study.

Reciprocity also implies that the transmit and receive properties of an antenna is similar to
each other. The transmit property of an antenna is governed by the gain function, while its
receive property is governed by the effective area or aperture. The effective aperture is also
a function of angle of the incident wave with respect to to the antenna. The gain function of
an antenna is related to its effective aperture by a constant as we shall argue.
Consider an antenna in the transmit mode. Then the time-average radiation power density
that it will yield around the antenna, in accordance to (25.3.5), is11
Pt
hSrad i = G(θ, φ) (30.6.1)
4πr2
where Pt is the total power radiated by the transmit antenna,

and G(θ, φ) is its directive gain
pattern or function. It is to be noted that in the above 4π dΩG(θ, φ) = 4π. The above is
valid when the antenna is lossless.

Effective Gain versus Directive Gain

At this juncture, it is important to introduce the concept of effective gain versus directive
gain. The effective gain, also called the power gain, is

Ge (θ, φ) = fe G(θ, φ) (30.6.2)

where fe is the efficiency of the antenna, a factor less than 1. It accounts for the fact that
not all power pumped into the antenna is delivered as radiated power. For instance, power
11 The author is indebted to inspiration from E. Kudeki of UIUC for this part of the lecture notes [139].
336 Electromagnetic Field Theory

can be lost in the circuits and mismatch of the antenna. Therefore, the correct formula the
radiated power density is

Pt Pt
hSrad i = 2
Ge (θ, φ) = fe G(θ, φ) (30.6.3)
4πr 4πr2

This radiated power resembles that of a plane wave when one is far away from the trans-
mitter. Thus if a receive antenna is placed in the far-field of the transmit antenna, it will
see this power density as coming from a plane wave. Thus the receive antenna will see an
incident power density as

Pt
hSinc i = hSrad i = Ge (θ, φ) (30.6.4)
4πr2

Effective Aperture

The effective area or the aperture of a receive antenna is used to characterize its receive
property. The power received by such an antenna is then, by using the concept of effective
aperture expounded in (26.1.23)

Pr = hSinc iAe (θ0 , φ0 ) (30.6.5)

where (θ0 , φ0 ) are the angles at which the plane wave is incident upon the receiving antenna
(see Figure 30.7). Combining the above formulas (30.6.4) and (30.6.5), we have

Pt
Pr = Ge (θ, φ)Ae (θ0 , φ0 ) (30.6.6)
4πr2

Now assuming that the transmit and receive antennas are identical. Next, we swap their
roles of transmit and receive, and also the circuitries involved in driving the transmit and
receive antennas. Then,

Pt
Pr = Ge (θ0 , φ0 )Ae (θ, φ) (30.6.7)
4πr2

We also assume that the receive antenna, that now acts as the transmit antenna is transmitting
in the (θ0 , φ0 ) direction. Moreover, the transmit antenna, that now acts as the receive antenna
is receiving in the (θ, φ) direction (see Figure 30.7).
By reciprocity, these two powers are the same, because Z12 = Z21 . Furthermore, since
these two antennas are identical, Z11 = Z22 . So by swapping the transmit and receive
electronics, the power transmitted and received will not change. A simple transmit-receive
circuit diagram is shown in Figure 30.8
Reciprocity Theorem 337

Figure 30.8: The schematic of the circuit for a transmit-receive antenna pair. Because the
mutual interaction between the two antennas can be described by the impedance matrix Z,
circuit theory can be applied to model their mutual interaction as indicated in (a). Moreover,
at the receive end, one can even further simplify the circuit by using a Thevenin equivalence
(b), or a Norton equivalence (c).

Consequently, we conclude that

Ge (θ, φ)Ae (θ0 , φ0 ) = Ge (θ0 , φ0 )Ae (θ, φ) (30.6.8)

The above implies that

Ae (θ, φ) Ae (θ0 , φ0 )
= = constant (30.6.9)
Ge (θ, φ) Ge (θ0 , φ0 )

The above Gedanken experiment is carried out for arbitrary angles. Therefore, the constant
is independent of angles. Moreover, this constant is independent of the size, shape, and
efficiency of the antenna, as we have not stipulated their shapes, sizes, and efficiency in the
above discussion.
To find this constant in (30.6.9), one can repeat the above for a Hertzian dipole, wherein
the mathematics of calculating Pr and Pt is a lot simpler. This constant is found to be
338 Electromagnetic Field Theory

λ2 /(4π).12 Therefore, an interesting relationship between the effective aperture (or area) and
the directive gain function is that

λ2
Ae (θ, φ) = Ge (θ, φ) (30.6.10)
4π
One amusing point about the above formula is that the effective aperture, say of a Hertzian
dipole, becomes very large when the frequency is low, or the wavelength is very long. Of
course, this cannot be physically true, and I will let you meditate on this paradox and muse
over this point.

12 See Kong [32][p. 700]. The derivation is for 100% efficient antenna. A thermal equilibrium argument is

used in [139] and Wikipedia [140] as well.

Lecture 31

Equivalence Theorems, Huygens’

Principle

Electromagnetic equivalence theorems are useful for simplifying solutions to many problems.
Also, they offer physical insight into the behaviour of electromagnetic fields of a Maxwellian
system. They are closely related to the uniqueness theorem and Huygens’ principle. One
application is their use in studying the radiation from an aperture antenna or from the output
of a lasing cavity. These theorems are discussed in many textbooks [32, 50, 54, 65, 181]. Some
authors also call it Love’s equivalence principles [182] and credit has been given to Schelkunoff
as well [172].
You may have heard of another equivalence theorem in special relativity. It was postulated
by Einstein to explain why light ray should bend around a star. The equivalence theorem
in special relativity is very different from those in electromagnetics. One thing they have in
common is that they are all derived by using Gedanken experiment (thought experiment)
, involving no math. But in this lecture, we will show that electromagnetic equivalence
theorems are also derivable using mathematics, albeit with more work.

31.1 Equivalence Theorems or Equivalence Principles

In this lecture, we will consider three equivalence theorems: (1) The inside out case. (2) The
outside in case. (3) The general case. As mentioned above, we will derive these theorems
using thought experiments or Gedanken experiments. As shall be shown later, they can also
be derived mathematically using Green’s theorem.

339
340 Electromagnetic Field Theory

31.1.1 Inside-Out Case

Figure 31.1: The inside-out problem where the two cases in (a) and (b) are equivalent. In (b),
equivalence currents are impressed on the surface S so as to produce the same fields outside
in Vo in both cases, cases (a) and (b).

In this case, we let J and M be the time-harmonic radiating sources inside a surface S
radiating into a region V = Vo ∪ Vi . They produce E and H everywhere. We can construct
an equivalence problem by first constructing an imaginary surface S. In this equivalence
problem, the fields outside S in Vo are the same in both (a) and (b). But in (b), the fields
inside S in Vi are zero.
Apparently, in case (b) in Figure 31.1, the tangential components of the fields are discon-
tinuous at S. This is not possible for a Maxwellian fields unless surface currents are impressed
on the surface S. We have learned from electromagnetic boundary conditions that electro-
magnetic fields are discontinuous across a current sheet. Then we ask ourselves what surface
currents are needed on surface S so that the boundary conditions for field discontinuities are
satisfied on S. Clearly, surface currents needed for these field discontinuities. By virtue of
the boundary conditions and the jump conditions in electromagnetics, these surface currents
to be impressed on S are

Js = n̂ × H, Ms = E × n̂ (31.1.1)

We have learnt from Section 4.3.3 that an electric current sheet in Ampere’s law produced a
jump discontinuity in the magnetic field. By the same token, when fictitious magnetic current
is added to Faraday’s law in Section 5.3 for mathematical symmetry, a magnetic current sheet
induces a jump discontinuity in the electric field. Because of the opposite polarity of the
magnetic current M in Faraday’s law compared to the electric current in Ampere’s law as is
ˆ
shown in Section 5.3, this magnetic current sheet is proportional to E × n̂ instead of ×H.
Consequetly, we can convince ourselves that n̂ × H and E × n̂ just outside S in both
cases are the same. Furthermore, we are persuaded that the above is a bona fide solution to
Maxwell’s equations.
Equivalence Theorems, Huygens’ Principle 341

ˆ The boundary conditions on the surface S satisfy the boundary conditions required of
Maxwell’s equations.

ˆ By the uniqueness theorem, only the equality of one of them E × n̂, or n̂ × H ons S,
will guarantee that E and H outside S are the same in both cases (a) and (b).

The fact that these equivalence currents generate zero fields inside S is known as the
extinction theorem. This equivalence theorem can also be proved mathematically, as shall
be shown.

31.1.2 Outside-in Case

Figure 31.2: The outside-in problem where equivalence currents are impressed on the surface
S to produce the same fields inside in both cases.

Similar to before, we find an equivalence problem (b) where the fields inside S in Vi is the
same as in (a), but the fields outside S in Vo in the equivalence problem is zero. The fields
are discontinuous across the surface S, and hence, impressed surface currents are needed to
account for these discontinuities.
Then by the uniqueness theorem,1 the fields Ei , Hi inside V in both cases are the same.
Again, by the extinction theorem, the fields produced by Ei × n̂ and n̂ × Hi are zero outside
S.

31.1.3 General Case

From these two cases, we can create a rich variety of equivalence problems. By linear su-
perposition of the inside-out problem, and the outside-in problem, then a third equivalence
problem is shown in Figure 31.3:

1 We can add infinitesimal loss to ensure that uniqueness theorem is satisfied in this enclosed volume Vi .
342 Electromagnetic Field Theory

Figure 31.3: The general case where the fields are non-zero both inside and outside the surface
S. Equivalence currents are needed on the surface S to support the jump discontinuities in
the fields.

31.2 Electric Current on a PEC

Using the equivalence problems in the previous section, we can derive other corollaries of
equivalence theorems. We shall show them next.
First, from reciprocity theorem, it is quite easy to prove that an impressed current on the
PEC cannot radiate. We can start with the inside-out equivalence problem. Since the fields
inside S is zero for the inside-out problem, using a Gedanken experiment, one can insert an
PEC object inside S without disturbing the fields E and H outside since the field is zero
inside S. As the PEC object grows to snugly fit the surface S, then the electric current
Js = n̂ × H does not radiate by reciprocity. Only one of the two currents is radiating, namely,
the magnetic current Ms = E × n̂ is radiating, and Js in Figure 31.4 can be removed. This
is commensurate with the uniqueness theorem that only the knowledge of E × n̂ is needed to
uniquely determine the fields outside S.

Figure 31.4: On a PEC surface, only one of the two currents is needed since an electric current
does not radiate on a PEC surface.
Equivalence Theorems, Huygens’ Principle 343

31.3 Magnetic Current on a PMC

Again, from reciprocity, an impressed magnetic current on a PMC cannot radiate. By the
same token, we can perform the Gedanken experiment as before by inserting a PMC object
inside S. It will not alter the fields outside S, as the fields inside S is zero. As the PMC
object grows to snugly fit the surface S, only the electric current Js = n̂ × H radiates, and
the magnetic current Ms = E × n̂ does not radiate and it can be removed. This is again
commensurate with the uniqueness theorem that only the knowledge of the n̂ × H is needed
to uniquely determine the fields outside S.

Figure 31.5: Similarly, on a PMC surface, only an electric current is needed to produce the
field outside the surface S.

31.4 Huygens’ Principle and Green’s Theorem

Huygens’ principle shows how a wave field on a surface determines the wave field outside the
surface S. This concept was expressed by Huygens heuristically in the 1600s [183]. But the
mathematical expression of this idea was due to George Green 2 in the 1800s. This concept
can be expressed mathematically for both scalar and vector waves. The derivation for the
vector wave case is homomorphic to the scalar wave case. But the algebra in the scalar wave
case is much simpler. Therefore, we shall discuss the scalar wave case first, followed by the
electromagnetic vector wave case.

2 George Green (1793-1841) was self educated and the son of a miller in Nottingham, England [184].
344 Electromagnetic Field Theory

31.4.1 Scalar Waves Case

Figure 31.6: The geometry for deriving Huygens’ principle for scalar wave equation.

For a ψ(r) that satisfies the scalar wave equation

(∇2 + k 2 ) ψ(r) = 0, (31.4.1)

the corresponding scalar Green’s function g(r, r0 ) satisfies

(∇2 + k 2 ) g(r, r0 ) = −δ(r − r0 ). (31.4.2)

Next, we multiply (31.4.1) by g(r, r0 ) and (31.4.2) by ψ(r). And then, we subtract the
resultant equations and integrating over a volume V as shown in Figure 31.6. There are two
cases to consider: when r0 is in V , or when r0 is outside V . Thus, we have
 
if r0 ∈ V , ψ(r0 )
= dr [g(r, r0 )∇2 ψ(r) − ψ(r)∇2 g(r, r0 )], (31.4.3)
if r0 6∈ V , 0
V

The left-hand side evaluates to different values depending on where r0 is due to the sifting
property of the delta function δ(r − r0 ). Since g∇2 ψ − ψ∇2 g = ∇ · (g∇ψ − ψ∇g), the left-hand
side of (31.4.3) can be rewritten using Gauss’ divergence theorem, giving3
 
if r0 ∈ V , ψ(r0 )
= dS n̂ · [g(r, r0 )∇ψ(r) − ψ(r)∇g(r, r0 )], (31.4.4)
if r0 6∈ V , 0
S

where S is the surface bounding V . The above is the Green’s theorem, or the mathematical
expression that once ψ(r) and n̂ · ∇ψ(r) are known on S, then ψ(r0 ) away from S could be
3 The equivalence of the volume integral in (31.4.3) to the surface integral in (31.4.4) is also known as

Green’s theorem [89].

Equivalence Theorems, Huygens’ Principle 345

found. This is similar to the expression of equivalence principle where n̂ · ∇ψ(r) and ψ(r) are
equivalence sources on the surface S. The first term on the right-hand side radiates via the
Green’s function g(r, r0 ). Since this is a monopole field, this source is also called a monolayer
or single layer source. The second term radiates, on the other hand, via the normal derivative
of the Green’s function, namely n̂·∇g(r, r0 ). Since the derivative of a Green’s function yields a
dipole field, the second term corresponds to sources that radiate like dipoles pointing normally
to the surface S. These sources ara also called double layer (or dipole layer) sources. These
terminologies are prevalent in acoustics. The above mathematical expression also embodies
the extinction theorem that says if r0 is outside V , the left-hand side evaluates to zero.

Figure 31.7: The geometry for deriving Huygens’ principle for scalar wave. The radiation
from the source can be replaced by equivalence sources on the surface S, and the field outside
S can be calculated using (31.4.4). Also, the field is zero inside S from (31.4.4). This is the
extinction theorem.

If the volume V is bounded by S and Sinf as shown in Figure 31.7, then the surface
integral in (31.4.4) should include an integral over Sinf . But when Sinf → ∞, all fields look
like plane wave, and ∇ → −r̂jk on Sinf . Furthermore, g(r − r0 ) ∼ O(1/r),4 when r → ∞, and
ψ(r) ∼ O(1/r), when r → ∞, if ψ(r) is due to a source of finite extent. Then, the integral
over Sinf in (31.4.4) vanishes, and (31.4.4) is valid for the case shown in Figure 31.7 as well
but with the surface integral over surface S only. Here, the field outside S at r0 is expressible
in terms of the field on S. This is similar to the inside-out equivalence principle we have
discussed previously Section 31.1.1, albeit this is for scalar wave case.
Notice that in deriving (31.4.4), g(r, r0 ) has only to satisfy (31.4.2) for both r and r0 in V
but no boundary condition has yet been imposed on g(r, r0 ). Therefore, if we further require
4 The symbol “O” means “of the order.”
346 Electromagnetic Field Theory

that g(r, r0 ) = 0 for r ∈ S, then (31.4.4) becomes


ψ(r0 ) = − dS ψ(r) n̂ · ∇g(r, r0 ), r0 ∈ V. (31.4.5)
S

On the other hand, if require additionally that g(r, r0 ) satisfies (31.4.2) with the boundary
condition n̂ · ∇g(r, r0 ) = 0 for r ∈ S, then (31.4.4) becomes

0
ψ(r ) = dS g(r, r0 ) n̂ · ∇ψ(r), r0 ∈ V. (31.4.6)
S

Equations (31.4.4), (31.4.5), and (31.4.6) are various forms of Huygens’ principle, or equiv-
alence principle for scalar waves (acoustic waves) depending on the definition of g(r, r0 ). Equa-
tions (31.4.5) and (31.4.6) stipulate that only ψ(r) or n̂ · ∇ψ(r) need be known on the surface
S in order to determine ψ(r0 ). The above are analogous to the PEC and PMC equivalence
principle considered previously in Sections 31.2 and 31.3. (Note that in the above derivation,
k 2 could be a function of position as well.)

31.4.2 Electromagnetic Waves Case

Figure 31.8: Derivation of the Huygens’ principle for the electromagnetic case. One only
needs to know the surface fields on surface S in order to determine the field at r0 inside V .

The derivation of Huygens’ principle and Green’s theorem for the electromagnetic case is
more complicated than the scalar wave case. But fortunately, this problem is mathematically
homomorphic to the scalar wave case. In dealing with the requisite vector algebra, we have
to remember to cross the t’s and dot the i’s, to carry ourselves carefully through the laborious
and complicated vector algebra!
Equivalence Theorems, Huygens’ Principle 347

In a source-free homogeneous region, an electromagnetic wave satisfies the vector wave

equation

∇ × ∇ × E(r) − k 2 E(r) = 0. (31.4.7)

The analogue of the scalar Green’s function for the scalar wave equation is the dyadic Green’s
function for the electromagnetic wave case [1, 32, 185, 186]. Moreover, the dyadic Green’s
function satisfies the equation5

∇ × ∇ × G(r, r0 ) − k 2 G(r, r0 ) = I δ(r − r0 ). (31.4.8)

It can be shown by direct back substitution that the dyadic Green’s function in free space
is [186]
 
∇∇
G(r, r ) = I + 2 g(r − r0 )
0
(31.4.9)
k

The above allows us to derive the vector Green’s theorem [1, 32, 185].
Then, after post-multiplying (31.4.7) by G(r, r0 ), pre-multiplying (31.4.8) by E(r), sub-
tracting the resultant equations and integrating the difference over volume V , considering
two cases as we did for the scalar wave case, we have
 
if r0 ∈ V , E(r0 )
= dV E(r) · ∇ × ∇ × G(r, r0 )

if r0 6∈ V , 0
V
−∇ × ∇ × E(r) · G(r, r0 ) .

(31.4.10)

Next, using the vector identity that6

−∇ · E(r) × ∇ × G(r, r0 ) + ∇ × E(r) × G(r, r0 )

 

= E(r) · ∇ × ∇ × G(r, r0 ) − ∇ × ∇ × E(r) · G(r, r0 ), (31.4.11)

then the integrand of (31.4.10) can be written as a divergence. With the help of Gauss’
divergence theorem, it can be written as

if r0 ∈ V , E(r0 )

= − dS n̂ · E(r) × ∇ × G(r, r0 ) + ∇ × E(r) × G(r, r0 )
 
if r0 6∈ V , 0
S

dS −E(r) × n̂ · ∇ × G(r, r0 ) + iωµ n̂ × H(r) · G(r, r0 ) .
 
=− (31.4.12)
S

5A dyad is an outer product between two vectors, and it behaves like a tensor, except that a tensor is more
general than a dyad. A purist will call the above a tensor Green’s function, as the above is not a dyad in its
strictest definition.
6 This identity can be established by using the identity ∇ · (A × B) = B · ∇ × A − A · ∇ × B. We will have

to let (31.4.11) act on an aribitrary constant vector to convert the dyad into a vector before applying this
identity. The equality of the volume integral in (31.4.10) to the surface integral in (31.4.12) is also known as
the vector Green’s theorem [32, 185]. Earlier form of this theorem was known as Franz formula [187].
348 Electromagnetic Field Theory

The above is just the vector analogue of (31.4.4). We have used the cyclic relation of
dot and cross products to rewrite the last expression. Since n̂ × E and n̂ × H are associated
with surface magnetic current Ms and surface electric current Js , respectively, the above
can be thought of having these equivalence surface currents radiating via the dyadic Green’s
function. Again, notice that (31.4.12) is derived via the use of (31.4.8), but no boundary
condition has yet been imposed on G(r, r0 ) on S even though we have given a closed form
solution for the free-space case. The above is similar to the outside-in equivalence theorem we
have derived in Section 31.1.2 using a Gedanken experiment. Now we have a mathematical
derivation of the same theorem.
Now, if we require the additional boundary condition that n̂ × G(r, r0 ) = 0 for r ∈ S. This
corresponds to a point source radiating in the presence of a PEC surface. Then (31.4.12)
becomes

E(r0 ) = − dS n̂ × E(r) · ∇ × G(r, r0 ), r0 ∈ V (31.4.13)
S

for it could be shown that n̂ × H · G = H · n̂ × G implying that the second term in (31.4.12)
is zero. On the other hand, if we require that n̂ × ∇ × G(r, r0 ) = 0 for r ∈ S, then (31.4.12)
becomes

E(r0 ) = −iωµ dS n̂ × H(r) · G(r, r0 ), r0 ∈ V (31.4.14)
S

Equations (31.4.13) and (31.4.14) state that E(r0 ) is determined if either n̂×E(r) or n̂×H(r) is
specified on S. This is in agreement with the uniqueness theorem. These are the mathematical
expressions of the PEC and PMC equivalence problems we have considered in the previous
sections, Sections 31.2 and 31.3.
The dyadic Green’s functions in (31.4.13) and (31.4.14) are for a closed cavity since bound-
ary conditions are imposed on S for them. But the dyadic Green’s function for an unbounded,
homogeneous medium, given in (31.4.10) can be written as
1
G(r, r0 ) = [∇ × ∇ × I g(r − r0 ) − I δ(r − r0 )], (31.4.15)
k2

∇ × G(r, r0 ) = ∇ × I g(r − r0 ). (31.4.16)

Then, (31.4.12), for r0 ∈ V and r0 6= r, becomes
 
1 0
E(r0 ) = −∇0 × dS g(r − r0 ) n̂×E(r) + ∇ ×∇0 × dS g(r − r0 ) n̂×H(r). (31.4.17)
iω
S S

The above can be applied to the geometry in Figure 31.7 where r0 is enclosed in S and Sinf .
However, the integral over Sinf vanishes by virtue of the radiation condition as for (31.4.4).
Then, (31.4.17) relates the field outside S at r0 in terms of only the equivalence surface
currents Ms = E × n̂ and Js = n̂ × H on S. This is similar to the inside-out problem in the
equivalence theorem (see Section 31.1.1). It is also related to the fact that if the radiation
condition is satisfied, then the field outside of the source region is uniquely satisfied. Hence,
this is also related to the uniqueness theorem.
Lecture 32

Shielding, Image Theory

The physics of electromagnetic shielding and electromagnetic image theory (also called image
theorem) go hand in hand. They work by the moving of charges around so as to cancel
the impinging fields. By understanding simple cases of shielding and image theory, we can
gain enough insight to solve some real-world problems. For instance, the art of shielding
is very important in the field of electromagnetic compatibility (EMC) and electromagnetic
interference (EMI). In the modern age where we have more electronic components working
side by side in a very compact environment, EMC/EMI become an increasingly challenging
issue. These problems have to be solved using heuristics with a high dosage of physical insight.

32.1 Shielding
We can understand shielding by understanding how electric charges move around in a conduc-
tive medium. They move around to shield out the electric field, or cancel the impinging field
inside the conductor. There are two cases to consider: the static case and the dynamic case.
The physical arguments needed to understand these two cases are quite different. Moreover,
since there are no magnetic charges around, the shielding of magnetic field is quite different
from the shielding of electric field, as shall be seen below.

32.1.1 A Note on Electrostatic Shielding

We begin with the simple case of electrostatic shielding. For electrostatic problems, a con-
ductive medium suffices to produce surface charges that shield out the electric field from the
conductive medium. If the electric field is not zero, then since J = σE, the electric current
inside the conductor will keep flowing. The current will produce charges on the surface of the
conductor to cancel the imping field, until inside the conductive medium E = 0. In this case,
electric current ceases to flow in the conductor.
In other words, when the field reaches the quiescent state, the charges redistribute them-
selves so as to shield out the electric field, and that the total internal electric field, E = 0 at

349
350 Electromagnetic Field Theory

equilibrium. And from Faraday’s law that tangential E field is continuous, then n̂ × E = 0 on
the conductor surface since n̂ × E = 0 inside the conductor. Figure 32.1 shows the static elec-
tric field, in the quiescent state, between two conductors (even though they are not PECs),
and the electric field has to be normal to the conductor surfaces. Moreover, since E = 0
inside the conductor, ∇Φ = 0 implying that the potential is a constant inside a conductor at
equilibrium.

Figure 32.1: The objects can just be conductors, and in the quiescent state (static state), the
tangential electric field will be zero on their surfaces. Also, E = 0 inside the conductor, or
∇Φ = 0, or Φ is a constant.

32.1.2 Relaxation Time

The time it takes for the charges to move around until they reach their quiescent distribution
such that E(t) = 0 is called the relaxation time. It is very much similar to the RC time
constant of an RC circuit consisting of a resistor in series with a capacitor (see Figure 32.2).
It can be proven that this relaxation time is related to ε/σ, but the proof is beyond the scope
of this course and it is worthwhile to note that this constant has the same unit as the RC
time constant of an RC circuit where a charged capacitor relaxes as exp(−t/τ ) where the
relaxation time τ = RC. Note that when σ → ∞, the relaxation time is zero. In other words,
in a perfect conductor or a superconductor, the charges reorient themselves instantaneously
if the external field is time-varying so that E(t) = 0 always.
Electrostatic shielding or low-frequency shielding is important at low frequencies. The
Faraday cage or Faraday shield is an important application of such a shielding (see Figure
32.3) [188].
However, if the conductor charges are induced by an external electric field that is time
varying, then the charges have to constantly redistribute/re-orient themselves to try to shield
out the incident time-varying electric field. Currents have to be constantly flowing around
the conductor. Then the electric field cannot be zero inside the conductors as shown in
Figure 32.4. In other words, an object with finite conductivity cannot shield out completely
a time-varying electric field. It can be shown that the depth of penetration of the field into
Shielding, Image Theory 351

Figure 32.2: The relaxation (or disappearance of accumulated charges) in a conductive object
is similar to the relaxation of charges from a charged capacitance in an RC circuit as shown.

p
the conductive object is about a skin depth δ = 2/(ωµσ). Or the lower the frequency ω or
the conductivity σ, the large the penetration depth.
For a perfect electric conductor (PEC) , E = 0 inside with the following argument: Because
if E 6= 0, then J = σE where σ → ∞. Let us assume an infinitesimally small time-varying
electric field in the PEC to begin with. It will induce an infinitely large electric current, and
hence an infinitely large time-varying magnetic field. An infinite time-varying magnetic field
in turn yields an infinite electric field that will drive an electric current, and these fields and
current will be infinitely large. This is an unstable chain of events if it is true. Moreover, it
will generate infinite energy in the system, which is not physical. Hence, the only possibility
for a stable solution is for the time-varying electromagnetic fields to be zero inside a PEC.
Thus, for the PEC, the charges can re-orient themselves instantaneously on the surface
when the inducing electric fields from outside are time varying. In other words, the relaxation
time ε/σ is zero. As a consequence, the time-varying electric field E is always zero inside
PEC, and therefore, n̂ × E = 0 on the surface of the PEC, even for time-varying fields.

32.2 Image Theory

The image theory here in electromagnetics is quite different from that in optics. As mentioned
before, when the frequency of the fields is high, the waves associated with the fields can be
described by rays. Therefore ray optics can be used to solve many high-frequency problems.
We can use ray optics to understand how an image is generated in a mirror. But the image
theory in electromagnetics is quite different from that in ray optics.
Image theory can be used to derived closed form solutions to boundary value problems
when the geometry is simple and has a lot of symmetry. These closed form solutions in turn
352 Electromagnetic Field Theory

Figure 32.3: Faraday cage demonstration on volunteers in the Palais de la Découverte in Paris
(courtesy of Wikipedia). When the cage is grounded, charges will surge from the ground to
the cage surface so as to make the field inside the cage zero, or the potential is constant.
Therefore, a grounded Faraday cage effectively shields the external fields from entering the
cage.

offer physical insight into the problems. This theory or method is also discussed in many
textbooks [1, 43, 54, 65, 79, 181, 189].

32.2.1 Electric Charges and Electric Dipoles

Image theory for a flat conductor surface or a half-space is quite easy to derive. To see
that, we can start with electro-static theory of putting a positive charge above a flat plane.
As mentioned before, for electrostatics, the plane or half-space does not have to be a perfect
conductor, but only a conductor (or a metal). From the previous Section 32.1.1, the tangential
static electric field on the surface of the conductor has to be zero.
By the principle of linear superposition, the tangential static electric field can be canceled
by putting an image charge of opposite sign at the mirror location of the original charge. This
is shown in Figure 32.5. Now we can mentally add the total field due to these two charges.
When the total static electric field due to the original charge and image charge is sketched, it
will look like that in Figure 32.6. It is seen that the static electric field satisfies the boundary
condition that n̂ × E = 0 at the conductor interface due to symmetry.
An electric dipole is made from a positive charge placed in close proximity to a negative
Shielding, Image Theory 353

Figure 32.4: If the source that induces the charges on the conductor is time varying, the
current in the conductor is always nonzero so that the charges can move around to respond
to the external time-varying charges. The two figures above show the orientation of the
charges for two snap-shot in time. In other words, p a time-varying field can penetrate the
conductor to approximately within a skin-depth δ = 2/(ωµσ).

charge. Using that an electric charge reflects to an electric charge of opposite polarity above
a conductor, one can easily see that a static horizontal electric dipole reflects to a static
horizontal electric dipole of opposite polarity. By the same token, a static vertical electric
dipole reflects to static vertical electric dipole of the same polarity as shown in Figure 32.7.
If this electric dipole is a Hertzian dipole whose field is time-varying, then one needs a
PEC half-space to shield out the electric field. Also, the image charges will follow the original
dipole charges instantaneously. Then the image theory for static electric dipoles over a half-
space still holds true if the dipoles now become Hertzian dipoles, but the surface will have to
be a PEC surface so that the fields can be shielded out instantaneously.

32.2.2 Magnetic Charges and Magnetic Dipoles

A static magnetic field can penetrate a conductive medium. This is apparent from our
experience when we play with a bar magnet over a copper sheet: the magnetic field from the
magnet can still be experienced by iron filings put on the other side of the copper sheet.
However, this is not the case for a time-varying magnetic field. Inside a conductive
medium, a time-varying magnetic field will produce a time-varying electric field, which in
turn produces the conduction current via J = σE. This is termed eddy current, which by
Lenz’s law, repels the magnetic field from the conductive medium.1
Now, consider a static magnetic field penetrating into a perfect electric conductor, an
minute amount of time variation will produce an electric field, which in turn produces an
infinitely large eddy current. So the stable state for a static magnetic field inside a PEC is
for it to be expelled from the perfect electric conductor. This in fact is what we observe when
a magnetic field is brought near a superconductor. Therefore, for the static magnetic field,
where B = 0 inside the PEC, then n̂ · B = 0 on the PEC surface (see Figure 32.8).
1 The repulsive force occurs by virtue of energy conservation. Since “work done” is needed to set the eddy

current in motion in the conductor, or to impart kinetic energy to the electrons forming the eddy current,
a repulsive force is felt in Lenz’s law so that work is done in pushing the magnetic field into the conductive
medium.
354 Electromagnetic Field Theory

Figure 32.5: The use of image theory to solve the BVP of a point charge on top of a conductor.
The boundary condition is that n̂ × E = 0 on the conductor surface. By placing a negative
charge with respect to the original charge, by the principle of linear superposition, both of
them produce a total field with no tangential component at the interface.

Now, assuming a magnetic monopole exists, it will reflect to itself on a PEC surface so
that n̂ · B = 0 as shown in Figure 32.8. Therefore, a magnetic charge reflects to a charge of
similar polarity on the PEC surface.
By extrapolating this to magnetic dipoles, they will reflect themselves to the magnetic
dipoles as shown in Figure 32.9. A horizontal magnetic dipole reflects to a horizontal magnetic
dipole of the same polarity, and a vertical magnetic dipole reflects to a vertical magnetic dipole
of opposite polarity. Hence, a dipolar bar magnet can be levitated by a superconductor when
this magnet is placed closed to it. This is also known as the Meissner effect [190], which is
shown in Figure 32.10.
A time-varying magnetic dipole can be made from a electric current loop. Over a PEC, a
time-varying magnetic dipole will reflect the same way as a static magnetic dipole as shown
in Figure 32.9.

32.2.3 Perfect Magnetic Conductor (PMC) Surfaces

Magnetic conductor does not come naturally in this world since there are no free-moving
magnetic charges around. Magnetic monopoles are yet to be discovered. On a PMC surface,
by duality, n̂ × H = 0. At low frequency, it can be mimicked by a high µ material. One can
see that for magnetostatics, at the interface of a high µ material and air, the magnetic flux
is approximately normal to the surface, resembling a PMC surface.
High µ materials are hard to find at higher frequencies. Since n̂ × H = 0 on such a
surface, no electric current can flow on such a surface. Hence, a PMC can be mimicked by
a surface where no surface electric current can flow. This has been achieved in microwave
engineering with a mushroom surface as shown in Figure 32.11 [192]. The mushroom structure
Shielding, Image Theory 355

Figure 32.6: By image theory, the total electric of the original problem and the equivalent
problem when we add the total electric field due to the original charge and the image charge.

consisting of a wire and an end-cap, can be thought of as forming an LC tank circuit. Close
to the resonance frequency of this tank circuit, the surface of mushroom structures essentially
becomes open circuits with no or little current flowing on the surface, or Js ∼ = 0. In other
words, n̂ × H ∼= 0. This resembles a PMC, because with no surface electric current on this
surface, the tangential magnetic field is small, the hallmark of a good magnetic conductor,
by the duality principle.
Mathematically, a surface that is dual to the PEC surface is the perfect magnetic conductor
(PMC) surface. The magnetic dipole is also dual to the electric dipole. Thus, over a PMC
surface, these electric and magnetic dipoles will reflect differently as shown in Figure 32.12.
One can go through Gedanken experiments and verify that the reflection rules are as shown
in the figure.

32.2.4 Multiple Images

For the geometry shown in Figure 32.13, one can start with electrostatic theory, and convince
oneself that n̂ × E = 0 on the metal surface with the placement of charges as shown. For
conducting media, the charges will relax to the quiescent distribution after the relaxation
time. For PEC surfaces, one can extend these cases to time-varying dipoles because the
charges in the PEC medium can re-orient instantaneously (i.e. with zero relaxation time) to
shield out or expel the E and H fields. Again, one can repeat the above exercise for magnetic
charges, magnetic dipoles, and PMC surfaces.
356 Electromagnetic Field Theory

Figure 32.7: By image theory, on a conductor surface, a horizontal static dipole reflects to
one of opposite polarity, while a static vertical dipole reflects to one of the same polarity. If
the dipoles are time-varying, then a PEC will have a same reflection rule.

Figure 32.8: On a PEC surface, the requisite boundary condition is n̂ · B = 0. Hence, a

magnetic monopole on top of a PEC surface will have magnetic field distributed as shown.
By image theory, such a field distribution can be obtained by adding a magnetic monopole
of the same polarity at its image point.

32.2.5 Some Special Cases—Spheres, Cylinders, and Dielectric In-

terfaces
One curious case is for a static charge placed near a conductive sphere (or cylinder) as shown in
Figure 32.14.2 A charge of +Q reflects to a charge of −QI inside the sphere. For electrostatics,
the sphere (or cylinder) need only be a conductor. However, this cannot be generalized to
electrodynamics or a time-varying problem, because of the retardation effect: A time-varying
dipole or charge will be felt at different points asymmetrically on the surface of the sphere
from the original and image charges. Exact cancelation of the tangential electric field on the
surface of the sphere or cylinder cannot occur for time-varying field.
When a static charge is placed over a dielectric interface, image theory can be used to
find the closed form solution. This solution can be derived using Fourier transform technique
which we shall learn later [35]. It can also be extended to multiple interfaces. But image
theory cannot be used for the electrodynamic case due to the different speed of light in
different media, giving rise to different retardation effects.

2 This is worked out in p. 48 and p. 49, Ramo et al [31].

Shielding, Image Theory 357

Figure 32.9: Using the rule of how magnetic monopole reflects itself on a PEC surface, the
reflection rules for magnetic dipoles can be ascertained. Magnetic dipoles are often denoted
by double arrows.

Figure 32.10: On a PEC (superconducting) surface, a vertical magnetic dipole (formed by a

small permanent bar magnet here) reflects to one of opposite polarity. Hence, the magnetic
dipoles repel each other displaying the Meissner effect. The magnet, because of the repulsive
force from its image, levitates above the superconductor (courtesy of Wikipedia [191]).
358 Electromagnetic Field Theory

Figure 32.11: A mushroom structure operates like an LC tank circuit. At the right res-
onant frequency, the surface resembles an open-circuit surface where no current can flow.
Hence, tangential magnetic field is zero resembling perfect magnetic conductor (courtesy of
Sievenpiper [192]).

Figure 32.12: Reflection rules for electric and magnetic dipoles over a PMC surface. Magnetic
dipoles are denoted by double arrows.
Shielding, Image Theory 359

Figure 32.13: Image theory for multiple images [31].

Figure 32.14: Image theory for a point charge near a cylinder or a sphere can be found in
closed form [31].
360 Electromagnetic Field Theory

Figure 32.15: A static charge over a dielectric interface can be found in closed form using
Fourier transform technique. The solution is beyond the scope of this course.
Lecture 33

High Frequency Solutions,

Gaussian Beams

When the frequency is very high, the wavelength of electromagnetic wave becomes very short.
In this limit, the solutions to Maxwell’s equations can be found approximately. These solutions
offer a very different physical picture of electromagnetic waves, and they are often used in
optics where the wavelength is short. So it was no surprise that for a while, optical fields
were thought to satisfy a very different equations from those of electricity and magnetism.
Thus it came as a surprise that when it was later revealed that in fact, optical fields satisfy
the same Maxwell’s equations as the fields from electricity and magnetism!
In this lecture, we shall seek approximate solutions to Maxwell’s equations or the wave
equations when the frequency is high or the wavelength is short. High frequency approximate
solutions are important in many real-world applications. This is possible when the wavelength
is much smaller than the size of the structure. This can occur even in the microwave regime
where the wavelength is not that small, but much smaller than the size of the structure.
This is the case when microwave interacts with reflector antennas for instance. It is also the
transition from waves regime to the optics regime in the solutions of Maxwell’s equations.
Often times, the term “quasi-optical” is used to describe the solutions in this regime.
In the high frequency regime, or when we are far away from a source much larger than
the Rayleigh distance (see Section 27.2.1), the field emanating from a source resembles a
spherical wave. Moreover, when the wavelength is much smaller than the radius of curvature
of the wavefront, the spherical wave can be approximated by a local plane wave. Thus we can
imagine rays to be emanating from a finite source forming the spherical wave. The spherical
wave will ultimately be approximated by plane waves locally at the observation point. This
will simplify the solutions in many instances. For instance, ray tracing can be used to track
how these rays can propagate, bounce, or “richohet” in a complex environment. In fact, it is
now done in a movie industry to give “realism” to simulate the nuances of how light ray will
bounce around in a room, and reflecting off objects.

361
362 Electromagnetic Field Theory

Figure 33.1: Ray-tracing technique can be used in the movie industry to produce realism in
synthetic images (courtesy of Wikipedia).

33.1 Tangent Plane Approximations

We have learnt that reflection and transmission of waves at a flat surface can be solved in
closed form. The important point here is the physics of phase matching. Due to phase
matching, we have the law of reflection, transmission and Snell’s law [58].1
When a surface is not flat anymore, there is no closed form solution. But when a surface is
curved, an approximate solution can be found. This is obtained by using a local tangent-plane
approximation when the radius of curvature is much larger than the wavelength. Hence, this
is a good approximation when the frequency is high or the wavelength is short. It is similar in
spirit that we can approximate a spherical wave by a local plane wave at the spherical wave
front when the wavelength is short compared to the radius of curvature of the wavefront.
When the wavelength is short, phase matching happens locally, and the law of reflection,
transmission, and Snell’s law are satisfied approximately as shown in Figure 33.2. The tangent
plane approximation is the basis for the geometrical optics (GO) approximation [32, 194]. In
GO, light waves are replaced by light rays. As mentioned before, a light ray is a part of a
spherical wave where locally, it can be approximated by a plane wave. The reflection and
transmission of these rays at an interface is then estimated using the local tangent plane
approximation and local Fresnel reflection and transmission coefficients. This is also the
basis for lens or ray optics from which lens technology is derived (see Figure 33.3). It is also
the basis for ray tracing for high-frequency solutions [195, 196].2
Many real world problems do not have closed-form solutions, and have to be treated
with approximate methods. In addition to geometrical approximations mentioned above,
asymptotic methods are also used to find approximate solutions. Asymptotic methods imply
finding a solution when there is a large parameter in the problem. In this case, it is usually
the frequency. Such high-frequency approximate methods are discussed in [197–201].

1 This law is also known in the Islamic world in 984 [193].

2 Please note that the tangent plane approximation is invalid near a sharp corner or an edge. The solution
has to be augmented by additional diffracted wave coming from the edge or the corner.
High Frequency Solutions, Gaussian Beams 363

Figure 33.2: In the tangent plane approximation, the surface where reflection and refraction
occur is assumed to be locally flat. Thus, phase-matching is approximately satisfied, and
hence, the law of reflection, transmittion, and Snell’s law are satisfied locally.

Figure 33.3: Tangent plane approximations can also be made at dielectric interfaces so that
Fresnel reflection and transmission coeffients can be used to ascertain the interaction of light
rays with a lens. Also, one can use ray tracing to understand the physics of an optical lens
(courtesy of Wikipedia).

33.2 Fermat’s Principle

Fermat’s principle (1600s) [58,202] says that a light ray follows the path that takes the shortest
time between two points.3 Since time delay is related to the phase shift, and that a light ray
can be locally approximated by a plane wave, this can be stated that a plane wave follows the
path that has a minimal phase shift. This principle can be used to derive law of reflection,
transmission, and refraction for light rays. It can be used as the guiding principle for ray
tracing as well.

3 This eventually give rise to the principle of least action.

364 Electromagnetic Field Theory

Figure 33.4: In Fermat’s principle, a light ray, when propagating from point A to point C,
takes the path of least delay in time in the time domain, and delay in phase in the frequency
domain.

Given two points A and C in two different half spaces as shown in Figure 33.4. Then the
phase delay between the two points, per Figure 33.4, can be written as4

P = ki · ri + kt · rt (33.2.1)

In the above, ki is parallel to ri , so is kt is parallel to rt . As this is the shortest path with

minimum phase shift or time delay, according to Fermat’s principle, another other path will
be longer. In other words, if B were to move slightly to another point, a longer path with
more phase shift or time delay will ensue, or that B is the stationary point of the path length
or phase shift. Specializing (33.2.1) to a 2D picture, then the phase shift as a function of xi
is stationary. In this Figure 33.4, we have xi + xt = const. Therefore, taking the derivative
of (33.2.1) or the phase change with respect to xi , assuming that ki and kt do not change as
B is moved slightly,5

∂P
= 0 = kix − ktx (33.2.2)
∂xi

The above yields the law of refraction that kix = ktx , which is just Snell’s law; it can also
be obtained by phase matching as have been shown earlier. This law was also known in the
Islamic world to Ibn Sahl in 984 [193].

4 In this course, for wavenumber, we use k and β interchangeably, where k is prevalent in optics and β used

in microwaves.
5 One can show that as the separations between A, B, and C are large, and the change in x is ∆x , the
i i
changes in ki and kt are small. The changeqin phase shift mainly
q comes from the change in xi . Alternatively,
we can write P = ki ri + kt rt , and let ri = x2i + zi2 , and rt = x2t + zt2 , and take the derivative with respect
to to xi , one would also get the same answer.
High Frequency Solutions, Gaussian Beams 365

33.2.1 Generalized Snell’s Law

Figure 33.5: A phase screen which is position dependent can be made using nano-fabrication
and design with commercial software for solving Maxwell’s equations. In such a case, one
can derive a generalized Snell’s law to describe the diffraction of a wave by such a surface
(courtesy of Capasso’s group [203]).

Metasurfaces are prevalent these days due to our ability for nano-fabrication and numerical
simulation. One of them is shown in Figure 33.5. Such a metasurface can be thought of as a
phase screen, providing additional phase shift for the light as it passes through it. Moreover,
the added phase shift can be controlled to be a function of position due to advent in nano-
fabrication technology and commercial software for numerical simulation.
To model this phase screen, we can add an additional function Φ(x, y) to (33.2.1), namely
that
P = ki · ri + kt · rt − Φ(xi , yi ) (33.2.3)
Now applying Fermat’s principle that there should be minimal phase delay, and taking the
derivative of the above with respect to xi , one gets
∂P ∂Φ(xi , yi )
= kix − ktx − =0 (33.2.4)
∂xi ∂xi
The above yields that the generalized Snell’s law [203] that
∂Φ(xi , yi )
kix − ktx = (33.2.5)
∂xi
It implies that the transmitted light can be directed to other angles due to the additional
phase screen. 6
6 Such research is being pursued by V. Shalaev and A. Boltasseva’s group at Purdue U [204].
366 Electromagnetic Field Theory

33.3 Gaussian Beam

We have seen previously that in a source-free medium
∇2 A + ω 2 µεA = 0 (33.3.1)

∇2 Φ + ω 2 µεΦ = 0 (33.3.2)
The above are four scalar equations; and the Lorenz gauge
∇ · A = −jωµεΦ (33.3.3)
connects A and Φ. We can examine the solution of A such that
A(r) = A0 (r)e−jβz (33.3.4)

where A0 (r) is a slowly varying function while e−jβz is rapidly varying in the z direction.
√
(Here, β = ω µ is the wavenumber.) This is primarily a quasi-plane wave propagating
predominantly in the z-direction. We know this to be the case in the far field of a source,
but let us assume that this form persists less than the far field, namely, in the Fresnel zone
as well. Taking the x component of (33.3.4), we have7
Ax (r) = Ψ(r)e−jβz (33.3.5)

where Ψ(r) = Ψ(x, y, z) is a slowly varying envelope function of x, y, and z, whereas e−jβz is
a rapidly varying function of z when β is large or the frequency is high.

33.3.1 Derivation of the Paraxial/Parabolic Wave Equation

Substituting (33.3.5) into (33.3.1), and taking the double z derivative first, we arrive at
 
∂2 
 2 
−jβz  ∂ ∂
Ψ(x, y, z) e = Ψ(x, y, z) − 2jβ Ψ(x, y, z) − β Ψ(x, y, z) e−jβz (33.3.6)
2
∂z 2 | {z } | {z } ∂z 2 ∂z
slow fast

Consequently, after substituting the above into the x component of (33.3.1), making use of
the definition of ∇2 , we obtain an equation for Ψ(r), the slowly varying envelope as
∂2 ∂2 ∂ ∂2
2
Ψ + 2 Ψ − 2jβ Ψ + 2 Ψ = 0 (33.3.7)
∂x ∂y ∂z ∂z
where the last term containing β 2 on the right-hand side of (33.3.6) cancels with the term
coming from ω 2 µεA of (33.3.1). So far, no approximation has been made in the above
equation. Since β is linearly proportional to frequency ω, when β → ∞ , or in the high
frequency limit,
   2 
2jβ ∂ Ψ   ∂ Ψ
   
2
(33.3.8)
 ∂z   ∂z 
7 Also, the wave becomes a transverse wave in the far field, and keeping the transverse component suffices.
High Frequency Solutions, Gaussian Beams 367

where we have assumed that Ψ is a slowly varying function of z such that βΨ  ∂/∂zΨ. In
other words, (33.3.7) can be approximated by

∂2Ψ ∂2Ψ ∂Ψ
2
+ 2
− 2jβ ≈0 (33.3.9)
∂x ∂y ∂z

The above is called the paraxial wave equation. It is also called the parabolic wave equation.8
It implies that the β vector of the wave is approximately parallel to the z axis, or βz ∼
= β to
be much greater than βx and βy , and hence, the name.

33.3.2 Finding a Closed Form Solution

A closed form solution to the paraxial wave equation can be obtained by a simple trick.9 It
is known that
0
e−jβ|r−r |
Ax (r) = (33.3.10)
4π|r − r0 |

is the exact solution to

∇2 Ax + β 2 Ax = 0 (33.3.11)

as long as r 6= r0 . One way to ensure that r 6= r0 always is to let r0 = −ẑjb, a complex number.
Then (33.3.10) is always a solution to (33.3.11) for all r, because |r − r0 | = 6 0 always as r0 is
complex. Then, we should next make a paraxial approximation to the solution (33.3.10) by
assuming that x2 + y 2  z 2 . By so doing, it follows that
p
|r − r0 | = x2 + y 2 + (z + jb)2
1/2
x2 + y 2

= (z + jb) 1 +
(z + jb)2
x2 + y 2
≈ (z + jb) + + ..., |z + jb| → ∞ (33.3.12)
2(z + jb)

where Taylor series has been used in approximating the last term. And then using the above
approximation in (33.3.10) yield

e−jβ(z+jb) −jβ 2(z+jb)

x2 +y 2
Ax (r) ≈ e ≈ e−jβz Ψ(r) (33.3.13)
4π(z + jb)

By comparing the above with (33.3.5), we can identify

jb −jβ 2(z+jb)
x2 +y 2
Ψ(x, y, z) ∼
= A0 e (33.3.14)
z + jb
8 The paraxial wave equation, the diffusion equation and the Schrodinger equation are all classified as

parabolic equations in mathematical parlance [35, 49, 205, 206].

9 Introduced by Georges A. Deschamps of UIUC [207].
368 Electromagnetic Field Theory

where A0 is used to absorbed the constant to simplify the expression. By separating the
exponential part into the real part and the imaginary part, viz.,

x2 + y 2 x2 + y 2
 
z b
= 2 2
−j 2 (33.3.15)
2(z + jb) 2 z +b z + b2

and writing the prefactor in terms of amplitude and phase, viz.,

jb 1 −1 z
=p ej tan ( b ) (33.3.16)
z + jb 2
1 + z /b2

we then have
2 2 2 2
A0 −1 z x +y
−jβ 2(z x +y
2 +b2 ) z −bβ 2(z 2 +b2 )
Ψ(x, y, z) ∼
=p ej tan (b)
e e (33.3.17)
1 + z 2 /b2

The above can be rewritten as

A0 x2 +y 2 x2 +y 2
Ψ(x, y, z) ∼
=p e−jβ 2R e− w2 ejψ (33.3.18)
1 + z 2 /b2

where A0 is a new constant introduced to absorb undesirable constants arising out of the
algebra, and

z2 z 2 + b2
 
2b z 
w2 = 1+ , R= , ψ = tan−1 (33.3.19)
β b2 z b

For a fixed z, the parameters w, R, and ψ are constants. It is seen that the beam is Gaussian
tapered in the x and y directions. q
Hence, w is the beam waist which varies with z, and it is
smallest when z = 0, or w = w0 = 2b β .
2 2
And the term exp(−jβ x 2R+y
) resembles the phase front of a spherical wave where R is its
radius of curvature. This can be appreciated by studying a spherical wave front e−jβR , and
make a paraxial wave approximation, namely, x2 + y 2  z 2 to get
 2 2 1/2
2
+y 2 +z 2 )1/2 −jβz 1+ x z+y
e−jβR = e−jβ(x =e 2

x2 +y 2 x2 +y 2
≈ e−jβz−jβ 2z ≈ e−jβz−jβ 2R (33.3.20)

In the last approximation, we assume that z ≈ R in the paraxial approximation.

The phase ψ defined in (33.3.19) changes linearly with z for small z, and saturates to a
constant for large z. Then the phase of the entire wave is due to the exp(−jβz) in (33.3.13).
A cross section of the electric field due to a Gaussian beam is shown in Figure 33.6.
High Frequency Solutions, Gaussian Beams 369

Figure 33.6: Electric field of a Gaussian beam in the x-z plane frozen in time. The wave
moves to the right as time increases; here, b/λ = 10/6 (courtesy of Haus, Electromagnetic
Noise and
p Quantum Optical Measurements [82]). The narrowest beam waist is given by
w0 /λ = b/(λπ).

33.3.3 Other solutions

In general, the paraxial wave equation in (33.3.9) is of the same form as the Schrödinger
equation which is of utmost importance in quantum theory. In recent years, the solution
of this equation has made use of spill-over knowledge and terms from quantum theory, such
as spin angular momentum (SAM) or orbital angular momentum (OAM). But it is a partial
differential equation which can be solved by the separation of variables just like the Helmholtz
wave equation. Therefore, in general, it has solutions of the form10
 1/2  
2 −N/2 1 2 2 2 β 2 2
Ψnm (x, y, z) = 2 e−(x +y )/w e−j 2R (x +y ) ej(m+n+1)Ψ
πn!m! w
 √   √ 
·Hn x 2/w Hm y 2/w (33.3.21)

where Hn (ξ) is a Hermite polynomial of order n. The solutions can also be expressed in terms
of Laguere polynomials , namely,
 1/2
2 1 β 2 2 2
Ψnm (x, y, z) = min(n, m)! e−j 2R ρ − e−ρ /w e+j(n+m+1)Ψ ejlφ
πn!m! w
√ !  2
2ρ 2ρ
(−1)min(n,m)
Ln−m
min(n,m) (33.3.22)
w w2

where Lkn (ξ) is the associated Laguerre polynomial.

These gaussian beams have rekindled recent excitement in the community because, in
addition to carrying spin angular momentum as in a plane wave, they can carry orbital
angular momentum due to the complex transverse field distribution of the beams.11 They
10 See F. Pampaloni and J. Enderlein [208].
11 See D.L. Andrew, Structured Light and Its Applications and articles therein [209].
370 Electromagnetic Field Theory

harbor potential for optical communications as well as optical tweezers to manipulate trapped
nano-particles. Figure 33.7 shows some examples of the cross section (xy plane) field plots
for some of these beams. They are richly endowed with patterns implying that they can be
used to encode information. These lights are also called structured lights [209].

Figure 33.7: Examples of structured light. It can be used in encoding more information in
optical communications (courtesy of L. Allen and M. Padgett’s chapter in J.L. Andrew’s book
on structured light [209]).
Lecture 34

Scattering of Electromagnetic
Field

The scattering of electromagnetic field is an important and fascinating topic. It especially

enriches our understanding of the interaction of light wave with matter. The wavelength of vis-
ible light is several hundred nanometers, with atoms and molecules ranging from nano-meters
onward, light-matter interaction is richly endowed with interesting physical phenomena! A
source radiates a field, and ultimately, in the far field of the source, the field resembles a
spherical wave which in turn resembles a plane wave. When a plane wave impinges on an
object or a scatterer, the energy carried by the plane wave is deflected to other directions
which is the process of scattering. In the optical regime, the scattered light allows us to see
objects, as well as admire all hues and colors that are observed of objects. In microwave, the
scatterers cause the loss of energy carried by a plane wave. A proper understanding of scat-
tering theory allows us to understand many physical phenomena around us. We will begin by
studying Rayleigh scattering, which is scattering by small objects compared to wavelength.
With Rayleigh scattering of a simple sphere, we can understand why the sky is blue and the
sunset is red!

34.1 Rayleigh Scattering

Rayleigh scattering is a solution to the scattering of light by small particles. These particles
are assumed to be much smaller than wavelength of light. The size of water molecule is
about 0.25 nm, while the wavelength of blue light is about 500 nm. Then a simple solution
can be found by using quasi-static analysis in the vicinity of the small particle. This simple
scattering solution can be used to explain a number of physical phenomena in nature (see
Figure 34.1). For instance, why the sky is blue, the sunset so magnificently beautiful, how
birds and insects can navigate themselves without the help of a compass. By the same token,
it can also be used to explain why the Vikings, as a seafaring people, could cross the Atlantic
Ocean over to Iceland without the help of a magnetic compass as the Chinese did in ancient
times.

371
372 Electromagnetic Field Theory

Figure 34.1: The magnificent beauty of nature can be partly explained by Rayleigh scattering
[210, 211].

When a ray of light impinges on an object, we model the incident light as a plane elec-
tromagnetic wave (see Figure 34.2). Without loss of generality, we can assume that the
electromagnetic wave is polarized in the z direction and propagating in the x direction. We
assume the particle to be a small spherical particle with permittivity εs and radius a. Essen-
tially, the particle sees a constant field as the plane wave impinges on it. In other words, the
particle feels an quasi-electrostatic field in the incident field. The incident field polarizes the
particle, making it radiate like a Hertzian dipole. This is the gist of a scattering process: an
incident field induces current (in this case, polarization current) on the scatterer. With the
induced current, the scatterer re-radiates (or scatters).

Figure 34.2: Geometry for studying the Rayleigh scattering problem.

Scattering of Electromagnetic Field 373

34.1.1 Scattering by a Small Spherical Particle

The incident field polarizes the particle making it look like an electric dipole. Since the
incident field is time harmonic, the small electric dipole will oscillate and radiate like a
Hertzian dipole in the far field. First, we will look at the solution in the vicinity of the
scatterer, namely, in the near field. Then we will motivate the form of the solution in the far
field of the scatterer. (Solving a boundary value problem by looking at the solutions in two
different physical regimes, and then matching the solutions together is known as asymptotic
matching.)
A Hertzian dipole can be approximated by a small current source so that

J(r) = ẑIlδ(r) (34.1.1)

Without loss of generality, we have assumed the Hertzian dipole to be at the origin. In the
above, we can let the time-harmonic current I = dq/dt = jωq, then

Il = jωql = jωp (34.1.2)

where the dipole moment p = ql. The vector potential A due to a Hertzian dipole, after
substituting (34.1.1), is

µ J(r0 ) −jβ|r−r0 |
A(r) = dr0 e
4π V |r − r0 |
µIl −jβr jωµql −jβr
= ẑ e = ẑ e (34.1.3)
4πr 4πr

where we have made use of the sifting property of the delta function in (34.1.1) when it is
substituted into the above integral.

Near Field

The above gives the vector potential A due to a Hertzian dipole. Since the dipole is infinites-
imally small, the above solution is both valid in the near field as well as the far field. Since
the dipole moment ql is induced by the incident field, we need to relate ql to the amplitude
of the incident electric field. To this end, we need to convert the above vector potential field
to the near electric field of a small dipole.
From prior knowledge, we know that the electric field is given by E = −jωA − ∇Φ. From
dimensional analysis, the scalar potential term dominates over the vector potential term in the
near field of the scatterer. Hence, we need to derive, for the corresponding scalar potential,
the approximate solution.
The scalar potential Φ(r) is obtained from the Lorenz gauge (see (23.2.23)) that ∇ · A =
−jωµεΦ. Therefore,

−1 Il ∂ 1 −jβr
Φ(r) = ∇·A=− e (34.1.4)
jωµε jωε4π ∂z r
374 Electromagnetic Field Theory

When we are close to the dipole, by assuming that βr  1, we can use a quasi-static approx-
imation about the potential.1 Then
∂ 1 −jβr ∂ 1 ∂r ∂ 1 z 1
e ≈ = =− 2 (34.1.5)
∂z r ∂z r ∂z ∂r r rr
or after using that z/r = cos θ,
ql
Φ(r) ≈ cos θ (34.1.6)
4πεr2
which is the static dipole potential because we are in the near field of the dipole. This dipole
induced in the small particle is formed in response to the incident field and its dipole potential
given by the previous expression. In other words, the incident field polarizes the small particle
into a small dipole.
The incident field can be approximated by a constant local static electric field,

Einc = ẑEi (34.1.7)

This is the field that will polarize the small particle. It can also be an electric field between
two parallel plates. The corresponding electrostatic potential for the incident field is then

Φinc = −zEi (34.1.8)

so that Einc ≈ −∇Φinc = ẑEi , as ω → 0. The scattered dipole potential from the spherical
particle in the vicinity of it is quasi-static and is given by

a3
Φsca = Es cos θ (34.1.9)
r2
which is the potential due to a static dipole. The electrostatic boundary value problem (BVP)
has been previously solved and2
εs − ε
Es = Ei (34.1.10)
εs + 2ε
Using (34.1.10) in (34.1.9), we get

εs − ε a3
Φsca = Ei 2 cos θ (34.1.11)
εs + 2ε r
On comparing with (34.1.6), one can see that the dipole moment induced by the incident field
is that
εs − ε 3
p = ql = 4πε a Ei = αEi (34.1.12)
εs + 2ε
where α is the polarizability of the small particle.
1 This is the same as ignoring retardation effect.
2 It was one of the homework problems. See also Section 8.3.6.
Scattering of Electromagnetic Field 375

Far Field
Now that we have learnt that a small particle is polarized by the incident field, which can be
treated as a constant E field in its vicinity. In other words, the incident field induces a small
dipole moment on the small particle. If the incident field is time-harmonic, the the small
dipole will be time-oscillating and it will radiate like a time-varying Hertzian dipole whose
far field is quite different from its near field (see Section 25.2). In the far field of the Hertzian
dipole, we can start with
1
E = −jωA − ∇Φ = −jωA − ∇∇ · A (34.1.13)
jωµε
In the above, Lorenz gauge (see (23.2.23)) has been used to relate Φ to the vector potential
A. But when we are in the far field, A behaves like a spherical wave which in turn behaves
like a local plane wave if one goes far enough. Therefore, ∇ → −jβ = −jβr̂. Using this
approximation in (34.1.13), we arrive at
 
ββ
E∼ = −jω A − · A = −jω(A − r̂r̂ · A) = −jω(θ̂Aθ + φ̂Aφ ) (34.1.14)
β2
where we have used r̂ = β/β. This is similar to the far field result we have derived in Section
26.1.2.

34.1.2 Scattering Cross Section

From (34.1.3), and making use of (34.1.2), we see that Aφ = 0 while
jωµql −jβr
Aθ = − e sin θ (34.1.15)
4πr
Consequently, using (34.1.12) for ql, we have in the far field that3
ω 2 µql −jβr εs − ε a3
 
Eθ ∼
= −jωAθ = − e sin θ = −ω 2 µε Ei e−jβr sin θ (34.1.16)
4πr εs + 2ε r
Using local plane-wave approximation that
r
ε 1
Hφ ∼ = Eθ = Eθ (34.1.17)
µ η
p
where η = µ/ε. The time-averaged Poynting vector is given by hSi = 1/2<e {E × H∗ }.
Therefore, the total scattered power is obtained by integrating the power density over a
spherical surface when r tends to infity. Thus, the total scattered power is
 π  2π  π  2π
1 1
Ps = r2 sin θdθ dφEθ Hφ∗ = r2 sin θdθ dφ|Eθ |2 (34.1.18)
2 0 0 2η 0 0
2  π
1 4  εs − ε  a6
 
2 2 3
= β |Ei | r sin θdθ 2π (34.1.19)
2η  εs + 2ε  r2 0
3 The ω 2 dependence of the following function implies that the radiated electric field in the far zone is

proportional to the acceleration of the charges on the dipole.

376 Electromagnetic Field Theory

But
 π  π  π
sin3 θdθ = − sin2 θd cos θ = − (1 − cos2 θ)d cos θ
0 0 0
 −1
4
=− (1 − x2 )dx = (34.1.20)
1 3

Therefore

 εs − ε 2 4 6
 
4π 2
Ps =  εs + 2εs  β a |Ei | (34.1.21)
 
3η

In the above, even though we have derived the equation using electrostatic theory, it is also
valid for complex permittivity defined in Section 7.1.2. One can take the divergence of (7.1.9)
to arrive at a Gauss’ law for lossy dispersive media, viz., ∇ · εE = 0 which is homomorphic
to the lossless case. Hurrah again to phasor technique! e
The scattering cross section is the effective area of a scatterer such that the total scattered
power is proportional to the incident power density times the scattering cross section. As such
it is defined as

2
8πa2  εs − ε 

Ps
Σs = = (βa)4 (34.1.22)
hSinc i 3  εs + 2ε 

where we have used the local plane-wave approximation that

1
hSinc i = |Ei |2 (34.1.23)
2η

The above also implies that

Ps = hSinc i · Σs

In other words, the scattering cross section Σs is an effective cross-sectional area of the
scatterer that will intercept the incident wave power hSinc i to produce the scattered power
Ps .
It is seen that the scattering cross section grows as the fourth power of frequency since
β = ω/c. The radiated field grows as the second power because it is proportional to the
acceleration of the charges on the particle. The higher the frequency, the more the scattered
power. This mechanism can be used to explain why the sky is blue. It also can be used to
explain why sunset has a brilliant hue of red and orange (see Figure 34.3).
Scattering of Electromagnetic Field 377

Figure 34.3: During the day time, when we look at the sky, we mainly see scattered sunlight.
Since high-frequency light is scattered more, the sky appears blue. At sunset, the sunlight
has to go through a thicker atmosphere. Thus, the blue light is scattered away, leaving the
red light that reaches the eyes. Hence, the sunset appears red.

The above also explains the brilliant glitter of gold plasmonic nano-particles as discovered
by ancient Roman artisans. For gold, the medium resembles a plasma, and hence, we can
have εs < 0, and the denominator can be very small giving rise to strongly scattered light
(see Section 8.3.6).

Furthermore, since the far field scattered power density of this particle is

1
hSi = Eθ Hφ∗ ∼ sin2 θ (34.1.24)
2η

the scattering pattern of this small particle is not isotropic. In other words, these dipoles ra-
diate predominantly in the broadside direction but not in their end-fire directions. Therefore,
insects and sailors can use this to figure out where the sun is even in a cloudy day. In fact,
it is like a rainbow: If the sun is rising or setting in the horizon, there will be a bow across
the sky where the scattered field is predominantly linearly polarized.4 Such a “sunstone” for
direction finding is shown in Figure 34.4.

4 You can go through a Gedanken experiment to convince yourself of such.

378 Electromagnetic Field Theory

Figure 34.4: A sunstone can indicate the polarization of the scattered light. From that, one
can deduce where the sun is located (courtesy of Wikipedia).

34.1.3 Small Conductive Particle

The above analysis is for a small dielectric particle. The quasi-static analysis may not be valid
for when the conductivity of the particle becomes very large. For instance, for a perfect electric
conductor immersed in a time varying electromagnetic field, the magnetic field in the long
wavelength limit induces eddy current in PEC sphere.5 Hence, in addition to an electric dipole
component, a PEC sphere also has a magnetic dipole component. The scattered field due to
a tiny PEC sphere is a linear superposition of an electric and magnetic dipole components.
These two dipolar components have electric fields that cancel precisely at certain observation
angle. It gives rise to deep null in the bi-static radar scattering cross-section (RCS)6 of a
PEC sphere as illustrated in Figure 34.5.

5 Note that there is no PEC at optics. A metal behaves more like a plasma medium at optical frequencies.
6 Scattering cross section in microwave range is called an RCS due to its prevalent use in radar technology.
Scattering of Electromagnetic Field 379

Figure 34.5: RCS (radar scattering cross section) of a small PEC scatterer (courtesy of Sheng
et al. [212]).

34.2 Mie Scattering

When the size of the dipole becomes larger compared to wavelength λ, quasi-static approxi-
mation is insufficient to approximate the solution. Then one has to solve the boundary value
problem in its full glory usually called the full-wave theory or Mie theory [213,214]. With this
theory, the scattering cross section does not grow indefinitely with frequency as in (34.1.22).
It has to saturate to a value for increasing frequency. For a sphere of radius a, the scattering
cross section becomes πa2 in the high-frequency limit. This physical feature of this plot is
shown in Figure 34.6, and it also explains why the sky is not purple.
380 Electromagnetic Field Theory

Figure 34.6: Radar cross section (RCS) calculated using Mie scattering theory [214].

34.2.1 Optical Theorem

Before we discuss the Mie scattering solution, let us discuss an amazing theorem called the
optical theorem. This theorem says that the scattering cross section of a scatterer depends
only on the forward scattering power density of the scatterer. In other words, if a plane wave
is incident on a scatterer, the scatterer will scatter the incident power in all directions. But
the total power scattered by the object is only dependent on the forward scattering power
density of the object or scatterer. This amazing theorem is called the optical theorem, and
the proof of this is given in J.D. Jackson’s book [48].
The true physical reason for this is power orthogonality. Two plane waves cannot interact
or exchange power with each other unless they share the same k or β vector, where β is both
the plane wave direction of the incident wave as well as the forward scattered wave. This
is similar to power orthogonality in a waveguide, and it happens for orthogonal modes in
waveguides [83, 178].
The scattering pattern of a scatterer for increasing frequency is shown in Figure 34.7. For
Rayleigh scattering where the wavelength is long, the scattered power is distributed isotrop-
ically save for the doughnut shape of the radiation pattern, namely, the sin2 (θ) dependence.
As the frequency increases, the power is scattered increasingly in the forward direction. The
reason being that for very short wavelength, the scatterer looks like a disc to the incident
Scattering of Electromagnetic Field 381

wave, casting a shadow in the forward direction. Hence, there has to be scattered field in the
forward direction to cancel the incident wave to cast this shadow.
In a nutshell, the scattering theorem is intuitively obvious for high-frequency scattering.
The amazing part about this theorem is that it is true for all frequencies.

Figure 34.7: A particle scatters increasingly more in the forward direction as the frequency
increases (Courtesy of hyperphysics.phy-astr.gsu.edu).

34.2.2 Mie Scattering by Spherical Harmonic Expansions

As mentioned before, as the wavelength becomes shorter, we need to solve the boundary
value problem in its full glory without making any approximations. This closed form solution
can be found for a sphere scattering by using separation of variables and spherical harmonic
expansions that will be discussed in the section.
The Mie scattering solution by a sphere is beyond the scope of this course.7 The separation
of variables in spherical coordinates is not the only useful for Mie scattering, it is also useful
for analyzing spherical cavity. So we will present the precursor knowledge so that you can
read further into Mie scattering theory if you need to in the future.

34.2.3 Separation of Variables in Spherical Coordinates8

To this end, we look at the scalar wave equation (∇2 +β 2 )Ψ(r) = 0 in spherical coordinates. A
lookup table can be used to evaluate ∇·∇, or divergence of a gradient in spherical coordinates.
Hence, the Helmholtz wave equation becomes9
∂2
 
1 ∂ 2 ∂ 1 ∂ ∂ 1 2
r + sin θ + + β Ψ(r) = 0 (34.2.1)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂φ2

Noting the ∂ 2 /∂φ2 derivative, by using separation of variables technique, we assume Ψ(r) to
be of the form

Ψ(r) = F (r, θ)ejmφ (34.2.2)

7 But it is treated in J.A. Kong’s book [32] and Chapter 3 of W.C. Chew, Waves and Fields in Inhomogeneous

Media [35] and many other textbooks [48, 65, 181].

8 May be skipped on first reading.
9 By quirk of mathematics, it turns out that the first term on the right-hand side below can be simplified

by observing that r12 ∂r ∂ 2

r = r1 ∂r
∂
r.
382 Electromagnetic Field Theory

∂ 2 jmφ
This will simplify the ∂/∂φ derivative in the partial differential equation since ∂φ2 e =
−m2 ejmφ . Then (34.2.1) becomes
m2
 
1 ∂ 2 ∂ 1 ∂ ∂ 2
r + sin θ − + β F (r, θ) = 0 (34.2.3)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ
Again, by using the separation of variables, and letting further that
F (r, θ) = bn (βr)Pnm (cos θ) (34.2.4)
where we require that
m2
  
1 d d
sin θ + n(n + 1) − Pnm (cos θ) = 0 (34.2.5)
sin θ dθ dθ sin2 θ
when Pnm (cos θ) is the associate Legendre polynomial. Note that (34.2.5) is an eigenvalue
problem with eigenvalue n(n + 1), and |m| ≤ |n|. The value n(n + 1) is also known as
separation constant.
Consequently, bn (kr) in (34.2.4) satisfies
 
1 d 2 d n(n + 1) 2
r − + β bn (βr) = 0 (34.2.6)
r2 dr dr r2
The above is the spherical Bessel equation where bn (βr) is either the spherical Bessel function
(1)
jn (βr), spherical Neumann function nn (βr), or the spherical Hankel functions, hn (βr) and
(2)
hn (βr). The spherical functions are the close cousins of the cylindrical functions. They are
related to the cylindrical functions via [35, 49] 10
r
π
bn (βr) = B 1 (βr) (34.2.7)
2βr n+ 2
It is customary to define the spherical harmonic as [48, 110]
s
2n + 1 (n − m)! m
Ynm (θ, φ) = P (cos θ)ejmφ (34.2.8)
4π (n + m)! n
The above is normalized such that
∗
Yn,−m (θ, φ) = (−1)m Ynm (θ, φ) (34.2.9)
and that
 2π  π
dφ sin θdθYn∗0 m0 (θ, φ)Ynm (θ, φ) = δn0 n δm0 m (34.2.10)
0 0

These functions are also complete11 like Fourier series, so that

∞ X
n
X 0 0
∗
Ynm (θ0 , φ0 )Ynm (θ, φ) = δ(φ − φ )δ(cos θ − cos θ ) (34.2.11)
n=0 m=−n
10 By a quirk of nature, the spherical Bessel functions needed for 3D wave equations are in fact simpler than

cylindrical Bessel functions needed for 2D wave equation. One can say that 3D is real, but 2D is surreal.
11 In a nutshell, a set of basis functions is complete in a subspace if any function in the same subspace can

be expanded as a sum of these basis functions.

Lecture 35

Spectral Expansions of Source

Fields—Sommerfeld Integrals

In previous lectures, we have assumed plane waves in finding closed form solutions. Plane
waves are simple waves, and their reflections off a flat surface or a planarly layered medium
can be found easily. When we have a source like a point source, it generates a spherical
wave. We do not know how to reflect exactly a spherical wave off a planar interface. But
by expanding a spherical wave in terms of sum of plane waves and evanescennt waves using
Fourier transform technique, we can solve for the solution of a point source over a layered
medium easily in terms of spectral integrals. Sommerfeld was the first person to have done
this, and hence, these integrals are often called Sommerfeld integrals. Finally, we shall apply
the method of stationary phase to approximate these integrals to elucidate their physics.
From this, we can see ray theory emerging from the complicated mathematics. It reminds
me of a lyric from the musical The Sound of Music—Ray, a drop of golden sun! Ray has
mesmerized the human mind, and it will be interesting to see if the mathematics behind it is
equally enchanting.
By this time, you probably feel inundated by the ocean of knowledge that you are imbibing.
If you can assimilate them, it will be an exhilarating experience.

35.1 Spectral Representations of Sources

A plane wave is a mathematical idealization that does not exist in the real world. In practice,
waves are nonplanar in nature as they are generated by finite sources, such as antennas
and scatterers: For example, a point source generates a spherical wave which is nonplanar.
Fortunately, these waves can be expanded in terms of sum of plane waves. Once this is done,
then the study of non-plane-wave reflections from a layered medium becomes routine. In the
following, we shall show how waves resulting from a point source can be expanded in terms of
plane waves summation. This topic is found in many textbooks [1, 32, 35, 95, 96, 181, 205, 215].

383
384 Electromagnetic Field Theory

35.1.1 A Point Source

There are a number of ways to derive the plane wave expansion of a point source. We will
illustrate one of the ways. The spectral decomposition or the plane-wave expansion of the
field due to a point source could be derived using Fourier transform technique. First, notice
that the scalar wave equation with a point source at the origin is
 2
∂2 ∂2

2 2

∂ 2
∇ + k0 φ(x, y, z) = + 2 + 2 + k0 φ(x, y, z) = −δ(x) δ(y) δ(z). (35.1.1)
∂x2 ∂y ∂z
The above equation could then be solved in the spherical coordinates, yielding the solution
given in the previous lecture, namely, Green’s function with the source point at the origin,
or1
eik0 r
φ(x, y, z) = φ(r) = . (35.1.2)
4πr
The solution is entirely spherically symmetric due to the symmetry of the point source.
Next, assuming that the Fourier transform of φ(x, y, z) exists,2 we can write
∞
1
φ(x, y, z) = dkx dky dkz φ̃(kx , ky , kz )eikx x+iky y+ikz z . (35.1.3)
(2π)3
−∞

Then we substitute the above into (35.1.1), after exchanging the order of differentiation and
integration,3 one can simplify the Laplacian operator in the Fourier space, or spectral domain,
to arrive at
∂2 ∂2 ∂2
∇2 = 2
+ 2 + 2 = −kx2 − ky2 − kz2
∂x ∂y ∂z
Then, together with the Fourier representation of the delta function, which is4
∞
1
δ(x) δ(y) δ(z) = dkx dky dkz eikx x+iky y+ikz z (35.1.4)
(2π)3
−∞

we convert (35.1.1) into

∞
dkx dky dkz [k02 − kx2 − ky2 − kz2 ]φ̃(kx , ky , kz )eikx x+iky y+ikz z (35.1.5)
−∞
∞
=− dkx dky dkz eikx x+iky y+ikz z . (35.1.6)
−∞
1 From this point onward, we will adopt the exp(−iωt) time convention to be commensurate with the optics

and physics literatures.

2 The Fourier transform of a function f (x) exists if it is absolutely integrable, namely that ∞ |f (x)|dx is
−∞
finite (see [110]).
3 Exchanging the order of differentiation and integration is allowed if the integral converges after the

exchange.

4 We have made use of that δ(x) = 1/(2π) ∞ dk exp(ik x) three times.
−∞ x x
Spectral Expansions of Source Fields—Sommerfeld Integrals 385

Since the above is equal for all x, y, and z, we can Fourier inverse transform the above to get
−1
φ̃(kx , ky , kz ) = . (35.1.7)
k02 − kx2 − ky2 − kz2

Consequently, we have
∞
−1 eikx x+iky y+ikz z
φ(x, y, z) = dk . (35.1.8)
(2π)3 k02 − kx2 − ky2 − kz2
−∞

where dk = dkx dky dkz . The above expresses the fact the φ(x, y, z) which is a spherical wave
by (35.1.2), is expressed as an integral summation of plane waves. But these plane waves are
not physical plane waves in free space since kx2 + ky2 + kz2 6= k02 .

Im [kz]

k20 – k 2x–
   k 2y
⊗
Re [kz]
×
Fourier Inversion Contour
– k20 – k 2x – k 2y

Figure 35.1: The integration along the real axis is equal to the integration along C plus the
residue of the pole at (k02 − kx2 − ky2 )1/2 , by invoking Jordan’s lemma.

Weyl Identity
To make the plane waves in (35.1.8) into physical plane waves, we have to massage it into
a different form. We rearrage the integrals in (35.1.8) so that the dkz integral is performed
first. In other words,
∞  ∞
1 eikz z
φ(r) = dkx dky eikx x+iky y dkz (35.1.9)
(2π)3 −∞ kz2 − (k02 − kx2 − ky2 )
−∞

where we have deliberately rearrange the denominator with kz being the variable in the inner
integral. Then the integrand has poles at kz = ±(k02 − kx2 − ky2 )1/2 .5 Moreover, for real k0 ,
and real values of kx and ky , these two poles lie on the real axis, rendering the integral in
5 In (35.1.8), the pole is located at k 2 + k 2 + k 2 = k 2 . This equation describes a sphere in k space, known
x y z 0
as the Ewald’s sphere [216].
386 Electromagnetic Field Theory

(35.1.8) undefined. However, if a small loss is assumed in k0 such that k0 = k00 + ik000 , then
the poles are off the real axis (see Figure 35.1), and the integrals in (35.1.8) are well-defined.
In actual fact, this is intimately related to the uniqueness principle we have studied before:
An infinitesimal loss is needed to guarantee uniqueness in an open space as shall be explained
below.
First, the reason is that without loss, |φ(r)| ∼ O(1/r), r → ∞ is not strictly absolutely
integrable, and hence, its Fourier transform does not exist [52]: The manipulation that leads
to (35.1.8) is not strictly correct. Second, the introduction of a small loss also guarantees the
radiation condition and the uniqueness of the solution to (35.1.1), and therefore, the equality
of (35.1.2) and (35.1.8) [35].
Observe that in (35.1.8), when z > 0, the integrand is exponentially small when =m[kz ] →
∞. Therefore, by Jordan’s lemma [89], the integration for kz over the contour C as shown
in Figure 35.1 vanishes. Then, by Cauchy’s theorem [89], the integration over the Fourier
inversion contour on the real axis is the same as integrating over the pole singularity located
at (k02 − kx2 − ky2 )1/2 , yielding the residue of the pole (see Figure 35.1). Consequently, after
doing the residue evaluation, we have

∞ 0
i eikx x+iky y+ikz z
φ(x, y, z) = dkx dky , z > 0, (35.1.10)
2(2π)2 kz0
−∞

where kz0 = (k02 − kx2 − ky2 )1/2 is the value of kz at the pole location.
Similarly, for z < 0, we can add a contour C in the lower-half plane that contributes zero
to the integral, one can deform the contour to pick up the pole contribution. Hence, the
integral is equal to the pole contribution at kz0 = −(k02 − kx2 − ky2 )1/2 (see Figure 35.1). As
such, the result for all z can be written as

∞ 0
i eikx x+iky y+ikz |z|
φ(x, y, z) = dkx dky , all z. (35.1.11)
2(2π)2 kz0
−∞

By the uniqueness of the solution to the partial differential equation (35.1.1) satisfying
radiation condition at infinity, we can equate (35.1.2) and (35.1.11), yielding the identity

∞
eik0 r i eikx x+iky y+ikz |z|
= dkx dky , (35.1.12)
r 2π kz
−∞

where kx2 + ky2 + kz2 = k02 , or kz = (k02 − kx2 − ky2 )1/2 . The above is known as the Weyl identity
(Weyl 1919). To ensure the radiation condition, we require that =m[kz ] > 0 and <e[kz ] > 0
over all values of kx and ky in the integration. Furthermore, Equation (35.1.12) could be
interpreted as an integral summation of plane waves propagating in all directions, including
evanescent waves. It is the plane-wave expansion (including evanescent wave) of a spherical
wave.
Spectral Expansions of Source Fields—Sommerfeld Integrals 387

Figure 35.2: The wave is propagating for kρ vectors inside the disk, while the wave is evanes-
cent for kρ outside the disk.

One can also interpret the above as a 2D surface integral in the Fourier space over the
kx and ky plane or variables. When kx2 + ky2 < k02 , or the spatial spectrum is inside a disk
of radius k0 , the waves are propagating waves. But for contributions outside this disk, the
waves are evanescent (see Figure 35.2). And the high Fourier (or spectral) components of
the Fourier spectrum correspond to evanescent waves. The high spectral components, which
are related to the evanescent waves, are important for reconstructing the singularity of the
Green’s function.

y
kρ

a ρ
φ x

Figure 35.3: The kρ and the ρ vector on the xy plane.

Sommerfeld Identity

The Weyl identity has double integral, and hence, is more difficult to integrate numerically.
Here, we shall derive the Sommerfeld identity which has only one integral. First, in (35.1.12),
we express the integral in cylindrical coordinates and write kρ = x̂kρ cos α + ŷkρ sin α, ρ =
x̂ρ cos φ + ŷρ sin φ (see Figure 35.3), and dkx dky = kρ dkρ dα. Then, kx x + ky y = kρ · ρ =
388 Electromagnetic Field Theory

kρ cos(α − φ), and with the appropriate change of variables, we have

∞  2π
eik0 r i eikρ ρ cos(α−φ)+ikz |z|
= kρ dkρ dα , (35.1.13)
r 2π 0 kz
0

where kz = (k02 − kx2 − ky2 )1/2 = (k02 − kρ2 )1/2 , where in cylindrical coordinates, in the kρ -space,
or the Fourier space, kρ2 = kx2 + ky2 . Then, using the integral identity for Bessel functions
given by6
2π
1
J0 (kρ ρ) = dα eikρ ρ cos(α−φ) , (35.1.14)
2π
0

(35.1.13) becomes
∞
eik0 r kρ
=i dkρ J0 (kρ ρ)eikz |z| . (35.1.15)
r kz
0

The above is also known as the Sommerfeld identity (Sommerfeld 1909 [109]; [205][p.
242]). Its physical interpretation is that a spherical wave can now be expanded as an integral
summation of conical waves or cylindrical waves in the ρ direction, times a plane wave in
the z direction over all wave numbers kρ . This wave is evanescent in the ±z direction when
kρ > k0 .
(1) (2)
By using the fact that J0 (kρ ρ) = 1/2[H0 (kρ ρ) + H0 (kρ ρ)], and the reflection formula
(1) iπ (2)
that H0 (e x) = −H0 (x), a variation of the above identity can be derived as
∞
eik0 r i kρ (1)
= dkρ H (kρ ρ)eikz |z| . (35.1.16)
r 2 kz 0
−∞

Im [kρ]

• +k0
Re [kρ]
•
–k0
Sommerfeld
Integration Path

Figure 35.4: Sommerfeld integration path.

(1)
Since H0 (x) has a logarithmic branch-point singularity at x = 0,7 and kz = (k02 − kρ2 )1/2
has algebraic branch-point singularities at kρ = ±k0 , the integral in Equation (35.1.16) is
6 See Chew [35], or Whitaker and Watson(1927) [217].
7 H (1) (x)∼ 2i ln(x), see Chew [35][p. 14], or Abromawitz or Stegun [117].
0 π
Spectral Expansions of Source Fields—Sommerfeld Integrals 389

undefined unless we stipulate also the path of integration. Hence, a path of integration
adopted by Sommerfeld, which is even good for a lossless medium, is shown in Figure 35.4.
Because of the manner in which we have selected the reflection formula for Hankel functions,
(1) (2)
i.e., H0 (eiπ x) = −H0 (x), the path of integration should be above the logarithmic branch-
point singularity at the origin. With this definition of the Sommerfeld integration, the integral
is well defined even when there is no loss, i.e., when the bronach points ±k0 are on the real
axis.

35.2 A Source on Top of a Layered Medium

Previously, we have studied the propagation of plane electromagnetic waves from a single
dielectric interface in Section 14.1 as well as through a layered medium in Section 16.1. It
can be shown that plane waves reflecting from a layered medium can be decomposed into
TE-type plane waves, where Ez = 0, Hz 6= 0, and TM-type plane waves, where Hz = 0,
Ez 6= 0.8 One also sees how the field due to a point source can be expanded into plane waves
in Section 35.1.
In view of the above observations, when a point source is on top of a layered medium, it
is then best to decompose its field in terms of plane waves of TE-type and TM-type. Then,
the nonzero component of Ez characterizes TM waves, while the nonzero component of Hz
characterizes TE waves. Hence, given a field, its TM and TE components can be extracted
readily. Furthermore, if these TM and TE components are expanded in terms of plane waves,
their propagations in a layered medium can be studied easily.
The problem of a vertical electric dipole on top of a half space was first solved by Som-
merfeld (1909) [109] using Hertzian potentials, which are related to the z components of the
electromagnetic field. The work is later generalized to layered media, as discussed in the liter-
ature. Later, Kong (1972) [218] suggested the use of the z components of the electromagnetic
field instead of the Hertzian potentials.

35.2.1 Electric Dipole Fields–Spectral Expansion

The representation of a spherical wave in terms of plane waves can be done using Weyl
identity or Sommerfeld identiy. Here, we will use Sommerfeld identity in anticipation of
simpler numerical integration, since only single integrals are involved. The E field in a
homogeneous medium due to a point current source or a Hertzian dipole directed in the α̂
direction, J = α̂I` δ(r), is derivable via the vector potential method or the dyadic Green’s
function approach. Then, using the dyadic Green’s function approach, or the vector/scalar
potential approach, the field due to a Hertzian dipole is given by

eikr
 
∇∇
E(r) = iωµ I + 2 · α̂I` , (35.2.1)
k 4πr
√
where I` is the current moment and k = ω µ , the wave number of the homogeneous
medium. Furthermore, from ∇ × E = iωµH, the magnetic field due to a Hertzian dipole is
8 Chew, Waves and Fields in Inhomogeneous Media [35]; Kong, Electromagnetic Wave Theory [32].
390 Electromagnetic Field Theory

shown to be given by

eikr
H(r) = ∇ × α̂I` . (35.2.2)
4πr
With the above fields, their TM and TE components can be extracted easily in anticipation
of their plane wave expansions for propagation through layered media.

(a) Vertical Electric Dipole (VED)

Region 1 x
–d1
Region i –di

Figure 35.5: A vertical electric dipole over a layered medium.

A vertical electric dipole shown in Figure 35.5 has α̂ = ẑ; hence, in anticipation of their plane
wave expansions, the TM component of the field is characterized by Ez 6= 0 or that

∂ 2 eikr
 
iωµI` 2
Ez = k + 2 , (35.2.3)
4πk 2 ∂z r

and the TE component of the field is characterized by

Hz = 0, (35.2.4)

implying the absence of the TE field.

Next, using the Sommerfeld identity (35.1.16) in the above, and after exchanging the order
of integration and differentiation, we have9
∞
−I` kρ3
Ez = dkρ J0 (kρ ρ)eikz |z| , |z| =
6 0 (35.2.5)
4πω kz
0

after noting that kρ2 + kz2 = k 2 . Notice that now Equation (35.2.5) expands the z component
of the electric field in terms of cylindrical waves in the ρ direction and a plane wave in the z
direction. Since cylindrical waves actually are linear superpositions of plane waves, because
we can work backward from (35.1.16) to (35.1.12) to see this. As such, the integrand in
9 By using (35.1.16) in (35.2.3), the ∂ 2 /∂z 2 operating on eikz |z| produces a Dirac delta function singularity.

Detail discussion on this can be found in the chapter on dyadic Green’s function in Chew, Waves and Fields
in Inhomogeneous Media [35].
Spectral Expansions of Source Fields—Sommerfeld Integrals 391

(35.2.5) in fact consists of a linear superposition of TM-type plane waves. The above is also
the primary field generated by the source.10
Consequently, for a VED on top of a stratified medium as shown, the downgoing plane
wave from the point source will be reflected like TM waves with the generalized reflection
TM
coefficient R̃12 . Hence, over a stratified medium, the field in region 1 can be written as

∞
−I` kρ3 h i
E1z = dkρ J0 (kρ ρ) eik1z |z| + R̃12
T M ik1z z+2ik1z d1
e , (35.2.6)
4πω1 k1z
0

1
where k1z = (k12 − kρ2 ) 2 , and k12 = ω 2 µ1 1 , the wave number in region 1.
The phase-matching condition dictates that the transverse variation of the field in all the
regions must be the same. Consequently, in the i-th region, the solution becomes

∞
−I` kρ3 h i
i Eiz = dkρ J0 (kρ ρ)Ai e−ikiz z + R̃i,i+1
T M ikiz z+2ikiz di
e . (35.2.7)
4πω k1z
0

Notice that Equation (35.2.7) is now expressed in terms of i Eiz because i Eiz reflects and
transmits like Hiy , the transverse component of the magnetic field or TM waves.11 Therefore,
TM
R̃i,i+1 and Ai could be obtained using the methods discussed in Chew, Waves and Fields in
Inhomogeneous Media [110].
This completes the derivation of the integral representation of the electric field everywhere
in the stratified medium. These integrals are known as Sommerfeld integrals. The case
when the source is embedded in a layered medium can be derived similarly.

(b) Horizontal Electric Dipole (HED)

The HED is more complicated. Unlike the VED that excites only the TM waves, an HED will
excite both TE and TM waves. For a horizontal electric dipole pointing in the x direction,
α̂ = x̂; hence, (35.2.1) and (35.2.2) give the TM and the TE components, in anticipation of
their plane wave expansions, as

iI` ∂ 2 eikr
Ez = , (35.2.8)
4πω ∂z∂x r
I` ∂ eikr
Hz = − . (35.2.9)
4π ∂y r

10 One can perform a sanity check on the odd and even symmetry of the fields’ z-component by sketching

the fields of a static horizontal electric dipole.

11 See Chew, Waves and Fields in Inhomogeneous Media [35], p. 46, (2.1.6) and (2.1.7). Or we can gather

from (14.1.6) to (14.1.7) that the µi Hiz transmits like Eiy at a dielectric interface, and by duality, i Eiz
transmits like Hiy .
392 Electromagnetic Field Theory

Then, with the Sommerfeld identity (35.1.16), we can expand the above as

∞
iI`
Ez = ± cos φ dkρ kρ2 J1 (kρ ρ)eikz |z| (35.2.10)
4πω
0
∞
I` kρ2
Hz = i sin φ dkρ J1 (kρ ρ)eikz |z| . (35.2.11)
4π kz
0

Now, Equation (35.2.10) represents the wave expansion of the TM field, while (35.2.11) repre-
sents the wave expansion of the TE field in terms of Sommerfeld integrals which are plane-wave
expansions in disguise. Observe that because Ez is odd about z = 0 in (35.2.10), the down-
going wave has an opposite sign from the upgoing wave. At this point, the above are just the
primary field generated by the source.
On top of a stratified medium, the downgoing wave is reflected accordingly, depending on
its wave type. Consequently, we have

∞
iI` h i
E1z = cos φ dkρ kρ2 J1 (kρ ρ) ±eik1z |z| − R̃12
T M ik1z (z+2d1 )
e , (35.2.12)
4πω1
0
∞
iI` kρ2 h i
H1z = sin φ dkρ J1 (kρ ρ) eik1z |z| + R̃12
T E ik1z (z+2d1 )
e . (35.2.13)
4π k1z
0

TM
Notice that the negative sign in front of R̃12 in (35.2.12) follows because the downgoing
wave in the primary field has a negative sign as shown in (35.2.10).

35.3 Stationary Phase Method—Fermat’s Principle

Sommerfeld integrals are rather complex, and by themselves, they do not offer much physical
insight into the physics of the field. To elucidate the physics, we can apply the stationary
phase method to find approximations of these integrals when the frequency is high, or kr is
large, or the observation point is many wavelengths away from the source point. It turns out
that this method is initmately related to Fermat’s principle.
In order to avoid having to work with special functions like Bessel functions, we convert
the Sommerfeld integrals back to spectral integrals in the cartesian coordinates. We could
have obtained the aforementioned integrals in cartesian coordinates were we to start with the
Weyl identity instead of the Sommerfeld identity. To do the back conversion, we make use of
the identity,

∞ ∞
eik0 r i eikx x+iky y+ikz |z| kρ
= dkx dky =i dkρ J0 (kρ ρ)eikz |z| . (35.3.1)
r 2π kz kz
−∞ 0
Spectral Expansions of Source Fields—Sommerfeld Integrals 393

We can just focus our attention on the reflected wave term in (35.2.6) and rewrite it in
cartesian coordinates to get

∞
R −I` kx2 + ky2 T M ikx x+iky y+ik1z (z+2d1 )
E1z = dkx dky R12 e
8π 2 ω1 k1z
−∞
∞
1
= dkx dky F (kx , ky )eikx x+iky y+ik1z (z+2d1 ) (35.3.2)
k1z
−∞

where

−I`
F (kx , ky ) = (k 2 + ky2 )R12
TM
8π 2 ω1 x

In the above, kx2 + ky2 + k1z

2
= k12 is the dispersion relation satisfied by the plane wave in region
q
TM
1. Also, R12 is dependent on kiz = ki2 − kx2 − ky2 in cartesian coordinates, where i = 1, 2.
Now the problem reduces to finding the approximation of the following integral:

∞
R 1
E1z = dkx dky F (kx , ky )eirh(kx ,ky ) (35.3.3)
k1z
−∞

where

 x y z
rh(kx , ky ) = r kx + ky + k1z , (35.3.4)
r r r

We want to approximate the above integral when kr is large. For simplicity, we have set
d1 = 0 to begin with.
394 Electromagnetic Field Theory

Figure 35.6: In this figure, t can represent kx or ky when one of them is varying. Around
the stationary phase point, the function h(t) is slowly varying. In this figure, λ = r, and
g(kx , ky , λ) = eiλh(kx ,ky ) = eirh(kx ,ky ) . When λ = r is large, the function g(λ, kx , ky ) is
rapidly varying with respect to either kx or ky . Hence, most of the contributions to the
integral comes from around the stationary phase point.

In the above, eirh(kx ,ky ) is a rapidly varying function of kx and ky when x, y, and z are
large, or r is large compared to wavelength.12 In other words, a small change in kx or ky
will cause a large change in the phase of the integrand, or the integrand will be a rapidly
varying function of kx and ky . Due to the cancellation of the integral when one integrates
a rapidly varying function, most of the contributions to the integral will come from around
the stationary point of h(kx , ky ) or where the function is least slowly varying. Otherwise,
the integrand is rapidly varying away from this point, and the integration will destructively
cancel with each other, while around the stationary point, they will add constructively.
The stationary point in the kx and ky plane is found by setting the derivatives of h(kx , ky )
with respect to to kx and ky to zero. By so doing

∂h x kx z ∂h y ky z
= − = 0, = − =0 (35.3.5)
∂kx r k1z r ∂ky r k1z r

The above represents two equations from which the two unknowns, kxs and kys , at the
stationary phase point can be solved for. By expressing the above in spherical coordinates,
x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ, the values of (kxs , kys ), that satisfy the above
equations are

kxs = k1 sin θ cos φ, kys = k1 sin θ sin φ (35.3.6)

with the corresponding k1zs = k1 cos θ.

12 The yardstick in wave physics is always wavelength. Large distance is also synonymous to increasing the

frequency or reducing the wavelength.

Spectral Expansions of Source Fields—Sommerfeld Integrals 395

When one integrates on the kx and ky plane, the dominant contribution to the integral will
come from the point in the vicinity of (kxs , kys ). Assuming that F (kx , ky ) is slowly varying,
we can equate F (kx , ky ) to a constant equal to its value at the stationary phase point, and
say that

∞
R 1 ikx x+iky y+ik1z z eik1 r
E1z ' F (kxs , kys ) e dkx dky = 2πF (kxs , kys ) (35.3.7)
k1z ir
−∞

In the above, the integral can be performed in closed form using the Weyl identity. The above
expression has two important physical interpretations.

(i) Even though a source is emanating plane waves in all directions in accordance to
(35.1.12), at the observation point r far away from the source point, only one or few
plane waves in the vicinity of the stationary phase point are important. They interfere
with each other constructively to form a spherical wave that represents the ray connect-
ing the source point to the observation point. Plane waves in other directions interfere
with each other destructively, and are not important. That is the reason that the source
point and the observation point is connected only by one ray, or one bundle of plane
waves in the vicinity of the stationary phase point. These bundle of plane waves are
also almost paraxial with respect to each other.

(ii) The function F (kx , ky ) could be a very complicated function like the reflection coefficient
RT M , but only its value at the stationary phase point matters. If we were to make d1 6= 0
again in thepabove analysis, the math remains similar except that now, we replace r
with rI = x2 + y 2 + (z + 2d1 )2 . Due to the reflecting half-space, the source point
has an image point as shown in Figure 35.7 This physical picture is shown in the figure
where rI now is the distance of the observation point to the image point. The stationary
phase method extract a ray that emanates from the source point, bounces off the half-
space, and the reflected ray reaches the observer modulated by the reflection coefficient
RT M . But the value of the reflection coefficient that matters is at the angle at which
the incident ray impinges on the half-space.

(iii) At the stationary point, the ray is formed by the k-vector where k = x̂k1 sin θ cos φ +
ŷk1 sin θ sin φ + ẑk1 cos theta. This ray points in the same direction as the position
vector of the observation point r = x̂r sin θ cos φ + ŷr sin θ sin φ + ẑr cos theta. In other
words, the k-vector and the r-vector point in the same direction. This is reminiscent
of Fermat principle, because when this happens, the ray propagates with the minimum
phase between the source point and the observation point. When z → z + 2d1 , the
ray for the image source is altered to that shown in Figure 35.7 where it the ray is
minimum phase from the image source to the obervation point. Hence, the stationary
phase method is initimately related to Fermat’s principle.
396 Electromagnetic Field Theory

Figure 35.7: At high frequencies, the source point and the observation point are connected
by a ray. The ray represents a bundle of plane waves that interfere constructively. This even
true for a bundle of plane waves that reflect off an interface. So ray theory or ray optics
prevails here, and the ray bounces off the interface according to the reflection coefficient of a
plane wave impinging at the interface with θI .

35.4 Riemann Sheets and Branch Cuts13

The Sommerfeld integrals will have integrands that are multi-value or double value. Proper
book keeping is needed so that the evaluation of these integrals can be performed unambigu-
ously. The function kz = (k02 − kρ2 )1/2 in (35.1.15) and (35.1.16) are double-value functions
because, in taking the square root of a number, two values are possible. In particular, kz is
a double-value function of kρ . Consequently, for every point on a complex kρ plane in Figure
35.4, there are two possible values of kz . Therefore, as an example, the integral (35.1.11)
is undefined unless we stipulate which of the two values of kz is adopted in performing the
integration.
A multivalue function is denoted on a complex plane with the help of Riemann sheets
[35, 89]. For instance, a double-value function such as kz is assigned two Riemann sheets to
a single complex plane. On one of these Riemann sheets, kz assumes a value just opposite in
sign to the value on the other Riemann sheet. The correct sign for kz is to pick the square
root solution so that =m(kz ) > 0. This will ensure a decaying wave from the source.

35.5 Some Remarks14

Even though we have arrived at the solutions of a point source on top of a layered medium
by heuristic arguments of plane waves propagating through layered media, they can also
be derived more rigorously. For example, Equation (35.2.6) can be arrived at by matching
boundary conditions at every interface. The reason why a more heuristic argument is still
valid is due to the completeness of Fourier transforms. It is best explained by putting a source
over a half space and a scalar problem.
13 This may be skipped on first reading.
14 This may be skipped on first reading.
Spectral Expansions of Source Fields—Sommerfeld Integrals 397

We can expand the scalar field in the upper region as

∞
Φ1 (x, y, z) = dkx dky Φ̃1 (kx , ky , z)eikx x+iky y (35.5.1)
−∞

and the scalar field in the lower region as

∞
Φ2 (x, y, z) = dkx dky Φ̃2 (kx , ky , z)eikx x+iky y (35.5.2)
−∞

If we require that the two fields be equal to each other at z = 0, then we have
∞ ∞
dkx dky Φ̃1 (kx , ky , z = 0)eikx x+iky y = dkx dky Φ̃2 (kx , ky , z = 0)eikx x+iky y (35.5.3)
−∞ −∞

In order to remove the integral, and replace it with a simple scalar problem, one has to impose
the above equation for all x and y. Then by the completeness of Fourier transform implies
that15

Φ̃1 (kx , ky , z = 0) = Φ̃2 (kx , ky , z = 0) (35.5.4)

The above equation is much simpler than that in (35.5.3). In other words, due to the com-
pleteness of Fourier transform, one can match a boundary condition spectral-component by
spectral-component. If the boundary condition is matched for all spectral components, than
(35.5.3) is also true.

15 Or that we can perform a Fourier inversion on the above integrals.

398 Electromagnetic Field Theory
Lecture 36

Computational
Electromagnetics, Numerical
Methods

Due to the advent of digital computers and the blinding speed at which computations can be
done, numerical methods to seek solutions of Maxwell’s equations have become vastly popular.
Massively parallel digital computers now can compute at breakneck speed of tera\peta\exa-
flops throughputs [219], where FLOPS stands for “floating operations per second”. They
have also spawn terms that we have not previously heard of (see also Figure 36.1).

399
400 Electromagnetic Field Theory

Figure 36.1: Nomenclature for measuring the speed of modern day computers (courtesy of
Wikipedia [219]).

We repeat a quote from Freeman Dyson—“Technology is a gift of God. After the gift of
life it is perhaps the greatest of God’s gifts. It is the mother of civilizations, of arts and of
sciences.” The spurr for computer advancement is due to the second world war. During then,
men went to war while women stayed back to work as computers, doing laborious numerical
computations manually (see Figure 36.2 [220]): The need for a faster computer is obvious.
Unfortunately, in the last half century or so, we have been using a large part of the gift of
technology to destroy God’s greatest gift, life, in warfare!

Figure 36.2: A woman working as a computer shortly after the second world war (courtesy
of Wikipedia [220]).
Computational Electromagnetics, Numerical Methods 401

36.1 Computational Electromagnetics, Numerical Meth-

ods
Due to the high fidelity of Maxwell’s equations in describing electromagnetic physics in nature,
and they have been validated to high accuracy (see Section 1.1), often time, a numerical
solution obtained by solving Maxwell’s equations is more reliable than laboratory experiments.
This field is also known as computational electromagnetics. Numerical methods exploit
the blinding speed of modern digital computers to perform calculations, and hence to solve
large system of equations.
Computational electromagnetics consists mainly of two classes of numerical solvers: one
that solves differential equations directly: the differential-equation solvers; and one that solves
integral equations: the integral equation solvers. Both these classes of equations are derived
from Maxwell’s equations.

36.2 Examples of Differential Equations

An example of differential equations written in terms of sources are the scalar wave equation:

(∇2 + k 2 (r)) φ(r) = Q(r), (36.2.1)

An example of vector differential equation for vector electromagnetic field is

∇ × µ −1 · ∇ × E(r) − ω 2 (r) · E(r) = iωJ(r) (36.2.2)

These equations are linear equations, but for inhomogeneous media where k 2 (r) and (r)
are functions of position vector r, generally, they do not have closed form solutions. They
have one commonality, i.e., they can be abstractly written as

Lf = g (36.2.3)

where L is the differential operator which is linear, and f is the unknown, and g is the
driving source. Differential equations, or partial differential equations, as mentioned before,
have to be solved with boundary conditions. Otherwise, there is no unique solution to these
equations.
In the case of the scalar wave equation (36.2.1), L = (∇2 + k 2 ) is a differential operator.
In the case of the electromagnetic vector wave equation (36.2.2), L = (∇ × µ −1 · ∇×) − ω 2 ·.
Furthermore, f will be φ(r) for the scalar wave equation (36.2.1), while it will be E(r) in the
case of vector wave equation for an electromagnetic system (36.2.2). The g on the right-hand
side can represent Q in (36.2.1) or iωJ(r) in (36.2.2).
402 Electromagnetic Field Theory

36.3 Examples of Integral Equations

Figure 36.3: Geometry for the derivation of the volume-integral equation for scalar waves.
The wavenumber k(r) is assumed to be inhomogeneous and hence, a function of position r
inside the scatterer, but a constant k0 outside the scatterer.

36.3.1 Volume Integral Equation

This course is replete with PDE’s, but we have not come across too many integral equations
as yet. In integral equations, the unknown is embedded in the integral. The simplest integral
equation to derive is the volume integral equation. Hence, we shall first derive the volume
integral equation for the scalar wave case. In this case, the pertinent scalar wave equation is

[∇2 + k 2 (r)]φ(r) = Q(r), (36.3.1)

where k 2 (r) represents an inhomogeneous medium over a finite domain V , and k 2 = k02 , which
is constant outside V (see Figure 36.3). Next, we define a Green’s function satisfying

[∇2 + k02 ]g(r, r0 ) = −δ(r − r0 ). (36.3.2)

Then, (36.3.1) can be rewritten as

[∇2 + k02 ]φ(r) = Q(r) − [k 2 (r) − k02 ]φ(r). (36.3.3)

Note that the right-hand side of (36.3.3) can be considered an equivalent source. Since the
Green’s function corresponding to the differential operator on the left-hand side of (36.3.3) is
known, by the principle of linear superposition, we can write the formal solution to (36.3.3)
as
 
φ(r) = − dV 0 g(r, r0 )Q(r0 ) + dV 0 g(r, r0 )[k 2 (r0 ) − k02 ]φ(r0 ). (36.3.4)
Vs V

The first term on the right-hand side is just the field due to the source in the absence of
the inhomogeneity or the scatterer, and hence, is the incident field. The second term is a
Computational Electromagnetics, Numerical Methods 403

volume integral over the space where k 2 (r0 ) − k02 6= 0, or inside the inhomogeneous scatterer.
Therefore, (36.3.4) becomes


φ(r) = φinc (r) + dV 0 g(r, r0 )[k 2 (r0 ) − k02 ]φ(r0 ). (36.3.5)
V

In the above equation, if the total field φ(r0 ) inside the volume V is known, then φ(r)
can be calculated everywhere. But φ(r) is unknown at this point. To solve for φ(r), an
integral equation has to be formulated for φ(r). To this end, we imposed (36.3.5) for r in V .
Then, φ(r) on the left-hand side and on the right-hand side are the same unknown defined
over the same domain. Consequently, (36.3.5) becomes the desired integral equation after
rearrangement as


φinc (r) = φ(r) − dV 0 g(r, r0 )[k 2 (r0 ) − kb2 ]φ(r0 ), r ∈ V. (36.3.6)
V

In the above, the unknown φ(r) is defined over a volume V , over which the integration is
performed, and hence the name, volume integral equation. Alternatively, the above can be
rewritten as

φinc (r) = φ(r) − G(r, r0 )φ(r0 ), r ∈ V, (36.3.7)

where G is the integral operator in (36.3.6).1 It is also a Fredholm integral equation of

the second kind because the unknown is both inside and outside the integral operator. In the
above, integration over repeated variable r0 is implied. Nevertheless, it can be written as

Lf = g (36.3.8)

where L is a linear operator, while f represents the unknown function φ(r) and g is the
known function φinc (r).

1 Sometimes, this is called the kernel of the integral equation.

404 Electromagnetic Field Theory

36.3.2 Surface Integral Equation

Figure 36.4: Geometry for the derivation of the surface-integral equation for vector electro-
magnetics waves. (a) The original electromagnetics scattering problem. (b) The equivalent
electromagnetics problem by invoking equivalence principle.

The surface integral equation method is rather popular in a number of applications, because it
employs a homogeneous-medium Green’s function which is simple in form, and the unknowns
reside on a surface rather than in a volume.2
The surface integral equation for vector electromagnetic field can be derived using the
equivalence theorem also called the Love’s equivalence theorem [182]. Given a scattering
problem shown in Figure 36.4(a), it can be replaced by an equivalence problem as shown in
Figure 36.4(b). One can verify this by performing a Gedanken experiment as we have done
for the other equivalence problems discussed in Section 31.1.
In this figure, the total field outside the scatterer, E = Einc + Esca and H = Hinc + Hsca .
The impressed equivalence currents are given by Ms = E × n̂, and Js = n̂ × H. These
impressed currents, together generate the scattered fields outside the scatterer, while they
generate zero field inside the scatterer! One can verify that this is the case by performing
Gedanken experiments.
As such, the scattered fields outside the scatterer can be found from the radiation of the
impressed currents Ms and Js . Notice that these currents are radiating via the free-space
Green’s function because the scatterer has been removed in this equivalence problem. Now
that if the scatterer is a PEC, then the tangential component of the total electric field is zero
on the PEC surface. Therefore, Ms = 0 and only Js is radiating via the free-space Green’s
function.
Note that this equivalence problem is very different from that of an impressed currents
on the PEC scatterer as discussed in Section 31.2. There, only the magnetic surface current
2 These are sometimes called boundary integral equations method [221, 222].
Computational Electromagnetics, Numerical Methods 405

is radiating in the presence of the PEC, and the Green’s function is that of a current source
radiating in the presence of the PEC scatterer, and it is not the free-space Green’s function.
Now we can write the fields outside the scatterer using (31.4.17)
 
1 1
Esca (r) = ∇×∇× dS 0 g(r − r0 ) n̂0 ×H(r0 ) = ∇×∇× dS 0 g(r − r0 ) Js (r0 ) (36.3.9)
iω iω
S S

In the above, we have swapped r0 and r compared to (31.4.17). Also, we have kept only
the electric current Js (r) due to n̂ × H(r). If we impose the boundary condition that the
tangential component of the total electric field is zero, then we arrive at n̂ × Esca = −n̂ × Einc

1
−n̂ × Einc (r) = n̂ × ∇×∇× dS 0 g(r − r0 ) Js (r0 ). r∈S (36.3.10)
iω
S

In the above, n̂ × Einc (r) is known on the left hand side on the scatterer’s surface, while
the right-hand side has embedded in it the unknown surface current Js (r) = n̂ × H(r) on
the surface the scatter. Therefore, the above is an integral equation for the unknown surface
current Js . It can be written as a form of L f = g just like other linear operator equations.

36.4 Function as a Vector

Several linear operator equations have been derived in the previous sections. They are all of
the form
Lf = g (36.4.1)

In the above, f is a functional vector which is the analogy of the vector f in matrix theory
or linear algebra. In linear algebra, the vector f is of length N in an N dimensional space. It
can be indexed by a set of countable index, say i, and we can described such a vector with
N numbers such as fi , i = 1, . . . , N explicitly. This is shown in Figure 36.5(a).
A function f (x), however, can be thought of as being indexed by x in the 1D case.
However, the index in this case is a continuum, and countably infinite. Hence, it corresponds
to a vector of infinite dimension and it lives in an infinite dimensional space.3
To make such functions economical in storage, for instance, we replace the function f (x)
by its sampled values at N locations, such that f (xi ), i = 1, . . . , N . Then the values of the
function in between the stored points f (xi ) can be obtained by interpolation.4 Therefore, a
function vector f (x), even though it is infinite dimensional, can be approximated by a finite
length vector, f . This concept is illustrated in Figure 36.5(b) and (c). This concept can be
generalized to a function of 3D space f (r). If r is sampled over a 3D volume, it can provide
an index to a vector fi = f (ri ), and hence, f (r) can be thought of as a vector as well.
3 When these functions are square integrable implying finite “energy”, these infinite dimensional spaces are

called Hilbert spaces.

4 This is in fact how special functions like sin(x), cos(x), exp(x), J (x), N (x), etc, are computed and
n n
stored in modern computers.
406 Electromagnetic Field Theory

Figure 36.5: A function can be thought of as a vector. (a) A continuum function f (x) plotted
as a function of x. (b) A digitally sampled values of the same function. (c) When stored in
a computer, it will be stored as an array vector.

36.5 Operator as a Map

36.5.1 Domain and Range Spaces

An operator like L above can be thought of as a map or a transformation. It maps a function

f defined in a Hilbert space V to g defined in another Hilbert space W . Mathematically, this
is written as

L :V →W (36.5.1)

indicating that L is a map of vectors in the space V to vectors in the space W . Here, V is
also called the domain space (or domain) of L while W is the range space (or range) of
L.
Computational Electromagnetics, Numerical Methods 407

36.6 Approximating Operator Equations with Matrix Equa-

tions
36.6.1 Subspace Projection Methods
One main task of numerical method is first to approximate an operator equation L f = g by
a matrix equation L · f = g. To achieve the above, we first let
N
f∼
X
= an fn = g (36.6.1)
n=1

In the above, fn , n, . . . , N are known functions called basis functions. Now, an ’s are the
new unknowns to be sought. Also the above is an approximation, and the accuracy of the
approximation depends very much on the original function f . A set of very popular basis
functions are functions that form a piece-wise linear interpolation of the function from its
nodes. These basis functions are shown in Figure 36.6.

Figure 36.6: Examples of basis function in (a) one dimension, (b) two dimension. Each of
these functions are define over a finite domain. Hence, they are also called sub-domain basis
functions. They can be thought of as interpolatory functions.

Upon substituting (36.6.1) into (36.4.1), we obtain

N
X
an L fn = g (36.6.2)
n=1

Then, upon multiplying (36.6.2) by wm and integrating over the space that wm (r) is defined,
408 Electromagnetic Field Theory

then we have
N
X
an hwm , L fn i = hwm , gi , m = 1, . . . , N (36.6.3)
n=1
In the above, the inner product is defined as

hf1 , f2 i = drf1 (r)f2 (r) (36.6.4)

where the integration is over the support of the functions, or the space over which the functions
are defined.5 For PDEs these functions are defined over a 3D coordinate space, while in SIEs,
these functions are defined over a surface. In a 1D problems, these functions are defined over
a 1D coordinate space.

36.6.2 Dual Spaces

The functions wm , m = 1, . . . , N is known as the weighting functions or testing functions.
The testing functions should be chosen so that they can approximate well a function that
lives in the range space W of the operator L . Such set of testing functions lives in the dual
space of the range space. For example, if fr lives in the range space of the operator L , the
set of function fd , such that the inner product hfd , fr i exists, forms the dual space of W .

36.6.3 Matrix and Vector Representations

The above equation (36.6.3) is a matrix equation of the form
L·a=g (36.6.5)
where
L mn = hwm , L fn i
 
(36.6.6)
[a]n = an , [g]m = hwm , gi
What has effectively happened here is that given an operator L that maps a function that lives
in an infinite dimensional Hilbert space V , to another function that lives in another infinite
dimensional Hilbert space W , via the operator equation L f = g, we have approximated
the Hilbert spaces with finite dimensional spaces (subspaces), and finally, obtain a finite
dimensional matrix equation that is the representation of the original infinite dimensional
operator equation. This is the spirit of the subspace projection method.
In the above, L is the matrix representation of the operator L in the subspaces, and a
and g are the vector representations of f and g, respectively, in their respective subspaces.
When such a method is applied to integral equations, it is usually called the method
of moments (MOM). (Surface integral equations are also called boundary integral equations
(BIEs) in other fields [222].) When finite discrete basis are used to represent the surface
unknowns, it is also called the boundary element method (BEM) [223]. But when this method
is applied to solve PDEs, it is called the finite element method (FEM) [224–227], which is a
rather popular method due to its simplicity.
5 This is known as the reaction inner product [35, 50, 131]. As oppose to most math and physics literature,


the energy inner product is used [131] where hf1 , f2 i = drf1∗ (r)f2 (r).
Computational Electromagnetics, Numerical Methods 409

36.6.4 Mesh Generation

In order to approximate a function defined on an arbitrary shaped surface or volume by a

finite sum of basis functions, it is best to mesh (tessellate or discretize) the surface and volume
by meshes. In 2D, all shapes can be tessellated by unions of triangles, while a 3D volume
can be meshed (tessellated) by unions of tetrahedrons. Such meshes are used not only in
CEM, but in other fields such as solid mechanics. Hence, there are many “solid modeling”
commercial software available to generate sophisticated meshes.
When a surface is curved, or of arbitrary shape, it can be meshed by union of triangles as
shown in Figure 36.7. When a volume is of arbitrary shape or a volume is around an arbitrary
shape object, it can be meshed by tetrahedrons as shown in Figure 36.8. Then basis functions
as used in (36.6.1) are defined to interpolate the field between nodal values or values defined
on the edges of a triangle or a tetrahedron.

Figure 36.7: An arbitrary surface can be meshed by a union of triangles.

Figure 36.8: A volume region can be meshed by a union of tetrahedra. But the surface of the
aircraft is meshed with a union of trianlges (courtesy of gmsh.info).
410 Electromagnetic Field Theory

36.6.5 Differential Equation Solvers versus Integral Equation Solvers

As have been shown, the two classes of numerical solvers for Maxwell’s equations consist of
differential equation solvers and integral equation solvers. Differential equation solvers are
generally easier to implement. As shall be shown in the next lecture, they can also be easily
implemented using finite difference solver. The unknowns in a differential equation solver
are the fields. The fields permeate all of space, and hence, the unknowns are volumetrically
distributed. When the fields are digitized by representing them by their point values in space,
they require a large number of unknowns to represent. The plus side is that the matrix system
associated with a differential equation solver is usually sparse, requiring less storage and less
time to solve.
As has been shown, integral equation solvers are formulated using Green’s functions. That
is integral equations are derived from Maxwell’s equations using Green’s function, where the
unknowns now are surface sources such as surface electric and magnetic currents. Therefore,
the unknowns are generally smaller, living only on the surface of a scatterer (or they occupy
a smaller part of space). Hence, they can be approximated by a smaller set of unknowns.
Thus, the matrix systems generally are smaller. Once the currents are found, then the fields
they generate can also be computed.
Since the derivation of integral equations requires the use of Green’s functions, they are
in general singular when r = r0 , or when the observation point (observation point) r and the
source point r0 coincide. Care has to be taken to discretize the integral equations. Moreover,
a Green’s function connects every current source point on the surface of a scatterer with every
other source points yielding a dense matrix system. But fast methods have been developed
to solve such dense matrix systems [9].

36.7 Solving Matrix Equation by Optimization

Given a matrix equation, there are many ways to seek its solution. The simplest way is to find
the inverse of the matrix operator by direct inversions (e.g., using Gaussian elimination [228]
or Kramer’s rule [229]) have computational complexity6 of O(N 3 ), and requiring storage of
O(N 2 ). Due to the poor computational and memory complexity of direct inversion, when N
is large, other methods have to be sought.
To this end, it is better to convert the solving of a matrix equation into an optimization
problem. These methods can be designed so that a much larger system can be solved with an
existing resource of a digital computer. Optimization problem results in finding the stationary
point of a functional.7 First, we will figure out how to find such a functional.
Consider a matrix equation given by

L·f =g (36.7.1)

6 The scaling of computer time with respect to the number of unknowns (degrees of freedom) is known in

the computer parlance as computational complexity.

7 Functional is usually defined as a function of a function [35, 49]. Here, we include a function of a vector

to be a functional as well.
Computational Electromagnetics, Numerical Methods 411

For simplicity, we consider L as a symmetric matrix.8 Then the corresponding functional is

I = f t · L · f − 2f t · g (36.7.2)

Such a functional is called a quadratic functional because it is analogous to I = Lx2 − 2xg,

which is quadratic, in its simplest 1D rendition.
Taking the first variation with respect to f , namely, we let f = fo + δf . Then we substitute
this into the above, and collect the leading order and first order terms. Then we find the first
order approximation of the functional I as

δI = δf t · L · fo + fot · L · δf − 2δf t · g (36.7.3)

If L is symmetric, the first two terms are the same, and the above becomes

δI = 2δf t · L · fo − 2δf t · g (36.7.4)

For fo to be the optimal point or the stationary point, then its first variation has to be zero,
or that δI = 0. Thus we conclude that at the optimal point (or the stationary point),

L · fo = g (36.7.5)

Hence, the optimal point to the functional I in (36.7.2) is the solution to (36.7.1) or (36.7.5).

36.7.1 Gradient of a Functional

The above method, when applied to an infinite dimensional Hilbert space problem, is called
variational method, but the main ideas are similar. The wonderful idea about such a method
is that instead of doing direct inversion of a matrix system (which is expensive), one can
search for the optimal point or stationary point of the quadratic functional using gradient
search or gradient descent methods or some optimization method.
It turns out that the gradient of a quadratic functional can be found quite easily. Also
it is cheaper to compute the gradient of a functional than to find the inverse of a matrix
operator. To do this, it is better to write out functional using index (or indicial, or Einstein)
notation [230]. In this notation, the functional first variation δI in (36.7.4) becomes

δI = 2δfj Lij fi − 2δfj gj (36.7.6)

Also, in this notation, the summation symbol is dropped, and summations over repeated
indices are implied. In the above, we neglect to distinguish between fo and f . It is implied
that f represents the optimal point. In this notation, it is easier to see what a functional
derivative is. We can differentiate the above with respect to fj easily to arrive at

∂I
= 2Lij fi − 2gj (36.7.7)
∂fj
8 Functional for the asymmetric case can be found in Chew, Waves and Fields in Inhomogeneous Media,

Chapter 5 [35].
412 Electromagnetic Field Theory

Notice that the remaining equation has one index j remaining in index notation, meaning
that it is a vector equation. We can reconstitute the above using our more familiar matrix
notation that

δI
= ∇f I = 2L · f − 2g (36.7.8)
δf

The left-hand side is a notation for the functional derivative or the gradient of a functional in
a multi-dimensional space whichi is a vector obviated by indicial notation. And the right-hand
side is the expression for calculating this gradient. One needs only to perform a matrix-vector
product to find this gradient. Hence, the computational complexity of finding this gradient
is O(N 2 ) at worst if L is a dense matrix, and as low as O(N ) if L is a sparse matrix.9 In
a gradient search method, such a gradient is calculated repeatedly until the optimal point is
found. Such methods are called iterative methods.

If the optimal point can be found in Niter iterations, then the CPU time scales as Niter N α
where 1 < α < 2. There is a clever gradient search algorithm, called the conjugate gradient
method that can find the exact optimal point in Niter = N in exact arithmetics. But exact
solution is not needed in an optimal solution: an approximate solution suffices. In many
gradient search solutions, to obtain an approximate solution where the error is acceptable,
Niter  N . The total solution time or solve time which is Niter N α  N N α  N 3 , resulting
in great savings in computer time, especially if α = 1. This is the case for FEM [227], [231],
[232], [233], [226], and fast multipole algorithm [234], [235].

What is more important is that this method does not require the storage of the matrix L,
but a computer code that produces the vector go = L · f as an output, with f as an input.
Both f and go require only O(N ) memory storage. Such methods are called matrix-free
methods. Even when L is a dense matrix, which is the case if it is the matrix representation
of matrix representation of some Green’s function, fast methods now exist to perform the
dense matrix-vector product in O(N log N ) operations.10

The value I is also called the cost function, and its minimum is sought in the seeking of
the solution by gradient search methods. Detail discussion of these methods is given in [236].
Figure 36.9 shows the contour plot of a cost function in 2D. When the condition number11
of the matrix L is large (implying that the matrix is ill-conditioned), the contour plot will
resemble a deep valley. And hence, the gradient search method will tend to zig-zag along the
way as it finds the optimal solution. Therefore, convergence is slow for matrices with large
condition numbers

9 This is the case for many differential equation solvers such as finite-element method or finite-difference

method.
10 Chew et al, Fast and Efficient Algorithms in CEM [9].
11 This is the ratio of the largest eigenvalue of the matrix to its smallest eigenvalue.
Computational Electromagnetics, Numerical Methods 413

Figure 36.9: Plot of a 2D cost function, I(x, y) for an ill-conditioned system (courtesy of
Numerical Recipe [236]). A higher dimensional plot of this cost function will be difficult.

Figure 36.10 shows a cartoon picture in 2D of the histories of different search paths from a
machine-learning example where a cost functional similar to I has to be minimized. Finding
the optimal point or the minimum point of a general functional is still a hot topic of research:
it is important in artificial intelligence.

Figure 36.10: Gradient search or gradient descent method is finding an optimal point (courtesy
of Y. Ioannou: https://blog.yani.io/sgd/).
414 Electromagnetic Field Theory
Lecture 37

Finite Difference Method, Yee

Algorithm

In this lecture, we are going to introduce one of the simplest methods to solve Maxwell’s
equations numerically. This is the finite-difference time-domain method first proposed by
Yee [237] and popularized by Taflove [238]. Because of its simplicity, and that a simple
Maxwell solver can be coded in one afternoon, almost every physics or electrical engineering
laboratory has a home-grown version of the finite-difference time-domain solver. This method
is the epitome of that “simplicity rules.”1 Professor Hermann Haus at MIT used to say: find
the simplest method to do things. Complicated methods will be forgotten, but the simplest
method will prevail. This is also reminiscent of Einstein’s saying, “Everything should be made
as simple as possible, but no simpler!”

37.1 Finite-Difference Time-Domain Method

To obtain the transient (time-domain) solution of the wave equation for a more general,
inhomogeneous medium, a numerical method has to be used. The finite-difference time-
domain (FDTD) method, a numerical method, is particularly suitable for solving transient
problems. Moreover, it is quite versatile, and given the present computer technology, it has
been used with great success in solving many practical problems. This method is based on a
simple Yee algorithm [237] and has been vastly popularized by Taflove [238, 239].
In the finite-difference method, continuous space-time is replaced with a discrete space-
time. Then, in the discrete space-time, partial differential equations are replaced with finite
difference equations. These finite difference equations are readily implemented on a digital
computer. Furthermore, an iterative or time-stepping scheme can be implemented without
having to solve large matrices, resulting in great savings in computer time. Moreover, the
matrix for the system of equations is never generated making this a matrix-free method.
There is no need for matrix management as one writes this numerical solver. More recently,
1 “rule” is used as a verb.

415
416 Electromagnetic Field Theory

the development of parallel processor architectures in computers has also further enhanced
the efficiency of the finite-difference time-domain scheme [240].

The finite-difference method is also described in numerous works (see, for example, Potter
1973 [241]; Taflove 1988 [238]; Ames 2014 [242]; and Morton 2019 [243].

37.1.1 The Finite-Difference Approximation

Consider first a scalar wave equation of the form

1 ∂2
φ(r, t) = µ(r)∇ · µ−1 (r)∇φ(r, t). (37.1.1)
c2 (r) ∂t2

The above equation appears in scalar acoustic waves or a 2D electromagnetic waves in inho-
mogeneous media [35, 244].

To convert the above into a form that can be solved by a digital computer easily, first,
one needs to find finite-difference approximations to the time derivatives. Then, the time
derivative can be approximated in many ways. For example, a derivative can be approximated
by forward, backward, and central finite difference formulas [245].

∂φ(r, t) φ(r, t + ∆t) − φ(r, t)

Forward difference: ≈ , (37.1.2)
∂t ∆t
∂φ(r, t) φ(r, t) − φ(r, t − ∆t)
Backward difference: ≈ , (37.1.3)
∂t ∆t
∂φ(r, t) φ(r, t + ∆t
2 ) − φ(r, t −
∆t
2 )
Central difference: ≈ , (37.1.4)
∂t ∆t
Finite Difference Method, Yee Algorithm 417

Figure 37.1: Different finite-difference approximations for the time derivative. One can eye-
ball that the central difference formula is the best. This can be further confirmed by a Taylor
series analysis.

where ∆t is a small number. Of the three methods of approximating the time derivative, the
central-difference scheme is the best approximation, as is evident from Figure 37.1. The errors
in the forward and backward differences are O(∆t) (or first-order error) while the central-
difference approximation has an error O[(∆t)2 ] (or second-order error). This can be easily
verified by Taylor series expanding the right-hand sides of (37.1.2) to (37.1.4).
Consequently, using the central-difference formula twice, we arrive at
" #
∂2 ∂ φ(r, t + ∆t ∆t
2 ) − φ(r, t − 2 )
φ(r, t) ≈ (37.1.5)
∂t2 ∂t ∆t
φ(r, t + ∆t) − 2φ(r, t) + φ(r, t − ∆t)
≈ . (37.1.6)
(∆t)2
418 Electromagnetic Field Theory

Next, if the function φ(r, t) is indexed on discrete time steps on the t axis, such that for
t = l∆t, then φ(r, t) = φ(r, l∆t) = φl (r), where l is an integer is used to count the time steps.
Using this notation, Equation (37.1.6) then becomes

∂2 φl+1 (r) − 2φl (r) + φl−1 (r)

φ(r, t) ≈ . (37.1.7)
∂t2 (∆t)2

37.1.2 Time Stepping or Time Marching

With this notation and approximations, (37.1.1) can be approximated by a time-stepping (or
time-marching) formula, namely,

φl+1 (r) ∼
= c2 (r)(∆t)2 µ(r)∇ · µ−1 (r)∇φl (r) + 2φl (r) − φl−1 (r). (37.1.8)

Therefore, given the knowledge of φ(r, t) at t = l∆t and t = (l − 1)∆t for all r, one can deduce
φ(r, t) at t = (l + 1)∆t. In other words, given the initial values of φ(r, t) at, for example, t = 0
and t = ∆t, φ(r, t) can be deduced for all subsequent times, provided that the time-stepping
formula is stable.
At this point, the right-hand side of (37.1.8) involves the space derivatives. There exist
a plethora of ways to approximate and calculate the right-hand side of (37.1.8) numerically.
Here, we shall illustrate again the use of the finite-difference method to calculate the right-
hand side of (37.1.8). Before proceeding further, note that the space derivatives on the
right-hand side in cartesian coordinates are
∂ −1 ∂ ∂ ∂ ∂ ∂
µ(r)∇ · µ−1 (r)∇φ(r) = µ µ φ + µ µ−1 φ + µ µ−1 φ. (37.1.9)
∂x ∂x ∂y ∂y ∂z ∂z
Then, one can approximate, using central differencing that
    
∂ 1 ∆z ∆z
φ(x, y, z) ≈ φ x, y, z + − φ x, y, z − , (37.1.10)
∂z ∆z 2 2
Consequently, using central differencing two times,
  
∂ −1 ∂ 1 −1 ∆z
µ φ(x, y, z) ≈ µ z + φ(x, y, z + ∆z)
∂z ∂z (∆z)2 2
    
∆z ∆z
− µ−1 z + + µ−1 z − φ(x, y, z)
2 2
  
∆z
+µ−1 z − φ(x, y, z − ∆z) . (37.1.11)
2
Furthermore, after denoting φ(x, y, z) = φm,n,p , µ(x, y, z) = µm,n,p , on a discretized grid
point at x = m∆x, y = n∆y, z = p∆z, we have (x, y, z) = (m∆x, n∆y, p∆z), and then
∂ −1 ∂ 1 h −1
µ φ(x, y, z) ≈ µ 1 φm,n,p+1
∂z ∂z (∆z)2 m,n,p+ 2
  i
− µ−1
m,n,p+ 1 + µ
−1
m,n,p− 1 φ m,n,p + µ−1
m,n,p− 1 φm,n,p−1 . (37.1.12)
2 2 2
Finite Difference Method, Yee Algorithm 419

This cumbersome and laborious looking equation can be abbreviated if we define a central
difference operator as2
1  
∂¯z φm = φm+ 12 − φm− 12 (37.1.13)
∆z
Then the right-hand side of the (37.1.12) can be written succinctly as
∂ −1 ∂
µ φ(x, y, z) ≈ ∂¯z µm,n,p ∂¯z φm,n,p (37.1.14)
∂z ∂z
With similar approximations to the other terms in (37.1.9), (37.1.8) is now compactly written
as
φl+1 2 2 ¯ ¯ ¯ ¯ ¯ ¯
 
m,n,p = (∆t) cm,n,p µm,n,p ∂x µm,n,p ∂x + ∂y µm,n,p ∂y + ∂z µm,n,p ∂z φm,n,p

+ 2φlm,n,p − φl−1
m,n,p . (37.1.15)
The above can be readily implemented on a computer for time stepping. Notice however,
that the use of central differencing results in the evaluation of medium property µ at half
grid points. This is inconvenient, as the introduction of material values at half grid points
increases computer memory. Hence, it is customary to store the medium property at the
integer grid points for ease of book-keeping, and to approximate
1
µm+ 12 ,n,p ' (µm+1,n,p + µm,n,p ), (37.1.16)
2

µm+ 21 ,n,p + µm− 21 ,n,p ' 2µm,n,p , (37.1.17)

and so on. Moreover, if µ is a smooth function of space, it is easy to show that the errors in
the above approximations are of second order by Taylor series expansions.
For a homogeneous medium, with ∆x = ∆y = ∆z = ∆s, namely, we assume the space
steps to be equal in all directions. (37.1.15) written explicitly becomes
 2
l+1 ∆t
c2 φlm+1,n,p + φlm−1,n,p + φlm,n+1,p + φlm,n−1,p + φlm,n,p+1

φm,n,p =
∆s
+ φlm,n,p−1 − 6φlm,n,p + 2φlm,n,p − φl−1

m,n,p . (37.1.18)

Notice then that with the central-difference approximation, the value of φl+1 m,n,p is dependent
only on φlm,n,p , and its nearest neighbors, φlm±1,n,p , φlm,n±1,p , φlm,n,p±1 , and φl−1
m,n,p , its value
at the previous time step. Moreover, in the finite-difference scheme outlined above, no matrix
inversion is required at each time step. Such a scheme is also known as an explicit scheme.
The use of an explicit scheme is a major advantage of the finite-difference method compared
to the finite-element methods. Consequently, in order to update N grid points using (37.1.15)
or (37.1.18), O(N ) multiplications are required for each time step. In comparison, O(N 3 )
multiplications are required to invert an N × N full matrix, e.g., using Gaussian elimination.
The simplicity and efficiency of these finite-difference algorithms have made them vastly
popular.
2 This is in the spirit of [246].
420 Electromagnetic Field Theory

37.1.3 Stability Analysis

The implementation of the finite-difference time-domain scheme using time-marching does
not always lead to a stable scheme. Hence, in order for the solution to converge, the time-
stepping scheme must at least be stable. Consequently, it is useful to find the condition
under which a numerical finite-difference scheme is stable. To do this, one performs the von
Neumann stability analysis (von Neumann 1943 [247]) on Equation (37.1.18). We will assume
the medium to be homogeneous to simplify the analysis.
As shown previously in Section 35.1, a point source gives rise to a spherical wave that
can be expanded in terms of sum of plane waves in different directions. It also implies that
any wave emerging from sources can be expanded in terms of sum of plane waves. This is
the spirit of the spectral expansion method. So if a scheme is not stable for a plane wave, it
would not be stable for any wave. Consequently, to perform the stability analysis, we assume
a propagating plane wave as a trial solution

φ(x, y, z, t) = A(t)eikx x+iky y+ikz z , (37.1.19)

In discretized form, by letting ∆x = ∆y = ∆z = ∆s, it is just

φlm,n,p = Al eikx m∆s+iky n∆s+ikz p∆s . (37.1.20)

We can imagine that Al = A0 e−iωl∆t , so that the above is actually a Fourier plane wave mode
in the frequency domain. Using (37.1.20), it is easy to show that for the x space derivative,

φlm+1,n,p − 2φlm,n,p + φlm−1,n,p = 2[cos(kx ∆s) − 1]φlm,n,p

 
2 kx ∆s
= −4 sin φlm,n,p . (37.1.21)
2

The space derivatives in y and z directions can be similarly derived.

The time derivative can be similarly approximated, and it is approximately equal to

∂2
φ(r, t)(∆t)2 ≈ φl+1 l l−1
m,n,p − 2φm,n,p + φm,n,p . (37.1.22)
∂t2
Substituting (37.1.20) into the above, we have the second time derivative being proportional
to
∂2
φ(r, t)(∆t)2 ≈ (Al+1 − 2Al + Al−1 )eikx m∆s+iky n∆s+ikz p∆s (37.1.23)
∂t2
To simplify further, one can assume that

Al+1 = gAl . (37.1.24)

This is commensurate with assuming that

A(t) = A0 e−iωt (37.1.25)

Finite Difference Method, Yee Algorithm 421

where ω can be complex. In other words, our trial solution (37.1.19) is also a time-harmonic
signal where ω can be complex. If the finite-difference scheme is unstable for such a signal, it
is unstable for all signals.
Consequently, the time derivative is proportional to
∂2
φ(r, t)(∆t)2 ≈ (g − 2 + g −1 )φlm,n,p (37.1.26)
∂t2
We need to find the value of g for which the solution (37.1.20) satisfies (37.1.18). To this end,
one uses (37.1.21) and (37.1.24) in (37.1.18), and repeating (37.1.21), which is for m variable
in the x direction, for the n and p variables in the x and y directions as well, one obtains
 2     
∆t kx ∆s ky ∆s
(g − 2 + g −1 )φlm,n,p = −4 c2 sin2 + sin2
∆s 2 2
 
k z ∆s
+ sin2 φlm,n,p
2
= −4r2 s2 φlm,n,p , (37.1.27)

where
       
∆t 2 2 kx ∆s 2 ky ∆s 2 kz ∆s
r= c, s = sin + sin + sin . (37.1.28)
∆s 2 2 2

Equation (37.1.27) implies that, for nonzero φlm,n,p ,3

g 2 − 2g + 4r2 s2 g + 1 = 0, (37.1.29)

or that
p
g = (1 − 2r2 s2 ) ± 2rs (r2 s2 − 1) . (37.1.30)

In order for the solution to be stable, it is necessary that |g| ≤ 1. But if

r2 s2 < 1, (37.1.31)

the second term in (37.1.30) is pure imaginary, and

|g|2 = (1 − 2r2 s2 )2 + 4r2 s2 (1 − r2 s2 ) = 1, (37.1.32)

when (37.1.31) is true. Therefore, stability is ensured. Since from (37.1.28), s2 ≤ 3 for all kx ,
ky , and kz , from (37.1.31). From (37.1.31), we conclude that
1
r<
s
But we also know that
1 1
≥√
s 3
3 For those who are more mathematically inclined, we are solving an eigenvalue problem in disguise. Re-

member that a function is a vector, even after it has been discretized:)

422 Electromagnetic Field Theory

The above two inequalities will be satisfied if the general condition is

1 ∆s
r< √ , or ∆t < √ . (37.1.33)
3 c 3

The above is the general condition for stability. The above analysis is for 3 dimensional
problems. It is clear from the above analysis that for an n-dimensional problem where n =
1, 2, 3, then

∆s
∆t < √ . (37.1.34)
c n

One may ponder on the meaning of this inequality further: but it is only natural that
the time step ∆t has to be bounded from above. Otherwise, one arrives at the ludicrous
notion that the time step can be arbitrarily large thus violating causality. Moreover, if the
grid points of√the finite-difference scheme are regarded as a simple cubic lattice, then the
distance ∆s/ n is also the distance between the closest lattice planes through the simple
cubic lattice.
√ Notice that the time for the wave to travel between these two lattice planes is
∆s/(c n ). Consequently, the stability criterion (37.1.34) implies that the time step ∆t has to
be less than the shortest travel time for the wave between the lattice planes in order to satisfy
causality. In other words, if the wave is time-stepped ahead of the time on the right-hand side
of (37.1.34), instability ensues. The above is also known as the CFL (Courant, Friedrichs,
and Lewy 1928 [248]) stability criterion. It could be easily modified for ∆x 6= ∆y 6= ∆z [239].
The above analysis implies that we can pick a larger time step if the space steps are
larger. A larger time step will allow one to complete generating a time-domain response
rapidly. However, one cannot arbitrary make the space step large due to grid-dispersion
error, as shall be discussed next.

37.1.4 Grid-Dispersion Error

When a finite-difference scheme is stable, it still may not be accurate to produce good results
due to the errors in the finite-difference approximations. Hence, it is useful to ascertain the
errors in terms of the size of the grid and the time step. An easy error to analyze is the grid-
dispersion error . In a homogeneous, dispersionless medium, all plane waves propagate with
the same phase velocity. However, in the finite-difference approximation, all plane waves will
not propagate at the same phase velocity due to the grid-dispersion error.
As a consequence, a pulse in the time domain, which is a linear superposition of plane waves
with different frequencies, will be distorted if the dispersion introduced by the finite-difference
scheme is intolerable. Therefore, to make things simpler, we will analyze the grid-dispersion
error in a homogeneous free space medium.
To ascertain the grid-dispersion error, we assume that the solution is time-harmonic, or
that Al = A0 e−iωl∆t in (37.1.20). In this case, the left-hand side of (37.1.27) becomes
 
−iω∆t +iω∆t 2 ω∆t
φlm,n,p φlm,n,p .


e −2+e = −4 sin (37.1.35)
2
Finite Difference Method, Yee Algorithm 423

Then, from (37.1.27), it follows that

 
ω∆t
sin = rs, (37.1.36)
2
where r and s are given in (37.1.28). Now, (37.1.36) governs the relationship between ω
and kx , ky , and kz in the finite-difference scheme, and hence, is a dispersion relation for the
approximate solution.
The above gives a rather complicated relationship between the frequency ω and the wave
numbers kx , ky , and kz . This is the result of the finite-difference approximation of the scalar
wave equation. As a sanity check, when the space and time discretizations become very
small, we should recover the dispersion relation of homogeneous medium or free space. But
if a medium is homogeneous, it is well known that (37.1.1) has a plane-wave solution of the
type given by (37.1.19) where
q
ω = c kx2 + ky2 + kz2 = c|k| = ck. (37.1.37)

where k = x̂kx + ŷky + ẑkz is the direction of propagation of the plane wave. Defining the
phase velocity to be ω/k = c, this phase velocity is isotropic, or the same in all directions.
Moreover, it is independent of frequency. But in (37.1.36), because of the definition of s as
given by (37.1.28), the dispersion relation between ω and k is not isotropic. This implies that
plane waves propagating in different directions will have different phase velocities.
Equation (37.1.36) is the dispersion relation for the approximate solution. It departs from
Equation (37.1.37), the exact dispersion relation, as a consequence of the finite-difference ap-
proximation. This departure gives rise to errors, which are the consequence of grid dispersion.
For example, when c is a constant, (37.1.37) states that the phase velocities of plane waves
of different wavelengths and directions are the same. However, this is not true for (37.1.36),
as shall be shown.
Assuming s small, (37.1.36), after using Taylor series expansion, can be written as

ω∆t r3 s3
= sin−1 rs ∼
= rs + . (37.1.38)
2 6
When ∆s is small, using the small argument approximation for the sine function, one obtains
from (37.1.28)
∆s 2
s' (k + ky2 + kz2 )1/2 (37.1.39)
2 x
Equation (37.1.38), by taking the higher-order Taylor expansion of (37.1.38), then becomes
ω∆t ∆s 2
'r (k + ky2 + kz2 )1/2 [1 − δ] (37.1.40)
2 2 x
where (see [35])

∆s2 kx4 + ky4 + kz4 r2 ∆s2 2

δ= 2 2 2
− (kx + ky2 + kz2 ) (37.1.41)
24 kx + ky + kz 24
424 Electromagnetic Field Theory

From the above, one can see that if δ = 0, we retrieve the dispersion relation of the homoge-
neous free-space medium. So δ is a measure of the departure of the dispersion relation from
that of free space due to our finite-difference approximation. A soothing fact is that when
∆s  1, δ is small.
Since k is inversely proportional to wavelength λ, then δ in the correction to the above
equation is proportional to ∆s2 /λ2 . Therefore, to reduce the grid dispersion error, it is
necessary for δ to be small or to have
 2
∆s
 1. (37.1.42)
λ

When this is true, using the fact that r = c∆t/∆s, then (37.1.40) becomes

ω q 2
≈ kx + ky2 + kz2 . (37.1.43)
c
which is close to the dispersion relation of free space as indicated in (37.1.37). Consequently, in
order for the finite-difference scheme to propagate a certain frequency content accurately, the
grid size must be much less than the wavelength of the corresponding frequency. Furthermore,
∆t must be chosen so that the CFL stability criterion is met. Therefore, the rule of thumb is
to choose about 10 to 20 grid points per wavelength. Also, for a plane wave propagating as
eik·r , an error δk in the vector k gives rise to cumulative error eiδk·r . The larger the distance
traveled, the larger the cumulative phase error, and hence the grid size must be smaller in
order to arrest such phase error due to the grid dispersion.

37.2 The Yee Algorithm

The Yee algorithm (Yee 1966 [237]) is a simple algorithm specially designed to solve vec-
tor electromagnetic field problems on a rectilinear grid. The finite-difference time-domain
(FDTD) method (Taflov 1988) when applied to solving electromagnetics problems, usually
uses this method. To derive it, Maxwell’s equations are first written in cartesian coordinates:

∂Bx ∂Ez ∂Ey

− = − , (37.2.1)
∂t ∂y ∂z
∂By ∂Ex ∂Ez
− = − , (37.2.2)
∂t ∂z ∂x
∂Bz ∂Ey ∂Ex
− = − , (37.2.3)
∂t ∂x ∂y
∂Dx ∂Hz ∂Hy
= − − Jx , (37.2.4)
∂t ∂y ∂z
∂Dy ∂Hx ∂Hz
= − − Jy , (37.2.5)
∂t ∂z ∂x
∂Dz ∂Hy ∂Hx
= − − Jz . (37.2.6)
∂t ∂x ∂y
Finite Difference Method, Yee Algorithm 425

Before proceeding any further, it is prudent to rewrite the differential equation form of
Maxwell’s equations in their integral form. The first equation above can be rewritten as
 
∂
− Bx dS = E · dl (37.2.7)
∂t ∆S ∆C

where ∆S = ∆x∆z. The approximation of this integral form will be applied to the face that
is closest to the observer in Figure 37.2. Hence, one can see that the curl of E is proportional
to the time-derivative of the magnetic flux through the suface enclosed by ∆C, which is ∆S.
One can see this relationship for the other surfaces of the cube in the figure as well: the
electric field is curling around the magnetic flux. For the second half of the above equations,
one can see that the magnetic fields are curling around the electric flux, but on a staggered
grid. These two staggered grids are intertwined with respect to each other. This is the spirit
with which the Yee algorithm is written. He was apparently motivated by fluid dynamics
when he did the work.

Figure 37.2: The assignment of fields on a grid in the Yee algorithm.

l
After denoting f (m∆x, n∆y, p∆z, l∆t) = fm,n,p , and replacing derivatives with central
finite-differences in accordance with Figure 37.2, (37.2.1) becomes
1 h l+ 21 l− 12
i 1 h l l
i
Bx,m,n+ 1 ,p+ 1 − Bx,m,n+ 1 1 = E y,m,n+ 1
,p+1 − E y,m,n+ 1
,p
∆t 2 2 2 ,p+ 2 ∆z 2 2

1 h l l
i
− Ez,m,n+1,p+ 1 − Ez,m,n,p+ 1 . (37.2.8)
∆y 2 2

where the above formula is evaluated at t = l∆t. Moreover, the above can be repeated for
(37.2.2) and (37.2.3). Notice that in Figure 37.2, the electric field is always assigned to the
edge center of a cube, whereas the magnetic field is always assigned to the face center of a
cube.4
4 This algorithm is intimately related to differential forms which has given rise to the area of discrete

exterior calculus [249].

426 Electromagnetic Field Theory

In fact, after multiplying (37.2.8) by ∆z∆y, (37.2.8) is also the approximation of the
integral forms of Maxwell’s equations when applied at a face of a cube. By doing so, the
left-hand side of (37.2.8), by (37.2.7), becomes

l+ 12 l− 21
h i
(∆y∆z/∆t) Bx,m,n+ 1 1 − B 1 1 , (37.2.9)
,p+ 2 x,m,n+ ,p+
2 2 2

which is the time variation of the total flux through an elemental area ∆y∆z. Moreover, by
summing this flux on the six faces of the cube shown in Figure 37.2, and using the right-
hand side of (37.2.8) and its equivalent, it can be shown that the magnetic flux adds up to
∂
zero. Hence, ∂t ∇ · B = 0 condition is satisfied within the numerical approximations of Yee
algorithm. The above shows that if the initial value implies that ∇ · B = 0, the algorithm
will preserve this condition. So even though we are solving Faraday’s law, Gauss’ law is also
satisfied. This is important in maintaining the stability of the numerical algorithm [250].

Furthermore, a similar approximation of (37.2.4) leads to

1 h l l−1
i 1 h l− 12 l− 21
i
Dx,m+ 1 ,n,p − Dx,m+ 1
,n,p
= Hz,m+ 1 ,n+ 1 ,p − Hz,m+ 1
,n− 1
,p
∆t 2 2 ∆y 2 2 2 2

1 h l− 12 l− 21
i
l− 12
− Hy,m+ 1 ,n,p+ 1 − Hy,m+ 1 ,n,p− 1 − Jx,m+ 1 ,n,p . (37.2.10)
∆z 2 2 2 2 2

Also, similar approximations apply for (37.2.5) and (37.2.6). In addition, the above has an
interpretation similar to (37.2.8) if one thinks in terms of a cube that is shifted by half a grid
point in each direction. Hence, the approximations of (37.2.4) to (37.2.6) are consistent with
∂
the approximation of ∂t ∇ · D = −∇ · J. This way of alternatively solving for the B and D
fields in tandem while the fields are placed on a staggered grid is also called the leap-frog
scheme.

In the above, D = E and B = µH. Since the magnetic field and the electric field are
assigned on staggered grids, µ and  may have to be assigned on staggered grids. This does
not usually lead to serious problems if the grid size is small. Alternatively, (37.1.16) and
(37.1.17) can be used to remove this problem, and to reduce storage.

By eliminating the E or the H field from the Yee algorithm, it can be shown that the Yee
algorithm is equivalent to finite differencing the vector wave equation directly. Hence, the
Yee algorithm is also constrained by the CFL stability criterion.

The following figures show some results of FDTD simulations. Because the answers are
in the time-domain, beautiful animations of the fields are also available online:

https://www.remcom.com/xfdtd-3d-em-simulation-software
Finite Difference Method, Yee Algorithm 427

Figure 37.3: The 2D FDTD simulation of complicated optical waveguides (courtesy of Math-
works).

Figure 37.4: FDTD simulation of human head in a squirrel cage of an MRI (magnetic reso-
nance imaging) system (courtesy of REMCOM).

37.2.1 Finite-Difference Frequency Domain Method

Unlike electrical engineering, in many fields, nonlinear problems are prevalent. But when
we have a linear time-invariant problem, it is simpler to solve the problem in the frequency
domain. This is analogous to perform a time Fourier transform of the pertinent linear equa-
tions.
Consequently, one can write (37.2.1) to (37.2.6) in the frequency domain to remove the
time derivatives. Then one can apply the finite difference approximation to the space deriva-
tives using the Yee grid. As a result, one arrives at a matrix equation
A·x=b (37.2.11)
where x is an unknown vector containing E and H fields, and b is a source vector that
drives the system containing J. Due to the near-neighbor interactions of the fields on the Yee
428 Electromagnetic Field Theory

grid, the matrix A is highly sparse and contains O(N ) non-zero elements. When an iterative
method is used to solve the above equation, the major cost is in performing a matrix-vector
product A · x. However, in practice, the matrix A is never generated nor stored making this
a matrix-free method. Because of the simplicity of the Yee algorithm, a code can be easily
written to produce the action of A on x.

37.3 Absorbing Boundary Conditions

It will not be complete to close this lecture without mentioning absorbing boundary condi-
tions. As computer has finite memory, space of infinitely large extent cannot be simulated
with finite computer memory. Hence, it is important to design absorbing boundary conditions
at the walls of the simulation domain or box, so that waves impinging on it are not reflected.
This mimicks the physics of an infinitely large box.

This is analogous to experments in microwave engineering. In order to perform experi-

ments in an infinite space, such experiments are usually done in an anechoic (non-echoing or
non-reflecting) chamber. An anechoic chamber has its walls padded with absorbing materials
or microwave absorbers so as to minimize the reflections off its walls (see Figure 37.5). Figure
37.6 shows an acoustic version of anechoic chamber.

Figure 37.5: An anechoic chamber for radio frequency. In such an electromagnetically quiet
chamber, interference from other RF equipment is minimized (courtesy of Panasonic).
Finite Difference Method, Yee Algorithm 429

Figure 37.6: An acoustic anechoic chamber. In such a chamber, even the breast-feeding sound
of a baby can be heard clearly (courtesy of AGH University, Poland).

By the same token, in order to simulate numerically an infinite box with a finite-size box,
absorbing boundary conditions (ABCs) are designed at its walls. The simplest of such ABCs
is the impedance boundary condition. (A transmission line terminated with an impedance
reflects less than one terminated with an open or a short circuit.) Another simple ABC is to
mimick the Sommerfeld radiation condition (much of this is reviewed in [35]).5
A recently invented ABC is the perfectly matched layers (PML) [251]. Also, another
similar ABC is the stretched coordinates PML [252]. Figure 37.7 shows simulation results
with and without stretched coordinates PMLs on the walls of the simulation domain [253].

Figure 37.7: Simulation of a source on top of a half-space (left) without stretched coordinates
PML; and (right) with stretched coordinats PML [253].

5 ABCs are beyond the scope of these lecture notes.

430 Electromagnetic Field Theory
Lecture 38

Quantum Theory of Light

The quantum theory of the world is the culmination of a series intellectual exercises. It is often
termed the intellectual triumph of the twentieth century. One often says that deciphering
the laws of nature is like watching two persons play a chess game with rules unbeknownst to
us. By watching the moves, we finally have the revelation about the perplexing rules. But
we are grateful that, aided by experimental data, these laws of nature are deciphered by our
predecessors.

It is important to know that with new quantum theory emerges the quantum theory of
light. This theory is intimately related to Maxwell’s equations as shall be seen. These new
theories spawn the possibility for quantum technologies, one of which is quantum computing.
Others are quantum communication, quantum crytography, quantum sensing and many more.

38.1 Historical Background on Quantum Theory

That light is a wave has been demonstrated by Newton’s ring phenomenon [254] in the
eighteenth century (1717) (see Figure 38.1). In 1801, Thomas Young demonstrated the double
slit experiment for light [255] that further confirmed its wave nature (see Figure 38.2). But
by the beginning of the 20-th century, one has to accept that light is both a particle, called
a photon, carrying a quantum of energy with momentum, as well as a particle endowed with
wave-like behavior. This is called wave-particle duality. We shall outline the historical reason
for this development.

431
432 Electromagnetic Field Theory

Figure 38.1: A Newton’s rings experiment (courtesy of [254]).

Figure 38.2: A Young’s double-slit experiment (courtesy of [256]).

As mentioned above, quantum theory is a major intellectual achievement of the twentieth

century, even though we are still discovering new knowledge in it. Several major experimental
findings led to the revelation of quantum theory of nature. In nature, we know that matter
is not infinitely divisible. This is vindicated by the atomic theory of John Dalton (1766-
1844) [257]. So fluid is not infinitely divisible: as when water is divided into smaller pieces,
Quantum Theory of Light 433

we will eventually arrive at water molecule, H2 O, which is the fundamental building block of
water.
In turns out that electromagnetic energy is not infinitely divisible either. The electromag-
netic radiation out of a heated cavity would have a very different spectrum if electromagnetic
energy is infinitely divisible. In order to fit experimental observation of radiation from a
heated electromagnetic cavity, Max Planck (1900s) [258] proposed that electromagnetic en-
ergy comes in packets or is quantized. Each packet of energy or a quantum of energy E is
associated with the frequency of electromagnetic wave, namely

E = ~ω = ~2πf = hf (38.1.1)

where ~ is now known as the Planck constant and ~ = h/2π = 6.626 × 10−34 J·s (Joule-
second). Since ~ is very small, this packet of energy is very small unless ω is large. So it
is no surprise that the quantization of electromagnetic field is first associated with light, a
very high frequency electromagnetic radiation. A red-light photon at a wavelength of 700 nm
corresponds to an energy of approximately 2 eV ≈ 3×10−19 J ≈ 75 kB T , where kB T denotes
the thermal energy from thermal law, and kB is Boltzmann’s constant. This is about 25 meV
at room temperature.1 A microwave photon has approximately 1 × 10−5 eV ≈ 10−2 meV.
The second experimental evidence that light is quantized is the photo-electric effect [259].
It was found that matter emitted electrons when light shined on it. First, the light frequency
has to correspond to the “resonant” frequency of the atom. Second, the number of electrons
emitted is proportional to the number of packets of energy ~ω that the light carries. This
was a clear indication that light energy traveled in packets or quanta as posited by Einstein
in 1905.
This wave-particle duality concept mentioned at the beginning of this section was not new
to quantum theory as electrons were known to behave both like a particle and a wave. The
particle nature of an electron was confirmed by the measurement of its charge by Millikan in
1913 in his oil-drop experiment. (The double slit experiment for electron was done in 1927 by
Davison and Germer, indicating that an electron has a wave nature as well [255].) In 1924,
De Broglie [260] suggested that there is a wave associated with an electron with momentum
p such that

p = ~k (38.1.2)

where k = 2π/λ, the wavenumber. All this knowledge gave hint to the quantum theorists of
that era to come up with a new way to describe nature.
Classically, particles like an electron moves through space obeying Newton’s laws of mo-
tion first established in 1687 [261]. The old way of describing particle motion is known as
classical mechanics, and the new way of describing particle motion is known as quantum
mechanics. Quantum mechanics is very much motivated by a branch of classical mechanics
called Hamiltonian mechanics. We will first use Hamiltonian mechanics to study a simple
pendulum and connect it with electromagnetic oscillations.
1 This is a number ought to be remembered by semi-conductor scientists as the size of the material bandgap

with respect to this thermal energy decides if a material is a semi-conductor at room temperature.
434 Electromagnetic Field Theory

38.2 Connecting Electromagnetic Oscillation to Simple

Pendulum

The theory for quantization of electromagnetic field was started by Dirac in 1927 [3]. In
the beginning, it was called quantum electrodynamics (QED) important for understanding
particle physics phenomena and light-matter interactions [262]. Later on, it became impor-
tant in quantum optics where quantum effects in electromagnetics technologies first emerged.
Now, microwave photons are measurable and are important in quantum computers. Hence,
quantum effects are important in the microwave regime as well.

Maxwell’s equations originally were inspired by experimental findings of Maxwell’s time,

and he beautifully put them together using mathematics known during his time. But Maxwell’s
equations can also be “derived” using Hamiltonian mechanics and energy conservation. First,
electromagnetic theory can be regarded as for describing an infinite set of coupled harmonic
oscillators. In one dimension, when a wave propagates on a string, or an electromagnetic wave
propagates on a transmission line, they can be regarded as propagating on a set of coupled
harmonic oscillators as shown in Figure 38.3. Maxwell’s equations describe the waves travel-
ling in 3D space due to the coupling between an infinite set of harmonic oscillators. In fact,
methods have been developed to solve Maxwell’s equations using transmission-line-matrix
(TLM) method [263], or the partial element equivalent circuit (PEEC) method [180]. In ma-
terials, these harmonic oscillators are atoms or molecules, but vacuum they can be thought of
as electron-positron pairs (e-p pairs). Electrons are matters, while positrons are anti-matters.
Together, in their quiescent state, they form vacuum or “nothingness”. Hence, vacuum can
support the propogation of electromagnetic waves through vast distances: we have received
light from galaxies many light-years away.
Quantum Theory of Light 435

Figure 38.3: Maxwell’s equations describe the coupling of harmonic oscillators in a 3D space.
This is similar to waves propagating on a string or a 1D transmission line, or a 2D array of
coupled oscillators. The saw-tooth symbol in the figure represents a spring.

The cavity modes in electromagnetics are similar to the oscillation of a pendulum in

simple harmonic motion. To understand the quantization of electromagnetic field, we start
by connecting these cavity-mode oscillations to the oscillations of a simple pendulum. It is
to be noted that fundamentally, electromagnetic oscillation exists because of displacement
current. Displacement current exists even in vacuum because vacuum is polarizable, namely
that D = εE where for vacuum, ε = ε0 . Furthermore, displacement current exists because of
the ∂D/∂t term in the generalized Ampere’s law added by Maxwell, namely,

∂D
∇×H= +J (38.2.1)
∂t

Together with Faraday’s law that

∂B
∇×E=− (38.2.2)
∂t

(38.2.1) and (38.2.2) together allow for the existence of wave. The coupling between the two
equations gives rise to the “springiness” of electromagnetic fields.
Wave exists due to the existence of coupled harmonic oscillators, and at a fundamental
level, these harmonic oscillators are electron-positron (e-p) pairs. The fact that they are
coupled allows waves to propagate through space, and even in vacuum.
436 Electromagnetic Field Theory

Figure 38.4: A one-dimensional cavity solution to Maxwell’s equations is one of the simplest
way to solve Maxwell’s equations.

To make the problem simpler, we can start by looking at a one dimensional cavity formed
by two PEC (perfect electric conductor) plates as shown in Figure 38.4. Assume source-free
Maxwell’s equations in between the plates and letting E = x̂Ex , H = ŷHy , then (38.2.1) and
(38.2.2) become
∂ ∂
Hy = −ε Ex (38.2.3)
∂z ∂t

∂ ∂
Ex = −µ Hy (38.2.4)
∂z ∂t
The above are similar to the telegrapher’s equations. We can combine them to arrive at

∂2 ∂2
E x = µε Ex (38.2.5)
∂z 2 ∂t2
There are infinitely many ways to solve the above partial differential equation. But here, we
use separation of variables to solve the above by letting Ex (z, t) = E0 (t)f (z). Then we arrive
at two separate equations that

d2 E0 (t)
= −ωl2 E0 (t) (38.2.6)
dt2
and
d2 f (z)
= −ωl2 µεf (z) (38.2.7)
dz 2
where ωl2 is the separation constant. There are infinitely many ways to solve the above
equations which are also eigenvalue equations where ωl2 and ωl2 µε are eigenvalues for the first
Quantum Theory of Light 437

and the second equations, respectively. The general solution for (38.2.7) is that

E0 (t) = E0 cos(ωl t + ψ) (38.2.8)

In the above, ωl , which is related to the separation constant, is yet indeterminate. To make ωl2
determinate, we need to impose boundary conditions. A simple way is to impose homogeneous
Dirchlet boundary conditions that f (z) = 0 at z = 0 and z = L. This implies that f (z) =
sin(kl z). In order to satisfy the boundary conditions at z = 0 and z = L, one deduces that

lπ
kl = , l = 1, 2, 3, . . . (38.2.9)
L
Then,

∂ 2 f (z)
= −kl2 f (z) (38.2.10)
∂z 2

where kl2 = ωl2 µ. Hence, kl = ωl /c, and the above solution can only exist for discrete
frequencies or that

lπ
ωl = c, l = 1, 2, 3, . . . (38.2.11)
L
These are the discrete resonant frequencies ωl of the modes of the 1D cavity.
The above solutions for Ex (z, t) can be thought of as the collective oscillations of coupled
harmonic oscillators forming the modes of the cavity. At the fundamental level, these oscil-
lations are oscillators made by electron-positron pairs. But macroscopically, their collective
resonances manifest themselves as giving rise to infinitely many electromagnetic cavity modes.
The amplitudes of these modes, E0 (t) are simple harmonic oscillations.
The resonance between two parallel PEC plates is similar to the resonance of a trans-
mission line of length L shorted at both ends. One can see that the resonance of a shorted
transmission line is similar to the coupling of infnitely many LC tank circuits. To see this, as
shown in Figure 38.3, we start with a single LC tank circuit as a simple harmonic oscillator
with only one resonant frequency. When two LC tank circuits are coupled to each other, they
will have two resonant frequencies. For N of them, they will have N resonant frequencies. For
a continuum of them, they will be infinitely many resonant frequencies or modes as indicated
by Equation (38.2.9).
What is more important is that the resonance of each of these modes is similar to the
resonance of a simple pendulum or a simple harmonic oscillator. For a fixed point in space,
the field due to this oscillation is similar to the oscillation of a simple pendulum.
As we have seen in the Drude-Lorentz-Sommerfeld mode, for a particle of mass m attached
to a spring connected to a wall, where the restoring force is like Hooke’s law, the equation of
motion of a pendulum by Newton’s law is

d2 x
m + κx = 0 (38.2.12)
dt2
438 Electromagnetic Field Theory

where κ is the spring constant, and we assume that the oscillator is not driven by an external
force, but is in natural or free oscillation. By letting2

x = x0 e−iωt (38.2.13)

the above becomes

−mω 2 x0 + κx0 = 0 (38.2.14)

Again, a non-trivial solution is possible only at the resonant frequency of the oscillator or
that when ω = ω0 where
r
κ
ω0 = (38.2.15)
m

This is the eigensolution of (38.2.12) with eigenvalue ω02 .

38.3 Hamiltonian Mechanics

Equation (38.2.12) can be derived by Newton’s law but it can also be derived via Hamiltonian
mechanics as well. Since Hamiltonian mechanics motivates quantum mechanics, we will look
at the Hamiltonian mechanics view of the equation of motion (EOM) of a simple pendulum
given by (38.2.12).
Hamiltonian mechanics, developed by Hamilton (1805-1865) [264], is motivated by energy
conservation [265]. The Hamiltonian H of a system is given by its total energy, namely that

H =T +V (38.3.1)

where T is the kinetic energy and V is the potential energy of the system.
For a simple pendulum, the kinetic energy is given by

1 1 2 2 p2
T = mv 2 = m v = (38.3.2)
2 2m 2m
where p = mv is the momentum of the particle. The potential energy, assuming that the
particle is attached to a spring with spring constant κ, is given by

1 2 1
V = κx = mω02 x2 (38.3.3)
2 2
Hence, the Hamiltonian is given by

p2 1
H =T +V = + mω02 x2 (38.3.4)
2m 2
2 For this part of the lecture, we will switch to using exp(−iωt) time convention as is commonly used in

optics and physics literatures.

Quantum Theory of Light 439

d
At any instant of time t, we assume that p(t) = mv(t) = m dt x(t) is independent of x(t).3
In other words, they can vary independently of each other. But p(t) and x(t) have to time
evolve to conserve energy to keep H, the total energy, constant or independent of time. In
other words,
d dp ∂H dx ∂H
H (p(t), x(t)) = 0 = + (38.3.5)
dt dt ∂p dt ∂x
Therefore, the Hamilton equations of motion are derived to be4
dp ∂H dx ∂H
=− , = (38.3.6)
dt ∂x dt ∂p
From (38.3.4), we gather that
∂H ∂H p
= mω02 x, = (38.3.7)
∂x ∂p m
Applying (38.3.6), we have5
dx p dp
= , = −mω02 x (38.3.8)
dt m dt
Combining the two equations in (38.3.8) above, we have

d2 x
m = −mω02 x = −κx (38.3.9)
dt2
which is also derivable by Newton’s law.
A typical harmonic oscillator solution to (38.3.9) is

x(t) = x0 cos(ω0 t + ψ) (38.3.10)

The corresponding p(t) = m dx

dt is

p(t) = −mx0 ω0 sin(ω0 t + ψ) (38.3.11)

Hence
1 1
H= mω02 x20 sin2 (ω0 t + ψ) + mω02 x20 cos2 (ω0 t + ψ)
2 2
1 2 2
= mω0 x0 = E (38.3.12)
2
And the total energy E is a constant of motion (physicists parlance for a time-independent
variable), it depends only on the amplitude x0 of the oscillation.
3 p(t) and x(t) are termed conjugate variables in many textbooks.
4 Note that the Hamilton equations are determined to within a multiplicative constant, because one has
not stipulated the connection between space and time, or we have not calibrated our clock [265].
5 We can also calibrate our clock here so that it agrees with our definition of momentum in the ensuing

equation.
440 Electromagnetic Field Theory

38.4 Schrödinger Equation (1925)

Having seen the Hamiltonian mechanics for describing a simple pendulum which is homomor-
phic to a cavity resonator, we shall next see the quantum mechanics description of the same
simple pendulum: In other words, we will look at the quantum pendulum. To this end, we
will invoke Schrödinger equation.
Schrödinger equation cannot be derived just as in the case Maxwell’s equations. It is
a wonderful result of a postulate and a guessing game based on experimental observations
[71, 72]. Hamiltonian mechanics says that

p2 1
H= + mω02 x2 = E (38.4.1)
2m 2
where E is the total energy of the oscillator, or pendulum. In classical mechanics, the position
x of the particle associated with the pendulum is known with great certainty. But in the
quantum world, this position x of the quantum particle is uncertain and is fuzzy. As shall be
seen later, x is a random variable.6
To build this uncertainty into a quantum harmonic oscillator, we have to look at it from
the quantum world. The position of the particle is described by a wave function,7 which
makes the location of the particle uncertain. To this end, Schrödinger proposed his equation
which is a partial differential equation. He was very much motivated by the experimental
revelation then that p = ~k from De Broglie and that E = ~ω from Planck’s law and the
photo-electric effect. Equation (38.4.1) can be written more suggestively as

~2 k 2 1
+ mω02 x2 = ~ω (38.4.2)
2m 2
To add more depth to the above equation, one lets the above become an operator equation
that operates on a wave function ψ(x, t) so that

~2 ∂ 2 1 ∂
− 2
ψ(x, t) + mω02 x2 ψ(x, t) = i~ ψ(x, t) (38.4.3)
2m ∂x 2 ∂t
If the wave function is of the form

ψ(x, t) ∼ eikx−iωt (38.4.4)

then upon substituting (38.4.4) back into (38.4.3), we retrieve (38.4.2).

Equation (38.4.3) is Schrödinger equation in one dimension for the quantum version of
the simple harmonic oscillator. In Schrödinger equation, we can further posit that the wave
function has the general form

ψ(x, t) = eikx−iωt A(x, t) (38.4.5)

6 For lack of a better notation, we will use x to both denote a position in classical mechanics as well as a

random variable in quantum theory.

7 Since a function is equivalent to a vector, and this wave function describes the state of the quantum

system, this is also called a state vector.

Quantum Theory of Light 441

where A(x, t) is a slowly varying function of x and t, compared to eikx−iωt .8 In other

words, this is the expression for a wave packet. With this wave packet, the ∂ 2 /∂x2 can
be again approximated by −k 2 in the short-wavelength limit, as has been done in the parax-
ial wave approximation. Furthermore, if the signal is assumed to be quasi-monochromatic,
then i~∂/∂t ψ(x, t) ≈ ~ω, we again retrieve the classical equation in (38.4.2) from (38.4.3).
Hence, the classical equation (38.4.2) is a short wavelength, monochromatic approximation
of Schrödinger equation. However, as we shall see, the solutions to Schrödinger equation are
not limited to just wave packets described by (38.4.5).
In classical mechanics, the position of a particle is described by the variable x, but in the
quantum world, the position of a particle x is a random variable. This property needs to be
related to the wavefunction that is the solution to Schrödinger equation.
For this course, we need only to study the one-dimensional Schrödinger equation. The
above can be converted into eigenvalue problem, just as in waveguide and cavity problems,
using separation of variables, by letting9

ψ(x, t) = ψn (x)e−iωn t (38.4.6)

By so doing, (38.4.3) becomes an eigenvalue problem

~2 d2
 
1 2 2
− + mω0 x ψn (x) = En ψn (x) (38.4.7)
2m dx2 2

where En = ~ωn is the eigenvalue while ψn (x) is the corresponding eigenfunction.

The parabolic x2 potential profile is also known as a potential well as it can provide
the restoring force to keep the particle bound to the well classically (see Section 38.3 and
(38.3.8)). The above equation is also similar to the electromagnetic equation for a dielectric
slab waveguide, where the second term is a dielectric profile (mind you, varying in the x
direction) that can trap a waveguide mode. Therefore, the potential well is a trap for the
particle both in classical mechanics or in wave physics.
The above equation (38.4.7) can be solved in closed form in terms of Hermite-Gaussian
functions (1864) [266], or that
s r r 
1 mω0 − mω0 x2 mω0
ψn (x) = e 2~ Hn x (38.4.8)
2n n! π~ ~

where Hn (y) is a Hermite polynomial, and the eigenvalues are found in closed form as
 
1
En = n + ~ω0 (38.4.9)
2

Here, the eigenfunction or eigenstate ψn (x) is known as the photon number state (or just a
number state) of the solution. It corresponds to having n “photons” in the oscillation. If
this is conceived as the collective oscillation of the e-p pairs in a cavity, there are n photons
8 Recall that this is similar in spirit when we study high frequency solutions of Maxwell’s equations and

paraxial wave approximation.

9 Mind you, the following is ω , not ω .
n 0
442 Electromagnetic Field Theory

corresponding to energy of n~ω0 embedded in the collective oscillation. The larger En is,
the larger the number of photons there is. However, there is a curious mode at n = 0. This
corresponds to no photon, and yet, there is a wave function ψ0 (x). This is the zero-point
energy state. This state is there even if the system is at its lowest energy state.

It is to be noted that in the quantum world, the position x of the pendulum is random.
Moreover, this position x(t) is mapped to the amplitude E0 (t) of the field. Hence, it is the
amplitude of an electromagnetic oscillation that becomes uncertain and fuzzy as shown in
Figure 38.5.

Figure 38.5: Schematic representation of the randomness of measured electric field. The elec-
tric field amplitude maps to the displacement (position) of the quantum harmonic oscillator,
which is a random variable (courtesy of Kira and Koch [267]).
Quantum Theory of Light 443

Figure 38.6: Plots of the eigensolutions of the quantum harmonic oscillator (courtesy of
Wikipedia [268]).

38.5 Some Quantum Interpretations–A Preview

Schrödinger used this equation with resounding success. He derived a three-dimensional ver-
sion of this to study the wave function and eigenvalues of a hydrogen atom. These eigenvalues
En for a hydrogen atom agreed well with experimental observations that had eluded scientists
for decades. Schrödinger did not actually understand what these wave functions meant. It
was Max Born (1926) who gave a physical interpretation of these wave functions.
As mentioned before, in the quantum world, a position x is now a random variable. There
is a probability distribution function (PDF) associated with this random variable x. This
PDF for x is related to the a wave function ψ(x, t), and it is given |ψ(x, t)|2 . Then according
to probability theory, the probability of finding the particles in the interval10 [x, x + ∆x] is
|ψ(x, t)|2 ∆x. Since |ψ(x, t)|2 is a probability density function (PDF), and it is necessary that
 ∞
dx|ψ(x, t)|2 = 1 (38.5.1)
−∞

The average value or expectation value of the random variable x is now given by
 ∞
dxx|ψ(x, t)|2 = hx(t)i = x̄(t) (38.5.2)
−∞
10 This is the math notation for an interval [, ].
444 Electromagnetic Field Theory

This is not the most ideal notation, since although x is not a function of time, its expectation
value with respect to a time-varying function, ψ(x, t), can be time-varying.
Notice that in going from (38.4.1) to (38.4.3), or from a classical picture to a quantum
picture, we have let the momentum become p, originally a scalar number in the classical
world, become a differential operator, namely that11
∂
p → p̂ = −i~ (38.5.3)
∂x
The momentum p of a particle now also becomes uncertain and is a random varible: its
expectation value is given by12
 ∞  ∞
∂
dxψ ∗ (x, t)p̂ψ(x, t) = −i~ dxψ ∗ (x, t) ψ(x, t) = hp̂(t)i = p̄(t) (38.5.4)
∞ −∞ ∂x
The expectation values of position x and the momentum operator p̂ are measurable in the
laboratory. Hence, they are also called observables.

38.5.1 Matrix or Operator Representations

We have seen in computational electromagnetics that an operator can be projected into a
smaller subspace and manifests itself in different representations. Hence, an operator in quan-
tum theory can have different representations depending on the space chosen. For instance,
given a matrix equation
P·x=b (38.5.5)
†
we can find a unitary operator U with the property U · U = I. Then the above equation
can be rewritten as
†
U·P·x=U·b→U·P·U ·U·x=U·b (38.5.6)

Then a new equation is obtained such that

0 0 †
P · x0 = b0 , P =U·P·U , x0 = U · x b0 = U · b (38.5.7)

The operators we have encountered thus far in Schrödinger equation are in coordinate
space representations.13 In coordinate space representation, the momentum operator p̂ =
∂
−i~ ∂x , and the variable x can be regarded as a position operator in coordinate space repre-
sentation. The operator p̂ and x do not commute. In other words, it can be shown that
 
∂
[p̂, x] = −i~ , x = −i~ (38.5.8)
∂x
In the classical world, [p, x] = 0, but not in the quantum world. In the equation above, we
ˆ where Iˆ is the identity operator. Then
can elevate x to become an operator by letting x̂ = xI,
11 We useˆto denote a quantum operator.
12 This concept of the average of an operator seldom has an analogue in an intro probality course, but it is
called the expectation value of an operator in quantum theory.
13 Or just coordinate representation.
Quantum Theory of Light 445

both p̂ and x̂ are now operators, and are on the same footing. In this manner, we can rewrite
equation (38.5.8) above as

[p̂, x̂] = −i~Iˆ (38.5.9)

By performing unitary transformation, it can be shown that the above identity is coordinate
independent: it is true in any representation of the operators.
It can be shown easily that when two operators share the same set of eigenfunctions,
they commute. When two operators p̂ and x̂ do not commute, it means that the expectation
values of quantities associated with the operators, hp̂i and hx̂i, cannot be determined to
arbitrary precision simultaneously. For instance, p̂ and x̂ correspond to random variables,
then the standard deviation of their measurable values, or their expectation values, obey the
uncertainty principle relationship that14

∆p∆x ≥ ~/2 (38.5.10)

where ∆p and ∆x are the standard deviation of the random variables p and x.

38.6 Bizarre Nature of the Photon Number States

The photon number states are successful in predicting that the collective e-p oscillations are
associated with n photons embedded in the energy of the oscillating modes. However, these
number states are bizarre: The expectation values of the position of the quantum pendulum
associated these states are always zero. To illustrate further, we form the wave function with
a photon-number state
ψ(x, t) = ψn (x)e−iωn t
Previously, since the ψn (x) are eigenfunctions, they are mutually orthogonal and they can be
orthonormalized meaning that
 ∞
dxψn∗ (x)ψn0 (x) = δnn0 (38.6.1)
−∞

Then one can easily show that the expectation value of the position of the quantum pendulum
in a photon number state is
 ∞  ∞
hx(t)i = x̄(t) = dxx|ψ(x, t)|2 = dxx|ψn (x)|2 = 0 (38.6.2)
−∞ −∞

because the integrand is always odd symmetric. In other words, the expectation value of the
position x of the pendulum is always zero. It can also be shown that the expectation value
of the momentum operator p̂ is also zero for these photon number states. Hence, there are
no classical oscillations that resemble them. Therefore, one has to form new wave functions
by linear superposing these photon number states into a coherent state. This will be the
discussion in the next lecture.
14 The proof of this is quite straightforward but is outside the scope of this course.
446 Electromagnetic Field Theory
Lecture 39

Quantum Coherent State of

Light

As mentioned in the previous lecture, the discovery of Schrödinger wave equation was a
resounding success. When it was discovered by Schrödinger, and appplied to a very simple
hydrogen atom, its eigensolutions, especially the eigenvalues En coincide beautifully with
spectroscopy experiment of the hydrogen atom. Since the electron wavefunctions inside a
hydrogen atom does not have a classical analog, less was known about these wavefunctions.
But in QED and quantum optics, the wavefunctions have to be connected with classical
electromagnetic oscillations. As seen previously, electromagnetic oscillations resemble those
of a pendulum. The original eigenstates of the quantum pendulum were the photon number
states also called the Fock states. The connection to the classical pendulum was tenuous, but
required by the correspondence principle–quantum phenomena resembles classical phenomena
in the high energy limit. This connection was finally established by the establishment of the
coherent state.

39.1 The Quantum Coherent State

We have seen that a photon number states1 of a quantum pendulum do not have a classical
correspondence as the average or expectation values of the position and momentum of the
pendulum are always zero for all time for this state. Therefore, we have to seek a time-
dependent quantum state that has the classical equivalence of a pendulum. This is the
coherent state, which is the contribution of many researchers, most notably, Roy Glauber
(1925–2018) [269] in 1963, and George Sudarshan (1931–2018) [270]. Glauber was awarded
the Nobel prize in 2005.
We like to emphasize again that the modes of an electromagnetic cavity oscillation are
homomorphic to the oscillation of classical pendulum. Hence, we first connect the oscillation
of a quantum pendulum to a classical pendulum. Then we can connect the oscillation of
1 In quantum theory, a “state” is synonymous with a state vector or a function.

447
448 Electromagnetic Field Theory

a quantum electromagnetic mode to the classical electromagnetic mode and then to the
quantum pendulum.

39.1.1 Quantum Harmonic Oscillator Revisited

To this end, we revisit the quantum harmonic oscillator or the quantum pendulum with more
mathematical depth. Rewriting Schrödinger equation as the eigenequation for the photon
number state for the quantum harmonic oscillator, we have
~2 d2
 
1 2 2
Ĥψ(x) = − + mω0 x ψ(x) = Eψ(x). (39.1.1)
2m dx2 2
where ψ(x) is the eigenfunction, and E is the eigenvalue. The p
above can be changed into
mω0
a dimensionless form first by dividing ~ω0 , and then let ξ = ~ x be a dimensionless
variable. The above then becomes
d2
 
1 2 E
− 2 + ξ ψ(ξ) = ψ(ξ) (39.1.2)
2 dξ ~ω0
d
We can define π̂ = −i dξ and ξˆ = Iξ
ˆ to rewrite the Hamiltonian as

1
Ĥ = ~ω0 (π̂ 2 + ξˆ2 ) (39.1.3)
2
Furthermore, the Hamiltonian in (39.1.2) looks almost like A2 − B 2 , and hence motivates
its factorization. To this end, we first show that
d2
       
1 d 1 d 1 2 1 d d
√ − +ξ √ +ξ = − 2 +ξ − ξ−ξ (39.1.4)
2 dξ 2 dξ 2 dξ 2 dξ dξ
It can be shown easily that as operators (meaning that they will act on a function to their
right), the last term on the right-hand side is an identity operator, namely that
 
d d
ξ−ξ = Iˆ (39.1.5)
dξ dξ
Therefore
d2
     
1 2 1 d 1 d 1
− 2 +ξ =√ − +ξ √ +ξ + (39.1.6)
2 dξ 2 dξ 2 dξ 2
We define the operator
 
1 d
↠= √ − +ξ (39.1.7)
2 dξ
The above is the creation, or raising operator and the reason for its name is obviated later.
Moreover, we define
 
1 d
â = √ +ξ (39.1.8)
2 dξ
Quantum Coherent State of Light 449

which represents the annihilation or lowering operator. With the above definitions of the
raising and lowering operators, it is easy to show that by straightforward substitution that
â, ↠= â↠− ↠â = Iˆ
 
(39.1.9)
Therefore, Schrödinger equation (39.1.2) for quantum harmonic oscillator can be rewritten
more concisely as
 
1 † 1 E
â â + â↠ψ = ↠â +


ψ= ψ (39.1.10)
2 2 ~ω0
In mathematics, a function is analogous to a vector. So ψ is the implicit representation of a
vector. The operator  
† 1
â â +
2
is an implicit2 representation of an operator, and in this case, a differential operator. So in
the above, (39.1.10), is analogous to the matrix eigenvalue equation A · x = λx.
Consequently, the Hamiltonian operator can be expressed concisely as
 
† 1
Ĥ = ~ω0 â â + (39.1.11)
2
Equation (39.1.10) above is in implicit math notation. In implicit Dirac notation, it is
 
† 1 E
â â + |ψi = |ψi (39.1.12)
2 ~ω0
In the above, ψ(ξ) is a function which is a vector in a functional space. It is denoted as ψ
in math notation and |ψi in Dirac notation. This is also known as the “ket”. The conjugate
transpose of a vector in Dirac notation is called a “bra” which is denoted as hψ|. Hence, the
inner product between two vectors is denoted as hψ1 |ψ2 i in Dirac notation.3
If we denote a photon number state by ψn (x) in explicit notation, ψn in math notation
or |ψn i in Dirac notation, then we have
   
1 En 1
↠â + |ψn i = |ψn i = n + |ψn i (39.1.13)
2 ~ω0 2
where we have used the fact that En = (n + 1/2)~ω0 . Therefore, by comparing terms in the
above, we have
↠â|ψn i = n|ψn i (39.1.14)

and the operator ↠â is also known as the number operator because of the above. It is often
denoted as
n̂ = ↠â (39.1.15)
2A
P
notation like A · x, we will call implicit, while a notation i,j Aij xj , we will call explicit.
3 There is a one-to-one correspondence of Dirac notation to matrix algebra notation. Â|xi ↔ A · x, hx| ↔
x† hx1 |x2 i ↔ x†1 · x2 .
450 Electromagnetic Field Theory

and |ψn i is an eigenvector of n̂ = ↠â operator with eigenvalue n. It can be further shown by
direct substitution that
√ √
â|ψn i = n|ψn−1 i ⇔ â|ni = n|n − 1i (39.1.16)
√ √
↠|ψn i = n + 1|ψn+1 i ⇔ ↠|ni = n + 1|n + 1i (39.1.17)

hence their names as lowering and raising operator.4

39.2 Some Words on Quantum Randomness and Quan-

tum Observables
We saw previously that in classical mechanics, the conjugate variables p and x are determin-
istic variables. But in the quantum world, they become random variables with means and
variance. It was quite easy to see that x is a random variable in the quantum world. But the
momentum p is elevated to become a differential operator p̂, and it is not clear that it is a
random variable anymore.
Quantum theory is a lot richer in content than classical theory. Hence, in quantum theory,
conjugate variables like p and x are observables endowed with the properties of mean and
variance. For them to be endowed with these properties, they are elevated to become quantum
operators, which are the representations of these observables. To be meaningful, a quantum
state |ψi has to be defined for a quantum system, and these operators representing observables
act on the quantum state.
Henceforth, we have to extend the concept of the average of a random variable to the
“average” of a quantum operator. Now that we know Dirac notation, we can write the
expectation value of the operator p̂ with respect to a quantum state ψ as

p̄ = hpi = hψ|p̂|ψi (39.2.1)

The above is the elevated way of taking the “average” of an operator which is related to the
mean of the random variable p. In the above, the two-some, {p̂, |ψi} endows the quantum
observable p with random properties. Here, p becomes a random variable whose mean is
given by the above formula.
As mentioned before, Dirac notation is homomorphic to matrix algebra notation. The
above is similar to ψ † · P · ψ = p̄. This quantity p̄ is always real if P is a Hermitian matrix.
Hence, in (39.2.1), the expectation value p̄ is always real if p̂ is Hermitian. In fact, it can be
proved that p̂ is Hermitian in the function space that it is defined.
Furthermore, the variance of the random variable p can be derived from the quantum
operator p̂ with respect to to a quantum state |ψi. It is defined as

σp2 = hψ|(p̂ − p̄)2 |ψi = hψ|p̂2 |ψi − p̄2 (39.2.2)

where σp is the standard deviation of the random variable p and σp2 is its variance [71, 72].
4 The above notation for a vector could appear cryptic or too terse to the uninitiated. To parse it, one can

always down-convert from an abstract notation to a more explicit notation. Namely, |ni → |ψn i → ψn (ξ).
Quantum Coherent State of Light 451

The above implies that the definition of the quantum operators and the quantum states
†
is not unique. One can define a unitary matrix or operator U such that U · U = I. Then
the new quantum state is now given by the unitary transform ψ 0 = U · ψ. With this, we can
easily show that
† †
p̄ = ψ † · P · ψ = ψ † · U · U · P · U · U · ψ
0
= ψ 0† · P · ψ 0 (39.2.3)
0 † 0
where P = U·P·U via unitary transform. Now, P is the new quantum operator representing
the observable p and ψ 0 is the new quantum state vector.
In the previous section, we have elevated the position variable or observable ξ to become an
operator ξˆ = ξ I.
ˆ This operator is clearly Hermitian, and hence, the expectation value of this
position operator is always real. Here, ξˆ is diagonal in the coordinate space representation,
but it need not be in other Hilbert space representations using unitary transformation shown
above.

39.3 Derivation of the Coherent States

As one cannot see the characteristics of a classical pendulum emerging from the photon
number states, one needs another way of bridging the quantum world with the classical
world. This is the role of the coherent state: It will show the correspondence principle, with
a classical pendulum emerging from a quantum pendulum when the energy of the pendulum
is large. Hence, it will be interesting to see how the coherent state is derived.
The derivation of the coherent state is more math than physics. Nevertheless, the deriva-
tion is interesting. We are going to present it according to the simplest way presented in
the literature. There are deeper mathematical methods to derive this coherent state like
Bogoliubov transform which is outside the scope of this course.
Now, endowed with the needed mathematical tools, we can derive the coherent state
simply. To say succinctly, the coherent state is the eigenstate of the annihilation operator,
namely that
â|αi = α|αi (39.3.1)
Here, we use α as an eigenvalue as well as an index or identifier of the state |αi.5 Since the
number state |ni is complete, the coherent state |αi can be expanded in terms of the number
state |ni. Or that
∞
X
|αi = Cn |ni (39.3.2)
n=0

When the annihilation operator is applied to the above, we have

∞ ∞ ∞ ∞
X X X √ X √
â|αi = Cn â|ni = Cn â|ni = Cn n|n − 1i = Cn+1 n + 1|ni (39.3.3)
n=0 n=1 n=1 n=0
5 This notation is cryptic and terse, but one can always down-convert it as |αi → |f i → f (ξ) to get a
α α
more explicit notation.
452 Electromagnetic Field Theory

The last equality follows from changing the variable of summation from n to n + 1. Equating
the above with α|αi on the right-hand side of (39.3.1), then
∞ ∞
X √ X
Cn+1 n + 1|ni = α Cn |ni (39.3.4)
n=0 n=0

By the orthonormality of the number states |ni and the completeness of the set,
√
Cn+1 = αCn / n + 1 (39.3.5)

Or recursively
√ p √
Cn = Cn−1 α/ n = Cn−2 α2 / n(n − 1) = . . . = C0 αn / n! (39.3.6)

Consequently, the coherent state |αi is

∞
X αn
|αi = C0 √ |ni (39.3.7)
n=0 n!

But due to the probabilistic interpretation of quantum mechanics, the state vector |αi is
normalized to one, or that6

hα|αi = 1 (39.3.8)

Then
∞ 0
X αn αn
hα|αi = C0∗ C0 √ √ hn0 |ni
n,n0
n! n0 !
∞
X |α|2n 2
= |C0 |2 = |C0 |2 e|α| = 1 (39.3.9)
n=0
n!
2
Therefore, C0 = e−|α| /2
for normalization, or that
∞
2 X αn
|αi = e−|α| /2
√ |ni (39.3.10)
n=0 n!

In the above, to reduce the double summations into a single summation, we have made use
of hn0 |ni = δn0 n , or that the photon-number states are orthonormal. Also since â is not a
Hermitian operator, its eigenvalue α can be a complex number.
Since the coherent state is a linear superposition of the photon number states, an average
number of photons can be associated with the coherent state. If the average number of
photons embedded in a coherent is N , then it can be shown that N = |α|2 . As shall be
shown, α is related to the amplitude of the quantum oscillation: The more photons there are
in a coherent state, the larger |α| is.

∞
6 The expression can be written more explicitly as hα|αi = hfα |fα i = −∞ dξfα∗ (ξ)fα (ξ) = 1.
Quantum Coherent State of Light 453

39.3.1 Time Evolution of a Quantum State

The Schrödinger equation can be written concisely as

Ĥ|ψi = i~∂t |ψi (39.3.11)

The above not only entails the form of Schrödinger equation, it is the form of the general
quantum state equation. Since Ĥ is time independent, the formal solution to the above is

|ψ(t)i = e−iĤt/~ |ψ(0)i (39.3.12)

Meaning of the Function of an Operator

At this juncture, it is prudent to digress to discuss the meaning of a function of an operator,
which occurs in (39.3.12): The exponential function is a function of the operator Ĥ. This
is best explained by expanding the pertinent function into a Taylor series, namely,
1 00 2 1 n
f (A) = f (0)I + f 0 (0)A + f (0)A + . . . + f (n) (0)A + . . . (39.3.13)
2! n!
Without loss of generality, we have used the matrix operator A as an illustration. The
above series has no meaning unless it acts on an eigenvector of the matrix operator A,
where Av = λv. Hence, by applying the above equation (39.3.13) to an eigenvector v of A,
we have
1 00 2 1 n
f (A)v = f (0)v + f 0 (0)Av + f (0)A v + . . . + f (n) (0)A v + . . .
2! n!
0 1 00 2 1 (n)
= f (0)v + f (0)λv + f (0)λ v + . . . + f (0)λn v + . . . = f (λ)v (39.3.14)
2! n!
The last equality follows by re-summing the Taylor series back into a function. Applying
this to an exponential function of an operator, we have

eA v = eλ v (39.3.15)
Applying this to the photon number state with Ĥ being that of the quantum pendulum
and that |ni is the eigenvector of Ĥ, then

e−iĤt/~ |ni = e−iωn t |ni (39.3.16)

where ωn = n + 21 ω0 . In particular, |ni is an eigenstate of the Hamiltonian Ĥ for the

quantum pendulum, or that from (39.1.11) and (39.1.14)

 
1
Ĥ|ni = ~ωn |ni = ~ω0 n + |ni (39.3.17)
2

In other words, |ni, a shorthand notation for |ψn i in (39.1.13), is an eigenvector of Ĥ.
454 Electromagnetic Field Theory

Time Evolution of the Coherent State

Using the above time-evolution operator, then the time dependent coherent state evolves in
time as7
∞
2 X αn e−iωn t
|α, ti = e−iĤt/~ |αi = e−|α| /2 √ |ni (39.3.18)
n=0 n!
By letting ωn = ω0 n + 12 , the above can be written as

∞
n
−iω0 t/2 −|α|2 /2
X αe−iω0 t
|α, ti = e e √ |ni (39.3.19)
n=0 n!
Now we see that the last factor in (39.3.19) is similar to the expression for a coherent state
in (39.3.10). Therefore, we can express the above more succinctly by replacing α in (39.3.10)
with α̃ = αe−iω0 t as
|α, ti = e−iω0 t/2 |αe−iω0 t i = e−iω0 t/2 |α̃i (39.3.20)
Consequently,
â|α, ti = âe−iω0 t/2 |αe−iω0 t i = e−iω0 t/2 αe−iω0 t |αe−iω0 t i = α̃|α, ti


(39.3.21)
Therefore, |α, ti is the eigenfunction of the â operator with eigenvalue α̃. But now, the
eigenvalue of the annihilation operator â is a complex number which is a function of time t.
It is to be noted that in the coherent state in (39.3.19), the photon number states time-evolve
coherently together in a manner to result in a phase shift e−iω0 t in the eigenvalue giving rise
to a new eigenvalue α̃!

39.4 More on the Creation and Annihilation Operator

As seen in the photon-number states, the oscillation of the pendulum does not emerge in
the quantum solutions to Schrödinger equation. Hence it is prudent to see if this physical
phenomenon emerge with the coherent state. In order to connect the quantum pendulum to
a classical pendulum via the coherent state, we will introduce some new operators. Since
 
† 1 d
â = √ − +ξ (39.4.1)
2 dξ
 
1 d
â = √ +ξ (39.4.2)
2 dξ
We can relate ↠and â, which are non-hermitian, to the momentum operator π̂ and position
operator ξˆ previously defined which are hermitian. Then
1  
↠= √ −iπ̂ + ξˆ (39.4.3)
2
1  
â = √ iπ̂ + ξˆ (39.4.4)
2
7 Note that |α, ti is a shorthand for fα (ξ, t).
Quantum Coherent State of Light 455

From the above, by subtracting and adding the two equations, we arrive at
1
ξˆ = √ ↠+ â = ξ Iˆ


(39.4.5)
2
i d
π̂ = √ ↠− â = −i


(39.4.6)
2 dξ

Notice that both ξˆ and π̂ are Hermitian operators in the above, with real expectation values
in accordance to (39.2.1). With this, the average or expectation value of the position of the
pendulum in normalized coordinate, ξ, can be found by taking expectation with respect to
the coherent state, or

ˆ = √1 hα|↠+ â|αi
hα|ξ|αi (39.4.7)
2
Since by taking the complex conjugate transpose of (39.3.1)8

hα|↠= hα|α∗ (39.4.8)

and (39.4.7) becomes

√
ˆ = √1 (α∗ + α) hα|αi = 2<e (α) 6= 0
ξ¯ = hξi = hα|ξ|αi (39.4.9)
2

Repeating the exercise for time-dependent case, when we let α → α̃(t) = αe−iω0 t , then,
letting α = |α|e−iψ yields
1 √
¯ = hξ(t)i = hα̃(t)|ξ|α̃(t)i
ξ(t) ˆ = √ (α̃∗ (t) + α̃(t)) hα̃(t)|α̃(t)i = 2<e (α̃(t)) 6= 0 (39.4.10)
2
where for the last equality, we have made used of that hα̃(t)|α̃(t)i = 1. Then
√
¯ = hξ(t)i = 2|α| cos(ω0 t + ψ)
ξ(t) (39.4.11)

In the above, we use ξ to denote the random variable. So hξ(t)i refers to the average of the
¯ that is a function of time.
random variable ξ, or ξ(t)
By the same token,
i √
π̄ = hπi = hα|π̂|αi = √ (α∗ − α) hα|αi = 2=m (α) 6= 0 (39.4.12)
2

For the time-dependent case, we let α → α̃(t) = αe−iω0 t ,

√
π̄(t) = hπ(t)i = − 2|α| sin(ω0 t + ψ) (39.4.13)

Hence, we see that the expectation values of the normalized coordinate and momentum just
behave like a classical pendulum. There is however a marked difference: These values have
8 Dirac notation is homomorphic with matrix algebra notation. (a · x)† = x† · (a)† .
456 Electromagnetic Field Theory

standard deviations or variances that are non-zero. Thus, they have quantum fluctuation or
quantum noise associated with them. Since the quantum pendulum is homomorphic with the
oscillation of a quantum electromagnetic mode, the amplitude of a quantum electromagnetic
mode will have a mean and a fluctuation as well.

Figure 39.1: The time evolution of the coherent state. It follows the motion of a classical
pendulum or harmonic oscillator (courtesy of Gerry and Knight [271]).
.

Figure 39.2: The time evolution of the coherent state for different α’s. The left figure is for
α = 5 while the right figure is for α = 10. Recall that N = |α|2 .
.
Quantum Coherent State of Light 457

39.4.1 Connecting Quantum Pendulum to Electromagnetic Oscilla-

tor9
We see that the electromagnetic oscillator in a cavity is similar or homomorphic to a pendu-
lum. The classical Hamiltonian is
p2 1 1 2
+ mω02 x2 = P (t) + Q2 (t) = E

H =T +V = (39.4.14)
2m 2 2
where E is the total energy of the system. In the above, ω0 is the resonant frequency of
the classical pendulum. To make the cavity mode homomorphic to a pendulum, we have to
replace ω0 with ωl , the resonant frequency of the cavity mode. In the above, P is a normalized
momentum and Q is a normalized coordinate, and their squares have the unit of energy. We
have also shown that when the classical
pendulum is elevated to be a quantum pendulum,
then H → Ĥ, where Ĥ = ~ωl ↠â + 12 . Then Schrödinger equation becomes
 
† 1
Ĥ|ψ, ti = ~ωl â â + |ψ, ti = i~∂t |ψ, ti (39.4.15)
2
Our next task is to connect the electromagnetic oscillator to this pendulum. In general,
the total energy or the Hamiltonian of an electromagnetic system is
  
1 2 1 2
H= dr εE (r, t) + B (r, t) . (39.4.16)
2 µ
V

It is customary to write this Hamiltonian in terms of scalar and vector potentials. For
simplicity, we use a 1D cavity, and let A = x̂Ax , ∇ · A = 0 so that ∂x Ax = 0, and letting
.
Φ = 0. Then B = ∇ × A and E = −A, and the classical Hamiltonian from (39.4.16) for a
Maxwellian system becomes
  
1 . 1
2 2
H= dr εA (r, t) + (∇ × A(r, t)) . (39.4.17)
2 µ
V
.
For the 1D case, the above implies that By = ∂z Ax , and Ex = −∂t Ax = −Ax . Hence, we
let

Ax = A0 (t) sin(kl z) (39.4.18)

.
Ex = −A0 (t) sin(kl z) = E0 (t) sin(kl z) (39.4.19)
By = kl A0 (t) cos(kl z). (39.4.20)
.

L
where E0 (t) = −A0 (t). After integrating over the volume such that V
dr = A 0
dz, the
Hamiltonian (39.4.17) then becomes
V0 ε  . 2 V
0 2 2
H= A0 (t) + k A (t). (39.4.21)
4 4µ l 0
9 May be skipped on first reading.
458 Electromagnetic Field Theory

where V0 = AL, is the mode volume. The form of (39.4.21) now resembles the pendulum
Hamiltonian. We can think of A0 (t) as being related to the displacement of the pendulum.
Hence, the second term resembles the potential energy. The first term has the time derivative
of A0 (t), and hence, can be connected to the kinetic energy of the system. Therefore, we can
rewrite the Hamiltonian as
1 2
P (t) + Q2 (t)

H= (39.4.22)
2
where r r s
V0 ε . V0 ε V0
P (t) = A0 (t) = − E0 (t), Q(t) = kl A0 (t) (39.4.23)
2 2 2µ
By elevating P and Q to be quantum operators,
p p
P (t) → P̂ = ~ωl π̂(t), ˆ
Q(t) → Q̂ = ~ωl ξ(t) (39.4.24)

so that the quantum Hamiltonian now is

1h 2 i 1  
Ĥ = P̂ + Q̂2 = ~ωl π̂ 2 + ξˆ2 (39.4.25)
2 2
similar to (39.1.3) as before except now that the resonant frequency of this mode is ωl instead
of ω0 because these are the cavity modes, each of which is homomorphic to a quantum
pendulum of frequency ωl . An equation of motion for the state of the quantum system can
be associated with the quantum Hamiltonian just as in the quantum pendulum case.
We have shown previously that
√
↠+ â = 2ξˆ (39.4.26)
†
√
â − â = − 2iπ̂ (39.4.27)

Then we can let r

V0 ε p
P̂ = − Ê0 = ~ωl π̂ (39.4.28)
2
Finally, we arrive at
r r
2~ωl 1 ~ωl †
Eˆ0 = −


π̂ = â − â (39.4.29)
εV0 i εV0

Now that E0 has been elevated to be a quantum operator Ê0 , from (39.4.19), we can put in
the space dependence in accordance to (39.4.19) to get

Êx (z) = Ê0 sin(kl z) (39.4.30)

Consequently, the fully quantized field is a field operator given by

r
1 ~ωl †

Êx (z) = â − â sin(kl z) (39.4.31)
i εV0
Quantum Coherent State of Light 459

Notice that in the above, Ê0 , and Êx (z) are all Hermitian operators and they correspond to
quantum observables that have randomness associated with them but with real mean values.
Also, the operators are independent of time because they are in the Schrödinger picture.
Heisenberg Picture and Schrödinger Picture
Because the quantum state equation can be solved formally in closed form, we can look
at the quantum world in two pictures: The Heisenberg picture or the Schrödinger picture.
A quantum observable is an operator, and the mean value of the observable, which is
measurable, is given by
Ō(t) = hψ(t)|Ôs |ψ(t)i. (39.4.32)
Alternatively, one can rewrite the above using (39.3.12) as

Ō(t) = hψ(0)|eiĤt/~ Ôs e−iĤt/~ |ψ(0)i. (39.4.33)

In other words, the above can be rewritten as

Ō(t) = hψ(0)|Ôh (t)|ψ(0)i. (39.4.34)

where
Ôh (t) = eiĤt/~ Ôs e−iĤt/~ . (39.4.35)
The above illustrates the two different ways to look at the quantum world that produce
the same expectation value of a quantum operator. The first way, called the Schrödinger
picture, keeps the quantum operator Ôs to be time independent, but retain the quantum
state vector |ψ(t)i as a function of time. The second way, called the Heisenberg picture, lets
the operator be time dependent, but the state vector is independent of time. Hence, one can
derive the equation of motion of an operator in the Heisenberg picture. Also, note that the
operators Ôh and Ôs in the Heisenberg picture and the Schrödinger picture, respectively,
are related to each other by a unitary transform via the time-evolution operator e−iĤt/~ .
In the Schrödinger picture, to get time dependence fields, one has to take the expectation
value of the operators with respect to time-varying quantum state vector like the time-varying
coherent state.
To let Eˆx have any meaning, it should act on a quantum state. For example,

|ψE i = Eˆx |ψi (39.4.36)

Notice that thus far, all the operators derived are independent of time. To derive time
dependence of these operators, one needs to find their expectation value with respect to
time-dependent state vectors.10
To illustrate this, we can take expectation value of the quantum operator Êx (z) with

10 This is known as the Schrödinger picture.

460 Electromagnetic Field Theory

respect to a time dependent state vector, like the time-dependent coherent state, Thus
r
1 ~ωl
hEx (z, t)i = hα, t|Êx (z)|α, ti = hα, t|↠− â|α, ti
i εV0
r r
1 ~ωl ∗ ~ωl
= (α̃ (t) − α̃(t)) hα, t|α, ti = −2 =m(α̃) (39.4.37)
i εV0 εV0

From (39.3.19), we can show that hα, t|α, ti = 1. Using the time-dependent α̃(t) = αe−iωl t =
|α|e−i(ωl t+ψ) in the above, similar to (39.4.13), we have
r
~ωl
hEx (z, t)i = 2 |α| sin(ωl t + ψ) (39.4.38)
εV0

where α̃(t) = αe−iωl t . The expectation value of the operator with respect to a time-varying
quantum state, in fact, gives rise to a time-varying quantity. The above, which is the average
of a random field, resembles a classical field. But since it is rooted in a random variable, it
has a standard deviation in addition to having a real-value mean.
We can also show that
r r
2µ~ωl ˆ µ~ωl †
B̂y (z) = kl Â0 cos(kl z) = ξ= (â + â) (39.4.39)
V0 V0
Again, these are time-independent operators in the Schrödinger picture. To get time-dependent
quantities, we have to take the expectation value of the above operator with respect to to a
time-varying quantum state.
39.5 Epilogue
In conclusion, the quantum theory of light is a rather complex subject. It cannot be taught
in just two lectures, but what we wish is to give you a peek into this theory. It takes much
longer to learn this subject well: after all, it is the by product of a century of intellectual
exercise. This knowledge is still very much in its infancy. Hopefully, the more we teach
this subject, the better we can articulate, understand, and explain this subject. When James
Clerk Maxwell completed the theory of electromagnetics over 150 years ago, and wrote a book
on the topic, rumor has it that most people could not read beyond the first 50 pages of his
book [39]. But after over a century and a half of regurgitation, we can now teach the subject
to undergraduate students! When Maxwell put his final stroke to the equations named after
him, he could never have foreseen that these equations are valid from nano-meter lengthscales
to galactic lengthscales, from static to ultra-violet frequencies. Now, these equations are even
valid from classical to the quantum world as well!
Hopefully, by introducing these frontier knowledge in electromagnetic field theory in this
course, it will pique your interest enough in this subject, so that you will take this as a
life-long learning experience.
Bibliography

[1] J. A. Kong, Theory of electromagnetic waves. New York, Wiley-Interscience, 1975.

[2] A. Einstein et al., “On the electrodynamics of moving bodies,” Annalen der Physik,
vol. 17, no. 891, p. 50, 1905.
[3] P. A. M. Dirac, “The quantum theory of the emission and absorption of radiation,” Pro-
ceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical
and Physical Character, vol. 114, no. 767, pp. 243–265, 1927.
[4] R. J. Glauber, “Coherent and incoherent states of the radiation field,” Physical Review,
vol. 131, no. 6, p. 2766, 1963.
[5] C.-N. Yang and R. L. Mills, “Conservation of isotopic spin and isotopic gauge invari-
ance,” Physical review, vol. 96, no. 1, p. 191, 1954.
[6] G. t’Hooft, 50 years of Yang-Mills theory. World Scientific, 2005.
[7] C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation. Princeton University
Press, 2017.
[8] F. Teixeira and W. C. Chew, “Differential forms, metrics, and the reflectionless absorp-
tion of electromagnetic waves,” Journal of Electromagnetic Waves and Applications,
vol. 13, no. 5, pp. 665–686, 1999.
[9] W. C. Chew, E. Michielssen, J.-M. Jin, and J. Song, Fast and efficient algorithms in
computational electromagnetics. Artech House, Inc., 2001.
[10] A. Volta, “On the electricity excited by the mere contact of conducting substances
of different kinds. in a letter from Mr. Alexander Volta, FRS Professor of Natural
Philosophy in the University of Pavia, to the Rt. Hon. Sir Joseph Banks, Bart. KBPR
S,” Philosophical transactions of the Royal Society of London, no. 90, pp. 403–431, 1800.
[11] A.-M. Ampère, Exposé méthodique des phénomènes électro-dynamiques, et des lois de
ces phénomènes. Bachelier, 1823.
[12] ——, Mémoire sur la théorie mathématique des phénomènes électro-dynamiques unique-
ment déduite de l’expérience: dans lequel se trouvent réunis les Mémoires que M.
Ampère a communiqués à l’Académie royale des Sciences, dans les séances des 4 et

461
462 Electromagnetic Field Theory

26 décembre 1820, 10 juin 1822, 22 décembre 1823, 12 septembre et 21 novembre 1825.

Bachelier, 1825.

[13] B. Jones and M. Faraday, The life and letters of Faraday. Cambridge University Press,
2010, vol. 2.

[14] G. Kirchhoff, “Ueber die auflösung der gleichungen, auf welche man bei der unter-
suchung der linearen vertheilung galvanischer ströme geführt wird,” Annalen der Physik,
vol. 148, no. 12, pp. 497–508, 1847.

[15] L. Weinberg, “Kirchhoff’s’ third and fourth laws’,” IRE Transactions on Circuit Theory,
vol. 5, no. 1, pp. 8–30, 1958.

[16] T. Standage, The Victorian Internet: The remarkable story of the telegraph and the
nineteenth century’s online pioneers. Phoenix, 1998.

[17] J. C. Maxwell, “A dynamical theory of the electromagnetic field,” Philosophical trans-

actions of the Royal Society of London, no. 155, pp. 459–512, 1865.

[18] H. Hertz, “On the finite velocity of propagation of electromagnetic actions,” Electric
Waves, vol. 110, 1888.

[19] M. Romer and I. B. Cohen, “Roemer and the first determination of the velocity of light
(1676),” Isis, vol. 31, no. 2, pp. 327–379, 1940.

[20] A. Arons and M. Peppard, “Einstein’s proposal of the photon concept–a translation of
the Annalen der Physik paper of 1905,” American Journal of Physics, vol. 33, no. 5,
pp. 367–374, 1965.

[21] A. Pais, “Einstein and the quantum theory,” Reviews of Modern Physics, vol. 51, no. 4,
p. 863, 1979.

[22] M. Planck, “On the law of distribution of energy in the normal spectrum,” Annalen der
physik, vol. 4, no. 553, p. 1, 1901.

[23] A. Houck, D. Schuster, J. Gambetta, J. Schreier, B. Johnson, J. Chow, L. Frunzio,

J. Majer, M. Devoret, S. Girvin et al., “Generating single microwave photons in a
circuit,” Nature, vol. 449, no. 7160, pp. 328–331, 2007.

[24] Z. Peng, S. De Graaf, J. Tsai, and O. Astafiev, “Tuneable on-demand single-photon

source in the microwave range,” Nature communications, vol. 7, p. 12588, 2016.

[25] B. D. Gates, Q. Xu, M. Stewart, D. Ryan, C. G. Willson, and G. M. Whitesides, “New

approaches to nanofabrication: molding, printing, and other techniques,” Chemical
reviews, vol. 105, no. 4, pp. 1171–1196, 2005.

[26] D. J. Griffiths and D. F. Schroeter, Introduction to quantum mechanics. Cambridge

University Press, 2018.
Quantum Coherent State of Light 463

[27] J. S. Bell, “The debate on the significance of his contributions to the foundations of
quantum mechanics, Bell’s Theorem and the Foundations of Modern Physics (A. van
der Merwe, F. Selleri, and G. Tarozzi, eds.),” 1992.

[28] C. Pickover, Archimedes to Hawking: Laws of science and the great minds behind them.
Oxford University Press, 2008.

[29] R. Resnick, J. Walker, and D. Halliday, Fundamentals of physics. John Wiley, 1988.

[30] J. L. De Lagrange, “Recherches d’arithmétique,” Nouveaux Mémoires de l’Académie de

Berlin, 1773.

[31] S. Ramo, J. R. Whinnery, and T. Duzer van, Fields and waves in communication
electronics, Third Edition. John Wiley & Sons, Inc., 1995, also 1965, 1984.

[32] J. A. Kong, Electromagnetic Wave Theory. EMW Publishing, 2008, also 1985.

[33] H. M. Schey, Div, grad, curl, and all that: an informal text on vector calculus. WW
Norton New York, 2005.

[34] R. P. Feynman, R. B. Leighton, and M. Sands, The Feynman lectures on physics, Vols.
I, II, & III: The new millennium edition. Basic books, 2011, also 1963, 2006, vol. 1,2,3.

[35] W. C. Chew, Waves and fields in inhomogeneous media. IEEE Press, 1995, also 1990.

[36] V. J. Katz, “The history of Stokes’ theorem,” Mathematics Magazine, vol. 52, no. 3,
pp. 146–156, 1979.

[37] J. C. Maxwell, A Treatise on Electricity and magnetism. Dover New York, 1954, first
published in 1873, vol. 1 and 2.

[38] A. D. Yaghjian, “Reflections on Maxwell’s treatise,” Progress In Electromagnetics Re-

search, vol. 149, pp. 217–249, 2014.

[39] P. J. Nahin, Oliver Heaviside: sage in solitude. IEEE Press New York, 1987.

[40] J. C. Maxwell, “Poems of James Clerk Maxwell,” https://mypoeticside.com/poets/

james-clerk-maxwell-poems.

[41] W. K. Panofsky and M. Phillips, Classical electricity and magnetism. Courier Corpo-
ration, 2005.

[42] T. Lancaster and S. J. Blundell, Quantum field theory for the gifted amateur. OUP
Oxford, 2014.

[43] J. D. Jackson, Classical Electrodynamics. John Wiley & Sons, 1962.

[44] W. C. Chew, “Fields and waves: Lecture notes for ECE 350 at UIUC,”
https://engineering.purdue.edu/wcchew/ece350.html, 1990.
464 Electromagnetic Field Theory

[45] C. M. Bender and S. A. Orszag, Advanced mathematical methods for scientists and
engineers I: Asymptotic methods and perturbation theory. Springer Science & Business
Media, 2013.
[46] J. M. Crowley, Fundamentals of applied electrostatics. Krieger Publishing Company,
1986.
[47] C. Balanis, Advanced Engineering Electromagnetics. Hoboken, NJ, USA: Wiley, 2012.
[48] J. D. Jackson, Classical electrodynamics. John Wiley & Sons, 1999.
[49] R. Courant and D. Hilbert, Methods of Mathematical Physics, Volumes 1 and 2. In-
terscience Publ., 1962.
[50] R. F. Harrington, Time-harmonic electromagnetic fields. McGraw-Hill, 1961.
[51] L. Esaki and R. Tsu, “Superlattice and negative differential conductivity in semicon-
ductors,” IBM Journal of Research and Development, vol. 14, no. 1, pp. 61–65, 1970.
[52] E. Kudeki and D. C. Munson, Analog Signals and Systems. Upper Saddle River, NJ,
USA: Pearson Prentice Hall, 2009.
[53] A. V. Oppenheim and R. W. Schafer, Discrete-time signal processing. Pearson Edu-
cation, 2014.
[54] E. C. Jordan and K. G. Balmain, Electromagnetic waves and radiating systems.
Prentice-Hall, 1968.
[55] G. Agarwal, D. Pattanayak, and E. Wolf, “Electromagnetic fields in spatially dispersive
media,” Physical Review B, vol. 10, no. 4, p. 1447, 1974.
[56] S. L. Chuang, Physics of photonic devices. John Wiley & Sons, 2012, vol. 80.
[57] B. E. Saleh and M. C. Teich, Fundamentals of photonics. John Wiley & Sons, 2019.
[58] M. Born and E. Wolf, Principles of optics: electromagnetic theory of propagation, in-
terference and diffraction of light. Elsevier, 2013, also 1959 to 1986.
[59] R. W. Boyd, Nonlinear optics. Elsevier, 2003.
[60] Y.-R. Shen, The principles of nonlinear optics. New York, Wiley-Interscience, 1984.
[61] N. Bloembergen, Nonlinear optics. World Scientific, 1996.
[62] P. C. Krause, O. Wasynczuk, and S. D. Sudhoff, Analysis of electric machinery.
McGraw-Hill New York, 1986.
[63] A. E. Fitzgerald, C. Kingsley, S. D. Umans, and B. James, Electric machinery.
McGraw-Hill New York, 2003, vol. 5.
[64] M. A. Brown and R. C. Semelka, MRI.: Basic Principles and Applications. John
Wiley & Sons, 2011.
Quantum Coherent State of Light 465

[65] C. A. Balanis, Advanced engineering electromagnetics. John Wiley & Sons, 1999, also
1989.
[66] Wikipedia, “Lorentz force,” https://en.wikipedia.org/wiki/Lorentz force/, accessed:
2019-09-06.
[67] R. O. Dendy, Plasma physics: an introductory course. Cambridge University Press,
1995.
[68] Wikipedia, “Kennelly-Heaviside Layer,” https://en.wikipedia.org/wiki/
Kennelly-Heaviside layer.
[69] ——, “Spectral line shape,” https://en.wikipedia.org/wiki/Spectral line shape.
[70] P. Sen and W. C. Chew, “The frequency dependent dielectric and conductivity response
of sedimentary rocks,” Journal of microwave power, vol. 18, no. 1, pp. 95–105, 1983.
[71] D. A. Miller, Quantum Mechanics for Scientists and Engineers. Cambridge, UK:
Cambridge University Press, 2008.
[72] W. C. Chew, “Quantum mechanics made simple: Lecture notes for ECE 487 at UIUC,”
http://wcchew.ece.illinois.edu/chew/course/QMAll20161206.pdf, 2016.
[73] B. G. Streetman and S. Banerjee, Solid state electronic devices. Prentice hall Englewood
Cliffs, NJ, 1995.
[74] Smithsonian, “This 1600-year-old goblet shows that the romans were
nanotechnology pioneers,” https://www.smithsonianmag.com/history/
this-1600-year-old-goblet-shows-that-the-romans-were-nanotechnology-pioneers-787224/,
accessed: 2019-09-06.
[75] K. G. Budden, Radio waves in the ionosphere. Cambridge University Press, 2009.
[76] R. Fitzpatrick, Plasma physics: an introduction. CRC Press, 2014.
[77] G. Strang, Introduction to linear algebra. Wellesley-Cambridge Press Wellesley, MA,
1993, vol. 3.
[78] K. C. Yeh and C.-H. Liu, “Radio wave scintillations in the ionosphere,” Proceedings of
the IEEE, vol. 70, no. 4, pp. 324–360, 1982.
[79] J. Kraus, Electromagnetics. McGraw-Hill, 1984, also 1953, 1973, 1981.
[80] Wikipedia, “Circular polarization,” https://en.wikipedia.org/wiki/Circular
polarization.
[81] Q. Zhan, “Cylindrical vector beams: from mathematical concepts to applications,”
Advances in Optics and Photonics, vol. 1, no. 1, pp. 1–57, 2009.
[82] H. Haus, Electromagnetic Noise and Quantum Optical Measurements, ser. Advanced
Texts in Physics. Springer Berlin Heidelberg, 2000.
466 Electromagnetic Field Theory

[83] W. C. Chew, “Lectures on theory of microwave and optical waveguides, for ECE 531
at UIUC,” https://engineering.purdue.edu/wcchew/course/tgwAll20160215.pdf, 2016.
[84] L. Brillouin, Wave propagation and group velocity. Academic Press, 1960.
[85] R. Plonsey and R. E. Collin, Principles and applications of electromagnetic fields.
McGraw-Hill, 1961.
[86] M. N. Sadiku, Elements of electromagnetics. Oxford University Press, 2014.
[87] A. Wadhwa, A. L. Dal, and N. Malhotra, “Transmission media,” https://www.
slideshare.net/abhishekwadhwa786/transmission-media-9416228.
[88] P. H. Smith, “Transmission line calculator,” Electronics, vol. 12, no. 1, pp. 29–31, 1939.
[89] F. B. Hildebrand, Advanced calculus for applications. Prentice-Hall, 1962.
[90] J. Schutt-Aine, “Experiment02-coaxial transmission line measurement using slotted
line,” http://emlab.uiuc.edu/ece451/ECE451Lab02.pdf.
[91] D. M. Pozar, E. J. K. Knapp, and J. B. Mead, “ECE 584 microwave engineering labora-
tory notebook,” http://www.ecs.umass.edu/ece/ece584/ECE584 lab manual.pdf, 2004.
[92] R. E. Collin, Field theory of guided waves. McGraw-Hill, 1960.
[93] Q. S. Liu, S. Sun, and W. C. Chew, “A potential-based integral equation method for
low-frequency electromagnetic problems,” IEEE Transactions on Antennas and Propa-
gation, vol. 66, no. 3, pp. 1413–1426, 2018.
[94] Wikipedia, “Snell’s law,” https://en.wikipedia.org/wiki/Snell’s law.
[95] G. Tyras, Radiation and propagation of electromagnetic waves. Academic Press, 1969.
[96] L. Brekhovskikh, Waves in layered media. Academic Press, 1980.
[97] Scholarpedia, “Goos-hanchen effect,” http://www.scholarpedia.org/article/
Goos-Hanchen effect.
[98] K. Kao and G. A. Hockham, “Dielectric-fibre surface waveguides for optical frequen-
cies,” in Proceedings of the Institution of Electrical Engineers, vol. 113, no. 7. IET,
1966, pp. 1151–1158.
[99] E. Glytsis, “Slab waveguide fundamentals,” http://users.ntua.gr/eglytsis/IO/Slab
Waveguides p.pdf, 2018.
[100] Wikipedia, “Optical fiber,” https://en.wikipedia.org/wiki/Optical fiber.
[101] Atlantic Cable, “1869 indo-european cable,” https://atlantic-cable.com/Cables/
1869IndoEur/index.htm.
[102] Wikipedia, “Submarine communications cable,” https://en.wikipedia.org/wiki/
Submarine communications cable.
Quantum Coherent State of Light 467

[103] D. Brewster, “On the laws which regulate the polarisation of light by reflexion from
transparent bodies,” Philosophical Transactions of the Royal Society of London, vol.
105, pp. 125–159, 1815.
[104] Wikipedia, “Brewster’s angle,” https://en.wikipedia.org/wiki/Brewster’s angle.
[105] H. Raether, “Surface plasmons on smooth surfaces,” in Surface plasmons on smooth
and rough surfaces and on gratings. Springer, 1988, pp. 4–39.
[106] E. Kretschmann and H. Raether, “Radiative decay of non radiative surface plasmons
excited by light,” Zeitschrift für Naturforschung A, vol. 23, no. 12, pp. 2135–2136, 1968.
[107] Wikipedia, “Homomorphic Encryption,” https://en.wikipedia.org/wiki/Homomorphic
encryption.
[108] ——, “Surface plasmon,” https://en.wikipedia.org/wiki/Surface plasmon.
[109] A. Sommerfeld, Über die Ausbreitung der Wellen in der drahtlosen Telegraphie. Verlag
der Königlich Bayerischen Akademie der Wissenschaften, 1909.
[110] W. Chew, Waves and Fields in Inhomogeneous Media, 378±381. Van Nostrand, 1990.
[111] Wikimedia, “Gaussian wave packet,” https://commons.wikimedia.org/wiki/File:
Gaussian wave packet.svg.
[112] Wikipedia, “Charles K. Kao,” https://en.wikipedia.org/wiki/Charles K. Kao.
[113] H. B. Callen and T. A. Welton, “Irreversibility and generalized noise,” Physical Review,
vol. 83, no. 1, p. 34, 1951.
[114] R. Kubo, “The fluctuation-dissipation theorem,” Reports on progress in physics, vol. 29,
no. 1, p. 255, 1966.
[115] C. Lee, S. Lee, and S. Chuang, “Plot of modal field distribution in rectangular and
circular waveguides,” IEEE transactions on microwave theory and techniques, vol. 33,
no. 3, pp. 271–274, 1985.
[116] W. C. Chew, Waves and Fields in Inhomogeneous Media. IEEE Press, 1996.
[117] M. Abramowitz and I. A. Stegun, Handbook of mathematical functions: with formulas,
graphs, and mathematical tables. Courier Corporation, 1965, vol. 55.
[118] ——, “Handbook of mathematical functions: with formulas, graphs, and mathematical
tables,” http://people.math.sfu.ca/∼cbm/aands/index.htm.
[119] W. C. Chew, W. Sha, and Q. I. Dai, “Green’s dyadic, spectral function, local density
of states, and fluctuation dissipation theorem,” arXiv preprint arXiv:1505.01586, 2015.
[120] Wikipedia, “Very Large Array,” https://en.wikipedia.org/wiki/Very Large Array.
[121] C. A. Balanis and E. Holzman, “Circular waveguides,” Encyclopedia of RF and Mi-
crowave Engineering, 2005.
468 Electromagnetic Field Theory

[122] M. Al-Hakkak and Y. Lo, “Circular waveguides with anisotropic walls,” Electronics
Letters, vol. 6, no. 24, pp. 786–789, 1970.

[123] Wikipedia, “Horn Antenna,” https://en.wikipedia.org/wiki/Horn antenna.

[124] P. Silvester and P. Benedek, “Microstrip discontinuity capacitances for right-angle

bends, t junctions, and crossings,” IEEE Transactions on Microwave Theory and Tech-
niques, vol. 21, no. 5, pp. 341–346, 1973.

[125] R. Garg and I. Bahl, “Microstrip discontinuities,” International Journal of Electronics

Theoretical and Experimental, vol. 45, no. 1, pp. 81–87, 1978.

[126] P. Smith and E. Turner, “A bistable fabry-perot resonator,” Applied Physics Letters,
vol. 30, no. 6, pp. 280–281, 1977.

[127] A. Yariv, Optical electronics. Saunders College Publ., 1991.

[128] Wikipedia, “Klystron,” https://en.wikipedia.org/wiki/Klystron.

[129] ——, “Magnetron,” https://en.wikipedia.org/wiki/Cavity magnetron.

[130] ——, “Absorption Wavemeter,” https://en.wikipedia.org/wiki/Absorption wavemeter.

[131] W. C. Chew, M. S. Tong, and B. Hu, “Integral equation methods for electromagnetic
and elastic waves,” Synthesis Lectures on Computational Electromagnetics, vol. 3, no. 1,
pp. 1–241, 2008.

[132] A. R. Choudhuri, The physics of fluids and plasmas: an introduction for astrophysicists.
Cambridge University Press, 1998.

[133] I. S. Gradshteyn and I. M. Ryzhik, Table of integrals, series, and products. Academic
press, 2014.

[134] L. Nagel and D. Pederson, “Simulation program with integrated circuit emphasis,” in
Midwest Symposium on Circuit Theory, 1973.

[135] T. M. Philip and M. J. Gilbert, “High-performance nanoscale topological inductor,” in

2017 75th Annual Device Research Conference (DRC). IEEE, 2017, pp. 1–2.

[136] Wikipedia, “Guglielmo Marconi,” https://en.wikipedia.org/wiki/Guglielmo Marconi.

[137] S. A. Schelkunoff and H. T. Friis, Antennas: theory and practice. Wiley New York,
1952, vol. 639.

[138] H. G. Schantz, “A brief history of UWB antennas,” IEEE Aerospace and Electronic
Systems Magazine, vol. 19, no. 4, pp. 22–26, 2004.

[139] E. Kudeki, “Fields and Waves,” http://remote2.ece.illinois.edu/∼erhan/FieldsWaves/

ECE350lectures.html.

[140] Wikipedia, “Antenna Aperture,” https://en.wikipedia.org/wiki/Antenna aperture.

Quantum Coherent State of Light 469

[141] L. Josefsson and P. Persson, Conformal array antenna theory and design. John Wiley
& Sons, 2006, vol. 29.
[142] R. J. Mailloux, Phased array antenna handbook. Artech House, 2017.
[143] J. G. Proakis, Digital signal processing: principles algorithms and applications. Pearson
Education India, 2001.
[144] R. W. P. King, G. S. Smith, M. Owens, and T. Wu, “Antennas in matter: Fundamentals,
theory, and applications,” NASA STI/Recon Technical Report A, vol. 81, 1981.
[145] B. Vaughn and D. Peroulis, “An updated applied formulation for the Goubau trans-
mission line,” Journal of Applied Physics, vol. 126, no. 19, p. 194902, 2019.
[146] Wikipedia, “Dipole Antenna,” https://en.wikipedia.org/wiki/Dipole antenna.
[147] ——, “Twin-Lead,” https://en.wikipedia.org/wiki/Twin-lead.
[148] H. Yagi and S. Uda, “Projector of the sharpest beam of electric waves,” Proceedings of
the Imperial Academy, vol. 2, no. 2, pp. 49–52, 1926.
[149] Wikipedia, “Yagi-Uda Antenna,” https://en.wikipedia.org/wiki/Yagi-Uda antenna.
[150] D. Dregely, R. Taubert, J. Dorfmüller, R. Vogelgesang, K. Kern, and H. Giessen, “3d
optical yagi–uda nanoantenna array,” Nature communications, vol. 2, no. 1, pp. 1–7,
2011.
[151] Antenna-theory.com, “Slot Antenna,” http://www.antenna-theory.com/antennas/
aperture/slot.php.
[152] A. D. Olver and P. J. Clarricoats, Microwave horns and feeds. IET, 1994, vol. 39.
[153] B. Thomas, “Design of corrugated conical horns,” IEEE Transactions on Antennas and
Propagation, vol. 26, no. 2, pp. 367–372, 1978.
[154] P. J. B. Clarricoats and A. D. Olver, Corrugated horns for microwave antennas. IET,
1984, no. 18.
[155] P. Gibson, “The Vivaldi aerial,” in 1979 9th European Microwave Conference. IEEE,
1979, pp. 101–105.
[156] Wikipedia, “Vivaldi Antenna,” https://en.wikipedia.org/wiki/Vivaldi antenna.
[157] ——, “Cassegrain Antenna,” https://en.wikipedia.org/wiki/Cassegrain antenna.
[158] ——, “Cassegrain Reflector,” https://en.wikipedia.org/wiki/Cassegrain reflector.
[159] W. A. Imbriale, S. S. Gao, and L. Boccia, Space antenna handbook. John Wiley &
Sons, 2012.
[160] J. A. Encinar, “Design of two-layer printed reflectarrays using patches of variable size,”
IEEE Transactions on Antennas and Propagation, vol. 49, no. 10, pp. 1403–1410, 2001.
470 Electromagnetic Field Theory

[161] D.-C. Chang and M.-C. Huang, “Microstrip reflectarray antenna with offset feed,” Elec-
tronics Letters, vol. 28, no. 16, pp. 1489–1491, 1992.

[162] G. Minatti, M. Faenzi, E. Martini, F. Caminita, P. De Vita, D. González-Ovejero,

M. Sabbadini, and S. Maci, “Modulated metasurface antennas for space: Synthesis,
analysis and realizations,” IEEE Transactions on Antennas and Propagation, vol. 63,
no. 4, pp. 1288–1300, 2014.

[163] X. Gao, X. Han, W.-P. Cao, H. O. Li, H. F. Ma, and T. J. Cui, “Ultrawideband and
high-efficiency linear polarization converter based on double v-shaped metasurface,”
IEEE Transactions on Antennas and Propagation, vol. 63, no. 8, pp. 3522–3530, 2015.

[164] D. De Schweinitz and T. L. Frey Jr, “Artificial dielectric lens antenna,” Nov. 13 2001,
US Patent 6,317,092.

[165] K.-L. Wong, “Planar antennas for wireless communications,” Microwave Journal,
vol. 46, no. 10, pp. 144–145, 2003.

[166] H. Nakano, M. Yamazaki, and J. Yamauchi, “Electromagnetically coupled curl an-

tenna,” Electronics Letters, vol. 33, no. 12, pp. 1003–1004, 1997.

[167] K. Lee, K. Luk, K.-F. Tong, S. Shum, T. Huynh, and R. Lee, “Experimental and simu-
lation studies of the coaxially fed U-slot rectangular patch antenna,” IEE Proceedings-
Microwaves, Antennas and Propagation, vol. 144, no. 5, pp. 354–358, 1997.

[168] K. Luk, C. Mak, Y. Chow, and K. Lee, “Broadband microstrip patch antenna,” Elec-
tronics letters, vol. 34, no. 15, pp. 1442–1443, 1998.

[169] M. Bolic, D. Simplot-Ryl, and I. Stojmenovic, RFID systems: research trends and
challenges. John Wiley & Sons, 2010.

[170] D. M. Dobkin, S. M. Weigand, and N. Iyer, “Segmented magnetic antennas for near-field
UHF RFID,” Microwave Journal, vol. 50, no. 6, p. 96, 2007.

[171] Z. N. Chen, X. Qing, and H. L. Chung, “A universal UHF RFID reader antenna,” IEEE
transactions on microwave theory and techniques, vol. 57, no. 5, pp. 1275–1282, 2009.

[172] S. Schelkunoff, “Some equivalence theorems of electromagnetics and their application

to radiation problems,” The Bell System Technical Journal, vol. 15, no. 1, pp. 92–112,
1936.

[173] C.-T. Chen, Linear system theory and design. Oxford University Press, Inc., 1998.

[174] S. H. Schot, “Eighty years of Sommerfeld’s radiation condition,” Historia mathematica,

vol. 19, no. 4, pp. 385–401, 1992.

[175] D. Hinton, Analects. Counterpoint, 2014.

[176] H. Bible, “New Revised Standard Version (NRSV),” Grand Rapids MI: Zondervan,
1989.
Quantum Coherent State of Light 471

[177] V. Rumsey, “Reaction concept in electromagnetic theory,” Physical Review, vol. 94,
no. 6, p. 1483, 1954.

[178] R. E. Collin, Foundations for microwave engineering. John Wiley & Sons, 2007, also
1966.

[179] W. J. Hoefer, “The transmission-line matrix method-theory and applications,” IEEE

Transactions on Microwave Theory and Techniques, vol. 33, no. 10, pp. 882–893, 1985.

[180] A. Ruehli, G. Antonini, and L. Jiang, Circuit oriented electromagnetic modeling using
the PEEC techniques. John Wiley & Sons, 2017.

[181] A. Ishimaru, Electromagnetic wave propagation, radiation, and scattering from funda-
mentals to applications. Wiley Online Library, 2017, also 1991.

[182] A. E. H. Love, “I. the integration of the equations of propagation of electric waves,”
Philosophical Transactions of the Royal Society of London. Series A, Containing Papers
of a Mathematical or Physical Character, vol. 197, no. 287-299, pp. 1–45, 1901.

[183] Wikipedia, “Christiaan Huygens,” https://en.wikipedia.org/wiki/Christiaan Huygens.

[184] ——, “George Green (mathematician),” https://en.wikipedia.org/wiki/George Green

(mathematician).

[185] C.-T. Tai, Dyadic Green’s Functions in Electromagnetic Theory. PA: International
Textbook, Scranton, 1971.

[186] ——, Dyadic Green functions in electromagnetic theory. Institute of Electrical &
Electronics Engineers (IEEE), 1994.

[187] W. Franz, “Zur formulierung des huygensschen prinzips,” Zeitschrift für Naturforschung
A, vol. 3, no. 8-11, pp. 500–506, 1948.

[188] Wikipedia, “Faraday Cage,” https://en.wikipedia.org/wiki/Faraday cage.

[189] J. A. Stratton, Electromagnetic Theory. McGraw-Hill Book Company, Inc., 1941.

[190] W. Meissner and R. Ochsenfeld, “Ein neuer effekt bei eintritt der supraleitfähigkeit,”
Naturwissenschaften, vol. 21, no. 44, pp. 787–788, 1933.

[191] Wikipedia, “Superconductivity,” https://en.wikipedia.org/wiki/Superconductivity.

[192] D. Sievenpiper, L. Zhang, R. F. Broas, N. G. Alexopolous, and E. Yablonovitch, “High-

impedance electromagnetic surfaces with a forbidden frequency band,” IEEE Transac-
tions on Microwave Theory and techniques, vol. 47, no. 11, pp. 2059–2074, 1999.

[193] Wikipedia, “Snell’s law,” https://en.wikipedia.org/wiki/Snell’s law.

[194] H. Lamb, “On sommerfeld’s diffraction problem; and on reflection by a parabolic mir-
ror,” Proceedings of the London Mathematical Society, vol. 2, no. 1, pp. 190–203, 1907.
472 Electromagnetic Field Theory

[195] W. J. Smith, Modern optical engineering. McGraw-Hill New York, 1966, vol. 3.

[196] D. C. O’Shea, T. J. Suleski, A. D. Kathman, and D. W. Prather, Diffractive optics:

design, fabrication, and test. Spie Press Bellingham, WA, 2004, vol. 62.

[197] J. B. Keller and H. B. Keller, “Determination of reflected and transmitted fields by

geometrical optics,” JOSA, vol. 40, no. 1, pp. 48–52, 1950.

[198] G. A. Deschamps, “Ray techniques in electromagnetics,” Proceedings of the IEEE,

vol. 60, no. 9, pp. 1022–1035, 1972.

[199] R. G. Kouyoumjian and P. H. Pathak, “A uniform geometrical theory of diffraction for

an edge in a perfectly conducting surface,” Proceedings of the IEEE, vol. 62, no. 11, pp.
1448–1461, 1974.

[200] R. Kouyoumjian, “The geometrical theory of diffraction and its application,” in Nu-
merical and Asymptotic Techniques in Electromagnetics. Springer, 1975, pp. 165–215.

[201] S.-W. Lee and G. Deschamps, “A uniform asymptotic theory of electromagnetic diffrac-
tion by a curved wedge,” IEEE Transactions on Antennas and Propagation, vol. 24,
no. 1, pp. 25–34, 1976.

[202] Wikipedia, “Fermat’s principle,” https://en.wikipedia.org/wiki/Fermat’s principle.

[203] N. Yu, P. Genevet, M. A. Kats, F. Aieta, J.-P. Tetienne, F. Capasso, and Z. Gaburro,
“Light propagation with phase discontinuities: generalized laws of reflection and refrac-
tion,” Science, vol. 334, no. 6054, pp. 333–337, 2011.

[204] X. Ni, N. K. Emani, A. V. Kildishev, A. Boltasseva, and V. M. Shalaev, “Broadband

light bending with plasmonic nanoantennas,” Science, vol. 335, no. 6067, pp. 427–427,
2012.

[205] A. Sommerfeld, Partial differential equations in physics. Academic Press, 1949, vol. 1.

[206] R. Haberman, Elementary applied partial differential equations. Prentice Hall Engle-
wood Cliffs, NJ, 1983, vol. 987.

[207] G. A. Deschamps, “Gaussian beam as a bundle of complex rays,” Electronics letters,

vol. 7, no. 23, pp. 684–685, 1971.

[208] J. Enderlein and F. Pampaloni, “Unified operator approach for deriving hermite–
gaussian and laguerre–gaussian laser modes,” JOSA A, vol. 21, no. 8, pp. 1553–1558,
2004.

[209] D. L. Andrews, Structured light and its applications: An introduction to phase-structured

beams and nanoscale optical forces. Academic Press, 2011.

[210] J. W. Strutt, “Xv. on the light from the sky, its polarization and colour,” The London,
Edinburgh, and Dublin Philosophical Magazine and Journal of Science, vol. 41, no. 271,
pp. 107–120, 1871.
Quantum Coherent State of Light 473

[211] L. Rayleigh, “X. on the electromagnetic theory of light,” The London, Edinburgh, and
Dublin Philosophical Magazine and Journal of Science, vol. 12, no. 73, pp. 81–101, 1881.
[212] S. Sun, Y. G. Liu, W. C. Chew, and Z. Ma, “Calderón multiplicative preconditioned efie
with perturbation method,” IEEE Transactions on Antennas and Propagation, vol. 61,
no. 1, pp. 247–255, 2012.
[213] G. Mie, “Beiträge zur optik trüber medien, speziell kolloidaler metallösungen,” Annalen
der physik, vol. 330, no. 3, pp. 377–445, 1908.
[214] Wikipedia, “Mie scattering,” https://en.wikipedia.org/wiki/Mie scattering.
[215] L. B. Felsen and N. Marcuvitz, Radiation and scattering of waves. John Wiley & Sons,
1994, also 1973, vol. 31.
[216] P. P. Ewald, “Die berechnung optischer und elektrostatischer gitterpotentiale,” Annalen
der physik, vol. 369, no. 3, pp. 253–287, 1921.
[217] E. Whitaker and G. Watson, A Course of Modern Analysis. Cambridge Mathematical
Library, 1927.
[218] J. Kong, “Electromagnetic fields due to dipole antennas over stratified anisotropic me-
dia,” Geophysics, vol. 37, no. 6, pp. 985–996, 1972.
[219] Wikipedia, “FLOPS,” https://en.wikipedia.org/wiki/FLOPS.
[220] ——, “Computer,” https://en.wikipedia.org/wiki/computer.
[221] W. C. H. McLean, Strongly elliptic systems and boundary integral equations. Cambridge
University Press, 2000.
[222] G. C. Hsiao and W. L. Wendland, Boundary integral equations. Springer, 2008.
[223] P. K. Banerjee and R. Butterfield, Boundary element methods in engineering science.
McGraw-Hill London, 1981, vol. 17.
[224] O. C. Zienkiewicz, R. L. Taylor, P. Nithiarasu, and J. Zhu, The finite element method.
McGraw-Hill London, 1977, vol. 3.
[225] J.-F. Lee, R. Lee, and A. Cangellaris, “Time-domain finite-element methods,” IEEE
Transactions on Antennas and Propagation, vol. 45, no. 3, pp. 430–442, 1997.
[226] J. L. Volakis, A. Chatterjee, and L. C. Kempel, Finite element method electromagnetics:
antennas, microwave circuits, and scattering applications. John Wiley & Sons, 1998,
vol. 6.
[227] J.-M. Jin, The finite element method in electromagnetics. John Wiley & Sons, 2015.
[228] G. Strang, Linear algebra and its applications. Academic Press, 1976.
[229] Cramer and Gabriel, Introduction a l’analyse des lignes courbes algebriques par Gabriel
Cramer... chez les freres Cramer & Cl. Philibert, 1750.
474 Electromagnetic Field Theory

[230] J. A. Schouten, Tensor analysis for physicists. Courier Corporation, 1989.

[231] A. C. Polycarpou, “Introduction to the finite element method in electromagnetics,”

Synthesis Lectures on Computational Electromagnetics, vol. 1, no. 1, pp. 1–126, 2005.

[232] J. P. A. Bastos and N. Sadowski, Electromagnetic modeling by finite element methods.

CRC press, 2003.

[233] Ö. Özgün and M. Kuzuoğlu, MATLAB-based Finite Element Programming in Electro-
magnetic Modeling. CRC Press, 2018.

[234] R. Coifman, V. Rokhlin, and S. Wandzura, “The fast multipole method for the wave
equation: A pedestrian prescription,” IEEE Antennas and Propagation magazine,
vol. 35, no. 3, pp. 7–12, 1993.

[235] J. Song, C.-C. Lu, and W. C. Chew, “Multilevel fast multipole algorithm for electro-
magnetic scattering by large complex objects,” IEEE Transactions on Antennas and
Propagation, vol. 45, no. 10, pp. 1488–1493, 1997.

[236] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical recipes

3rd edition: The art of scientific computing. Cambridge University Press, 2007.

[237] K. Yee, “Numerical solution of initial boundary value problems involving maxwell’s
equations in isotropic media,” IEEE Transactions on Antennas and Propagation,
vol. 14, no. 3, pp. 302–307, 1966.

[238] A. Taflove, “Review of the formulation and applications of the finite-difference time-
domain method for numerical modeling of electromagnetic wave interactions with arbi-
trary structures,” Wave Motion, vol. 10, no. 6, pp. 547–582, 1988.

[239] A. Taflove and S. C. Hagness, Computational electrodynamics: the finite-difference time-

domain method. Artech house, 2005, also 1995.

[240] W. Yu, R. Mittra, T. Su, Y. Liu, and X. Yang, Parallel finite-difference time-domain
method. Artech House Norwood, 2006.

[241] D. Potter, “Computational physics,” 1973.

[242] W. F. Ames, Numerical methods for partial differential equations. Academic press,
2014, also 1977.

[243] K. W. Morton, Revival: Numerical Solution Of Convection-Diffusion Problems (1996).

CRC Press, 2019.

[244] K. Aki and P. G. Richards, Quantitative seismology, 2002.

[245] F. B. Hildebrand, Introduction to numerical analysis. Courier Corporation, 1987.

[246] W. C. Chew, “Electromagnetic theory on a lattice,” Journal of Applied Physics, vol. 75,
no. 10, pp. 4843–4850, 1994.
Quantum Coherent State of Light 475

[247] J. v. Neumann, Mathematische Grundlagen der Quantenmechanik, Berlın. Springer,

New York, Dover Publications, 1943.

[248] R. Courant, K. Friedrichs, and H. Lewy, “Über die partiellen differenzengleichungen der
mathematischen physik,” Mathematische annalen, vol. 100, no. 1, pp. 32–74, 1928.

[249] M. Desbrun, A. N. Hirani, M. Leok, and J. E. Marsden, “Discrete exterior calculus,”

arXiv preprint math/0508341, 2005.

[250] W. C. Chew, “Vector potential electromagnetics with generalized gauge for inhomoge-
neous media: Formulation,” Progress In Electromagnetics Research, vol. 149, pp. 69–84,
2014.

[251] J.-P. Berenger, “A perfectly matched layer for the absorption of electromagnetic waves,”
Journal of computational physics, vol. 114, no. 2, pp. 185–200, 1994.

[252] W. C. Chew and W. H. Weedon, “A 3d perfectly matched medium from modified

maxwell’s equations with stretched coordinates,” Microwave and optical technology let-
ters, vol. 7, no. 13, pp. 599–604, 1994.

[253] W. C. Chew, J. Jin, and E. Michielssen, “Complex coordinate system as a generalized

absorbing boundary condition,” in IEEE Antennas and Propagation Society Interna-
tional Symposium 1997. Digest, vol. 3. IEEE, 1997, pp. 2060–2063.

[254] B. Bapat, “Newton’s rings,” http://www.iiserpune.ac.in/∼bhasbapat/phy221 files/

NewtonsRing.pdf.

[255] Wikipedia, “Double-slit experiment,” https://en.wikipedia.org/wiki/Double-slit

experiment.

[256] Shmoop.Com, “Young’s double-slit,” https://www.shmoop.com/optics/

young-double-slit.html.

[257] Wikipedia, “John Dalton,” https://en.wikipedia.org/wiki/John Dalton.

[258] ——, “Max Planck,” https://en.wikipedia.org/wiki/Max Planck.

[259] ——, “Photoelectric effect,” https://en.wikipedia.org/wiki/Photoelectric effect.

[260] ——, “Louis de Broglie,” https://en.wikipedia.org/wiki/Louis de Broglie.

[261] ——, “Newton’s laws of motion,” https://en.wikipedia.org/wiki/Newton’s laws of

motion.

[262] ——, “Quantum electrodynamics,” https://en.wikipedia.org/wiki/Quantum

electrodynamics.

[263] C. Christopoulos, The transmission-line modeling method: TLM. IEEE New York,
1995, vol. 221.
476 Electromagnetic Field Theory

[264] Wikipedia, “William Rowan Hamilton,” https://en.wikipedia.org/wiki/William

Rowan Hamilton.
[265] W. C. Chew, A. Y. Liu, C. Salazar-Lazaro, D. Na, and W. E. I. Sha, “Hamilton equation,
commutator, and energy conservation,” Quantum Report, pp. 295–303, Dec 2019.

[266] Wikipedia, “Gaussian beam,” https://en.wikipedia.org/wiki/Gaussian beam.

[267] M. Kira and S. W. Koch, Semiconductor quantum optics. Cambridge University Press,
2011.
[268] Wikipedia, “Quantum harmonic oscillator,” https://en.wikipedia.org/wiki/Quantum
harmonic oscillator.
[269] ——, “Roy J. Glauber,” https://en.wikipedia.org/wiki/Roy J. Glauber.
[270] ——, “E.C. George Sudarshan,” https://en.wikipedia.org/wiki/E. C. George
Sudarshan.
[271] C. Gerry, P. Knight, and P. L. Knight, Introductory quantum optics. Cambridge
University Press, 2005.
Index

A source on top of a layered medium, 389 Bound electron case

horizontal electric dipole, 391 Drude-Lorentz-Sommerfeld model, 85
vertical electric dipole, 390 Boundary conditions, 41
Absorbing boundary conditions, 428 Ampere’s law, 46
Advective equations, 32 conductive media, 52
AM 920 station, 74 Faraday’s law, 44
Ampere’s law, 7, 186 Gauss’s law, 45, 48
Andrew Lloyd Weber Boundary-value problems
Cats, 84 BVP, 39
Anisotropic media, 69 Branch cuts, 396
Anisotropic medium, 93 Brewster’s angle, 155
Antenna Brief history
net gain, 335 electromagnetics, 6
radiation power density, 335 optics, 6
resonance tunneling, 304
transmit and receive patterns, 335 Cats
Aperture antenna, 208 Andrew Lloyd Weber, 84
Array antennas, 293 Cavity resonator, 219
broadside array, 296 cylindrical waveguide resonator, 221
endfire array, 297 full-width half maximum bandwidth,
far-field approximation, 294, 298 237
linear phase array, 295 pole location, 236
lobes, 296 quality factor, 233
radiation pattern, 295 quarter wavelength resonator, 221
Axial ratio, 96 transfer function, 236
transmission line model, 219
Back-of-cab formula, 40 Characteristic impedance, 214
Back-of-the-cab formula, 31, 187, 188 Circuit theory, 257
Bell’s theorem, 7 capacitance, 263
Bessel equation, 199 energy storage method, 264
Bessel function, 187, 199, 205 inductor, 262
Bi-anisotropic media, 69 Kirchhoff current law, 257
Biot-Savart law, 50 Kirchhoff voltage law, 257
Born, Max, 443 resistor, 263

477
478 Electromagnetic Field Theory

Circular waveguide, 187, 198, 205 Dirichlet boundary condition, 189, 194,
cut-off frequency, 200 201
eigenmode, 199 Discrete exterior calculus, 425
Circularly polarized wave, 208 Dispersion relation, 80, 147, 151
Coherent state, 451 plane wave, 75
Compass, 4 Divergence operator, physical meaning, 21
Complex power, 65 Divergence, definition of, 17
Computational electromagnetics, 401 Drude-Lorentz-Sommerfeld model, 82,
Conductive media, 69 169, 437
highly conductive case, 80 bound electron case, 85
lowly conductive case, 81 broad applicability of, 86
Conjugate gradient method, 412 damping or dissipation case, 85
Constitutive relations, 29, 69 Duality principle, 140, 149, 162, 212, 214
anisotropic media, 71
bi-anisotropic media, 72 Effective aperture, 336, 338
inhomogeneous media, 72 Effective mass, 88, 89
isotropic frequency dispersive media, Einstein, 2
69 Electric current, 342
nonlinear media, 73 Electric dipole, 142
uniaxial and biaxial media, 72 Electric susceptibility, 30
Constructive interference, 175, 176, 178 Electromagnetic compatibility, 349
Copenhagen school Electromagnetic interference, 349
quantum interpretation, 7 Electromagnetic theory
Corpuscular nature of light, 103 unification, 6
Coulomb gauge, 41, 247 Electromagnetic wave, 73
Coulomb’s Law, 8 triumph of Maxwell’s equations, 30
Curl operator, definition, 22 Electromagnetics, 1
Curl operator, physical meaning, 26 history, 1
importance, 1
Damping or dissipation case Electrostatics, 34
Drude-Lorentz-Sommerfeld model, 85 Equivalence theorems, 339
de Broglie, Louis, 433 general case, 341
Derivation inside-out case, 340
Gauss’s law from Coulomb’s law, 12 outside-in case, 341
Diagonalization, 92 Euler’s formula, 62
Dielectric slab, 152 Evanescent wave, 152, 159
Dielectric waveguide, 175 plasma medium, 84
cut-off frequency, 181, 182, 184 Extinction theorem, 341, 345
TE polarization, 178
TM polarization, 183 Far zone, 300
Differential equation, 401 Faraday rotation, 91
Differential forms, 425 Faraday’s law, 7
Dirac, Paul, 2 Fermat’s principle, 363, 392
Directive gain pattern, 276, 289 Finite-Difference Time-Domain, 415
directivity, 290 grid-dispersion error, 422
Quantum Coherent State of Light 479

stability analysis, 420 Gyrotropic medium, 93

Yee algorithm, 424
Folded dipole, 305 Hamilton, William Rowan, 438
Fourier transform technique, 64, 65 Hamiltonian mechanics, 438
Frequency dispersive media, 69, 88 Hankel function, 199, 205
Frequency domain analysis, 61 Heaviside, 28
Fresnel reflection coefficient, 145 Heaviside layer
Fresnel zone, 300 Kennelly-Heaviside layer, 84
Fresnel, Austin-Jean, 145 Heinrich Hertz, 271
Functional, 411 experiment, 5
Functional space, 241 Helmholtz equation, 116, 170, 187, 188
Hilbert space, 242 Hermite polynomial, 369
Fundamental mode, 192 Hermite-Gaussian functions, 441
Hermitian matrix, 240
Gauge theory, 2 Hertzian dipole, 271, 301
Gauss’s divergence theorem, 17, 19 far field, 276
Gauss’s Law, 11 near field, 275
Gauss’s law point source approximation, 273
differential operator form, 21 radiation resistance, 279
Gaussian beam, 366 Hidden variable theorey, 7
Gauss’s law, 7 Hollow waveguide, 185, 190, 211
Gedanken experiment, 339 cut-off frequency, 192
General reflection coefficient, 122 cut-off mode, 190
Generalized impedance, 123 cut-off wavelength, 192
Generalized reflection coefficient, 137, eigenmode, 191, 194
166, 175 eigenvalue, 191
Generalized transverse resonance eigenvector, 191
condition, 175, 176, 223 TE mode, 186, 190, 212
Geometrical optics, 362 TM mode, 186, 189
Goos-Hanschen shift, 152, 178 Homogeneous solution, 159, 173, 219, 234,
Goubau line, 305, 313 235, 318, 321–323
Green’s function Hooke’s law, 437
dyadic, 347 Horn antenna, 208, 307
scalar, 344 Huygens’ principle
static, 36 electromagnetic waves case, 346
Green’s function method, 248 scalar waves case, 344
Green’s theorem, 344 Huygens, Christiaan, 343
vector, 347 Hybrid modes, 210
Green, George, 343
Group velocity, 170, 182, 192, 194 Image theory, 351
Guidance condition, 175, 177, 178, 183, electric charges, 352
191 electric dipoles, 352
Guided mode, 173, 213 magnetic charges, 353
Gyrotropic media magnetic dipoles, 353
Faraday rotation, 91 multiple images, 355
480 Electromagnetic Field Theory

perfect magnetic conductor surfaces, Marconi, Guglielmo, 271

354 Matrix eigenvalue problem, 191, 240
Impressed current, 56–58, 293, 328, Maxwell, 28
331–334, 342 lifespan, 28
Inhomogeneous media, 69 marriage, 28
Initial value proble, IVP, 323 poems, 28
Integral equation, 402 Maxwell’s equations
Intrinsic impedance, 77, 99 accuracy, 2
broad impact, 2
James Clerk Maxwell differential operator form, 26
displacement current, 5 integral form, 7
Jump condition, 41, 43, 48 quantum regime, 2
Jump discontinuity, 43, 45–48 relativistic invariance, 2
valid over vast length scale, 1
Kao, Charles, 176
Yang-Mills theory, 2
Kernel, 403
Maxwell’s theory
Kinetic inductance, 212
wave phenomena, 5
Laguere polynomials, 369 Maxwell, James Clerk, 1
associated, 369 Metasurfaces, 365
Laplace’s equation, 36 Microstrip patch antenna, 307
Laplacian operator, 31, 35 Mie scattering, 379
Lateral wave, 152 Mode conversion, 216
Layered media, 166 Mode matching method, 216
LC tank, 437 Mode orthogonality, 240
Leap-frog scheme, 426 resonant cavity modes, 244
Legendre polynomial waveguide modes, 242
associate, 382 Momentum density, 103
Lenz’s law, 262 Mono-chromatic signal, 168
Linear phase array, 295
Linear time-invariant system, 88 Nano-fabrication, 6
Lode stone, 4 Nanoparticles, 89
Lorentz force law, 82, 91 Natural solution, 219, 235, 321, 322
Lorenz gauge, 248 Near zone, 300
Lossless conditions, 105, 106 Negative resistance, 58
Lossy media, 69 Neumann boundary condition, 188, 190,
Love’s equivalence theorem, 404 191, 200
Neumann function, 199
Magnetic charge, 142, 249 Nonlinear media, 69
Magnetic current, 142, 249, 343 Normalized generalized impedance, 123
Magnetic dipole, 142 Null space solution, 173, 322
Magnetic monopole, 142 Numerical methods, 401
Magnetic source, 140
Magnetostatics, 39 Oliver Heaviside, 84
Marconi One-way wave equations, 32
experiment, 84 Optical fiber, 152, 210
Quantum Coherent State of Light 481

Optical theorem, 380 complex, 104

Optimization, 410 instantaneous, 56
Poynting’s vector
Paraxial wave, 206 complex, 66, 99
Paraxial wave equation, 366 instantaneous, 58, 65
Pendulum, 434 Probability distribution function, 443
Perfect electric conductor, 351
Perfectly matched layers, 429 Quantum electrodynamics, 434
Permeability, 30 Quantum harmonic oscillator, 440, 448
Permittivity, 30 Quantum theory, 2
Phase matching condition, 147, 150, 157, Quasi-optical antennas, 309
175 Quasi-plane wave, 366
Phase shift, 152 Quasi-static electromagnetic theory, 250,
Phase velocity, 77, 88, 168, 182, 192, 210 255
Phasor technique, 61, 62, 79, 83 Quasi-TEM mode, 210
phasor technique, 65
Photo-electric effect, 6 Radiation condition; Sommerfeld
Photoelectric effect, 433 radiation condition, 325
Photon number state, 441 Radiation fields, 284
Pilot potential, 196, 205 effective aperture, 290
Pilot potential method, 186 effective area, 290
Pilot vector, 187 far-field approximation, 284
Planck radiation law, 6 locally plane wave approximation,
Planck, Max, 433 286
Plane waves Rayleigh distance, 299
in lossy conductive media, 79 Rayleigh scattering, 371
uniform, 75 Reaction
Plasma medium, 191 Rumsey, 330
cold collisionless, 83 Reactive power, 67
Plasmonics, 89 Reciprocity theorem, 342
Poisson’s equation, 35 conditions, 331
Polarizability, 374 mathematical derivation, 328
Polarization, 94, 102 two-port network, 331
CP, 94 Rectangular cavity, 222
LHCP, 95 Rectangular waveguide, 187, 190, 194, 195
LHEP, 95 cut-off frequency, 194
RHCP, 95 TE mode, 214
RHEP, 95 TM mode, 214
Polarization density, 29 Reduced wave equation, 189
Power flow Reflection coefficient, 122, 135, 136, 145,
LHCP, 98 156, 158, 162
LHEP, 98 Reflector antenna, 278
RHCP, 99 Refractive index, 177
RHEP, 99 Relativistic invariance, 2
Poynting’s theorem Resonant solution, 219, 320–324
482 Electromagnetic Field Theory

Resonant solution, homogeneous solution, Surface plasmonic mode, 212

322 Surface plasmonic waveguide, 212
Resonator, 219
Retardation, 116 Tangent plane approximations, 362
RFID, 312 TE polarization, 146
Riemann sheets, 396 reflection coefficient, 148
Rumsey reaction theorem, 330 transmission coefficient, 148
Telegrapher’s equations, 118, 122, 214,
Scalar potential, 187, 190 215
electrodynamics, 247 frequency domain, 116
statics, 246 time domain, 115
Schrödinger equation, 440 Telegraphy
Separation of variables, 190, 198 lack of understanding, 5
Shielding, 349 TEM mode, 186
electrostatic, 349 Time-Harmonic fields, 62
relaxation time, 350 TM mode, 194
Shielding by conductive media, 52 TM polarization, 149
Slot antenna, 306 reflection coefficient, 149
Snell’s law, 147, 157, 364 Transmission coefficient, 149
generalized, 365 Toroidal antenna, 142
Sommerfeld identity, 387, 392 Total internal reflection, 150, 178, 181
Sommerfeld radiation condition, 325 Transmission coefficient, 156, 162, 163
Spatially dispersive, 69 Transmission line, 112, 121, 214
Special relativity, 2 capacitive effects, 113
Spectral representations, 383 characteristic impedance, 115, 119
point source, 384 current, 113, 122
Spherical Bessel equation, 382 distributed lumped element model,
Spherical Bessel function, 382 113
Spherical Hankel functions, 382 equivalent circuit, 135, 136, 138
Spherical harmonic, 382 frequency analysis, 116
Spherical Neumann function, 382 inductive effects, 113
Spin angular momentum, 102 load, 122
Standing wave, 190, 195 lossy, 118
Static electricity, 4 matched load, 123
Static electromagnetics parasitic circuit elements, 140
differential operator form, 34 stray/parasitic capacitances, 139
integral form, 8 stray/parasitic inductance, 139
Statics time-domain analysis, 113
electric field, 8 voltage, 113
Stationary phase method, 392 Transmission line matrix, 333, 434
Stokes’s theorem, 22 Transverse resonance condition, 174, 175,
structured lights, 370 220
Subspace projection, 407 Transverse wave number, 213
Surface current, 340 Trapped wave, 159
Surface plasmon, 158, 173 Traveling wave, 195
Quantum Coherent State of Light 483

Twin-lead transmission line, 305 statics, 246

Voltaic cell
Uncertainty principle, 445 telegraphy, 4
Uniqueness theorem, 317, 324, 339,
341–343, 348
Wave equations, 115
conditions, 320
Wave packet, 169
connection to poles of a linear
Wave Phenomenon
system, 323
frequency domain, 73
radiation from antenna sources, 324
Wave-particle duality, 431
Vector Poisson’s equation, 40 Weyl identity, 385, 392
Vector potential
electrodynamics, 247 Yagi-Uda antenna, 306
non-unique, 41 Young, Thomas, 431