Plasma Waves and Instabilities

Lecture notes (under development)

Last modified: February 28, 2022

Contents

Preface vi

I Introduction 1
1 Electromagnetic dispersion 2
1.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Electrodynamics in a linear medium . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Dispersion operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Waves in homogeneous linear media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Dispersion relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 Quasimonochromatic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 A sneak preview: waves in cold nonmagnetized plasma 12

2.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.1 Plasma model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.2 Dispersion properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Homogeneous plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 General considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Static magnetic-field mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Electrostatic Langmuir oscillations . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.4 Transverse electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Wave transformations in inhomogeneous plasma . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 WKB approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Field structure near a cutoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.3 Oblique incidence on a cutoff region . . . . . . . . . . . . . . . . . . . . . . . . 19

Appendices for Part I 23

AI.1 Analytic properties of response functions . . . . . . . . . . . . . . . . . . . . . . . . . 23

Problems for Part I 25

PI.1 Electrostatic approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
PI.2 Photon wave function in cold magnetized plasma . . . . . . . . . . . . . . . . . . . . . 25
PI.3 Beam–plasma instability (cold electrostatic limit) . . . . . . . . . . . . . . . . . . . . . 26
PI.4 Surface waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

i
CONTENTS ii

II Basic theory of quasimonochromatic waves 28

3 Asymptotic expansion of dispersion operators 29
3.1 Problem setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Wigner–Weyl transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Envelope equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 How to use the envelope equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4 Equations of geometrical optics 37

4.1 Scalar-wave model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Ray equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.1 Consistency relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.2 Hamilton’s equations for rays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.3 Alternative forms of the ray equations . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Amplitude equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4 ∗ Spin Hall effect of light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5 Wave action, energy, and momentum 45

5.1 Wave action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Wave energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Wave momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.4 Example: α channeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

Problems for Part II 53

PII.1 Single-wave dynamics within geometrical optics . . . . . . . . . . . . . . . . . . . . . 53
PII.2 Coupling of resonant waves, mode conversion . . . . . . . . . . . . . . . . . . . . . . 53

III Waves in plasmas: fluid theory 56

6 Waves in cold magnetized plasma 57
6.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.2 Susceptibility and dielectric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.3 General dispersion relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.4 Eigenmodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.4.1 Cutoffs and resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.4.2 Low-frequency limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.3 Parallel propagation (θ = 0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.4.4 Perpendicular propagation (θ = π/2) . . . . . . . . . . . . . . . . . . . . . . . . 67
6.4.5 Propagation at a general angle . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.4.6 ∗ Level repulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

7 Waves in warm fluid plasma 71

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.2 Nonmagnetized plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.2.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.2.2 Dielectric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.2.3 High-frequency oscillations: Langmuir waves . . . . . . . . . . . . . . . . . . . 75
7.2.4 Low-frequency oscillations: Debye shielding and ion sound . . . . . . . . . . . . 75
7.3 Magnetized plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

ii
CONTENTS iii

Problems for Part III 78

PIII.1 Methods of cold-plasma diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
PIII.2 Wave transformations in the ionosphere . . . . . . . . . . . . . . . . . . . . . . . . . 79
PIII.3 MHD waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

IV Waves in plasmas: kinetic theory 82

8 Introduction to kinetic theory of plasma waves 83
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.1.1 Distribution function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.1.2 Liouville’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.2 Vlasov equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2.1 Macroscopic fields and collision operator . . . . . . . . . . . . . . . . . . . . . . 85
8.2.2 Linearized Vlasov equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.3 Phase mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

9 Eigenmodes in kinetic theory 88

9.1 Case–van Kampen modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
9.2 Initial-value problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
9.2.1 Equation for the field spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
9.2.2 Dispersion relation and quasimodes . . . . . . . . . . . . . . . . . . . . . . . . . 92

10 Dispersion properties of nonmagnetized plasma 94

10.1 Dielectric properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
10.1.1 Susceptibility at Im ω > 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
10.1.2 Susceptibility at any Im ω: Landau’s rule . . . . . . . . . . . . . . . . . . . . . 94
10.1.3 General dielectric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.2 Dielectric properties of isotropic plasma . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.2.1 Dielectric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.2.2 Transverse waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.2.3 Longitudinal waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
10.3 Stability of electrostatic oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
10.3.1 Nyquist theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.3.2 Single-peak distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.3.3 Double-peak distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

11 Electrostatic waves in isotropic Maxwellian plasma 102

11.1 Susceptibility of Maxwellian plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.1.1 Plasma dispersion function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.1.2 An alternative derivation using Landau’s rule . . . . . . . . . . . . . . . . . . . 104
11.1.3 General susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.2 Asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.2.1 Warm species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
11.2.2 Hot species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
11.3 Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
11.3.1 Langmuir waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
11.3.2 Ion acoustic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

iii
CONTENTS iv

12 Landau damping and kinetic instabilities 110

12.1 Passing and trapped particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
12.2 Wave–particle energy exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
12.2.1 Direct Landau damping: wave dissipation . . . . . . . . . . . . . . . . . . . . . 112
12.2.2 Inverse Landau damping: wave amplification . . . . . . . . . . . . . . . . . . . 113
12.2.3 Saturated states, BGK waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

13 Dispersion properties of magnetized plasma 116

13.1 General dispersion operator from kinetic theory . . . . . . . . . . . . . . . . . . . . . . 116

14 Waves in magnetized plasma 118

14.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
14.2 Perpendicular propagation: general considerations . . . . . . . . . . . . . . . . . . . . 119
14.3 Waves in the upper-hybrid frequency range . . . . . . . . . . . . . . . . . . . . . . . . 119
14.3.1 Electrostatic approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
14.3.2 Electromagnetic dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
14.3.3 EBW application to plasma heating . . . . . . . . . . . . . . . . . . . . . . . . 122
14.4 Waves in the lower-hybrid frequency range . . . . . . . . . . . . . . . . . . . . . . . . . 123
14.4.1 Electrostatic dispersion relation . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
14.4.2 IBW application to plasma heating . . . . . . . . . . . . . . . . . . . . . . . . . 125

15 Collisionless dissipation in magnetized plasma 126

15.1 Single-particle picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
15.2 Power absorption in Maxwellian plasma: basic formulas . . . . . . . . . . . . . . . . . 127
15.3 Landau damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
15.4 Transit-time magnetic pumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
15.5 Cyclotron damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

16 Quasilinear theory 133

16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
16.1.1 One wave: nonlinearities due to trapped particles . . . . . . . . . . . . . . . . . 133
16.1.2 Two waves: Chirikov criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
16.1.3 Many waves: statistical quasilinear approach . . . . . . . . . . . . . . . . . . . 134
16.2 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
16.2.1 Equation for the distribution function . . . . . . . . . . . . . . . . . . . . . . . 135
16.2.2 Diffusion coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
16.2.3 Field equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
16.3 Properties and applications of quasilinear theory . . . . . . . . . . . . . . . . . . . . . 138
16.3.1 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
16.3.2 Quasilinear evolution: broadband bump-on-tail instability . . . . . . . . . . . . 139

17 Dressed particles and ponderomotive forces 141

17.1 Quasilinear theory for inhomogeneous plasma . . . . . . . . . . . . . . . . . . . . . . . 141
17.1.1 Equation for the average distribution . . . . . . . . . . . . . . . . . . . . . . . . 141
17.1.2 Oscillation centers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
17.2 Interaction with on-shell waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Appendices for Part IV 145

AIV.1 Plasma dispersion function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

iv
CONTENTS v

Problems for Part IV 147

PIV.1 Wave propagation: initial-value problem . . . . . . . . . . . . . . . . . . . . . . . . . 147
PIV.2 Longitudinal waves in Lorentzian plasma . . . . . . . . . . . . . . . . . . . . . . . . 147
PIV.3 Alfvén resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
PIV.4 Two-stream instability in Lorentzian plasma . . . . . . . . . . . . . . . . . . . . . . 149
PIV.5 Weibel instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
PIV.6 Plasma susceptibility in magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . . 150
PIV.7 Kinetic waves propagating parallel to magnetic field . . . . . . . . . . . . . . . . . . 151
PIV.8 Kinetic whistler waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
PIV.9 Cyclotron heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Appendices 154
A Abbreviations 154

B Notation 155

C Conventions 158
C.1 Numbers and matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
C.2 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

Bibliography 160

v
Preface

Plasmas support a wide variety of waves. These waves significantly determine plasma dynamics, and
they are also indispensable for plasma manipulation and diagnostics.

Studying plasma waves is an essential part of studying plasma physics and its applications. It is also
useful for understanding waves in general, because the complexity of waves in plasmas demands a
particularly systematic approach to wave theory.

This course is intended as an introduction into physics of (mostly linear) plasma waves and briefly
covers the following general topics:
• concept of linear dispersion;
• dispersion operators and their symbols;

• geometrical-optics approximation;
• envelope equation, ray tracing, mode conversion;
• transport of the wave action, energy, and momentum;

• dispersion properties of nonmagnetized and magnetized plasma within fluid and kinetic models;
• basic types of plasma waves and their applications to plasma manipulation and diagnostics;
• basic instabilities and mechanisms of collisionless dissipation;
• nonlinear saturation of kinetic instabilities, elements of quasilinear theory.

For in-depth discussions of these topics, see additional literature, for example, Refs. [1–8].

vi
 Part I

Introduction

The purpose of this first, intentionally haphazard, part of the course is to familiarize
readers with some basic vocabulary of plasma-wave theory. The concepts introduced in
this part will be used later for developing a more systematic theory.

1
Lecture 1

Electromagnetic dispersion

In this introductory lecture, we introduce the concept of electromagnetic dispersion in application to

a general linear medium, without focusing on effects specific to plasmas.

1.1 Basic equations

1.1.1 Maxwell’s equations
Any wave propagating in a medium that contains electric charges involves oscillations of electric
currents, which cause oscillations of electromagnetic fields. Because of this, studying waves in such
media usually starts with considering Maxwell’s equations,

∂t E = c∇ × B − 4πj (Ampere’s law), (1.1a)

∂t B = −c∇ × E (Faraday’s law), (1.1b)

∇ · E = 4πρ (electric Gauss’s law), (1.1c)
∇·B =0 (magnetic Gauss’s law), (1.1d)

where E and B are the electric and magnetic field, respectively, j is the current density, ρ is the
charge density, and c is the speed of light. (Gaussian units will be used throughout the course.)
Magnetic Gauss’s law can be considered as an initial condition for Faraday’s law, because

∂t (∇ · B) = ∇ · ∂t B = −c∇ · (∇ × E) = 0, (1.2)

where we used the fact that the divergence of a curl is identically zero. Similarly, electric Gauss’s law
can be considered as an initial condition for Ampere’s law, because

∂t (∇ · E − 4πρ) = ∂t (∇ · E) + 4π∇ · j = c∇ · (∇ × B) = 0, (1.3)

where we used charge conservation:

∂t ρ + ∇ · j = 0. (1.4)

In this sense, not all of Eqs. (1.1) are entirely independent from each other, and it will be sufficient
for us to use only a subset of them for studying waves. We will return to this subject later.

2
LECTURE 1. ELECTROMAGNETIC DISPERSION 3

1.1.2 Electrodynamics in a linear medium

Effects caused by a medium enter Eqs. (1.1) only through ρ and j, and ρ can be expressed through j
via Eq. (1.4); thus, a model for j of needed. In this course, we will mostly assume that the medium
(in our case, plasma) is linear. This means that j will be assumed to depend on E linearly,1 i.e.,

j = j (i) + j (f) , j (i) = σ̂E. (1.5)

The “induced” current density j (i) is determined by medium’s response to the electric field, and the
linear operator σ̂ that determines this response is called conductivity.2 The remaining, “free” current
density, j (f) , is independent of E; it includes the current density that is prescribed externally, and
it can also include a (generally time-dependent) current density that is determined by the initial
conditions. For this reason, specifying the representation (1.5) requires specifying the initial moment
of time since which the dynamics is considered. We will denote this moment as t0 ; then, by definition,

j(t = t0 ) = j (f) (t = t0 ), j (i) (t = t0 ) = 0. (1.6)

Below, we consider j (f) as prescribed and discuss how to model the induced current density j (i) .
In some cases, the latter can be as simple as

j (i) (t, x) = σ(t, x)E(t, x), (1.7)

where σ is some matrix function or even a scalar. This model is commonly known as Ohm’s law. It
is also called the local-response model, because it assumes that the current at a given location (t, x)
is determined by the field only at the same location (t, x). Such an approximation can be reasonable,
for example, for modeling strongly collisional media. However, in general, j (i) (t, x) can also depend
on E(t0 , x0 ) at (t0 , x0 ) other than (t, x). Such media are called dispersive. Nonlocality in time is called
temporal dispersion, and nonlocality in space is called spatial dispersion.
The general induced current in a dispersive medium can be expressed as a functional
 t  ∞
j (i) (t, x) = d t0 d x0 Σ(t, x, t0 , x0 )E(t0 , x0 ). (1.8)
t0 −∞

The tensor Σ(t, x, t0 , x0 ), which is the “coordinate” (time-space) representation of σ̂ (see also Box 1.1),
can be formally defined as a functional derivative
δj(t, x)
Σ(t, x, t0 , x0 ) = . (1.9)
δE(t0 , x0 )

In other words, it serves as the weight function that determines how much the field E(t0 , x0 ) contributes
to the current j (i) (t, x). The fact that the integration domain in Eq. (1.10) is limited to t0 < t reflects
causality; that is, the current at a given time t can be affected by past fields (t0 < t) but not by future
fields (t0 > t). One can also replace the upper integration limit t in Eq. (1.8) with ∞,
 ∞  ∞
j (i) (t, x) = d t0 d x0 Σ(t, x, t0 , x0 )E(t0 , x0 ), (1.10)
t0 −∞

assuming the convention that Σ(t, x, t0 , x0 ) = 0 for all t0 > t.

Below, we will often express σ̂ and related quantities through the frequency operator ω̂ and the
wavevector operator k̂:

ω̂ = i ∂t , k̂ = −i ∇,
. .
(1.11)
1 We do not involve B here because for oscillatory fields that we are interested in, B can always be expressed through E
using Faraday’s law. For stationary fields, the general form of the induced current density is j (i) = σ̂E + κ̂B.
2 We use caret ˆ to denote integral operators, including differential operators as a special case.

3
LECTURE 1. ELECTROMAGNETIC DISPERSION 4

Box 1.1: An alternative coordinate representation of σ̂

The current density j (i) is also often expressed in the following alternative form:
 ∞  ∞
t + t0 x + x0
 
(i)
j (t, x) = dt 0
d x Σ̄ 2 , 2 , t − t , x − x E(t0 , x0 ),
0 0 0
t0 −∞

where the “symmetrized” kernel Σ̄ is connected with the previously introduced Σ as follows:

Σ̄(t̄, x̄, τ, s) = Σ(t̄ + τ /2, x̄ + s/2, t̄ − τ /2, x̄ − s/2)

(and X̄ is defined similarly through X ). As to be discussed in Lecture 3, the Fourier transform

of Σ̄ with respect to τ and s is the Weyl symbol of σ̂, which is a more natural quantity than the
Fourier transform of Σ.

.
where = denotes definitions. In particular, using Eq. (1.4), one can express ρ through E via ω̂:

−i ω̂ρ + ∇ · j (f) + ∇ · (σ̂E) = 0. (1.12)

Although ω̂ is not invertible, one can define its right inverse ω̂ −1 as

 t
(ω̂ −1 f )(t) = −i d t0 f (t0 ),
.
ω̂ −1 ω̂ 6= ω̂ ω̂ −1 = 1̂. (1.13)
t0

Using this notation, one can write the solution of Eq. (1.12) as

ρ = ρ(f) − ∇ · (i ω̂ −1 σ̂E), (1.14)

where ρ(f) may include contributions from initial conditions and j (f) but is independent of E. It is
also convenient to introduce the susceptibility operator
4π i
χ̂ = σ̂ (1.15)
ω̂
(we assume the notation 1/ω̂ ≡ ω̂ −1 ), which we also represent as
 t  ∞
(χ̂E)(t, x) = d t0 d x0 X (t, x, t0 , x0 )E(t0 , x0 ), (1.16)
t0 −∞

and the dielectric operator

ˆ = 1̂ + χ̂. (1.17)

where 1̂ is a unit matrix operator. Then, Eq. (1.14) can be written as ρ = ρ(f) + ρ(i) . Here, the
“induced” charge density is given by

ρ(i) = −∇ · P (1.18)

and P is the so-called electric polarization, which is understood as the electric dipole moment per
unit volume [9] and is defined as3
. χ̂
P = E. (1.19)
4π
3 In this course, we define χ̂ using Stix’s notation [1]. Sometimes, though, the factor 4π is absorbed in the definition
.
of χ̂; then, ˆ
 = 1̂ + 4π χ̂ and P = χ̂E.

4
LECTURE 1. ELECTROMAGNETIC DISPERSION 5

Also, the field ˆE = E +4πP is called the electric displacement field. With these, Maxwell’s equations
(1.1) can be re-written as follows:

∂t (ˆE) = c∇ × B − 4πj (f) , (1.20a)

∂t B = −c∇ × E, (1.20b)

∇ · (ˆE) = 4πρ(f) , (1.20c)

∇ · B = 0. (1.20d)

1.1.3 Dispersion operators

Waves are usually defined as fields that are periodic or quasiperiodic in time and (or) space.4 For
the purpose of this course, we adopt a different definition; namely, the term “(linear) wave” will be
applied to any solution of Eqs. (1.20). In particular, we will discuss superpositions of quasiperiodic
fields, as well as rapidly growing and dissipating fields, which are generally not quasiperiodic. In this
sense, Eqs. (1.20) will be called wave equations.
We now seek to represent these equations in a more compact form that contains fewer independent
variables. Because the electric and magnetic Gauss’s laws can be considered as initial conditions
(Sec. 1.1.1), it will be enough for us to limit our consideration to Ampere’s and Faraday’s laws. Let
us consider the curl of Eq. (1.20b) and substitute Eq. (1.20a) for ∇ × B. Then, one obtains

1 ∂ 2 (ˆE) 4π ∂j (f)
∇×∇×E+ 2 2
=− 2 . (1.21)
c ∂t c ∂t
We will limit our consideration to regions where the external currents are zero. This does not auto-
matically mean that j (f) is zero, because the initial conditions can give rise to time-dependent currents
(Sec. 1.1.2), including currents resonant to waves of interest. Such currents can significantly affect
the wave propagation. However, we will postpone a detailed discussion of this subject until Part IV.
For the time being, let us simply assume that: (i) the integral (1.8) that defines j (i) converges at
t0 → −∞ and (ii) the initial conditions at t0 → −∞ do not affect waves of interest at finite t. One
can expect this to be a reasonable assumption, for example, in collisional media, because collisions
destroy information about the initial conditions naturally. Also, if the wave amplitude grows, then
j (i) is determined by E mainly from the recent past and, again, the initial conditions at t0 → −∞
should not matter. Hence, we adopt t0 → −∞, so σ̂E is given by the following integral that converges
by our assumption:
 t  ∞
(σ̂E)(t, x) = d t0 d x0 Σ(t, x, t0 , x0 )E(t0 , x0 ), (1.22)
−∞ −∞

and we also adopt j (f) = 0. Then, Eq. (1.21) can be written as

. c2
D̂ E E = 0, D̂ E = 2 (k̂k̂† − 1 k̂ 2 ) + ˆ. (1.23)
ω̂
The operator D̂ E will be called the (electromagnetic) dispersion operator.
Note that other representations of the wave equation are also possible. For example, it can be
convenient to use the electrostatic potential (Problem PI.1) or the magnetic field (Sec. 2.3.3) instead
of E, or even give up the very notion of the dielectric operator (Problem PI.2).
4 Global modes, which are described by discrete variables ψ(t), are also waves in the sense that ψ(t) can be considered

as time-dependent fields ψ(t, x) on zero-dimensional coordinate space (dim x = 0).

5
LECTURE 1. ELECTROMAGNETIC DISPERSION 6

1.2 Waves in homogeneous linear media

Even without source terms, Eq. (1.23) remains a complicated integro-differential equation that in
practice can be solved only numerically. However, it can be made tractable if the underlying medium
is homogeneous (both in time and in space) or weakly inhomogeneous. The case of weakly inhomo-
geneous media is more relevant for practical applications, but the corresponding theory is relatively
complicated, so it is left to future lectures (Part II). Here, we consider waves in strictly homogeneous
media in order to introduce some basic concepts that we will need to use later.

1.2.1 Basic concepts

A homogeneous linear medium is defined as a medium where Σ(t, x, t0 , x0 ) may depend on (t, x) only
relatively to (t0 , x0 ). Specifically, this means that Σ is representable as5

Σ(t, x, t0 , x0 ) = Σ̄(t − t0 , x − x0 ), (1.24)

which automatically leads to a similar expression for X :

X (t, x, t0 , x0 ) = X̄ (t − t0 , x − x0 ). (1.25)

Then, from Eq. (1.22),

 t  ∞
(σ̂E)(t, x) = d t0 d x0 Σ̄(t − t0 , x − x0 )E(t0 , x0 )
−∞ −∞
 ∞  ∞
= dτ d s Σ̄(τ, s)E(t − τ, x − s). (1.26)
0 −∞

Accordingly, Σ̄ab (t, x) can be understood as the ath component of the current, ja (t, x), produced by
the bth component of the electric field of the form6 Eb (t0 , x0 ) = δ(t0 − 0)δ(x0 ).
It is easy to see from here that homogeneous linear media support monochromatic waves, that is,
waves of the form (Box 1.2)

E(t, x) = E e −i ωt+i k·x , (1.27)

where the amplitude E , the frequency ω, and the wavevector k are complex constants. Indeed, in this
case, Eq. (1.26) leads to
 ∞  ∞
(σ̂E)(t, x) = dτ d s Σ̄(τ, s)E e −i ω(t−τ )+i k·(x−s) (1.28)
0 −∞
 ∞  ∞ 
= dτ d s Σ̄(τ, s) e i ωτ −i k·s E(t, x). (1.29)
0 −∞

Let us denote the expression in the square brackets as σ(ω, k),

 ∞  ∞
.
σ(ω, k) = dt d x Σ̄(t, x) e i ωt−i k·x . (1.30)
0 −∞

This function is called the spectral representation of σ̂, or simply the spectral conductivity, because
σ(ω, k) is obtained from Σ̄ by applying the Fourier transform in space and the Laplace transform
in time. The physical meaning of this quantity can be understood by noticing that according to
5 To put it differently, the function Σ̄ introduced in Box 1.1 is independent of t̄ and x̄.
6 The added 0 in the argument of the delta function
denotes an infinitesimal positive value.
This shift ensures that


the

t time integral over

the domain (t0 , t) is well defined, tt d t0 δ(t − t0 − 0)= 0t−t0 d τ δ(τ − 0)= −∞∞
d τ δ(τ ) = 1, while
t
0
d t0 δ(t − t0 )= ∞ d τ δ(τ ) is undefined. In our special 0case, t → −∞, but this is not essential.
0 0

6
LECTURE 1. ELECTROMAGNETIC DISPERSION 7

Box 1.2: Complex representation of real fields

Although the actual electric field is real, the sourceless wave equations that we work with allow
solutions in a complex form. These equations have real coefficients, so for any complex field E c
that is a solution (the index c stands for “complex”), the complex-conjugate field E ∗c is a solution
too, and so is any linear combination of the two. Hence, one can search for real E in the form

E = Re E c = 1/2 (E c + E ∗c ).

To simplify the notation, we will not distinguish E and E c where it is not essential.

Eq. (1.26), j is a convolution of Σ̄ and E; thus, its spectral representation j(ω, k) is simply propor-
tional to the spectral representation E(ω, k) of the electric field with the proportionality coefficient
σ(ω, k):7
j(ω, k) = σ(ω, k)E(ω, k). (1.31)
Since Σ̄(t, x) = 0 at t < 0, the integral over t in Eq. (1.30) can be formally extended to −∞,
 ∞  ∞
σ(ω, k) = dt d x Σ̄(t, x) e i ωt−i k·x . (1.32)
−∞ −∞

Still, it should not be confused with the Fourier transform, because ω is generally complex. Since
|Σ̄(t, x)e −i ωt+i k·x | = |Σ̄(t, x)|e −ω t , either ωi = Im ω should be sufficiently large or the response func-
i
.
tion Σ̄(t, x) should diminish rapidly enough at t → ∞ for the integral (1.30) to converge. In particular,
if Σ̄(t, x) ∝ e −νt with constant ν, then the integrals (1.30) and (1.32) exist provided that
ωi + ν > 0. (1.33)
These are the same requirements as those adopted in Sec. 1.1.3; that is, applicability of our theory is
facilitated by collisions and by exponential growth of the wave amplitude. See also Appendix AI.1 for
basic properties of response functions like σ.
From the spectral representation of σ̂, one readily finds the spectral representation of ˆ, which is
known as the dielectric tensor :
(ω, k) = 1 + χ(ω, k), (1.34)
with 1 being the unit matrix. Here, χ is the spectral representation of χ̂; it is defined by analogy
with Eq. (1.30) and given by
4π i
χ(ω, k) = σ(ω, k), (1.35)
ω
as shown in Box 1.3. Then, the wave equation (1.23) can be written as follows:8
D E (ω, k)E = 0, D E (ω, k) = N N † − 1 N 2 + (ω, k). (1.36)
The matrix D E is called a dispersion matrix or dispersion tensor, and N is the refractive-index vector:
.
N = ck/ω. (1.37)

7 See theory of the Fourier and Laplace transforms. We will revisit this topic in Part IV, where we will also discuss

properties of the Laplace transform in more detail.

8 Keep in mind that all functions of (ω, k) are introduced here strictly for homogeneous media. The only fundamental

objects in inhomogeneous media are operators, because their representations can be defined in more than one way and
can differ significantly. For example, “the dielectric tensor” of inhomogeneous plasma is undefined until a convention is
specified for the mapping ˆ  → (t, x, ω, k). Except in special settings, such as cold stationary plasmas, one should not
expect (t, x, ω, k) to be a meaningful quantity if it is obtained simply by taking (ω, k) of homogeneous plasma and
replacing the constant plasma parameters with functions of (t, x). We will revisit this issue in Part II (Box 3.2).

7
LECTURE 1. ELECTROMAGNETIC DISPERSION 8

Box 1.3: Derivation of Eq. (1.35)

From Eqs. (1.5), (1.15), (1.16), and (1.25), one finds

 t  ∞
4πj (i) = ∂t (χ̂E) = F (t, x) + d t0 d x0 [∂t X̄ (t − t0 , x − x0 )]E(t0 , x0 ),
t0 −∞
 ∞
.
F (t, x) = d x0 X̄ (0, x − x0 )E(t, x0 ).
t0

For t = t0 , this gives F (t0 , x) = 4πj (i) (t0 , x) = 0 for any E; thus, one also has X̄ (0, x) = 0 at
all x. Then, the above equation can be written as
 t  ∞
4π(σ̂E)(t, x) = dt 0
d x0 [∂t X̄ (t − t0 , x − x0 )]E(t0 , x0 ),
t0 −∞

which means that ∂t X̄ = 4π Σ̄. Then, from Eq. (1.30), one finds
 ∞  ∞
4πσ(ω, k) = dt d x [∂t X̄ (t, x)] e i ωt−i k·x = i ωχ(ω, k),
0 −∞

where we used integration by parts and X̄ (0, x) = 0. This leads to Eq. (1.35).

1.2.2 Dispersion relations

As seen from the previous section, equations describing electromagnetic waves in linear media can be
expressed as
Dab (ω, k)ψb = 0. (1.38)
The function ψ is an electric field in Eq. (1.36), but in general it can also be a different field (for
example, see Problems PI.1 and PI.2), so we leave it unspecified in this section. Equation (1.38)
indicates that ψ can be nonzero only if

det D(ω, k) = 0, (1.39)

or simply D = 0 when D = D is scalar. This can be considered as an equation for ω(k), which is called
a dispersion relation. The solutions for ω at given k, ω = ωq (k), are called dispersion branches. Note
that they are generally complex, and we will assume the following notation throughout the course:

ω = ωr + i ωi ,
. .
ωr = Re ω, ωi = Im ω. (1.40)

Depending on the number of these branches b,9 or the order of the wave equation, the waves are
attributed as scalar waves (b = 1) or as vector waves (b > 1); see also Box 1.4.
A spatially monochromatic field with a given wavevector k can contain contributions from multiple
branches, called (eigen)modes, and can be expressed as follows:10

ψ̄ q e −i ωq (k)t+i k·x ,
X
ψ(t, x) = (1.41)
q

9 Integral
wave equations generally yield infinitely many branches and will be discussed in Part IV.
10 When solutions for ω are degenerate, it is in principle possible to have solutions ∝ ta e −i ωq (k)t+i k·x with natural a.
However, this is not typical for waves of our interest, so we will not consider such cases.

8
LECTURE 1. ELECTROMAGNETIC DISPERSION 9

Box 1.4: Scalar waves vs. vector waves

Any differential equation ∂tm ψ = F[ψ, ∂t ψ, . . . ∂tm−1 ψ] (where F may be an integral transform
in x) can be represented as ∂t ψ̃ = F̄(ψ̃), where ψ̃ = (ψ̃ 1 , . . . ψ̃ m−1 )| is a vector of dimension
. . .
m × dim ψ, with ψ̃ 1 = ψ, ψ̃ 2 = ∂t ψ, . . . , ψ̃ m = ∂tm−1 ψ̃. Assuming F is linear, such an equa-
tion generally has b = m × dim ψ roots, so it describes vector waves unless ψ is scalar and m = 1.

For example, consider the Schrödinger equation and the Klein–Gordon equation from quantum
mechanics. (As will be discussed later, they also emerge in plasma-wave theory.) The Schrödinger
equation has dim ψ = m = 1, so b = 1; i.e., it has only one dispersion branch and thus the
corresponding waves are true scalar waves. The Klein–Gordon equation also has dim ψ = 1; but
in this case, m = 2, so b = 2, and the corresponding waves are vector waves. The representation
of the Klein–Gordon equation as a first-order equation for a two-dimensional vector is known as
the Feshbach–Villars representation. A similar reduction of plasma-wave equations to first-order
vector equations is discussed, for example, in Problem PI.2.

where the coefficients ψ̄ q are determined by the initial conditions. Each of these coefficients must
satisfy D[ωq (k), k]ψ̄ q = 0. Thus, it can be expressed as ψ̄ q = Aq hq (k), where Aq is a scalar amplitude
and hq (k) is a unit polarization vector 11 defined via

D[ωq (k), k] hq (k) = 0, |hq | = 1. (1.42)

In general, dim hq = dim ψ can be different from dim x (Problem PI.2), in which case hq may not
have a clear physical meaning. But let us mention the important special case when ψ is a complexified
(Box 1.2) single-mode electric field on a three-dimensional space, with the real field being

E(t, x) = Re E e −i ωq (k)t+i k·x ,

 
E = hq E . (1.43)

If hq is parallel to some fixed coordinate axis a (hq = ēa , where ēa is the unit vector along the a
axis), such a field√ is called
√linearly polarized, because E remains parallel to the a axis at all times. If
hq = (ēa ±i ēb )/ 2 (here 2 is added only to ensure the normalization |hq | = 1), then Ea2 +Eb2 = const.
This means that E at given x rotates in the (a, b) plane with a constant amplitude, √ so the electric
field is called circularly polarized in this plane. Similarly, if hq = (ēa + i ς ēb )/ 2 with general real ς,
the trajectory of E in the (a, b) plane at fixed x is an ellipse, so the field is called elliptically polarized.

1.2.3 Quasimonochromatic waves

Let us consider localized wave packets, which can be viewed as superpositions of monochromatic waves
with various real k:
 ∞
ψ(t, x) =
X
ψ q (t, x), ψq =
d k ψ̄ (k) e −i ωq (k)t+i k·x . (1.44)
n q
q −∞ (2π)
.
Here, n = dim x is the dimension of the underlying coordinate space (dim x and dim ψ are generally
unrelated) and the factor (2π)n has been added for convenience; see below. Equation (1.44) shows
that individual branches of the dispersion relation contribute to the total field independently.12 Hence,
from now on, we will consider only one branch and omit the branch index q for brevity:

ψ(t, x) =
∞
d k ψ̄(k) e −i ω(k)t+i k·x . (1.45)
−∞ (2π)n
11 A polarization vector should not be confused with the electric polarization introduced in Sec. 1.1.2.
12 This is true only in a homogeneous medium, as will be discussed later.

9
LECTURE 1. ELECTROMAGNETIC DISPERSION 10

Suppose a quasimonochromatic wave, i.e., a wave such that ψ(k) is localized around some k = k0 ,
.
so only small κ = k − k0 contribute to the integral (1.45). For simplicity, let us assume that ωi is
small enough, so we can expand ω(k) as follows:
ω = ω0 + i γ + v g · κ, (1.46)
. .
where ω0 = ωr (k0 ), γ = ωi (k0 ), and v g is the group velocity given by
 
. ∂ωr (k)
vg = . (1.47)
∂k k=k0

This leads to

ψ(t, x) = e i k0 ·(x−vp t)+γt
∞
d κ ψ̄(k + κ) e i κ·(x−vg t) , (1.48)
0
−∞ (2π)n
where we introduced the “phase velocity”
. k0 ω0
vp = . (1.49)
k0 k0
(The corresponding phase speed is vp = ω/k.) This can also be written equivalently as

ψ(t, x) = Ψ(t, x) e i k0 ·(x−vp t) , (1.50)

 ∞
Ψ(t, x) = e γt
. d κ ψ̄(k + κ) e i κ·(x−vg t) , (1.51)
n 0
−∞ (2π)

and as can be checked by direct substitution, Ψ satisfies

(∂t + v g · ∇ − γ)Ψ = 0. (1.52)
Aside from the uniform amplification factor e γt , the wave envelope Ψ(t, x) depends on (t, x) only
through the combination x − v g t, so the envelope is stationary in the frame moving with velocity v g .
This mIn Lecture 5, we will explicitly show that for linear waves, v g is also the velocity of the wave
action (“photons”), energy, and momentum.] Likewise, the phase factor in Eq. (1.50) depends on
(t, x) only through the combination x − v p t, so v p serves as the velocity of phase fronts; hence the
name phase velocity. The geometrical meaning of v g and v p , which can be very different, is illustrated
2
in Fig. 1.1. Also note that we neglected ∂kk ωr . This term can be important at large t, when the
pulse starts to experience distortion due to the difference in the group velocities that correspond to
different k (Exercise 1.1). Confusingly enough, this distortion is called dispersion too (cf. Sec. 1.1.2),
2
and waves with zero ∂kk ωr are called dispersionless. Such waves include light in vacuum and sound
waves in neutral gases.

Exercise 1.1: Show that if the first-order expansion of ωr in κ [Eq. (1.46)] is replaced with
the second-order expansion of ωr in κ, then Eq. (1.52) acquires a form similar to the Schrödinger
equation of a free quantum particle,

(∂t + v g · ∇ − γ)Ψ = i/2 Θ : ∇∇Ψ, (1.53)

. 2
where Θ = ∂kk ωr is a symmetric matrix and : denotes double contraction; i.e., A : B =
tr (AB) ≡ Aab Bba . What new effect does the term Θ bring in? Estimate the time scale on
which this effect becomes noticeable for a pulse with a given initial width.

The above discussion shows that dispersion relations ω(k) contain information not only about
monochromatic oscillations but also about the evolution of wave packets. Because of this, dispersion
relations are particularly important and will be the primary focus of this course.

10
LECTURE 1. ELECTROMAGNETIC DISPERSION 11

Figure 1.1: A schematic of a localized quasimonochromatic scalar wave ψ(t, x) on a one-dimensional

space x (lighter regions correspond to larger ψ 2 ). The straight colored lines show the directions of the
corresponding group velocity (yellow) and phase velocity (cyan).

11
Lecture 2

A sneak preview: waves in cold

nonmagnetized plasma

In this lecture, we illustrate the key concepts introduced in Lecture 1 and also give a sneak preview
of the generic transformations of waves in inhomogeneous plasmas, which will be discussed more
systematically in Part II. To do this, we describe waves in cold nonmagnetized plasmas as a simple
example. We will also discuss these waves later in a broader context.

2.1 Basic equations

2.1.1 Plasma model
As discussed in Lecture 1, describing electromagnetic waves in a dispersive medium starts with devel-
oping a model for the electric current density j. In classical plasma, j can be written as
X X
j= es ns v s = es (n0s + ñs )(v 0s + ṽ s ). (2.1)
s s

Here, the summation is taken over all species s, es are charges of these species (e.g., for electrons,
ee = −e < 0), ns are the corresponding densities, and v s are the corresponding flow velocities. Tildes
denote that the corresponding quantities are small linear perturbations, while background currents
and fields are denoted with index 0. For simplicity, we assume that there are no background flows,
v 0s = 0, so ns0 may not depend on t but may depend on x. In this case,
X X X
j = j̃ = es n0s ṽ s + es ñṽ s ≈ es ns0 ṽ s , (2.2)
s s s

where ñs ṽ s is omitted because it is of the second order in Ẽ.

To calculate ṽ s , let us consider the momentum equation
 
∂ ṽ s es 1 ∇Ps
+ (ṽ s · ∇) ṽ s = Ẽ + ṽ s × (B 0 + B̃) − + C s. (2.3)
∂t ms c ms ns

Let us assume for now that there is no background magnetic field (B 0 = 0); then, the magnetic part
of the Lorentz force is quadratic in Ẽ and thus can be neglected. The term (ṽ s · ∇) ṽ s is negligible for
the same reason. Let us also assume that the pressure term ∇Ps can be neglected too. (The validity
conditions of this approximation will become clear when we study kinetic theory.) For the collision

12
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 13

term C s , we adopt a simple model C s = −νs ṽ s , where νs = νs (x) serves as the collision rate. Then,
Eq. (2.3) becomes

∂ ṽ s es
= −νs ṽ s + Ẽ. (2.4)
∂t ms
This leads to the following equation for the current density:
2
∂ j̃ s ωps
= −νs j̃ s + Ẽ, (2.5)
∂t 4π
where we introduced the so-called plasma frequencies
s
. 4πns0 e2s
ωps = . (2.6)
ms

2.1.2 Dispersion properties

Equation (2.5) is an inhomogeneous partial differential equation (PDE), so its general solution is
(f)
a general solution j̃ s (t, x) of the corresponding homogeneous equation plus a particular solution
(i)
j̃ s (t, x) of the inhomogeneous equation:

i ωps
2 (f)
∂ j̃ s
j̃ s = j̃ (f)
s + Ẽ, = −νs j̃ (f)
s . (2.7)
4π(ω̂ + i νs ) ∂t

Since the total current density is a sum of j̃ s , one can introduce the conductivity σ̂ s for each species
independently, with the total conductivity σ̂ being the sum over those of the individual species,
X
σ̂ = σ̂ s . (2.8)
s

Also note that in our problem, all these operators are scalar operators, σ̂ s = σ̂ s , and the same applies
to χ̂ and ˆ. In summary then, an inhomogeneous stationary cold nonmagnetized plasma without
average flows is characterized by the following operators:

i ωps
2 2
ωps X 2
ωps
σ̂ s = , χ̂s = − , ˆ = 1 − . (2.9)
4π(ω̂ + i νs ) ω̂(ω̂ + i νs ) s
ω̂(ω̂ + i νs )

In order to find the operators (ω̂ + i νs )−1 explicitly, let us directly integrate Eq. (2.5). This leads
to (Exercise 2.1)
2  t
ωps
s (t, x) = j̃ s (t0 , x) e d t0 e ν (x)(t −t) Ẽ(t0 , x).
0
−νs (x)(t−t0 )
j̃ (f) (f)
, j̃ (i)
s (t, x) =
s
(2.10)
4π t0

Exercise 2.1: Derive Eq. (2.10).

From here, we can readily infer the coordinate representation (1.24) of the conductivity operator:
2
ωps (x) −νs (x)(t−t0 )
Σs = Σs , Σs (t, x, t0 , x0 ) =
4π
e δ(x − x0 )H (t − t0 ), (2.11)

13
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 14

where we have added the Heaviside step function H (t − t0 ) to emphasize that Σ(t, x, t0 , x0 ) ≡ 0 for
t0 > t. The presence of the delta function in Eq. (2.11) signifies that the plasma response is local in
space, so there is no spatial dispersion.1
The effect of the conductivity operator on the electric field can now be written as
 t  ∞ 2
ωps (x) −νs (x)(t−t0 )
(σ̂ s E)(t, x) = dt 0
d x0 4π
e δ(x − x0 )E(t0 , x0 )
t0 −∞
2  t
ωps (x)
=
4π
d t0 e −ν (x)(t−t ) E(t0 , x).
s
0
(2.12)
t0

For fields monochromatic in time with frequency ω > −iνs , this yields

i ωps
2
(x)
(σ̂ s E)(t, x) = E(t, x). (2.13)
4π(ω + i νs (x))

Because x enters here only as a parameter, one can introduce the local conductivities, the local
susceptibilities, and the local dielectric tensor as functions of (ω; x):

i ωps
2 2
ωps X 2
ωps
σs (ω; x) = , χs (ω; x) = − , (ω; x) = 1 − , (2.14)
4π(ω + i νs ) ω(ω + i νs ) s
ω(ω + i νs )

2
where the dependence on x enters through ωps and νs .

2.2 Homogeneous plasma

2.2.1 General considerations

In plasma that is spatially homogeneous, the functions (2.14) are independent of x, so there exist waves
that are monochromatic in x, i.e., have a well-defined wavevector k. Let us search for such waves
assuming the coordinates x = {x, y, z} such that k is directed along the x axis. The corresponding
dispersion tensor [Eq. (1.36)] can be written as
 
(ω) 0 0
D E (ω, k) =  0 (ω) − N 2 0 , (2.15)
0 0 (ω) − N 2
.
and as a reminder, N = ck/ω. The corresponding dispersion relation det D E (ω, k) = 0 can be written
as

(ω) [(ω) − N 2 ]2 = 0, (2.16)

so it has solutions of two types:

(ω) = 0 and (ω) = N 2 . (2.17)

2
Since ωps ∝ m−1s and mi  me , the ion contribution to the dispersion of cold nonmagnetized
plasma is typically negligible. Assuming that this is the case, we obtain
2
ωpe
(ω) ≈ 1 − . (2.18)
ω(ω + i νe )
1 The absence of spatial dispersion is due to the fact that we have neglected the plasma temperature and average

flows. These effects will be discussed later.

14
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 15

Then, (ω) = 0 leads to a quadratic equation for ω and thus has two solutions, while (ω) = N 2 leads
to a cubic equation for ω and thus has three solutions, so there are five modes overall. Although exact
solutions for ω(k) are possible in both cases, it is more instructive to consider the regime when νe is
small. By assuming ω  νe and justifying this assumption a posteriori one obtains

ω ≈ ±ωpe −
i νe , (2.19a)
2
q
2 + c2 k 2 −
ω ≈ ± ωpe
i νe
2
ωpe
. (2.19b)
2 c2 k 2 + ωpe
2

If one assumes that ω is of order νe instead, one similarly obtains

c2 k 2
ω ≈ −i νe . (2.19c)
c2 k 2 + ωpe
2

Because Eq. (2.16) has exactly five roots, no solutions other solutions are possible.
Note that all five modes (2.19) satisfy ωi > −νe . Thus, the applicability condition of our theory,
νe + ωi > 0, is satisfied and the result (2.19) can be trusted. One might wonder if this happened
by accident, and the answer is no. As seen from Eq. (1.8), Σ̄ is just the current density produced
by a field that is delta-shaped in time (and in space). In contrast, the solutions (2.19) describe
collective oscillations, in which particles remain coupled with nonzero self-consistent field at all t. As
collisions dissipate the electron energy, this loss is partially compensated from the energy of the field,
so the current dissipates slower than electrons would slow down without the wave. This argument
also extends to general collisional plasmas, so in such plasmas one can always obtain σ(ω, k) from σ̂
simply by replacing ω̂ with ω (assuming that the plasma parameters are time-independent). However,
this is not the case in collisionless plasmas,2 where the total macroscopic field can dissipate while the
response of individual particles does not. For the time being, we will always assume some amount
of collisional dissipation to ensure the applicability of our general approach; i.e., collisionless plasmas
will be considered as collisional plasmas with infinitesimally small but positive νs :

νs → 0 + . (2.20)

In what follows, Eq. (2.20) will always be assumed for simplicity, in which case  is simplified as

ωp2
s
. X
2 .
(ω) = 1 − 2 , ωp = ωps (2.21)
ω s

“The” plasma frequency ωp absorbs information about both electrons and ions, so we do not have
to neglect the ion response to treat dispersion relations analytically. Then the resulting three wave
modes are as follows.

2.2.2 Static magnetic-field mode

Let us discuss Eq. (2.19c) first, because this is the simplest case. In the collisionless limit, it has zero
frequency, ω = 0, so often it is not even considered as a wave per se. This mode consists of static
currents, which create a nonuniform magnetic field but no electric field. Such static magnetic-field
“waves” are of interest in some cases [10] but will not be discussed further.
2 Since no plasma is purely collisionless, the distinction between “collisionless” and “collisional” can be tricky. In a

nutshell, (relevant) wave modes are defined differently when the collision rates are small enough, and then, in a way,
small collision rates cease to matter. We will return to this subject in the second half of the course.

15
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 16

2.2.3 Electrostatic Langmuir oscillations

Equation (2.19a), which corresponds to (ω) = 0, in the collisionless limit becomes
ω 2 = ωp2 . (2.22)
This shows that ωp is the frequency of natural oscillations of cold plasma. Consequently, it is also
common to refer to the mode (2.22) as “the” plasma oscillations, as opposed to other waves and
oscillations that can also exist in plasma.
The waves governed by Eq. (2.22) are nondissipative and can have any phase velocity v p =
(k/k)(ωp /k), because while ω is fixed by Eq. (2.22), k can be any vector. In contrast, the group
.
velocity v g = ∂k ω(k) is identically zero for these waves. Also, the wave polarization h is found from
the field equation D E (ω, k)h = 0, where E is the complex amplitude of the electric field (Sec. 1.2.2).
At ω 2 = ωp2 , this equation becomes
  
0 0 0 hx
 0 N 2 0   hy  = 0, (2.23)
0 0 N2 hz
so hy = hz = 0. In other words, the field is linearly polarized (Sec. 1.2.2) parallel to k, so the wave is
longitudinal and therefore electrostatic (Exercise 2.2).

Exercise 2.2: Derive a PDE for the electron-density perturbations in this wave using the
linearized continuity equation for the electron density, the linearized momentum equation for the
electron velocity, and the electrostatic dispersion relation derived in Lecture 1.

In electron–ion plasma, one has ωp ≈ ωpe (due to mi  me ), in which case the plasma oscillations
can be interpreted as oscillations of electrons relative to the stationary neutralizing ion background.
These oscillations were discovered by Langmuir and Tonks in the 1920s [11], so they are commonly
called electron Langmuir oscillations. The terms is also often shortened to just “Langmuir oscillations”,
but note that ion Langmuir oscillations are possible too. They occur at ωp ≈ ωpi when electrons are
hot enough to have negligible susceptibility. We will discuss this subject in Lecture 7.

2.2.4 Transverse electromagnetic waves

Equation (2.19b), which corresponds to (ω) = N 2 , in the collisionless limit becomes
ω 2 = ωp2 + k 2 c2 . (2.24)
Unlike Langmuir oscillations, these are electromagnetic waves with transverse polarization. This
means that their electric field is perpendicular to the wavevector, so there are two independent trans-
verse modes overall. They can be chosen as the modes with the polarization vector h along, say, the
y axis and the z axis, respectively. Alternatively, one can define them as the two circularly polarized
(Sec. 1.2.2) modes with opposite, left and right, polarization. (This flexibility is removed when plasma
is magnetized; see Lecture 6.)
The transverse electromagnetic waves have a superluminal phase velocity,
ω kc
vp = > = c, (2.25)
k k
and their group velocity is given by
c2
vg = < c. (2.26)
vp
At ω  ωp , these waves become the usual vacuum light waves.

16
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 17

2.3 Wave transformations in inhomogeneous plasma

Now let us discuss how waves propagate in inhomogeneous cold nonmagnetized stationary plasma.
Because there is no spatial dispersion in cold stationary plasma, the dispersion operator in this case is
the same as the matrix (2.15) except k must be replaced with the operator k̂ = −i ∇. We will focus on
transverse electromagnetic waves described by Eq. (2.24) and consider how they transform in simple
geometries. The transformations that we will discuss are of interest in that they are characteristic of
generic waves in inhomogeneous media. Hence, the material below is intended as a motivation for a
more general theory, a subject that will be discussed in Part II.

2.3.1 WKB approximation

Let us start with considering a stationary (∂t = −i ω) transverse wave propagating along the x axis in
collisionless stationary plasma. Finding the wave field in this case is a boundary-value problem. We
will assume that the field at the boundary has no dependence on y or z and that ωp = ωp (x), so we
.
can adopt ∂y = ∂z = 0 everywhere. The direction of the field polarization h = Ẽ/Ẽ is perpendicular
to the x axis but otherwise unimportant. In any case, ∇ · Ẽ = 0, so Eq. (1.23) becomes

d 2E + q(x)E = 0, (2.27)
d x2
where we have adopted the notation Ẽ = Re [hE (x)e −i ωt ] and introduced

. ω2 ω 2 − ωp2 (x)
q(x) = 2 (ω; x) = . (2.28)
c c2
Now that the coefficients of the wave equation explicitly depend on x, a spatially monochromatic
wave is no longer a solution. But let us assume (until Sec. 2.3.2) that the plasma parameters change
along x slowly, the condition for which is yet to be specified. Then locally, q can be considered
√
constant, so q serves as the local wavenumber k. [This can be seen by comparing Eq. (2.28) with
the homogeneous-plasma dispersion relation (2.24) or simply by solving Eq. (2.27) locally under the
approximation of constant q.] This interpretation is justified provided that k is reasonably well defined,
.
i.e., if the corresponding wavelength λ = 2π/k has a small gradient (which is a dimensionless quantity):

d λ  1, 2π
λ= √ . (2.29)
dx q

Then, it is possible to construct an asymptotic solution of Eq. (2.27) using the so-called Wentzel–
Kramers–Brillouin (WKB) approximation. We will not review the general WKB method here, but
will rather present an ad hoc solution of Eq. (2.27) sufficient for our purposes. (In Part II, we will
adopt an alternative approach, which is easier to apply in the general case.)
√
Let ε  1 be the corresponding characteristic value of d λ/d x ∼ (Lc q)−1 , where Lc is some
constant characteristic scale of the plasma density. Then, Eq. (2.27) can be written as

ε2 E 00 + Q(ξ)E = 0, (2.30)

where the prime denotes d /d ξ, ξ = x/Lc , and Q(ξ) = (εLc )2 q(x) is an order-one function with
. .
Q0 ∼ Q. Let us search for a solution in the form E = e i S(ξ)/ε , where S is a complex function of the
form S = S0 + εS1 + ε2 S2 . . . with order-one Sn . Then, Eq. (2.30) yields3

Q − S002 + ε(i S000 − 2S00 S10 ) + O(ε2 ) = 0. (2.31)

3f = O(εα ) means “f is of order εα ”, or more precisely, there is constant M such that |f | ≤ M εα for small enough ε.

17
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 18

Since the equality is supposed to hold for all small ε, this leads to

S002 = Q, S10 = i/2 S000 /S00 , (2.32)

and the terms Sn>1 will be ignored because including them changes E only by O(ε). Then,

S0 = ± dξ
p
Q(ξ), S1 =
i ln Q(ξ) + const, (2.33)
4
so the field can be expressed as follows:
  
exp ± i dx
A p
E (x) = p
4
q(x) , (2.34)
q(x)

where A is an arbitrary complex constant and ± correspond to the waves propagating in the ±x
directions, respectively. Since Eq. (2.27) is linear, any linear combination of the solutions with plus
and minus sign is also a solution.
The WKB solution (2.34) corroborates our expectation that the local wavevector k(x), defined
as the gradient of the phase, ispthe same4 as that predicted by the homogeneous-plasma dispersion
relation (2.24); i.e., k(x) = ± q(x). Equation (2.34) also shows the field amplitude changes as
|E | ∝ [q(x)]−1/4 , so
p
|E (x)|2 q(x) = const. (2.35)

This coincides with the well-known adiabatic invariant of a harmonic oscillator [our original Eq. (2.27)
is the equation of a harmonic oscillator], which is the ratio of the oscillator’s energy and frequency [12].
A more general form of this conservation law will be discussed in Lecture 5.

2.3.2 Field structure near a cutoff

Equations (2.34) and (2.35) predict |E | → ∞ at q → 0, or at the “critical density”, which corresponds
to ω 2 = ωp2 . This is an artifact of the WKB approximation, which is inapplicable at small q, as
seen from Eq. (2.29). The critical point q = 0 corresponds to a cutoff (a type of caustic), where
the homogeneous-plasma dispersion relation (2.24) predicts k 2 to change sign (Box 2.1). Beyond the
cutoff (i.e., at ωp2 > ω 2 ), the wavenumber is imaginary. Such waves are called evanescent and do not
transport energy (Exercise 2.3). This means that a wave has to experience reflection near a cutoff.

Exercise 2.3: Show that the time-average Poynting vector of an evanescent wave with a single
imaginary wavevector is zero. Explain how waves can carry electromagnetic energy across a
finite-width region where they are evanescent.

Let us assume a linear approximation for ωp2 (x) and choose coordinates such that the cutoff cor-
responds to x = 0, with waves propagating at x < 0 and evanescent at x > 0. Then locally,
 
x
ωp2 (x) = ω 2 1 + , Lc > 0. (2.36)
Lc

Equation (2.27) becomes

d 2E
1/3
Lc c2

. x .
− η E = 0, η= , `= . (2.37)
d η2 ` ω2
4 Although this is always true approximately, we will show in Lecture 4 that a small deviation from the homogeneous-

plasma dispersion relation is possible in some cases.

18
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 19

E
0.8

0.6

0.4

0.2

η
-20 -15 -10 -5
-0.2

-0.4

Figure 2.1: The Airy function Ai (η) (blue) vs. the WKB approximation (2.40) (red), with x = η`,
q = −η`2 , and C = 1. The field is a standing wave at η < 0 and an evanescent wave at η > 0.

This is the so-called Airy equation, and its general solution is a linear superposition of the Airy
functions of the first and second kind, Ai (η) and Bi (η), respectively. Assuming that a wave is
incident from the left, we are interested in a solution that corresponds to E (x → ∞) = 0. This leaves
us with E = CAi (η), where C is an arbitrary constant that determines the maximum near-cutoff
magnitude, E max ∼ C (Fig. 2.1). We can connect C with the constants A+ and A− of the general
WKB solution,

A e i θ(x) + A− e −i θ(x) x
E (x) = +
π
d x0
p
, θ(x) = + q(x0 ), (2.38)
[q(x)]1/4 4 0

by comparing its asymptotic behavior with the known asymptotic behavior of Ai (η) at η → −∞:
cos[θA (η)] π 2
Ai (η) ≈ √ , θA (η) = − (−η)3/2 . (2.39)
π (−η)1/4 4 3
Near the cutoff, q ≈ −η/`2 , which leads to θ(x) ≈ θA (η). Then, for the
√ solution (2.38) to match
E = CAi (η) at not-too-large negative η, one should adopt A+ = A− = 2 π`C, which leads to
C cos[θ(x)]
E (x) = √ . (2.40)
π[q(x)`2 ]1/4
In other words, the WKB approximation
p predicts that at large negative η, the field is a standing wave
with the local wavenumber k(x) = q(x) [same as predicted by the homogeneous-plasma dispersion
relation (2.24)] and the amplitude |E | that satisfies
1/6
|E |

1 1 c
∼ = √ . (2.41)
E max (q`2 )1/4 4
 ωLc
Since the right-hand side depends on the plasma parameters slowly, the field amplification near the
cutoff is not as dramatic as one might imagine based on the WKB approximation. Nevertheless, the
WKB approximation describes the field well all the way up to η ∼ −1 (Fig. 2.1).

2.3.3 Oblique incidence on a cutoff region

Let us also consider oblique incidence of a transverse wave on a cutoff. As in the previous section, we
assume a plane incident wave and ωp = ωp (x). However, now we allow for a nonzero constant ky , so

19
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 20

Box 2.1: Reinstating the WKB approximation near cutoffs

In principle, the WKB approximation can be reinstated near a cutoff by changing the represen-
tation of the field equation. For example, consider the Airy equation (2.37) in the operator form,
k̂ 2 E + x̂E = 0. In the usual, coordinate representation, one has x̂ = x and k̂ = −i ∂x . (Here, we
blur the distinction between the dimensionless η and −i ∂η on one side and the dimensional x and
−i ∂x on the other side for simplicity.) However, we can adopt the wavevector representation (cf.
momentum representation in quantum mechanics), i.e., take the Fourier transform of the Airy
equation. Then, x̂ = i ∂k and k̂ = k, so the equation becomes

k 2 Ē + i ∂k Ē = 0,

where Ē is the Fourier spectrum of E . This new equation can be solved using the WKB
method. In fact, the corresponding WKB approximation, Ē = const × exp(i k 3 /3), is an exact
solution.

Mapping it back to the coordinate space is only a matter of calculating the integral
d k exp(i kx + i k3 /3), which can be done at various levels of accuracy.
One can also use a more general class of transforms called metaplectic transforms. The latter
can rotate (more generally, perform symplectic transformations of) the “phase space” (x, k)
such that in the new phase space (x0 , k 0 ), the wavenumber k 0 is never zero and thus cutoffs are
eliminated [13] (Fig. 2.2). The Fourier transform can be considered as a particular case of the
metaplectic transform that corresponds to 90◦ rotation.

the electric and magnetic fields can be written as follows:

Ẽ = Re [E (x)e −i ωt+i ky y ], B̃ = Re [B (x)e −i ωt+i ky y ]. (2.42)
Two distinct wave patterns can be formed in this case, depending on the wave polarization. Below,
we briefly discuss both cases.

TE wave
Let us start with the transverse-electric (TE) polarization (Fig. 2.3), which corresponds to E x = E y =
0. Since ∂z E z = 0, one has ∇ · Ẽ = 0, which leads to the following equation for E z :

0 = ∇2 E z +
ω2
(ω; x)E z =
d 2 E z + 1 [ω2 − c2 k2 − ω2 (x)]E . (2.43)
c2 d x2 c2 y p z

This equation is similar to Eq. (2.27) up to replacing ω 2 → ω 2 − c2 ky2 . Thus, the field forms the same
WKB/Airy profile (Fig. 2.1), except the cutoff is shifted to the location where ωp2 (x) = ω 2 − ky2 c2 , in
agreement with the homogeneous-plasma dispersion relation (2.24).

TM wave
Let us also consider the transverse-magnetic (TM) polarization (Fig. 2.3). In this case, ∇ · Ẽ is
nonzero, which means that a transverse wave is coupled to the oscillations of the charge density, ρ̃,
through Gauss’s law (1.1c). Similarly, ρ̃ is coupled to Ẽ via (Exercise 2.4)
e2
∂t2 ρ̃ + ωp2 ρ̃ = f˜, f˜ = − ∇n0e · Ẽ. (2.44)
me
Equation (2.44) can be understood as the electrostatic wave equation for free Langmuir oscillations
driven by an external force f˜ ∝ Ẽ. If this force has a resonant frequency ω = ±ωp , the magnitude of
ρ̃ can become large. Then, several scenarios are possible.

20
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 21

Figure 2.2: An illustration of cutoff elimination using the metaplectic transform (Box 2.1): the phase
space is continuously transformed such that the wavenumber remains constant and thus a wave never
sees a cutoff in this transforming frame. In the simplest case, the transformation is a rotation by some
nonzero angle α, with α = 90◦ corresponding to the Fourier transform.

Figure 2.3: A schematic of transverse-wave oblique incidence on plasma with background electron
density n0e = n0e (x) illustrating the difference between a TE wave and a TM wave. The vertical red
line indicates the critical-density region, where ωp2 (x) = ω 2 . The dashed red line indicates the cutoff
region, where ωp2 (x) = ω 2 − ky2 c2 . Waves are propagating on the left and evanescent on the right.
The background color intensity illustrates the background density, with darker colors corresponding
to higher densities.

21
LECTURE 2. A SNEAK PREVIEW: WAVES IN COLD NONMAGNETIZED PLASMA 22

Ex

x
0

Figure 2.4: A typical E x [Eq. (2.46), numerical solution] for intermediate β and weak dissipation,
with the boundary conditions corresponding to vanishing B z at x → ∞. Here, (ω; x) is given by
Eq. (2.18), with ωp2 (x) given by Eq. (2.36).

When the angle β between k and ∇n0e approaches zero, one recovers the usual WKB/Airy profile
discussed above. Likewise, if β is large, one can expect from Eq. (2.24) that the wave is reflected
much earlier than it reaches the critical-density region. Then, f˜ is extremely small in the resonance
region, so Langmuir oscillations are not excited (unless the collision rate is also extremely small) and
a TM wave behaves just like a TE wave. However, at intermediate β, the coupling of a TM wave with
Langmuir oscillations can be substantial and results in a significant peak of the wave field near the
critical region. This can be described by the following equations (Exercise 2.4)

d 2 B z − ∂ ln (ω; x) d B z + ω2 (ω; x) − k2 B = 0,
 
(2.45)
d x2 ∂x dx c2 y z

ck B
Ex = − y z , Ey =
ic dBz . (2.46)
ω(ω; x) ω(ω; x) d x

A typical E x [Eq. (2.46)] for intermediate β is shown in Fig. 2.4, where the Airy pattern is seen on the
left and a large near-singular field is seen near the Langmuir resonance. (For details and asymptotic
analytic solutions, see Ref. [14].) Note that a small collision rate has to be added to keep the wave
field finite. This is due to the fact that energy is continuously deposited by the incoming wave into
Langmuir oscillations and cannot escape the critical-density region, because Langmuir waves are tied
to this region and because they cannot dissipate collisionlessly in cold plasma. One can show that
adding thermal effects removes both these limitations.
The coupling of TM waves with Langmuir oscillations is an example of mode conversion, which is
mathematically similar to quantum tunneling. Note that this effect is completely missed within the
homogeneous-plasma analysis (Sec. 2.2), where the two transverse modes have identical properties
and are completely uncoupled from Langmuir oscillations. To capture polarization effects robustly
and in a general geometry, a more systematic theory is needed. Later, we will present such theory,
which will be readily applicable to almost any dispersion operators.

Exercise 2.4: Derive Eqs. (2.44)–(2.46).

22
Appendices for Part I

AI.1 Analytic properties of response functions

Here, we briefly summarize some important properties of response functions that commonly emerge
in theory of dispersion. We will be concerned only with temporal dynamics, so possible dependence
of these functions on spatial coordinates and wavevectors will be ignored.
Let us consider a general system whose response j (which may or may not be the current density)
to an external force E (which may or may not be the electric field) is linear, i.e., can be expressed in
terms of some Green’s function Σ̄:
 t
j(t) = d t0 Σ̄(t − t0 )E(t0 ). (2.47)
0

Let us assume that our system is stable in the sense that a response to a delta-shaped field eventually
fades away, i.e., Σ̄(t → +∞) = 0.5 Then, the integral that determines its Laplace image,
 ∞
σ(ω) = d t e i ωt Σ̄(t), (2.48)
0

converges for all ω that satisfy Im ω ≥ 0. Similarly, if Σ̄(t) fades away faster than any power of t at
t → ∞ (which is typically the case, as discussed in Lecture 8), then all integrals of the form
 ∞
.
Jn [Σ̄](ω) = d t e i ωt (i t)n Σ̄(t) (2.49)
0

converge too. [The square brackets denote that Jn is a functional of Σ̄, and (ω) denotes that Jn
depends on ω, as usual.] Since Jn [Σ̄](ω) = σ (n) (ω), where (n) is the nth derivative, this means that
all derivatives of σ(ω) are well defined at Im ω ≥ 0. This proves the following theorem:

Theorem: σ(ω) is analytic in the upper half of the complex-ω plane.

∞
Next, note that the function σ(ω) ≡ J0 [Σ̄](ω) = 0
d t e i ωt Σ̄(t) can be expressed as follows:

d t ddt ei ω Σ̄(t) − ei ω Σ̄0 (t)
∞  i ωt  i ωt  
J0 [Σ̄](ω) =
0
 ∞
e
1  i ωt 1
d t e i ωt Σ̄0 (t)
 ∞
= Σ̄(t) 0 −
iω iω 0
0
Σ̄(0) J0 [Σ̄ ](ω)
=−
iω − iω . (2.50)
5 This does not rule out collective instabilities, where the field itself grows in time. In the sense assumed here, the

response of a current to an external field, which is described by the conductivity operator σ̂, is typically stable. (Systems
exhibiting avalanche ionization are exceptions.) In contrast, the response of a self-consistent field to an external current,
which is described by D̂ −1
E , is often not.

23
APPENDICES FOR PART I 24

[Here, we have assumed that the function Σ̄ is also sufficiently well behaved so that Σ̄(0) and J[Σ̄0 ] are
finite.] The function Σ̄(t) is determined by microscopic processes that have some nonzero minimum
time scale T . Then, roughly, |J[Σ̄0 ]| . |J[Σ̄]|/T . Hence, at ω  T −1 , the second term on the right-
hand side of Eq. (2.50) can be neglected compared to the term on the left-hand side. Thus, at large
enough ω, one has σ(ω) ≈ i Σ̄(0)/ω, and in particular,

σ(ω → ∞) = 0. (2.51)

In combination with Eq. (2.51), the analyticity of σ(ω) in the upper half of the complex-ω plane
(proven above) leads to some interesting properties of the functions
. .
σr (ω) = Re σ(ω), σi (ω) = Im σ(ω). (2.52)

These properties are derived as follows. Let us consider real ω0 and the integral

.
I(ω0 ) = d ω ωσ(ω)
−ω
, (2.53)
C 0

where C is a contour that goes along the real axis and encircles the pole at ω = ω0 from above. On
one hand, one can rewrite I(ω0 ) as
∞
I(ω0 ) = d ω ωσ(ω)
−ω
− i πσ(ω0 ), (2.54)
−∞ 0

where denotes the Cauchy principal value of the corresponding integral. On the other hand, one
can close the contour C through a semicircle C+ with radius R → ∞ at Im ω > 0, because the integral
over C+ is zero. [This is because at ω → ∞, the integrand decreases faster than 1/ω due to σ(ω) → 0.]
But σ(ω)/(ω − ω0 ) is analytic everywhere within the closed contour, so the integral over this closed
contour must be zero. Therefore, I(ω0 ) = 0, which means

σ(ω0 ) = −
i ∞
d ω ωσ(ω) . (2.55)
π −∞ −ω 0

By taking the real and imaginary parts of Eq. (2.55), one obtains that σr and σi are connected via
the Hilbert transform; namely,
∞
σr (ω0 ) =
1
π
d ω ωσ−
i (ω)
ω
, (2.56a)
−∞ 0
∞
σi (ω0 ) = −
1
π
d ω ωσr−(ω)
ω
. (2.56b)
−∞ 0

These are known as the Kramers–Kronig relations. They show that having nonzero σi at some frequen-
cies implies having nonzero σr at some (possibly, different) frequencies, and vice versa. Notably, the
collisionless dissipation discussed in Part IV can be anticipated from the Kramers–Kronig relations.

24
Problems for Part I

PI.1 Electrostatic approximation

In some cases, a wave can be considered electrostatic, meaning that its electric field is approximately
representable as (minus) the gradient of an electrostatic potential, E ≈ −∇ϕ. Such waves can be
easier to describe, because instead of the multiple components of E, one can work with a single
function ϕ. Here, you are asked to explore the electrostatic-wave approximation for a homogeneous
stationary medium.

(a) Consider the components of E parallel and perpendicular to the wavevector, E k and E ⊥ . Using
Eq. (1.36), estimate the ratio of |E k | and |E ⊥ | at sufficiently large N and argue that

N 2  ab for all a and b (2.57)

is a sufficient condition for a wave to be electrostatic.

(b) Assuming that the field is electrostatic and that the medium is described by an unspecified
dielectric tensor , derive the corresponding dispersion relation from Gauss’s law. Show that the
same result is obtained from Eq. (1.36) if one takes for granted that the field is electrostatic.
(c) In a gyrotropic medium with no spatial dispersion, the dielectric tensor has the form

⊥ (ω) −i g(ω)
 
0
(ω, k) =  i g(ω) ⊥ (ω) 0 . (2.58)
0 0 k (ω)

Consider an electrostatic wave propagating with a such medium with k = {k⊥ , 0, kk }. (One can
always choose axes such that ky = 0.) Substitute this k and Eq. (2.58) for  into the dispersion
relation derived in problem (b). Using your result, show that the group velocity of such a wave
is orthogonal to the wave phase velocity.

PI.2 Photon wave function in cold magnetized plasma

Consider a cold plasma formed by N species with charges es , masses ms , and unperturbed densities
n0s = n0s (x). Assume that this plasma is immersed in a stationary magnetic field B 0 (x) and has no
average flows.

(a) Show that the combination of the linearized momentum equation for the fluid velocities ṽ s ,
Ampere’s law for the wave electric field Ẽ, and Faraday’s law for the wave magnetic field Ẽ can
be represented together in the form
i ∂t ψ = Ĥψ, (2.59)

25
PROBLEMS FOR PART I 26

or equivalently, D̂ ψ ψ = 0 with D̂ ψ = ω̂ − Ĥ, where ψ is a 3(N + 2)-dimensional real vector

field (the same equation can be expressed without i ) and the Hamiltonian Ĥ being the following
Hermitian operator:

i ωp1 (x)
 
−α · Ω1 (x) 0 ... 0 0

 0 −α · Ω2 (x) . . . 0 i ωp2 (x) 0 

 .. .. .. .. .. .. 
Ĥ = 
 . . . . . . .

(2.60)
 0 0 . . . −α · Ω N (x) i ω pN (x) 0 
 −i ωp1 (x) −i ωp2 (x) . . . −i ωpN (x) i cα · k̂ 
 
0
0 0 ... 0 −i cα · k̂ 0
. . p
Here, Ωs = es B 0 (x)/(ms c), ωps = es 4πn0s (x)/ms (note that this definition is slightly differ-
.
ent from the one that we used earlier), and α = (αx , αy , αz )| is the column vector comprised of
the following Hermitian matrices:6

0 0 i 0 −i 0
     
0 0 0
0 0 −i  , αy =  0 0 0  , αz =  i
.  . .
αx = 0 0 . (2.61)
0 i 0 −i 0 0 0 0 0

Note that Eq. (2.59) is similar to the Schrödinger equation for vector particles. (This becomes
even more evident if one multiplies it by ~ and expresses the right-hand side through the mo-
.
mentum operator p̂ = ~k̂ instead of the wavevector operator k̂.) Thus, ψ can be understood as
the photon wave function in cold plasma (cf. the photon wave function in vacuum [15]).

Hint: Use ψ = (ζ̃ 1 , ζ̃ 2 , . . . , ζ̃ N , Ẽ, B̃)| , where ζ̃ s is a rescaled v s , with the rescaling
factor that you are asked to find. Also use that for any three-component column
vectors A and B, one has A×B = −i (α·A)B, as can be verified by direct calculation.
2 2
(b) Calculate |ψ|

, which is 2the same as ψ here. What is the physical meaning of this quantity?
Show that d x |ψ(t, x)| ≡ hψ|ψi is conserved.

Hint: In the last question, use Eq. (2.59) and the fact that Ĥ is Hermitian. Do not
use the explicit formula (2.60), or your calculations will be much longer than necessary.

(c) Argue that cold-plasma modes governed by Eq. (2.59) cannot be unstable, irrespective of how
inhomogeneous the plasma and B 0 are.7

Hint: This can be argued in one sentence.

PI.3 Beam–plasma instability (cold electrostatic limit)

Consider one-dimensional nonmagnetized collisionless cold homogeneous stationary electron plasma
with motionless ions. Suppose that bulk electrons have the average density n0 and zero average
velocity. Consider also a cold electron beam with some average density nb and some average velocity vb .

(a) Consider bulk electrons and beam electrons as different species. Show that the beam-electron
.
susceptibility in the spectral representation is given by χb (ω, k) = −ωb2 /(ω − kvb )2 , where ωb =
p
4πnb e2 /me . (You may adopt ω̂ = ω and k̂ = k.)
6 Notably, αa belong to the family of Gell–Mann matrices, which serve as infinitesimal generators of SU(3).
7 This is true only for cold plasma without flows. Otherwise, Ĥ is different and generally has different properties.

26
PROBLEMS FOR PART I 27

(b) Calculate the plasma dielectric function (ω, k) and write down the general dispersion relation
of electrostatic oscillations.
(c) How many branches does this dispersion relation have?
(d) Show graphically, by qualitatively analyzing the function (ω, k) derived in (b) and using your
answer from (c), that such plasma is unstable at small enough k.
(e) For the unstable regime, plot Re ω and Im ω as functions of k. (You may do it qualitatively or
by solving the dispersion relation numerically, but make sure to plot all branches.)
.
(f) Assuming η = nb /n0 is a small parameter, show that the maximum of the growth rate γ scales
as γmax ∝ η 1/3 . (If a beam is warm, which case will be studied later in this course, the scaling
for γ can be different.)

Hint: Use the fact that, to the zeroth order in η, the unstable branch has ω = kvb
p from (e)]. Show that to the leading order, γ → ∞ at
[as you should be able to see
2 .
k = ±ωp0 /vb , where ωp0 = 4πn0 e2 /me . Hence, adopt k = ±ωp0 /vb as the optimum
wavenumber for the instability and then find γ more accurately.

PI.4 Surface waves

Here, you are asked to study azimuthally symmetric electromagnetic waves propagating along a cold-
plasma column of finite radius a. Assume the cylindrical coordinates (r, θ, z), where z is the axis of
symmetry, and Ẽ = Re (E e −i ωt+i kz z ), with constant ω and kz . For simplicity, consider only TM
waves, i.e., waves with B̃z = 0.

Hint: Remember that the tangential component of the electric field is always continuous
at the plasma boundary. Make sure you understand why.

(a) Using Maxwell’s equations for the three components of the electric field Ẽ and the three com-
ponents of the magnetic field B̃, show that the above assumptions lead to Ẽθ = B̃r = B̃z = 0
at r 6= a. (Assume that  6= 0.) Argue that in this problem, B̃θ is continuous at the plasma
boundary.
(b) Show that ∇ · Ẽ = 0 at r 6= a. Using this, show that E z satisfies the modified Bessel equation
d 2 E z + 1 d E z − E = 0. (2.62)
d ζ2 ζ d ζ z

. .
Here ζ = κr, κ2 = kz2 +ωp2 /c2 −ω 2 /c2 , ωp (r < a) = ωp0 is a nonzero constant, and ωp (r > a) = 0.
Using the continuity of Ẽz at r = a, find E z (r) as a piecewise-analytic function. (Remember
that your solution is supposed to be finite everywhere, including r → 0 and r → ∞.)
(c) Using the continuity of B̃θ at r = a (which can be expressed through d E z /d ζ), show that the
dispersion relation is
ωp2
 
1 I1 (κin a) 1 K1 (κout a)
1− 2 + = 0. (2.63)
ω κin a I0 (κin a) κout a K0 (κout a)
Here, In and Kn are modified Bessel functions of the first and second kind, respectively; also,
. .
κin = κ(r < a) and κout = κ(r > a).
.
(d) Solve Eq. (2.63) numerically. Plot ω(kz ) and E z (r) for several values of a/δc , where δp = c/ωp0 .
Explain the results qualitatively.

27
 Part II

Basic theory of
quasimonochromatic waves

In this part of the course, we outline a general asymptotic theory of quasimonochro-

matic waves in linear dispersive media. In the context of fusion research, our discussion
is particularly relevant to waves in the electron-cyclotron and lower-hybrid ranges of fre-
quencies, which correspond to mm and cm wavelengths, respectively (Part III). For waves
with longer wavelengths, such as those in the ion-cyclotron range, our discussion is relevant
qualitatively.

28
Lecture 3

Asymptotic expansion of dispersion

operators

In this lecture, we derive an approximate envelope equation for a general quasimonochromatic wave
by asymptotically expanding the wave dispersion operator. In doing so, we also introduce the Wigner–
Weyl transform, which is a central element of modern wave theory and will be used also in other parts
of the course.

3.1 Problem setup

Let us consider a general wave ψ governed by
D̂ψ = 0, (3.1)
where D̂ is a linear dispersion operator. We will assume that the wave remains quasimonochromatic,
so ψ can be split into a rapid real phase θ (“eikonal”, to be specified in Lecture 4) and a slow complex
envelope Ψ:
ψ = e i θ Ψ. (3.2)
Then, the quantities
. .
w̄ = −∂t θ, k̄ = ∇θ (3.3)
represent the local frequency and the local wavevector (note that they are real by definition), and
. 2π . 2π
τ= , λ= (3.4)
w̄ k̄
are the local temporal period and the local spatial period (wavelength). Suppose the characteristic
temporal and spatial scales of Ψ, w̄, k̄, and of the medium are limited from below by some values Tc
and Lc , respectively. Then, we introduce a “geometrical-optics parameter”
 
. τ λ
ε = max ,  1. (3.5)
Tc Lc
Our goal is to develop an approximate geometrical-optics (GO)1 theory that captures the leading-
order effects caused by nonzero ε. This will be done by approximating the dispersion operator D̂ to
the first order in ε, which involves three main steps:
1 The name “GO” is determined by the fact that the regime ε  1 was initially studied for optical waves. Higher-order

theories include the quasioptical approximation and the paraxial approximation as a special case (Box 3.1).

29
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 30

(i) map the operator D̂ to some function D,

(ii) approximate the function D using ε as a small parameter,
(iii) and map the approximated function back to the operator space.
To ensure that D be efficiently expandable in ε, the mapping will be done using the Wigner–Weyl
transform (WWT).2 But let us introduce some basic notation first.

3.2 Notation
To shorten the calculations, we will describe the wave propagation in terms of spacetime coordi-
.
nates xα , or in the invariant notation, x = {x0 , x}, where x0 = ct and xa with a > 0 are spatial
.
coordinates. The dimension of the coordinate space n = dim x can be any n ≥ 0. The spacetime
.
dimension is n = dim x = n + 1 and restricted to n ≥ 1. For simplicity, we restrict our consideration
to spacetimes Mn with the Minkowski metric gαβ = g αβ = diag {−1, 1, 1, . . .}.3 Then, w̄ and k̄ can be
expressed through the components of the row vector
.
k̄α = ∂α θ = (−w̄/c, k̄)α , (3.6)
.
where ∂α = ∂/∂xα ; in particular, k̄0 = −w̄/c. The corresponding column vector has components
k = g kβ , where g αβ is the inverse metric. In case of the Minkowski metric, g αβ = gαβ , so
α αβ

k 0 = −k̄0 = w̄/c and k̄ a = k̄a , or in the invariant form, k̄ = (w̄/c, k̄)| . This implies

k̄ · x = gαβ k̄ α xβ = k̄α xα = −k̄ 0 x0 + k̄ a xa = −w̄t + k̄ · x. (3.7)

The operators ω̂ = i ∂t and k̂ = −i ∇ that we introduced earlier can be similarly expressed through
. .

k̂ α = −i ∂α .
.
(3.8)

By analogy with quantum mechanics, let us consider scalar fields on Mn as vectors |ψi in the
corresponding Hilbert space. Then, any scalar function ψ(x) can be understood as the x-representation
of the corresponding vector |ψi; namely, ψ(x0 ) = hx0 |ψi. Here, |x0 i is the normalized eigenvector of
the coordinate operator x̂ that corresponds to the eigenvalue x0 ; i.e.,

x̂ |x0 i = x0 |x0 i , hx0 |x00 i = δ(x0 − x00 ), (3.9)

so dx0 |x0 i hx0 | is a unit operator. (All integrals are taken on (−∞, +∞) unless specified otherwise.)
Accordingly, any (M̂ ψ)(x) can be expressed as
 
(M̂ ψ)(x) = hx|M̂ ψi ≡ hx|M̂ |ψi = 0 0 0
dx hx|M̂ |x i hx |ψi = d x0 M(x, x0 )ψ(x0 ), (3.10)

.
where M(x, x0 ) = hx|M̂ |x0 i is the x-representation of M̂ . Similarly, one can introduce the normalized
eigenvectors of the wavevector operator k̂,

k̂ |k0 i = k0 |k0 i , hk0 |k00 i = δ(k0 − k00 ), (3.11)

and define the k-representation of vectors as hk|ψi and of operators as hk|M̂ |k0 i (Exercise 3.1).
2 Reported here is a modern approach to GO [16, 17]. Earlier formulations of GO [18] do not use the term WWT

explicitly but still follow similar steps.

3 Readers who are not familiar with tensor analysis may consider upper-index and lower-index quantities as indepen-

dent functions formally defined through the formulas below.

30
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 31

Exercise 3.1: Show that hx|ki = hk|xi = (2π)−n/2 e i k·x .

∗

For N-component vector fields



ψ1
 ψ2 
ψ=
 ... ,
 (3.12)
ψN
where ψu are scalar fields on Mn and N may or may not coincide with n, we similarly introduce
 
|ψ1 i
 |ψ2 i 
|ψi = 
 ...

 (3.13)
|ψN i
via ψu (x0 ) = hx0 |ψu i. Then, operators on vector fields are considered as matrices of operators that act
on the field components; i.e.,

(M̂ uv ψv )(x) = d x0 Muv (x, x0 )ψv (x0 ), Muv (x, x0 ) = hx|M̂ uv |x0 i , (3.14)

the k-representations of matrix operators are introduced similarly.

3.3 Wigner–Weyl transform

The WWT4 can be understood as a “mixed”, or “phase-space”, representation of an operator. Unlike
the x- and k-representations, which involve projections either on |xi or on |ki, the WWT projects an
operator on both |xi and |ki simultaneously such that the resulting projection has “natural” properties.
This is more easily seen from the definition of the WWT that is formulated in terms of the “phase-
.
space” coordinates z = {x, k} [2, 19, 20]. Here, though, we will use a less elegant definition that is
perhaps easier to absorb. This definition is as follows.
For any given scalar operator M̂ , its Weyl image M (x, k) ≡ W [M̂ ], also called the (Weyl) symbol
of M̂ , is a function defined on the 2n-dimensional real space (x, k) as follows:

.
M (x, k) = d s hx + s/2|M̂ |x − s/2i e −i k·s . (3.15)
.
One can also understand M (x, k) as the Fourier image of M̄(x̄, s) = M(x̄ + s/2, x̄ − s/2) with respect
to s at fixed x̄ = x (cf. Box 1.1):

M (x, k) = d s M̄(x, s) e −i k·s . (3.16)

The inverse WWT4 W −1 maps any given function M (x, k) to an operator M̂ via [19]

M̂ =
1
(2π)n
d x d k d s |x + s/2i M (x, k) e i k·s hx − s/2| . (3.17)

For M̂ that is a matrix of operators M̂ uv , the corresponding symbol M (x, k) is defined as the matrix
whose elements are the symbols of M̂ uv . Also note that the WWT commutes with the Hermitian
conjugation:
W [M̂ † ] = M † (x, k). (3.18)
4 The direct WWT is also called the Wigner transform. The inverse WWT is also called the Weyl transform.

31
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 32

As a corollary (Problem 3.2), the Hermitian and anti-Hermitian parts [Eq. (C.3)] of M̂ and M satisfy

MH (x, k) = W [M̂ H ], MA (x, k) = W [M̂ A ]. (3.19)

Exercise 3.2: Prove Eqs. (3.18) and (3.19). Show that the symbol of a Hermitian operator is a
Hermitian matrix or, in case of a scalar operator, a real function.

The WWT is a natural mapping between operators and functions on phase space in the following
sense. The symbol of a unit operator is unity:

W [1̂] = d s hx + s/2|1̂|x − s/2i e −i k·s

= d s hx + s/2|x − s/2i e −i k·s

= d s δ(s) e −i k·s
= 1. (3.20)

The symbol of an operator that can be represented as a function of the coordinate operator x̂ is the
same function of x:

W [F (x̂)] = d s hx + s/2|F (x̂)|x − s/2i e −i k·s

= d s F (x − s/2) hx + s/2|x − s/2i e −i k·s

= d s F (x − s/2)δ(s) e −i k·s
= F (x). (3.21)

Similarly, the symbol of an operator that can be represented as a function of the wavevector operator
k̂ is the same function of k (Exercise 3.3):

W [G(k̂)] = d s hx + s/2|G(k̂)|x − s/2i e −i k·s = G(k). (3.22)

Also, the symbol of an operator representable as F (x̂) + G(k̂) is F (x) + G(k). In summary then,

1̂ ⇔ 1, x̂ ⇔ x, k̂ ⇔ k, F (x̂) + G(k̂) ⇔ F (x) + G(k), (3.23)

where ⇔ denotes the correspondence between operators and their symbols.

Exercise 3.3: Prove Eq. (3.22) using the result from Exercise 3.1.

In the general case, though, the correspondence ⇔ is more complicated than swapping (x̂, k̂) and
(x, k). For example,

W [k̂F (x̂)] = d s e −i k·s hx + s/2|k̂F (x̂)|x − s/2i

= d k0 d s e −i k·s hx + s/2|k0 i k̂F (x − s/2) hk0 |x − s/2i

32
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 33


= d k0 d s e −i k·s hx + s/2|k0 i k0 F (x − s/2) hk0 |x − s/2i

=
d k0 d s k0 F (x − s/2) e i (k −k)·s 0

(2π)n

=
d k0 d s (k0 + k)F (x − s/2) e i k ·s 0

(2π)n

=k
d k0 d s F (x − s/2) e i k ·s − i  d k0 d s F (x − s/2) ∂ e i k ·s
0 0
s
(2π)n (2π)n
 
= k d s F (x − s/2)δ(s) + i
d k0 d s ∂ [F (x − s/2)] e i k ·s 0
s
(2π)n

= kF (x) −
i ∂x
d k0 d s F (x − s/2) e i k ·s 0

2 (2π)n

= kF (x) −
i ∂x F (x). (3.24)
2
Using Eq. (3.18), one also obtains

†
W [F (x̂)k̂] = W [(F (x̂)k̂)† ]

†
= W [k̂F † (x̂)]
†
= kF † (x) − i/2 ∂x F † (x)


= F (x)k + i/2 ∂x F (x), (3.25)

and as a corollary,

kF (x), F (x)k ⇔ 1/2 [F (x̂)k̂ + k̂F (x̂)] . (3.26)

The terms ∂α F in Eqs. (3.24) and (3.25) emerge because x̂ and k̂ do not commute,

[x̂α , k̂ β ] = i δβα . (3.27)

These terms are of order ε and vanish in the GO limit (ε → 0). Similarly, for a general M̂ = M (0) (x̂, k̂),
where M (0) is any combination of x̂ and k̂, one has

W [M̂ ] = M (0) (x, k) + O εM (0) .

(3.28)

3.4 Envelope equation

Now, let us return to Eq. (3.1), multiply it by e −i θ(x) , and rewrite the result as follows:
D̂ = e −i θ(x) D̂ e i θ(x) .
.
D̂Ψ = 0, (3.29)

The operator D̂ serves as the envelope dispersion operator and can also be expressed in the following
invariant form:

D̂ = e −i θ(x̂) D̂ e i θ(x̂) . (3.30)

By definition [Eq. (3.15)], the symbol of D̂ is


D(x, k) = d s hx + s/2|e −i θ(x̂) D̂ e i θ(x̂) |x − s/2i e −i k·s

= d s hx + s/2|e −i θ(x+s/2) D̂ e i θ(x−s/2) |x − s/2i e −i k·s

33
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 34


= d s hx + s/2|D̂|x − s/2i e −i [θ(x+s/2)−θ(x−s/2)+k·s] . (3.31)

Since the gradient of θ is assumed to be a slow function, we can use

sα ∂θ sα sβ ∂ 2 θ sα sβ sγ ∂3θ
θ(x ± s/2) = θ(x) ± α
+ α β
± + ..., (3.32)
2 ∂x 8 ∂x ∂x 48 ∂x ∂xβ ∂xγ
α

where summation over repeated indexes is assumed. Then,

e −i [θ(x+s/2)−θ(x−s/2)+k·s] = 1 − 24i sα sβ sγ ∂x∂αk̄∂xγ β + . . . e −i [k̄(x)+k]·s

2
 

1 ∂ 2 k̄γ ∂3
 
= 1−
24 ∂xα ∂xβ ∂kα ∂kβ ∂kγ
+ . . . e −i [k̄(x)+k]·s , (3.33)

where we assumed that the second term in the parentheses is small compared to one. (A more explicit
estimate of the error is presented below.) This leads to

1 ∂ 2 k̄γ ∂3

D(x, k) = 1 −
24 ∂xα ∂xβ ∂kα ∂kβ ∂kγ
+ ... d s hx + s/2|D̂|x − s/2i e −i [k̄(x)+k]·s
1 ∂ 2 k̄γ ∂3
 


= 1− + . . . D x, k̄(x) + k
24 ∂xα ∂xβ ∂kα ∂kβ ∂kγ

= D 0 x, k̄(x) + k ,


(3.34)

where we have introduced

. 1 ∂ 2 k̄γ ∂ 3 D(x, k)
D 0 (x, k) = D(x, k) − + ... (3.35)
24 ∂xα ∂xβ ∂kα ∂kβ ∂kγ

Note that the inverse WWT will, loosely speaking, turn the coordinate k into the operator k̂. The
latter will act on the wave envelope, which is considered slow in the coordinate representation, so
k̂ |Ψi = O(ε). In this sense, k = O(ε) in Eq. (3.34). Then, D 0 (x, k̄(x) + k) can be Taylor-expanded
in k, which corresponds to expanding D̂ in ε. Next, notice that

. ∂ 2 k̄γ ∂ 3 D[x, k̄(x) + k]

ε0 = D[x, k̄(x) + k]−1 α β
∂x ∂x ∂kα ∂kβ ∂kγ
∂ 2 k̄γ ∂ 3 D(x, k̄)
≈ D(x, k̄)−1
∂xα ∂xβ ∂ k̄α ∂ k̄β ∂ k̄γ
  
k̄ D
∼ = O(ε2 ), (3.36)
L2c D k̄ 3

where we have assumed ∂/∂xα ∼ 1/Lc and ∂/∂ k̄ ∼ 1/k̄.5 Because we are interested only in O(ε)
corrections, this means that on the right-hand side of Eq. (3.35), the second term can be neglected
compared to the first term. In other words, D 0 (x, k) ≈ D(x, k) and thus


D(x, k) ≈ D x, k̄(x) + k , (3.37)


5A more accurate
estimate is as follows. Let us consider D(x, k) in the form (3.16), i.e., D(x, k) = d s D̄(x, s) e −i k·s .
Then, ∂D/∂kα = −i d s D̄(x, s)sα e −i k·s ∼ sα D, where sα is the characteristic scale on which D̄(x, s) fades away along
the sα axis. Then, ε0 ∼ (k̄Lc )−2 (k̄s)3 ∼ ε2 (k̄s)3 . To ensure that ε0  ε, one must require (k̄s)3 ε  1. Strictly
speaking, this is an additional requirement of GO, independent of Eq. (3.5). The scale s can be estimated in the
homogeneous-plasma limit, when D(x, k) is just the dispersion tensor, D(x, k) → D(k).

34
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 35

Box 3.1: Quasioptical and paraxial approximation

In practice, waves often propagate as narrow beams whose transverse scale L⊥ is much less than
the longitudinal scale Lk , which is comparable to that of the medium. For diffraction-limited
. .
beams, one has εk ∼ ε2⊥ , where εk = (k̄Lk )−1 and ε⊥ = (k̄L⊥ )−1 , so when retaining terms of the
first order in εk , one must retain effects of the second order in ε⊥ . Because ε0 [as in Eq. (3.36)] is
of order ε2k , Eq. (3.37) is still applicable, but Eq. (3.38) must be replaced with the second order
expansion in k⊥ . The corresponding envelope equation is
 
D x, k̄(x) Ψ −
i (∂α V α )Ψ − i V α ∂α Ψ −
1
(Φαβ : ∇⊥α ∇⊥β )Ψ = 0,
2 2
. 2
where Φαβ = ∂ D/∂kα ∂kβ evaluated at [x, k̄(x)]. This model is called the paraxial approximation
if the beam axis is a straight line, or in the general case, the quasioptical approximation [17]. It is
widely used, for example, in optics and also for modeling mm- and cm-wave beams in magnetically
confined plasmas.

or using first-order Taylor expansion,

 
. ∂D(x, k)
D(x, k) ≈ D x, k̄(x) + kα V α (x), V α (x) =


. (3.38)
∂kα k=k̄(x)

By Eq. (3.26), the inverse WWT of this expression is

1
D̂ ≈ D x̂, k̄(x̂) + [k̂α V α (x̂) + V α (x̂)k̂α ], (3.39)
2
or in the coordinate representation,

D̂Ψ ≈ D(x, k̄(x)]Ψ(x) −

i ∂α [V α (x)Ψ(x)) −
i V α (x)∂α Ψ(x). (3.40)
2 2
Then, the envelope equation D̂Ψ = 0 can be approximately expressed as follows:


D x, k̄(x) Ψ −
i (∂α V α )Ψ − i V α ∂α Ψ = 0, (3.41)
2
which can be considered as an extension of Eq. (1.52) to a general inhomogeneous medium (Box 3.1).
We will further simplify this equation in Lecture 4.

3.5 How to use the envelope equation

In order to use Eq. (3.41), one must know D to the first order in ε, which can be challenging to calculate
in practice. Fortunately, this problem is often alleviated by the following. As seen from Lecture 2 and
Problem PI.2, collisionless cold static (CCS) plasmas are naturally described by Hermitian dispersion
operators that satisfy (Exercise 3.4)
(0)
D CCS = D CCS , (3.42)

where the notation is the same as in Sec. 3.3. Deviations from CCS models due to thermal effects and
ˆ to the dispersion operator D̂. Then, the symbol of
collisions often result in only small corrections ∆
this operator, D, satisfies
(0)
D − D (0) = D CCS + ∆ − D CCS + ∆(0) = ∆ − ∆(0) = O(ε∆),



(3.43)

35
LECTURE 3. ASYMPTOTIC EXPANSION OF DISPERSION OPERATORS 36

Box 3.2: Subtleties with approximating D in other formulations of GO

Although GO can be constructed based on a transform other than the WWT, the approxima-
tion (3.45) relies on the WWT property (3.19) that the other transform may not have. Then,
one might not be able to approximate the corresponding symbol of D̂ with the (well-known)
dispersion tensor of the wave in a homogeneous medium.

For example, sometimes GO is constructed based on the transform


.
W¯ [M̂ ] = d s M(x, x − s) e −i k·s = hx|hx|ki
M̂ |ki

(e.g., see Ref. [21]). To the extent that the commutator [x̂, k̂] is negligible, one has W¯ [M̂ ] ≈
M (0) (x, k), so W¯ satisfies Eq. (3.28) just like the WWT. However, symbols produced by the
transform W¯ do not have the property (3.18). As a result,
(0) (0)
(W¯ [D̂])A = D A + [O(εD (0) )]A = D A + O(ε).

Both terms on the right-hand side are of the same order, so the anti-Hermitian part of the symbol
(0)
differs from the corresponding matrix in homogeneous plasma by ∼ ∂ 2 D H /∂xα ∂kα . This leads
to alternative GO equations, which often results in confusion and stirred a controversy [22, 23].

where we have used Eq. (3.28) to obtain the last equality. Because both ε and ∆ are small, the term
O(ε∆) is often negligible [as opposed to O(ε) and O(∆)], i.e., one can adopt

D ≈ D (0) . (3.44)
This means that the symbol of the dispersion operator can be approximated with the dispersion tensor
of homogeneous plasma. In fact, the two are often not even distinguished explicitly in literature.

Exercise 3.4: Calculate the symbols of: (a) D̂ E for waves in CCS nonmagnetized plasma
(Lecture 2) and (b) D̂ ψ for waves in CCS magnetized plasma (Problem PI.2).

(0)
Because D CCS is Hermitian, Eq. (3.44) is often further reduced to (Box 3.2)
(0) (0) (0) (0) (0)
D H ≈ D CCS + ∆H ≈ D CCS , D A ≈ 0 + ∆A = ∆A . (3.45)
This is justified in the following sense. In D H , which determines wave propagation (Lecture 5), the
main contribution is provided by D CCS , so ∆H is only a small correction, which is often negligible.
In contrast, D A , which determines wave dissipation (Lecture 5), is entirely determined by ∆A , so
(0)
unlike ∆H , the function ∆A must be kept. The advantage of the model (3.45) is that D CCS (Part III)
(0)
is simpler than D H (Part IV) and thus easier to implement numerically. (However, note that this
model can be insufficient when dissipation is strongly inhomogeneous. Sometimes this happens for
resonant absorption, which is naturally localized in space.)
The arguments presented in this section do not apply to corrections ∆ ˆ m caused by plasma motion.
ˆ (0) (0)
In this case, ∆m = 0, so Eq. (3.43) becomes D − D = ∆m = O(). Ignoring this correction can
lead to significant violation of energy conservation on times t = O(−1 ) and is typically not acceptable.
One must actually recalculate the dispersion operator for moving plasma to fix this issue. (But if one
needs an equation just for the energy, the problem can be bypassed as discussed in the next lectures.)
This asymmetry in how GO tolerates various corrections is due to the fact that plasma temporal
dispersion is typically much stronger than plasma spatial dispersion.

36
Lecture 4

Equations of geometrical optics

In this lecture, we derive a complete set of equations that describe scalar waves: the local dispersion
relation, the polarization equation, the consistency relations, the ray equations, and scalar-amplitude
equations, including an equation for the wave action. These are known as the equations of GO.

4.1 Scalar-wave model

Based on the reasoning presented in Sec. 3.5, let us assume the following ordering:

D H (x, k) = O(1), D A (x, k) = O(ε). (4.1)

Then, since ∂α = O(ε) and D A = O(ε), the envelope equation (3.41) can be simplified as follows:

D H x, k̄(x) Ψ + i D A x, k̄(x) Ψ −
i
H )Ψ − i V H ∂α Ψ = 0,
(∂α V α α
   
(4.2)
2
where V α α
H is the Hermitian part of V . It is readily seen from here that

D H [x, k̄(x)]Ψ = O(ε). (4.3)

This means that the field polarization must approximately satisfy the same equation as in the cor-
responding homogeneous medium, D H [x, k̄(x)]Ψ ≈ 0. Hence, it is convenient to represent Ψ in the
basis of the orthonormal eigenvectors hv (x) of D H [x, k̄(x)],
N
X
Ψ(x) = hv (x)Ψv (x), hu · hv ≡ h†u (x)hv (x) = δuv (4.4)
v=1
.
(as earlier, N = dim Ψ), and the corresponding eigenvalues Λv (x), which satisfy

D H [x, k̄(x)]hv (x) = Λv (x)hv (x). (4.5)

Let us consider the scalar product of Eq. (4.3) with hu (x):

N
X
O(ε) = h†u (x)D H [x, k̄(x)]hv (x)Ψv (x)
v=1
N
X
= Λv (x)h†u (x)hv (x)Ψv (x)
v=1

= Λu (x)Ψu (x). (4.6)

37
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 38

Equation (4.6) shows that Ψu can be order-one only if the corresponding eigenvalue is small,
Λu = O(ε). This can be satisfied for more than one mode at a time, in which case “mode conversion”
is possible (Problem PII.2). Here, we will consider a simpler case, when only one, say uth, mode
has a noticeable amplitude. Specifically, suppose Λu = O(ε), and Λv6=u = O(1); then, Ψu = O(1)
is allowed, and Ψv6=u = O(ε). Let us re-examine the wave equation under these conditions more
carefully. Consider the product of Eq. (4.2) with h†u :
N 
i

Λv h†u hv Ψv + i h†u D A hv Ψv − −i
X
0= h†u (∂α V α
H )hv Ψv h†u V α
H ∂α (hv Ψv )
v=1
2
N
(Λv δuv Ψv ) + i (h†u D A hu )Ψu −
i
H )hu Ψu − i hu V H ∂α (hu Ψu )
X
≈ h†u (∂α V α † α

v=1
2

= Λu Ψu + i Γu Ψu −
i
H )hu Ψu − i (hu V H ∂α hu )Ψu − i Vu ∂α Ψu .
h†u (∂α V α † α α
(4.7)
2
(Although u is a repeating index, no summation over u is assumed.) Here,

Λu = h†u (x)D H [x, k̄(x)]hu (x), (4.8a)

.
Γu = h†u (x)D A [x, k̄(x)]hu (x), (4.8b)
.
Vuα = h†u (x)V α
H (x)hu (x) (4.8c)

are real scalar functions. Also note that

i
H )hu + i hu V H ∂α hu
h†u (∂α V α † α
2
i∂ i i
H (∂α hu ) + i hu V H (∂α hu )
† α
= α (hu V H hu ) − (∂α h†u )V α hu − h†u V α † α
2 2 2
=
i (∂α Vuα ) +
i h
h†u V α (∂ h ) − (∂ h†
)V α
h
i
H α u α u H u
2 2
=
i (∂α Vuα ) + Qu , (4.9)
2
where Qu is a real function given by1

.
Qu =
i h
h†u V α (∂ h ) − (∂ h†
)V α
h
i
= Im
h
(∂ h†
)V α
i
H α u α u H u α u h
H u . (4.10)
2
Then, we obtain the following scalar equation:

i
 
Λu − Qu + i Γu − (∂α Vuα ) − i Vuα ∂α Ψu = 0. (4.11)
2

Recall now that k̄ was introduced as the gradient of θ [Eq. (3.3)] and θ has not been defined yet.
Let us define it now by requiring that2

Λu (x, ∂θ) = 0, i.e., Λu (x, k̄) = 0, (4.12)

which can be understood as a Hamilton–Jacobi equation for θ [12]. This leads to the following:
1 If N = 1, then there is only one normalized polarization vector, hu ≡ 1. Accordingly, ∂α h†u = 0, so Qu = 0.
2 This is the most common approach to formulating GO. Another natural way to define θ is to require that Ψ be
real; then Eq. (4.12) is replaced with Λu − Qu = 0. This is further discussed in Sec. 4.4.

38
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 39

• Equation (4.12) connects the local frequency w̄(t, x) with the local wavevector k̄(t, x). This
means that w̄ can be expressed through k̄:

w̄(t, x) = ω[t, x, k̄(t, x)], (4.13)

where ω(t, x, k̄) is a solution of the local dispersion relation

Λu [t, x, ω(t, x, k̄), k̄] = 0. (4.14)

Q
Also note that because det D H = v Λv , Eq. (4.14) can be considered as a particular solution
of the more general dispersion relation

det D H [t, x, ω(t, x, k̄), k̄] = 0, (4.15)

which is not specific to v = u but describes all modes simultaneously.

• By Eq. (4.5) with v = u, Eq. (4.12) also yields

D H hu = 0, (4.16)

which determines the mode polarization. This equation is not in violation of the assumed order-
ing (4.1), because the latter characterizes D H (x, k) at generic (x, k) while Eq. (4.16) characterizes
h specifically on solutions of Eq. (4.12).
• By combining Eqs. (4.11) and (4.12), one obtains the following equation for Ψu (Box 4.1):

(∂α Vuα )Ψu = (Γu + i Qu )Ψu .

1
Vuα ∂α Ψu + (4.17)
2
Since the phase of Ψu is often not of interest, let us transform this into an equation for the real
field |Ψ|2 . To do this, let us multiply Eq. (4.21c) by Ψ∗ and sum up the resulting equation with
its complex conjugate,

Ψ∗u (Vuα ∂α )Ψu + (∂α Vuα )|Ψu |2 = (Γu + i Qu )|Ψu |2 ,

1
(4.18)
2
Ψu (Vuα ∂α )Ψ∗u + (∂α Vuα )|Ψu |2 = (Γu − i Qu )|Ψu |2 .
1
(4.19)
2
This leads to

∂α (Vuα |Ψu |2 ) = 2Γu |Ψu |2 . (4.20)

Let us summarize these results and simplify the notation. Specifically, we now omit the bar in k̄
and drop the mode index u . Then, our main equations can be written as follows (Exercise 4.1):

Λ[t, x, ω(t, x, k), k] = 0, (4.21a)

D H h = 0, (4.21b)
α 2 2
∂α (V |Ψ| ) = 2Γ |Ψ| . (4.21c)

This is a complete set of GO equations that allows one to find the whole vector field ψ = hΨe i θ +O(ε)
up to the O(ε) term, which is often considered negligible and otherwise can be found perturbatively.
The effect of other modes is completely ignored in this approximation, so effectively, any wave is
modeled as a scalar wave. (A more general model, which allows for linear resonant coupling of
multiple GO modes, is addressed in Problem PII.2.) Below, we discuss how to solve these equations
in practice.

39
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 40

Box 4.1: Amplitude equation for narrow beams

If a wave beam is narrow enough such that variations of Vuα across the beam can be ignored,
then Vuα ≈ Vuα (l), Vuα ∂α Ψu ≈ Vul ∂l Ψu , and ∂α Vuα ≈ dl Vul , where l is the coordinate along the
.
beam axis. In this case, Eq. (4.17) yields the following equation for Ψ̄u = Ψu (Vul )1/2 :

Vul ∂l Ψ̄u = (Γu + i Qu )Ψ̄u .

This equation does not contain ∂α Vuα , so may be easier to implement numerically than Eq. (4.17).
Also notably, when Vul turns to zero, the field is singular. Beyond the GO approximation, this
corresponds to a caustic like those discussed in Sec. 2.3.

Exercise 4.1: Show that at Γ = 0, Eq. (4.21c) can
be derived, along with Eq. (4.21a), from
the least-action principle with the action integral S = d x Λ(x, ∂θ)A. Here, θ and A = |Ψ|2 are
.
considered as independent functions of x and Λ is a given function.

4.2 Ray equations

4.2.1 Consistency relations

As the first step, one needs to solve Eq. (4.21a) for ω and Eq. (4.21b) for h, so one knows

ω = ω(t, x, k), h = h(t, x, k). (4.22)

The next step is to calculate k(t, x). This can be done by revisiting the definitions of w and k [see
Eq. (3.3), where the bars have been dropped to simplify the notation]:
. .
w(t, x) = −∂t θ(t, x), k(t, x) = ∇θ(t, x). (4.23)

Since ∇ × ∇ = 0 and ∂t ∇θ = ∇∂t θ, one obtains the following “consistency relations”:

∇ × k = 0, ∂t k(t, x) = −∇w(t, x). (4.24)

The first consistency relation yields ∂kb /∂xa = ∂ka /∂xb . With this and Eq. (4.13), the second
consistency relation yields the following nonlinear PDE for k(t, x):

∂ka ∂w
=− a
∂t ∂x
∂ω ∂ω ∂kb
=− a −
∂x ∂kb ∂xa
∂ω ∂ω ∂ka
=− a −
∂x ∂kb ∂xb

Box 4.2: Conservation of wave crests

In a one-dimensional system, the first consistency relation can be written as a continuity equation,
∂t k+∂x (vp k) = 0, where vp = ω/k is the phase velocity. This can be understood as a conservation
of wave crests (or zeros) in GO [24].

40
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 41

Box 4.3: Connection between the two consistency relations

By taking the curl of ∂t k = −∇w, one obtains ∂t (∇ × k) = 0. Therefore, the second consistency
relation can be considered as the initial condition for the first consistency relation, much like
magnetic Gauss’s law serves as the initial condition for Faraday’s law (Lecture 1.1.1).

∂ω ∂ka
=− − vg · , (4.25)
∂xa ∂x
where the spatial derivative of ω on the right-hand side is taken at fixed k. The same can be written
in the following vector form:

[∂t + (v g · ∇)]k = −∂x ω. (4.26)

In other words, the evolution of the field k(t, x) in the frame moving at the group velocity v g is
determined by the explicit dependence of ω(t, x, k) on x.

4.2.2 Hamilton’s equations for rays

Consider trajectories x(t), called “characteristics” or “GO rays”, that are governed by

dt x(t) = vg [t, x(t), k(t)] ≡ ∂k ω[t, x(t), k(t)], (4.27)

.
with k(t) = k[t, x(t)]. Then, Eq. (4.26) yields

dt k(t) = [∂t + (vg · ∇)]k(t, x) x=x(t) = −∂x ω[t, x(t), k(t)],


(4.28)

which is an ordinary differential equation (ODE) for k(t). Also notice that the instantaneous frequency
.
ω(t) = ω[t, x(t), k(t)] is governed by

dt ω = ∂t ω + ∂x ω · dt x + ∂k ω · dt k = ∂t ω + ∂x ω · ∂k ω − ∂k ω · ∂x ω = ∂t ω. (4.29)

In summary then,

dt x = ∂k ω(t, x, k), dt k = −∂x ω(t, x, k), dt ω = ∂t ω(t, x, k). (4.30)

Notably, Eqs. (4.30) remain applicable near reflection points as well, even though the amplitude
equation breaks down there (Box 4.1) and the GO parameter is not small. The reason for this is that
the GO approximation can be reinstated near cutoffs using phase-space rotation (Box 2.1) that does
not affect the ray equations.
. .
To better understand Eqs. (4.30), notice the following. The quantities p = ~k and H = ~ω are
commonly interpreted as the photon canonical momentum and energy. Assuming this interpretation,
the ray equations can be viewed as Hamilton’s equations of the photon motion in phase space (x, p),
with H(t, x, p) serving as the Hamiltonian:3
d x = ∂ω = ∂(~ω) = ∂H , (4.31a)
d t ∂k ∂(~k) ∂p
d p = d (~k) = − ∂(~ω) = − ∂H , (4.31b)
dt dt ∂x ∂x
d H = ~ d ω = ~ ∂ω = ∂H . (4.31c)
3 Strictly
dt dt ∂t ∂t
speaking, photons are defined through global modes and thus cannot be assigned specific coordinates. For
example, in a stationary GO wave, each photon is spread out along the whole propagation distance. However, this does
not limit the applicability of the ray equations, because they are derived from classical considerations and independently
from the photon analogy.

41
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 42

Box 4.4: Covariant form of the ray equations

Equations (4.34) can also be derived as follows. Let us introduce some auxiliary time τ and
consider solutions of the original equation D̂ψ = 0 as “stationary” (∂τ ψ = 0) solutions of
(D̂ − i ∂τ )ψ = 0. The latter can be considered as D̂ ext ψ = 0, where D̂ ext = D̂ − $̂ is an
.
“extended” dispersion operator and $̂ = i ∂τ is the frequency operator associated with the new
.
time τ . Then, one can build GO for D̂ ext , in which case x0 ≡ ct and k0 ≡ −ω/c are treated
on the same footing as x and k. The corresponding dispersion relation is readily found to be
$ = Λ(t, x, ω, k), and Eqs. (4.34) emerge as the covariant ray equations in spacetime:

d xα =
∂Λ(x, k)
,
d kα =−
∂Λ(x, k)
.
dτ ∂kα dτ ∂xα

4.2.3 Alternative forms of the ray equations

Another useful form of the ray equations is derived as follows. Because Eq. (4.21a) holds for all t, x,
and k, one can differentiate it with respect to all these variables to obtain

0 = dt Λ[t, x, ω(t, x, k), k] = (∂ω Λ)(∂t ω) + ∂t Λ, (4.32a)

0 = dx Λ[t, x, ω(t, x, k), k] = (∂ω Λ)(∂x ω) + ∂x Λ, (4.32b)
0 = dk Λ[t, x, ω(t, x, k), k] = (∂ω Λ)(∂k ω) + ∂k Λ. (4.32c)

These equations lead to

∂t Λ ∂ω ∂x Λ ∂ω ∂k Λ ∂ω
=− , =− , =− = vg . (4.33)
∂ω Λ ∂t ∂ω Λ ∂x ∂ω Λ ∂k

Let us define an auxiliary time variable τ via dt τ = −1/∂ω Λ.4 Using dt = (dt τ )dτ and Eqs. (4.33),
one can rewrite Eqs. (4.30) as follows:

dt ∂Λ(t, x, ω, k) d ω = + ∂Λ(t, x, ω, k) ,
dτ = − ∂ω
,
dτ ∂t
(4.34a)
d x = + ∂Λ(t, x, ω, k) , d k = − ∂Λ(t, x, ω, k) , (4.34b)
dτ ∂k dτ ∂x
where the former equation is the definition of τ included for completeness. (An alternative path to
these equations is presented in Box 4.4.) As a corollary, k is conserved if the medium is spatially
homogeneous (∂x Λ = 0), and ω is conserved if the medium is stationary (∂t Λ = 0).

4.3 Amplitude equation

Now, let us figure out how to marry the ray equations with the amplitude equation (4.21c). Note that

∂D H
V α = h† h
∂kα
 †  
∂ ∂h ∂h
= (h† D H h) − D H h − h† D H
∂kα ∂kα ∂kα
 †  
∂Λ ∂h ∂h
= − ×0−0×
∂kα ∂kα ∂kα
4 Other definitions of τ can be used as well and lead to different but equivalent ray equations.

42
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 43

∂Λ
= , (4.35)
∂kα
where we have used Eq. (4.21b) and its conjugate. Then, we obtain
   
∂ ∂Λ 2 ∂ ∂Λ 2
|Ψ| + · |Ψ| = 2Γ |Ψ|2 , (4.36)
∂x0 ∂k0 ∂x ∂k
where the first term can as well be expressed as follows:
     
∂ ∂Λ 2 ∂ ∂Λ 2 ∂ ∂Λ 2
|Ψ| = |Ψ| = − |Ψ| . (4.37)
∂x0 ∂k0 ∂(ct) ∂(−ω/c) ∂t ∂ω
Let us introduce
. ∂Λ 2 ∂D H ∂D H
I= |Ψ| = h† h|Ψ|2 = Ψ† Ψ, (4.38)
∂ω ∂ω ∂ω
. ∂Λ ∂D H ∂D H
Ja = − |Ψ|2 = −h† h|Ψ|2 = −Ψ† Ψ. (4.39)
∂ka ∂ka ∂k
Importantly, one can also express J through the eigenvalue Λ as
∂k Λ
J =− I = v g I, (4.40)
∂ω Λ
where we have used Eq. (4.33). Then, Eq. (4.36) can be expressed as a continuity equation for I,
∂t I + ∇ · (v g I) = 2γI, (4.41)
where γ is given by

. Γ h† D A h h† D A h
γ=− =− † =− . (4.42)
∂ω Λ h (∂ω D H ) h ∂ω Λ
The quantity I that enters this equation [and is given by Eq. (4.38)] is called the action density,
and accordingly, J can be understood as the action flux density. The physical meaning of these
quantities will be discussed in detail in Lecture 5. Meanwhile, note that Eq. (4.41) can be rewritten as
[∂t + (v g · ∇)] I = −I∇ · v g + 2γI. (4.43)
.
This readily leads to an ODE for the action density on a ray, I(t) = I[t, x(t)], namely,5
dt I = 2γeff (t)I, .
γeff (t) = (γ − ∇ · v g /2)[t,x(t),k(t)] . (4.44)
By integrating Eq. (4.44), one obtains
 t 
I(t) = exp d t 2γeff (t ) I0 ,
0 0
(4.45)
0

where I0 is determined by initial conditions.

In summary then, GO equations are applied as follows. First, one finds ω(t, x, k) from
Λ(t, x, ω, k) = 0. (4.46)
Then, one calculates k(t, x) by solving either the consistency relation (4.26) as a PDE or the ray
equations (4.30) as ODEs. Next, one finds the wave polarization h from D H h = 0. After that, one
solves the equation for the action density, which also can be viewed either as a PDE [Eq. (4.41)] or
as an ODE on the rays [Eq. (4.44)]. Finally, one calculates the wave amplitude |Ψ| by definition of I
[Eq. (4.38)]. In Lecture 5, we will show how to do this for electromagnetic waves in particular.
5 For narrow wave beams, ∇ · v g can be removed from Eqs. (4.43) and (4.44) using the same idea as in Box 4.1.

43
LECTURE 4. EQUATIONS OF GEOMETRICAL OPTICS 44

∗
4.4 Spin Hall effect of light
The dispersion relation (4.46) is postulated in a somewhat arbitrary manner and is not entirely
consistent. The term i Q in the amplitude equation (4.17) can, with enough time, produce an arbitrarily
large gradient of arg Ψ. Because Q does not enter Eq. (4.41), such a gradient has no effect on |Ψ| within
the GO approximation, However, it can eventually undermine the validity of the GO approximation,
because the latter requires that Ψ be a slow function.
To prevent this, one can define θ such that Ψ be real. Then, by taking the real part of Eq. (4.11),
one arrives at a dispersion relation

Λ(t, x, ω, k) − Q(t, x, ω, k) = 0 (4.47)

instead of Eq. (4.12). This complicates the polarization equation, which now becomes

D H h = Λh = Qh (4.48)

[cf. Eq. (4.21b)]. However, the amplitude equation is simplified:

1
V α ∂α Ψ + (∂α V α )Ψ = Γ Ψ. (4.49)
2
The expression for V α is not affected:
∂D H
V α = h† h
∂kα
 †  
∂ ∂h ∂h
= (h† D H h) − D H h − h† D H
∂kα ∂kα ∂kα
 †  
∂Λ ∂h ∂h
= − Qh − h† Q
∂kα ∂kα ∂kα
∂Λ ∂(h† h)
= −Q
∂kα ∂kα
∂Λ
= (4.50)
∂kα

[cf. Eq. (4.35); here we have substituted Eq. (4.48) and h† h = 1]. However, remember that redefining
θ entails modification of k, so V α is now evaluated on a different k and thus is affected. Thus in
reality, a wave propagates somewhat differently than as predicted by the conventional ray equations
(4.30). This effect is known as the spin Hall effect of light and has been observed experimentally [25].
The spin Hall effect is typically negligible in practical applications, so it is usually ignored or not
even recognized. Still, this effect is interesting because of its analogy with the spin–orbital interaction
known for quantum particles. (The GO limit in quantum mechanics is known as the semiclassical
approximation, and ray equations for quantum waves are known as classical mechanics.) When placed
in an external magnetic field, quantum particles exhibit different trajectories depending on their spin
state, for example, as seen in the famous Stern–Gerlach experiment. The analog of the spin for
electromagnetic waves is their polarization state, and parameters of the medium serve as the vector
potential, whose derivatives create an effective “magnetic field” for the “spin” to interact with. This
analogy can be made quantitative and in fact the spin Hall effect of light and the Stern–Gerlach effect
are mathematically identical [20, 26, 27].

44
Lecture 5

Wave action, energy, and

momentum

In this lecture, we discuss the physical meaning of the wave action and also introduce the wave energy,
the wave momentum, and the corresponding transport equations.

5.1 Wave action

As shown in Lecture 4, the amplitude equation for GO waves can be represented in the form

∂t I + ∇ · (v g I) = 2γI, (5.1)

where I = Ψ† (∂ω D H )Ψ is the wave action density and γ = −h† D A h/∂ω Λ. By integrating Eq. (5.1)
over the whole space, one obtains

d  d x I +  d x ∇ · (v I) =  d x 2γI. (5.2)
dt g

By Gauss’s theorem, the second term equals the flux through the infinite surface and thus is zero,
.

assuming that the field is localized. Then, if γ = 0, the total action I = d x I is conserved:

dI = 0. (5.3)
dt
The reason for this conservation law is that at zero γ, the wave is a Lagrangian system whose La-
grangian density does not depend on θ explicitly (Box 5.1). This makes I a Noether invariant of an
asymptotic theory, an adiabatic invariant. The well-known adiabatic invariant of a harmonic oscillator
can be understood as a special case of I (Box 5.2). In a broader context, Eqs. (5.1)–(5.3) generalize the
WKB conservation law (2.35) that was derived in Lecture 2 for a specific wave in a simple geometry.
Because I satisfies a continuity equation (modulo the local dissipation term γ), it can be interpreted
as the density of quasiparticles that travel with velocity v g . With the same reservations as in Lecture 4,
these particles can be identified as photons. We will return to the photon analogy later in this lecture.

45
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 46

Box 5.1: Variational approach

At γ = 0, the envelope equation (4.11) becomes

0 = [Λ − Q − i/2 (∂α V α ) − i V α ∂α ] Ψ ≡ LˆΨ, (5.4)

where the mode index is omitted and Γ ∝ γ has vanished. The operator
Lˆ is Hermitian, so this
equation satisfies the least-action principle δS[Ψ , Ψ] = 0 with S = d t d x Ψ∗ LˆΨ, where Ψ and
∗

Ψ∗ are independent variables.a At ε → 0, when Q and ∂α vanish, one obtains

S≈ d t d x Λ(t, x, −∂t θ, ∇θ)|Ψ|2 , (5.5)
.
where we used w = −∂t θ and k = ∇θ. (See also Exercise 5.2.) If θ and A = |Ψ|2 are treated as
independent variables, this S readily leads to the following Euler–Lagrange equations:

0 = δA S = Λ(t, x, w, k), (5.6)

0 = δθ S = ∂t I + ∇ · (v g I), (5.7)

which are the anticipated GO dispersion relation and the action-conservation theorem. Note that
there is no need to retain O(ε) corrections when deriving the action-conservation theorem from
.
a variational

principle. Also note that one obtains a coninuity equation for I = ∂ω L for any
S ≈ d t d x L such that the wave Lagrangian density has the form L = L(x, ∂θ). However, if L
explicitly depends on θ rather than ∂θ, then the wave action is not conserved.
a Because Ψ and Ψ∗ are linearly independent combinations of Re Ψ and Im Ψ, and because Re Ψ and Im Ψ can

be treated as independent, the complex functions Ψ and Ψ∗ can be treated as independent too.

Application to electromagnetic waves: action density

Let us now apply the above machinery to electromagnetic waves as a special case. In this case, Ψ = E
is the electric-field complex amplitude and1
 2 
1 c † 2
D̂(t, x, ω, k) = (k̂k̂ − 1 k̂ ) + ˆ . (5.8)
16π ω̂ 2

The corresponding symbol is

 2 
1 c † 2
D(t, x, ω, k) = (kk − 1 k ) + (t, x, ω, k) , (5.9)
16π ω 2

where  is the symbol of ˆ. Also, from Faraday’s law the magnetic-field complex envelope is, to the
leading order, B = (c/ω)(k × E ).
The corresponding action density can be calculated as follows:
 2
2c2
  
1 † c † 2 1 † † 2
I= E ∂ω 2 (kk − 1 k ) + H E = E − 3 (kk − 1 k ) + ∂ω H E . (5.10)
16π ω 16π ω

Using D HE = 0, one can rewrite this as follows:

1
I= E † [−2(DH − H ) + ω∂ω H ]E
16πω
1 A dispersion operator of a free wave is defined only up to a constant factor. Here, we have introduced an additional

factor (16π)−1 (compared to Lecture 1) in order to simplify the interpretation of I in later sections.

46
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 47

Box 5.2: Example: linear oscillator

Our formulation of GO applies not just to electromagnetic waves but also to any linear oscil-
lating systems that satisfy the assumed orderings. For example, consider a dissipative harmonic
oscillator with time-dependent frequency ω0 (t):

q̈ + 2ν q̇ + ω02 (t)q = 0, q = Re (a e i θ ),

assuming the damping coefficient ν is a small constant. We can rewrite this as D̂q = 0, where

D̂ = ω̂ 2 + 2i ν ω̂ − ω02 (t̂), ω̂ = i ∂t .
.

The symbol of this D̂ is

D(t, ω) = ω 2 + 2i νω − ω02 (t).

Since D is a scalar, one has Λ = DH = ω 2 − ω02 (t), so the local dispersion relation is ω 2 = ω02 (t).
The action density I, which is the same as the total action I here, is given by
.
I = (∂ω Λ) |a|2 = 2ω0 |a|2

and satisfies İ = 2γI, where

DA
γ=− = −ν.
∂ω Λ
Note that this I = I is proportional to U/ω, where U is the time-averaged energy:
1 2 1 1
U= hq̇ i + ω02 hq 2 i = ω02 hq 2 i = ω02 |a|2 .
2 2 2
Thus, the action I of a harmonic oscillator is the same as the well-known adiabatic invariant [12]
that is conserved at ν = 0.

1
= E † (ω∂ω H + 2H )E
16πω
1
= E † ∂ω (ω 2 H )E . (5.11)
16πω 2
Using that

E † (kk† − 1 k2 )E = E † [k(k · E ) − E k2 ] = E · [k × (k × E )] = (E × k) · (k × E ) = −|k × E |2

(assuming the notation as described in Appendix B), one can also express I as follows:

1 h c2 i
I= E † − 2 (kk† − 1 k2 ) − (DH − H ) + ω∂ω H E
16πω ω
1 †
h c2 i
= E ∂ω (ωH ) − 2 (kk† − 1 k2 ) E
16πω ω
1 h † c2 i
= E ∂ω (ωH )E + 2 |k × E |2
16πω ω
1 h † i
= E ∂ω (ωH )E + |B |2 . (5.12)
16πω

47
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 48

Then in summary,
1 1
I= 2
E † ∂ω (ω 2 H )E = [E † ∂ω (ωH )E + B †B ], (5.13)
16πω 16πω

or in terms of real fields Ẽ = Re (E e i θ ) and B̃ = Re (B e i θ ) (Exercise 5.1),

1 † 1  † 2 
I= 2
hẼ ∂ω (ω 2 H )Ẽi = hẼ ∂ω (ωH )Ẽi + hB̃ i , (5.14)
8πω 8πω
where the angular brackets denote average over the fast oscillations.

Exercise 5.1: Consider any a = Re (ac e i θ ) and b = Re (bc e i θ ) with slow complex envelopes ac
and bc and fast real phase θ. Show that their θ-averaged product can be expressed as follows:

habi = 1/2 Re (ac b∗c ) = 1/2 Re (a∗c bc ).

Application to electromagnetic waves: action flux density

The action flux density can be calculated either via J = v g I (Sec. 5.1) or directly as follows:
 2 
1 † c † 2 S+K
J =− E ∂k 2 (kk − 1 k ) + H (t, x, ω, k) E = . (5.15)
16π ω ω

Then, the gth component of the vector S is defined as

. c2 ∂
Sg = − (ka kb − δab kc kc )E ∗a E b
16πω ∂kg
c2
=− (δag kb + ka δbg − 2δab δcg kc )E ∗a E b
16πω
c2
= (2kg E ∗a E a − E ∗g kb E b − E g ka E ∗a )
16πω
c2
= [E × (k × E ∗ ) + E ∗ × (k × E )]g
16πω
c2
= Re [E × (k × E ∗ )]g
8πω
c
= Re (E × B ∗ )g , (5.16)
8π
so S is the average Poynting vector,
c c
S= Re (E × B ∗ ) = hẼ × B̃i. (5.17)
8π 4π
Then K, whose components are

. ω ∂
Ka = − E† HE, (5.18)
16π ∂ka
must be the kinetic action flux density. We will symbolically express it in the vector form as follows:
ω ω †
K=− E † ∂k HE = − hẼ ∂k H Ẽi. (5.19)
16π 8π

48
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 49

5.2 Wave energy

Equation (4.41) for the wave action density can be used to generate various corollaries. For example,
for any function X(t, x), one obtains
∂t (XI) + ∇ · (XIv g ) = (∂t X)I + X(∂t I) + [∇ · (Iv g )]X + I(v g · ∇)X
= I[∂t + (v g · ∇)]X + X[∂t I + ∇ · (Iv g )]
= I dt X + 2γXI. (5.20)
Below, we consider the choices of X that are particularly important.
First, consider X = ω. In this case, Eq. (5.20) becomes
∂t U + ∇ · (U v g ) = I dt ω + 2γωI, (5.21)
where we used the ray equation dt ω = ∂t ω and introduced
U =. ωI. (5.22)
Then, Eq. (5.21) can be written as follows:
∂t U + ∇ · (U v g ) = I∂t ω + 2γ U . (5.23)

Adiabatic waves in stationary media (γ = 0, ∂t ω = 0)

First, suppose that there is no dissipation (γ = 0), in which case the system becomes Lagrangian.
Also suppose that the medium is stationary (∂t ω = 0). Then, this equation becomes conservative [cf.
Eq. (5.3)]:


∂t U + ∇ · (U v g ) = 0, d x U = const. (5.24)
In other words, the
invariance of a Lagrangian system with respect to translations in time leads to
the conservation of d x U . This means that by definition [12], U is the density of the wave canonical
energy, at least up to a constant factor α. This also means that the energy of a linear wave propagates
at the group velocity.2
In order to determine α for electromagnetic waves, consider the following. From Eq. (5.13),
1 1
U= E † ∂ω (ω 2 H )E = [E † ∂ω (ωH )E + B †B ], (5.25)
16πω 16π
or in terms of the reals fields,
1 † 1  † 2 
U= hẼ ∂ω (ω 2 H )Ẽi = hẼ ∂ω (ωH )Ẽi + hB̃ i . (5.26)
8πω 8π
Consider a wave in vacuum, in which case  = 1. Then, U exactly coincides with the known formula
for the vacuum-wave energy density [29], and thus α = 1. Accordingly, U v g has the meaning of the
energy flux density. Due to Eq. (5.15), it can be expressed as
U v g = S + K, (5.27)
where S [Eq. (5.17)] is the Poynting vector and K [Eq. (5.18)] has the meaning of the kinetic-energy
flux density. Notably, Eq. (5.27) also leads to an alternative formula for the group velocity:
S+K
vg = . (5.28)
U
2 Thisconclusion is invalid for nonlinear waves, which have different propagation velocities for the action, energy, and
momentum [24, 28]. Some authors define the group velocity v g for nonlinear waves as the energy velocity, but this is
an arbitrary definition that can lead to confusion. In particular, such v g will not be the velocity on the characteristics.
Moreover, there are two families of characteristics in this case, which have different velocities [24, 28].

49
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 50

Adiabatic waves in nonstationary media (γ = 0, ∂t ω 6= 0)

Let us compare Eq. (5.22) with the expression for the wave energy density expected from quantum
mechanics,

U = ~ωnph , (5.29)

where nph is the photon density. It is seen from here that I = ~nph ; i.e., I is just the photon density
in units ~−1 (cf. Exercise 5.2). Accordingly, Eq. (5.21) can be understood as follows. Let us assume
for now that γ = 0. Then, the action conservation (5.1) says that the number of photons (action/~) is
conserved, and thus the wave energy can evolve only through the evolution of the energies of individual
photons, H = ~ω. This regime is called adiabatic (cf. Box 5.2). The corresponding power density is

nph dt H = (I/~) dt (~ω) = I dt ω = I∂t ω, (5.30)

which is in agreement with Eq. (5.25) at zero γ.

Exercise 5.2: One can use the Taylor expansion of Eq. (5.6) around the solution of the
dispersion relation w = ω(t, x, k) to obtain Λ(t, x, w, k) ≈ (w − ω(t, x, k))∂ω Λ. Then, using
∂ω Λ|Ψ|2 = I, one arrives at the “canonical” form of the wave Lagrangian density [24, 30]

L = −(∂t θ + ω(t, x, k))I, (5.31)

where the independent functions are θ and I. Show that
in the “point-particle limit”, when
I(t, x) = ~δ[x − X(t)],a this action becomes S[X, P ] = d t [P · Ẋ − H(t, X, P )] and leads to
Hamilton’s equations (4.31), with H = ~ω(t, X, P ) as the Hamiltonian, X(t) as the coordinate,
.
and P (t) = ∇θ[t, X(t)] as the canonical momentum [31].
a Here, “δ[x − X(t)]” denotes a profile that is narrow compared to the scale of the background medium but

still wide compared to the wavelength, so GO is still applicable.

Dissipative waves (γ 6= 0)
Now, let us allow for non-conservation of photons, i.e., nonzero γ. To the extent that U still can be
.
understood as the wave energy density,3 the function Pabs = −2γ U can be interpreted as the wave
power density absorbed irreversibly. (The sign is chosen such that Pabs > 0 when a wave is losing
action.) This is understood because dissipation is due to the loss of photons; the loss of a single
photon results in the energy loss equal to ~ω, so the loss of 2γI/~ photons per unit time per unit
volume results in the loss of energy 2γ U .
Using Eq. (4.42), one finds that

h† A h h† A h
   
1 ω ω
γU = − ωI = − ∂ω Λ |E |2 = − E † AE . (5.32)
16π ∂ω Λ 16π ∂ω Λ 16π

Hence,
ω †
Pabs = E AE . (5.33)
8π
3 Strictly speaking, the concept of energy is undefined for dissipative systems. However, if such a system transitions

from a conservative state into another conservative one, then the change of its energy is well defined. The function
U can be used as a means to calculate this change, because it represents the true energy density before and after the
transition and its governing equation (5.21) holds at all times. The physical meaning of U during the transition is
irrelevant, but one might as well call U energy density for the lack of a better definition of the energy density.

50
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 51

Also note that

 
ω † 4πiσ
Pabs = E 1+ E
8π ω A
1
= E † σ HE
2
1
= Re (E † σ HE )
2
1
= Re (E † σE )
2
= hj̃ ind · Ẽi, (5.34)

where j̃ ind and Ẽ = Re (E e i θ ) are the real current density and the real electric field (Exercise 5.1).
Therefore, Pabs represents the Joule-heating power.

5.3 Wave momentum

Let us return to Eq. (5.20) and substitute X = ka , where ka is any of the components of the wave
vector. This leads to

∂t (ka I) + ∇ · (ka Iv g ) = I dt ka + 2γka I. (5.35)

Let us use the ray equation dt k = −∂x ω and introduce

P =. kI. (5.36)

Then, Eq. (5.35) can be written as follows:

∂t P a + ∇ · (P a v g ) = −I∂a ω + 2γ P a . (5.37)

Suppose there is no dissipation (γ = 0), in which case the system becomes Lagrangian. Also suppose
that the medium is homogeneous (∂a ω = 0). Then, this equation becomes conservative:


∂t P a + ∇ · (P a v g ) = 0, d x P a = const. (5.38)

In other words, the
invariance of a Lagrangian system with respect to translations in time leads to
the conservation of d x P . This means that by definition [12], P is the density of the wave canonical
energy, at least up to a constant factor. This also means that the energy of a linear wave propagates
at the group velocity. One can also extend this discussion by analogy with Sec. 5.2.
Finally, notice that together, Eqs. (5.36) with Eq. (5.22) lead to the following fundamental relation
between the wave momentum density and the wave energy density:
k
P= U. (5.39)
ω
Equation (5.39) is similar to the relation between photon momentum p = ~k and the photon energy
H = ~ω. Because of this, some authors derive Eq. (5.39) from the quantum analogy. However, as
shown above, Eq. (5.39) can be derived using purely classical arguments as well.

5.4 Example: α channeling

Equation (5.39) can be useful for understanding basic physics of wave absorption. For example,
consider the interaction between a wave and a particle in a static magnetic field B 0 = ∇ × A. We will

51
LECTURE 5. WAVE ACTION, ENERGY, AND MOMENTUM 52

k k
A

B
z
r k
Figure 5.1: Schematic of the geometry assumed in Sec. 5.4.

assume the planar geometry but adopt the notation for the coordinates as in the cylindrical geometry
such as that in a tokamak with a high aspect ratio. Assume that the vector potential A has the form
A = eθ Ψ(r), so B 0 = ez Ψ0 (r) (Fig. 5.1). A radial static electric field is also allowed but will not be
important. Suppose that a wave has deposited energy ∆U through the interaction with a particle with
charge q. Since the overall system is symmetric in θ, the total canonical momentum in the θ direction
must be conserved. Hence, the particle must change its canonical momentum in the θ direction by
k⊥ ∆U/ω. This leads to
h q i k
⊥
∆ pθ + Ψ(r) = ∆U, (5.40)
c ω
where p is the particle kinetic momentum. After the interaction, the left-hand side does not change,
so it is equal to its own time average. But hpθ i = 0 (the average perpendicular velocity of a particle
that experiences stationary Larmor rotation in a dc magnetic field is zero), so one obtains

k⊥ qΨ0 (r) qB0

∆U ≈ h∆ri = ∆rgc , (5.41)
ω c c
where ∆rgc = h∆ri is the displacement of the particle guiding center. This shows that whenever a
particle extracts energy ∆U from the wave, it also gets shifted radially by
ck⊥
∆rgc = ∆U. (5.42)
qB0 ω
This equation plays an important role in tokamak plasmas. In particular, it underlies the α channeling
effect, and it is also relevant in the context of wave-induced plasma rotation.

52
Problems for Part II

PII.1 Single-wave dynamics within geometrical optics

Consider one-dimensional propagation of a stationary transverse electromagnetic wave in inhomoge-
neous nonmagnetized CCS plasma with unperturbed density ne (x) = (1 + x/Lc )n0 , where n0 and Lc
are positive constants. Assume that the wave is launched at x = 0 with a wavevector k0 > 0 and that
the GO approximation holds at all x.

(a) Write down the ray equations explicitly. Find the wave frequency ω(x), the wavevector k(x),
and the group velocity vg (x). At what x is the wave reflected?

(b) Find and plot the ray trajectory x(t). Find the time at which the envelope arrives at the
reflection point.
(c) Calculate the wave energy density. How much of it is stored in: (i) the electric field, (ii) the
magnetic field, and (iii) plasma oscillations?

(d) Write down the conservation law for the wave energy in a differential form. Using the above
results, find the electric field amplitude as a function of x for a stationary wave. Compare the
result with Eq. (2.35) obtained from the WKB approximation.
(e) Suppose now that the plasma is homogeneous but undergoes mechanical compression trans-
versely to the wavevector. What is conserved in this case? Assuming for simplicity that ∂x ne = 0,
find the wave total energy as a function of the instantaneous density ne (t).

PII.2 Coupling of resonant waves, mode conversion

Suppose a wave field ψ that is quasimonochromatic in the region of interest, i.e., can be expressed as
ψ = Ψe i θ , where Ψ and kα = ∂α θ may slowly depend on spacetime coordinates x. Unlike in Lecture 4,
.
here we consider the case when the field consists of two modes,

Ψ = h1 Ψ1 + h2 Ψ2 + O(ε) ≡ ΞΨ̄ + O(ε), (5.43)

.
where h1,2 are normalized eigenvectors of D H [x, k̄(x)], Ξ = (h1 , h2 ) is a 3 × 2 matrix that has h1
and h2 as its columns, and Ψ̄ is a two-dimensional column vector with components Ψ1 and Ψ2 .
The derivation of the approximate equation for Ψ̄ is identical to that presented in Lecture 4 up to
replacing Ψ with Ψ̄, h with Ξ, and h† with Ξ† , and one obtains

i
 
Λ − Q + i Γ − (∂α V ) − i V ∂α Ψ̄ ≈ 0,
α α
(5.44)
2

53
PROBLEMS FOR PART II 54

which you may consider as given. Here, Λ and V α are 2 × 2 real diagonal matrices given by
   
Λ1 0 ∂Λ1 /∂kα 0
Λ= , Vα≈ , (5.45)
0 Λ2 0 ∂Λ2 /∂kα

Q is a nondiagonal Hermitian matrix that determines mode coupling, and Γ is a (generally, also non-
diagonal) Hermitian matrix that determines dissipation. For simplicity, let us suppose that dissipation
is negligible (Γ ≈ 0) and that the problem is one-dimensional, i.e., all functions depend on only one
spacetime coordinate l (which can be the time, a spatial coordinate, or a linear combination thereof).
Suppose also that ∂Λ1 /∂kl and ∂Λ2 /∂kl have the same sign, say, are positive.
.
p p
(a) Consider ζ = V l Ψ̄, where V l is a (diagonal) matrix whose square equals V l . Show that ζ
satisfies the following equation, where H is a Hermitian 2 × 2 matrix:4

i ddζl = Hζ. (5.46)

(b) Because H is Hermitian, Eq. (5.46) has a conservation law. Write down this law in terms of ζ,
then express it in terms of Ψ̄. What is the physical meaning of the conserved quantity?
(c) Like in Eq. (4.11), k̄(x) must be chosen such that the right side of Eq. (5.46) be small, namely, O(ε).
We cannot choose k̄(x) to make the whole Λ zero, because Λ is now a matrix that is determined
by two independent scalar functions, Λ1 and Λ2 . But by our assumption that ψ is quasi-
monochromatic, the two modes must be approximately in resonance; thus, if we choose k̄(x),
say, such that Λ1 be small, then Λ2 will be small automatically, and vice versa. Like in Lecture 4,
the specific choice of k̄(x) is a matter of convenience; for example, one can choose k̄(x) such that
c1 Λ1 + c2 Λ2 = 0, where c1,2 can have any values of order one or less. It is conventional to adopt
tr H[x, k̄(x)] = 0; then, H can be parametrized as follows:5

−α −i β
!
H=
i β∗ α , (5.47)

where α is real and α, β = O(ε). Show that the variable transformation a1 = ζ1 e −i γ/2 and
.
a2 = ζ2 e
. i γ/2 .
, where γ = arg β, leads to the following equation for a1,2 :

i ddτ aa1 = −u(τ ) −i

    
a1
2 i u(τ ) a2
, (5.48)

.

where τ = d l |β| and u = α/|β|− γ̇/2, and the dot denotes d /d τ . Equation (5.48) is the canon-
.
ical form of the envelope equation describing one-dimensional coupling of two (nondissipative)
resonant modes. Examples of such “mode conversion” in plasma will be discussed in Part III.
(d) For constant u, search for the eigenmodes of Eq. (5.48) in the form a1,2 = A1,2 e −i Ωτ . The
values of A1 and A2 characterize how close these eigenmodes are to Mode I and Mode II of the
corresponding homogeneous medium (Q = 0). [For example, if A = (1, 0)| , then the value of a2
in such eigenmode is zero, i.e., the eigenmode is purely Mode I.] Plot Ω and A1,2 as functions
of u for both eigenmodes and explain where each eigenmode is close to Mode I and where it is
close to Mode II.
4 Notably, Eq. (5.46) is similar to the equation governing a two-level quantum system.
5 Choosing an alternative convention leads to equations that have are equivalent to Eq. (5.48) but a different form.
This is due to the fact that redefining k̄(x) implies that the total phase of ψ is split differently between Ψ and θ,
resulting in a different definition of the envelope functions a1,2 . In other words, the alternative equations are different
because they describe the same total field ψ in different variables.

54
PROBLEMS FOR PART II 55

(e) Suppose u(τ ) = µτ , where µ is constant, and assume µ > 0 for simplicity. Here, τ = 0 is
understood as the moment when Mode I and Mode II are exactly in resonance. In quantum
mechanics, this is known as the Landau–Zener problem [32, 33]. One way to solve Eq. (5.48) in
this case is to rewrite it as the Weber equation, ä1 + (1 + µ2 τ 2 − i µ)a1 = 0, and then explore the
asymptotic behavior of its solutions, which are given by parabolic cylinder functions. But let us
adopt a more intuitive approach (which is an abridgment of the approach used in Ref. [34]).
First, show that Eq. (5.48) can be expressed as follows:

a1 . s − i ∂s . s + i ∂s . √
ξˆη̂ a1 = − , ξˆ = √ , η̂ = √ , s = τ µ. (5.49)
2µ 2 2
For clarity, let us call s a “time” variable, so i ∂s can be viewed as the s-representation of the
corresponding “frequency” operator ω̂ s . (Alternatively, one can view s as a spatial variable
and −i ∂s as the corresponding wavevector operator, or momentum √ operator.) Then, √ one can
write ξˆ and η̂ in the following invariant form: ξˆ = (ŝ − ω̂ s )/ 2 and η̂ = (ŝ + ω̂ s )/ 2. Show
ˆ η̂] = −i . This is identical to [ŝ, ω̂ s ], so one can view ξˆ and η̂ as the new time and
that [ξ,
frequency operators in the phase space rotated by 45◦ with respect to (s, ωs ).6 Accordingly, in
the ξ-representation, Eq. (5.49) has the form

i ξ ddaξ1 =−
a1
2µ
. (5.50)

(You are not asked to prove this formally but you can try to do it if you are interested; otherwise,
.
see the solutions later.) Integrate Eq. (5.50) and show that the “transmission coefficient” T =
a1 (+∞)/a1 (−∞) is given by7

T = e −π/(2µ) . (5.51)

Hint: In order to encircle the pole at ξ = 0 correctly, you will need to introduce
infinitesimal positive dissipation, i.e., replace ∓u with ∓u − i 0 [cf. the discussion
preceding Eq. (2.20)]. If you did not study complex analysis, know that for any g,
 b b
lim
ν→0+
d ξ ξg(ξ)
+ iν
= −i πg(0) + d ξ g(ξ)
ξ
, (5.52)
a a

where d ξ (...) is the Cauchy principal value of the corresponding integral. This is
known as the Sokhotski–Plemelj theorem. It will be discussed in detail in Part IV.

(f) Using Eq. (5.51) and the result of part (b), qualitatively explain the result of mode conversion
at |µ|  1 (“adiabatic” regime) and at |µ|  1 (“diabatic” regime).
.
(g) Show that µ is of the order of the largest relative rate at which the beat period Tb = 2π/|ΩI −ΩII |
evolves with τ ; i.e., show that |µ| ∼ max |d Tb /d τ |. Using this, formulate the condition under
8

which the interaction between two modes can be neglected, i.e., the single-mode model studied
in Lecture 4 can be used instead of the more complicated two-mode theory discussed above.

6 Thisis an example of the metaplectic transform mentioned in Box 2.1.

7 Remember that µ > 0 is assumed here. For general µ, Eq. (5.51) becomes T = e −π/(2|µ|) .
8 Because d T /d τ is dimensionless, this derivative can just as well be calculated in the original dimensional units.
b

55
 Part III

Waves in plasmas: fluid theory

In this part of the course, we overview basic waves in plasmas within fluid models. Our
overview is intended as introductory rather than exhaustive. For more information, see,
for example, Refs. [1, 3, 35].

56
Lecture 6

Waves in cold magnetized plasma

In this lecture, we study waves in cold magnetized plasma in the same manner as we studied waves
in cold nonmagnetized plasma in Lecture 2.

6.1 Basic equations

Let us explore dispersion properties of cold collisionless plasma with nonzero background magnetic
field B 0 . This topic has been partly discussed in Problem PI.2, but here we will formulate it in a
more conventional form, namely, in terms of the dielectric operator.
Let us assume the same plasma model as in Sec. 2.1.1, except now the background magnetic field
is nonzero and we assume coordinates such that B 0 = ēz B0 , where ēz is a unit vector along the z
axis. Then, the fluid velocity of any given species s is governed by
∂ ṽ s es
= Ẽ + ṽ s × Ωs − νs ṽ s , (6.1)
∂t ms
. .
where Ωs = ez Ωs and Ωs = es B0 /(ms c) is the gyrofrequency. This leads to the following equations
for the velocity components:
∂ṽs,x es
= Ẽx + Ωs ṽs,y − νs ṽs,x , (6.2a)
∂t ms
∂ṽs,y es
= Ẽy − Ωs ṽs,x − νs ṽs,y , (6.2b)
∂t ms
∂ṽs,z es
= Ẽz − νs ṽs,z . (6.2c)
∂t ms
The equation for ṽs,z can be approached just like in Lecture 2, so let us focus on the equations for ṽs,x
and ṽs,y . Let us multiply Eq. (6.2b) by the imaginary unit and add it to (subtract it from) Eq. (6.2a).
Assuming the notation

j̃s,± = j̃s,x ± i j̃s,y , Ẽ± = Ẽx ± i Ẽy ,

. .
(6.3)

one can write the result as follows:

2
ωps
= −(νs ± i Ωs )j̃s,± +
∂ j̃s,±
Ẽ± , (6.4)
∂t 4π
and the real current densities are found from
(i) (i)
j̃ (i)
s,x = Re j̃ s,± , j̃ (i)
s,y = ±Im j̃ s,± . (6.5)

57
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 58

In case of a homogeneous plasma, one can assume all the “tilded” quantities to be ∝ e −i ωt+i k·x and
infer the spectral conductivity as described in Lecture 1. More generally, though, we are interested
in the Weyl symbol of σ̂, which is derived as follows. Equation (6.4) has the same form as Eq. (2.5),
(f) (i)
and similarly, its general solution of Eq. (6.4) can be written as j̃s,± = j̃ s,± + j̃ s,± ,

j̃ s,± (t, x) = j̃ s,± (t0 , x) e −(νs ±i Ω)(t−t0 ) ,

(f) (f)
(6.6a)
2  t
ωps
(i)
j̃ s,± (t, x) =
4π
d t0 e (ν ±i Ω)(t −t) Ẽ± (t0 , x).
s
0
(6.6b)
t0

(Here and further, ωps , Ωs , and νs are evaluated at location x.) Then, from Eq. (6.5), one finds
2  t
ωps
d t0 + cos[Ωs (t − t0 )] Ẽx (t0 , x) + sin[Ωs (t − t0 )] Ẽy (t0 , x) e −νs (t−t ) ,
n o 0
j̃ (i)
s,x (t, x) = (6.7a)
4π t0
2  t
ωps
d t0 − sin[Ωs (t − t0 )] Ẽx (t0 , x) + cos[Ωs (t − t0 )] Ẽy (t0 , x) e −νs (t−t ) ,
n o 0
j̃ (i)
s,y (t, x) = (6.7b)
4π t0

where Ωs and νs are evaluated at location x. From here,

2
ωps
δ(x − x0 ) cos[Ωs (t − t0 )] e −νs (t−t ) ,
0
Σs,xx (t, x, t0 , x0 ) = +Σs,yy (t, x, t0 , x0 ) = (6.8a)
4π
2
ωps
δ(x − x0 ) sin[Ωs (t − t0 )] e −νs (t−t ) ,
0
Σs,xy (t, x, t0 , x0 ) = −Σs,yx (t, x, t0 , x0 ) = (6.8b)
4π
2
ω
δ(x − x0 ) e −νs (t−t ) ,
ps 0
Σs,zz (t, x, t0 , x0 ) = (6.8c)
4π
and the remaining elements of Σs are zero. Like in nonmagnetized plasma, the presence of δ(x − x0 )
signifies that the plasma response is local in space, so there is no spatial dispersion.
Remember that Σ(t, x, t0 , x0 ) ≡ 0 for t0 > t. This means that for general t0 , the Heaviside step
function H (t − t0 ) should be added as a factor on the right-hand side in all Eqs. (6.8). Then, each Σab
has a form

Σab (t, x, t0 , x0 ) = fab (t − t0 , x) H (t − t0 ) δ(x − x0 ). (6.9)

The corresponding symbol is, by definition [Eq. (3.15)], as follows:

 ∞  ∞
σab (t, x, ω, k) = dτ d s Σab (t + τ /2, x + s/2, t − τ /2, x − s/2) e i ωτ −i k·s
−∞ −∞
 ∞  ∞
= dτ d s fab (τ, x + s/2) δ(s) e i ωτ −i k·s
0 −∞
 ∞
= d τ fab (τ, x) e i ωτ . (6.10)
0

The right-hand side is independent of t, which is due to the fact that the plasma is assumed stationary;
however, it generally depends on x through ωps , Ωs , and νs . After substituting the corresponding fab
and performing the integration, one obtains

σs,xx = +σs,yy = +
2
ωps i (ω + i νs ) , (6.11a)
4π (ω + i νs )2 − Ω2s
2
ωps Ωs
σs,xy = −σs,yx =− , (6.11b)
4π (ω + i νs )2 − Ω2s

58
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 59

2
ωps
σs,zz = − , (6.11c)
4π i (ω + i νs )
where the time integrals (6.10) converge because Im ω = 0 and νs > 0. Because this result extends to
homogeneous plasma as is, we will ignore the distinction between the symbol of σ̂ from the spectral
conductivity in this lecture. For the same reason, we will not distinguish the symbol of χ̂ from the
spectral susceptibility and the symbol of ˆ from the dielectric tensor. We will also assume for simplicity
that the plasma is collisionless, i.e., νs → 0+ [cf. Eq. (2.20)].

6.2 Susceptibility and dielectric tensor

In the collisionless limit, the susceptibility χ = 4π i σ/ω corresponding to the conductivity (6.11) is
2
ωps
χs,xx = +χs,yy = − , (6.12a)
ω 2 − Ω2s

χs,xy = −χs,yx =−
i Ωs ωps 2
, (6.12b)
ω ω 2 − Ω2s
2
ωps
χs,zz =− 2. (6.12c)
ω
Using Eqs. (6.12), one readily obtains the corresponding dielectric tensor (Box 6.1):
S −iD 0
 

 =  iD S 0 , (6.13)
0 0 P
which is Hermitian.1 The notation assumed here is the traditional notation from Ref. [1]:
2
ωps
S =1− = (R + L),
1
X
(6.14a)
s
ω 2 − Ω2s 2
2
ωps
D= = (R − L),
X Ωs 1
(6.14b)
s
ω ω2 − Ω2s 2
2
X ωps
P =1− ω2
, (6.14c)
s
2
ωps
R, L = 1 −
X
. (6.14d)
s
ω(ω ± Ωs )

If ω is much larger than all Ωs and ωps , then  ≈ 1. In the somewhat more general case when
ω  Ωe yet ω ∼ ωpe , the tensor  is the same as in nonmagnetized plasma. Let us also consider the
low-frequency limit, ω  Ωi , which is known as the magnetohydrodynamic (MHD) limit. There, one
has S ≈ 1 + γA , where
2 X 4πns0 e2 m2 c2
. X ωps s s 4π X c2
γA = 2
= 2 2 = c2 2 ns0 ms = 2 . (6.16)
s
Ωs s
m s es B0 B0 s VA

Here, we have also introduced the so-called Alfvén speed,

. B0
VA = √ , (6.17)
4πρm
1 The regions ω → 0 and ω → ±Ω are exceptions. There, one should account for nonzero ν , or other dissipation,
s s
even in the limit when νs is infinitesimally small (Problem PIV.9).

59
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 60

Box 6.1: Alternative form of 

Using the Gell–Mann matrices α introduced in Problem PI.2, one can also express  in the
following invariant form that allows for an arbitrary orientation of B 0 :
" #
2 2 2
X ωps ωps (α · Ωs ) ωps (α · Ωs )2
=1+ −1 2 + − 2 2 . (6.15)
s
ω ω(ω 2 − Ω2s ) ω ω − Ω2s

. P
and the mass density ρm = s ns0 ms . Also,
2
ωps
D=−
X Ωs

s
ω Ω2s
(1 − ω 2 /Ω2s )
2 
X Ωs ωps ω2

≈− 1+ 2
s
ω Ω2s Ωs
2 2
1 X ωps X ω ωps
=− − . (6.18)
ω s Ωs s
Ωs Ω2s

Here,
2
X ωps X 4πn0,s e2 ms c
s 4πc X
= = n0,s es = 0 (6.19)
s
Ωs s
ms es B0 B0 s

due to plasma neutrality, so

2
X ω ωps
D=− Ωs Ωs2
∼
ω
Ωi
γA  S . (6.20)
s

Hence, we obtain the following low-frequency limit of , which we will use later (Exercise 6.1):
 
1 + γA 0 0
≈ 0 1 + γA 0  . (6.21)
0 0 P

Exercise 6.1: At ω → 0, the wave field becomes stationary, and one might expect that a
stationary field cannot create a current perpendicular to B 0 . Nevertheless, the above calculation
shows that xx = yy → 1 + γA , which is not unity. (In fact, γA is often large.) This means that
plasma does respond to such field. What is the physical nature of this response?

6.3 General dispersion relation

Now let us explore the general dispersion relation (4.15) and the equation for the field polarization
(4.16), which we copy here in the following form (since D E is Hermitian):

det D E (x, ω, k) = 0, D E (x, ω, k)h = 0. (6.22)

60
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 61

Without loss of generality, one can choose the coordinate axes such that k = (k⊥ , 0, kk ), where
k⊥ = k sin θ, kk = k cos θ, and θ is the angle between k and B 0 . A similar notation will be assumed
for the refractive index N . Then, the dispersion tensor (or the symbol of D̂) is given by

S − Nk2 −iD N⊥ Nk
 

DE =  iD S − N2 0 , (6.23)
N⊥ Nk 0 P − N⊥2
and thus,

det D E
= (S − N 2 cos2 θ)(S − N 2 )(P − N 2 sin2 θ) − (−iD )(iD )(P − N 2 sin2 θ) − (N 2 sin θ cos θ)2 (S − N 2 )
= (S − N 2 )(SP − S N 2 sin2 θ − P N 2 cos2 θ + N 4 sin2 θ cos2 θ − N 4 sin2 θ cos2 θ) − D 2 (P − N 2 sin2 θ)
= (S − N 2 )(SP − S N 2 sin2 θ − P N 2 cos2 θ) − D 2 (P − N 2 sin2 θ)
= S 2 P − S 2 N 2 sin2 θ − SP N 2 cos2 θ − SP N 2 + S N 4 sin2 θ + P N 4 cos2 θ − D 2 P + D 2 N 2 sin2 θ
= N 4 (S sin2 θ + P cos2 θ) + N 2 (−S 2 sin2 θ − SP cos2 θ − SP + D 2 sin2 θ) + S 2 P − D 2 P
= N 4 (S sin2 θ + P cos2 θ) − N 2 [(S 2 − D 2 ) sin2 θ + PS (1 + cos2 θ)] + P (S 2 − D 2 ).

Let us notice that

S 2 − D 2 = 41 (R + L)2 − 14 (R − L)2 = RL (6.24)

and introduce

A = S sin2 θ + P cos2 θ, B = RL sin2 θ + PS 1 + cos2 θ , C = PRL.

. .
.
(6.25)

Then, we can also rewrite our equation as follows:

AN 4 − BN 2 + C = 0. (6.26)

This biquadratic equation for N has the following solutions:

B±F
N2 = , (6.27)
2A
where
.
F 2 = B 2 − 4AC
= [RL sin2 θ + PS (1 + cos2 θ)]2 − 4(S sin2 θ + P cos2 θ)PRL
= [(RL − PS ) sin2 θ + 2PS ]2 − 4PRL(S sin2 θ + P cos2 θ)
= (RL − PS )2 sin4 θ + 4PS (RL − PS ) sin2 θ + 4P 2 S 2 − 4PRLS sin2 θ − 4P 2 RL cos2 θ
= (RL − PS )2 sin4 θ + 4P 2 S 2 cos2 θ − 4P 2 RL cos2 θ
= (RL − PS )2 sin4 θ + P 2 (R + L)2 cos2 θ − 4P 2 RL cos2 θ
= (RL − PS )2 sin4 θ + P 2 (R − L)2 cos2 θ
= (RL − PS )2 sin4 θ + 4P 2 D 2 cos2 θ.

Note that F 2 > 0 at all real frequencies, so N 2 is real too. Thus, N is either real or imaginary.
.
Consider also τ = tan2 θ and note that
1 τ
cos2 θ = , sin2 θ = . (6.28)
1+τ 1+τ

61
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 62

Then, we obtain

0 = (S τ + P )N 4 − [RLτ + PS (2 + τ )]N 2 + PRL(1 + τ )

= τ (S N 4 − RLN 2 − PS N 2 + PRL) + P N 4 − 2PS N 2 + PRL
= τ [N 2 (N 2 − P )S − RL(N 2 − P )] + P (N 4 − 2S N 2 + RL)
= τ (N 2 S − RL)(N 2 − P ) + P [N 4 − (R + L)N 2 + RL]
= τ (N 2 S − RL)(N 2 − P ) + P (N 2 − R )(N 2 − L). (6.29)

This leads to the following handy formula, which will be used below:

τ =−
P (N 2 − R )(N 2 − L) . (6.30)
(N 2 S − RL)(N 2 − P )

6.4 Eigenmodes
Solutions of the dispersion relation (6.27) can be qualitatively understood by analyzing the charac-
teristic frequencies and limits, which are as follows.

6.4.1 Cutoffs and resonances

Let us start with cutoffs, where N = 0.2 According to Eq. (6.27), cutoffs correspond to C = 0, or
PRL = 0, which requires
P=0 or R=0 or L = 0. (6.31)

The former is satisfied when ω 2 = ωp2 . The other two equations generally have multiple solutions
depending on the number of species; but there are only two solutions in case of electron–ion plasma
with single type of ions, ω = ωR ,L , and ωR  ωL .
Other notable frequencies are resonances, which correspond to N → ∞. Resonances can be found
by taking the corresponding limit in Eq. (6.30) or, equivalently, by noticing that our solution for N 2
predicts infinite refractive index at A = 0. In either case, one finds that such points are located where

S sin2 θ + P cos2 θ = 0. (6.32)

This equation coincides with the electrostatic dispersion relation discussed in Problem PI.1. How-
ever, note that the resonance condition is necessary but insufficient for a field to be electrostatic
(Exercise 6.2).

Exercise 6.2: Show that the electrostatic dispersion relation discussed in Problem PI.1 leads
to Eq. (6.32). Using the condition (2.57), explain which resonances in cold magnetized plasma
(discussed below) are electrostatic and which are not.

At θ = 0, the resonance condition can be satisfied at P = 0, or at ω 2 = ωp2 . Alternatively, as seen

from τ (N 2 → ∞) → −P /S , the resonance condition can be satisfied at S → ∞, which corresponds
to cyclotron resonances, ω = Ωs . At θ = π/2, the resonance condition becomes S = 0. Solutions of
this equation are called hybrid resonances. One of them is the upper-hybrid resonance, ω = ωuh . It
2 Cutoffs can also be defined depending on the problem. For example, consider a problem where N is conserved and
z
Nx is not, as in a tokamak with a large aspect ratio or as in Problem PIII.2. Then, Nz can be considered as a constant
parameter of the problem and cutoffs are defined as locations where Nx = 0. Examples will be discussed later.

62
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 63

ω ω

Ωe  ωp

ωp Ωe 

Ωi Ωi
θ θ
π/8 π/4 3π/8 π/2 π/8 π/4 3π/8 π/2

Figure 6.1: Resonance frequencies in electron–ion plasma with one type of ions [numerical solution of
Eq. (6.32)]. Left – underdense plasma (ωp < |Ωe |). Right – overdense plasma (ωp > |Ωe |). In both
cases, there are three resonances at θ = 0 (at frequencies ωp , Ωe , and Ωi ; all dashed), and there are
two resonances at θ = π/2 (at frequencies ωlh and ωuh ).

corresponds to oscillations at frequencies high enough for the ion response to be negligible, so ωuh can
be found from
X ωps 2 2
ωpe
0=1− ≈ 1 − . (6.33)
s
ω 2 − Ω2s ω 2 − Ω2e

This leads to the following expression for the upper-hybrid frequency:

q
ωuh = ωpe 2 + Ω2 . (6.34)
e

In electron–ion plasma with only one type of ions, the only other hybrid resonance is the lower-hybrid
resonance that corresponds to the frequency (Exercise 6.3)
!−1/2
1 1
ωlh = 2 + Ω2 + |Ω Ω | . (6.35)
ωpi i e i

At intermediate θ, the corresponding resonances are shown in Fig. 6.1. Plasmas with multiple ion
species also have ion–ion hybrid resonances.

Exercise 6.3: Consider electron–ion plasma with single type of ions with charge ei = Ze. Show
that up to terms of order Zme /mi , S can be approximated as follows:
2 2 2 2
− ωuh )(ω − ωlh )
S ≈ (ω
(ω 2 − Ω2 )(ω 2 − Ω2 )
,
e i

where ωuh is defined in Eq. (6.34) and ωlh is defined in Eq. (6.35).

6.4.2 Low-frequency limit

At ω  Ωi , Eq. (6.23) can be simplified using the low-frequency limit of  given by Eq. (6.21):
1 + γA − Nk2
  
0 N⊥ Nk hx
 0 1 + γA − N 2 0   hy  = 0. (6.36)
N⊥ Nk 0 P − N⊥2 hz

63
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 64

At small enough ω, we also have large P 3 while the refraction index remains finite, as we will find
shortly. Then, hz ≈ −(N⊥ Nk /P )hx  hx , so the first two field equations can be written as follows:
1 + γA − Nk2
  
0 hx
= 0. (6.37)
0 1 + γA − N 2 hy

Equation (6.37) indicates that there are two modes in this limit. The first one, known as the shear
Alfvén wave, is x-polarized (hy = 0) and satisfies
Nk2 = 1 + γA , (6.38)
or equivalently,
kk2 VA2
ω2 = −1 . (6.39)
1 + γA
The second mode, known as the compressional Alfvén wave, is y-polarized (hx = 0) and satisfies
N 2 = 1 + γA , (6.40)
or equivalently,
k 2 c2 k 2 c2 k 2 VA2
ω2 = = −1 = −1 . (6.41)
1 + γA γA (1 + γA ) 1 + γA
Many plasmas of practical interest, including magnetically confined fusion plasmas, have γA  1,
−1
so the term γA in Eqs. (6.39) and (6.41) is often neglected. These waves are further studied in
Problem PIII.3.

6.4.3 Parallel propagation (θ = 0)

Now, let us allow for general frequencies but focus on the case of parallel propagation, i.e., θ = 0. In
this case, Eq. (6.22) becomes
S − N 2 −iD 0
  
hx
 iD S − N 2 0   hy  = 0 (6.42)
0 0 P hz
and Eq. (6.30) has three solutions (Figs. 6.2 and 6.3). The first one is P = 0, which corresponds to
the Langmuir oscillations with E k B 0 . The other two solutions are
N 2 = R, N 2 = L, (6.43)
so they are called the R wave and the L wave, respectively. The R wave satisfies
S − R −iD 0
  
hx
 iD S − R 0   hy  = 0. (6.44)
0 0 P hz
This gives hz = 0 and
hx
=−
S − R = − R + L − 2R = −i , (6.45)
hy iD i (R − L)
so this wave is circularly polarized in the (x, y) plane.4 For the L wave, one similarly obtains hx /hy = i ,
which corresponds to the opposite circular polarization (Exercise 6.4).
3 The low-ω limit of P is determined by thermal corrections discussed in Lecture 7 and later.
4 In the regime Ωi  ω . Ωe , the R branch is known as the whistler wave and will be studied separately.

64
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 65

N2 N2

ω p < Ωe  ω p < Ωe 

θ=0 ◦
θ = 90 ◦

1 + γA 1 + γA

1 1

ω ω
0 Ωi ωlh ωp Ωe  ωuh 0 Ωi ωlh ωp Ωe  ωuh

ω ω

ωuh ωuh
Ωe  Ωe 

ωp ωp

ωlh ωlh
Ωi Ωi
k k
0 0

Figure 6.2: The dispersion curves of underdense plasma: N 2 (ω) (upper row) and ω(k) (lower row).
The left column corresponds to θ = 0◦ : red – L mode, blue – R mode, green – Langmuir oscillations.
The right column corresponds to θ = 90◦ : red – X mode, blue – O mode. The dashed diagonals
correspond to ω = ck.

65
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 66

N2 N2

ω p > Ωe  ω p > Ωe 

θ=0◦ θ = 90 ◦

1 + γA 1 + γA

1 1
ω ω
0 Ωi ωlh Ωe  ωp ωuh 0 Ωi ωlh Ωe  ωp ωuh

ω ω

ωuh ωuh

ωp ωp

Ωe  Ωe 

ωlh ωlh
Ωi Ωi
k k
0 0

Figure 6.3: The dispersion curves of overdense plasma: N 2 (ω) (upper row) and ω(k) (lower row). The
left column corresponds to θ = 0◦ : red – L mode, blue – R mode, green – Langmuir oscillations. The
right column corresponds to θ = 90◦ : red – X mode, blue – O mode. The dashed diagonals correspond
to ω = ck.

66
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 67

Exercise 6.4: Which direction do the R and L waves rotate relative to the particle rotation?
Can you answer this without doing calculations?

6.4.4 Perpendicular propagation (θ = π/2)

At perpendicular propagation, which corresponds to θ = π/2, Eq. (6.23) takes the form

S −iD
  
0 hx
 iD S − N 2 0   hy  = 0, (6.46)
0 0 P − N2 hz

and the equation for τ has two solutions (Figs. 6.2 and 6.3). The first one,

N 2 − P = 0, (6.47)

corresponds to the same dispersion as in nonmagnetized plasma,

ω 2 = ωp2 + c2 k 2 , (6.48)

with a cutoff at P = 0. The corresponding mode is called the O wave (“ordinary wave”). The fact
that the O wave is insensitive to B0 is explained by the wave polarization. One can see from

S −iD 0
  
hx
 iD S − N 2 0   hy  = 0 (6.49)
0 0 0 hz

that the corresponding E is parallel to B 0 . Such field causes oscillations of (cold) particles parallel
to B 0 , so the magnetic Lorentz force on the particles in the O wave is zero.
The other mode, called the X wave (“extraordinary wave”), corresponds to

N 2 = RL/S . (6.50)

It has cutoffs at R = 0 and L = 0. The resonances are zeros of S, i.e., hybrid resonances. The
polarization is found from

S −iD
  
0 hx
 iD (S 2 − RL)/S 0   hy  = 0. (6.51)
0 0 P − (S − RL)/S
2
hz

This shows that hz = 0 and hx /hy = iD /S , so the X-wave polarization is elliptic in the (x, y) plane. At
the electron cyclotron resonance in particular, one has S ≈ R /2 ≈ D , so hx /hy ≈ i , which corresponds
to a circular polarization. Similarly, at ion cyclotron resonances, one has a circular polarization in the
opposite direction, hx /hy = −i .

6.4.5 Propagation at a general angle

At general θ, the structure of the dispersion curves is qualitatively similar except the branches usually
do not cross but rather “repel” each other (Fig. 6.4), forming frequency gaps (Exercise 6.5). This
is a generic feature of dispersion curves (Sec. 6.4.6), which fact helps plotting the dispersion curves
qualitatively without actually calculating them. One needs to determine only the limits and the
asymptotes, then one can connect the corresponding curves by continuity such that they do not
intersect.

67
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 68

N2 N2

ω p < Ωe  ω p > Ωe 

1 + γA

1 + γA 1
ω
0 Ωi ωlh Ωe  ωp ωuh

ω
0 Ωi ωlh ωp Ωe  ωuh

ω
ω

ωuh
ωuh
Ωe  ωp

Ωe 
ωp

ωlh ωlh
Ωi Ωi
k k
0 0

Figure 6.4: The dispersion curves for various θ (colder colors correspond to smaller θ): N 2 (ω) (upper
row) and ω(k) (lower row). The left column corresponds to underdense plasma, the right column
corresponds to overdense plasma. The dashed diagonals correspond to ω = ck.

68
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 69

Exercise 6.5: Qualitatively plot the dispersion curves for plasma that has two types of ions.

∗
6.4.6 Level repulsion
In this (optional) section, we explain the cause of level repulsion as a generic effect. Consider a general
wave system governed by DΨ = 0 with Hermitian D. Suppose there are two waves that are in exact
resonance (have their dispersion curves crossed) at some ω = ω0 , k = k0 , and θ = θ0 . Let us perturb
. .
k by some small δk = k − k0 and θ by some small δθ = θ − θ0 and consider the resulting frequency
.
shifts of the eigenmodes, δω = ω − ω0 . These shifts are determined by the field equation that for
homogeneous waves can be written as follows:

(D̄ + δD)Ψ = 0. (6.52)

. .
Here, D̄ = D(ω, k, θ0 ) and δD = D(ω, k, θ) − D̄ is a small matrix induced by δθ. Let us search for
the field in the form Ψ = h1 Ψ1 + h2 Ψ2 + O(δθ), where h1,2 are the eigenvectors of D̄ that correspond
to the two resonant modes of interest. This can also be written symbolically as Ψ = ΞΨ̄ + O(δθ),
where Ξ is a (non-square) matrix that has h1,2 as its columns and Ψ̄ is a column vector with elements
Ψ1,2 (cf. Problem PII.2). Upon multiplying Eq. (6.52) by Ξ† , one obtains

Λ + Ξ† (δD) Ξ Ψ̄ = 0,
 
(6.53)

where Λ = diag {Λ1 , Λ2 } is a diagonal 2 × 2 matrix. By Taylor-expanding this matrix in δω and in

. .
δk, one obtains Λ ≈ V ω δω + V k δk, where V ω = ∂ω Λ(ω0 , k0 , θ0 ) and V k = ∂k Λ(ω0 , k0 , θ0 ) are also
. 1/2
diagonal matrices. Then, assuming the notation ζ = V ω Ψ̄, one can rewrite Eq. (6.53) as follows:


H̄ − 1 δω ζ = 0, (6.54)

where H̄ is the following matrix:

   
. v1 δk 0 ∆1 β1
H̄ = + (6.55)
0 v2 δk β2 ∆2
| {z } | {z }
−1/2 −1/2 −1/2 −1/2
−V ω V kV ω δk −V ω Ξ† (δD) ΞV ω

and the second term is small so it can be evaluated at ω = ω0 . Because V ω and V k are diagonal and
real, and because Ξ† (δD) Ξ is Hermitian, one can readily show that v1,2 and ∆1,2 are real and

β1 β2 = |β|2 sign (Vω,1 Vω,2 ), (6.56)

where β is a complex number. The corresponding frequency shifts δω are found from det(H̄−1 δω) = 0.
The latter is a quadratic equation for δω, so it has two roots that satisfy δω1 = δω2∗ :
1 1p
δω1,2 = [∆1 + ∆2 + (v1 + v2 )δk] ± β1 β2 + [∆1 − ∆2 + (v1 − v2 )δk]2 . (6.57)
2 2
If δk is sufficiently large such that β1 β2 and ∆1,2 are negligible, then Eq. (6.57) leads to

1 v1 δk,
δω1,2 ≈ (v1 + v2 )δk ± |(v1 − v2 )δk| → (6.58)
2 v2 δk,

which is just the linear approximation to the unperturbed dispersion curves that correspond to δD ≈ 0.
These asymptotics cross at δk = 0; but when δk is small, the terms β1 β2 and ∆1,2 are not negligible.
To understand what happens then, consider the following. If the signs of Vω,1 and Vω,2 are opposite,
then β1 β2 = −|β|2 < 0, so the square root in Eq. (6.57) becomes imaginary at δk close enough to

69
LECTURE 6. WAVES IN COLD MAGNETIZED PLASMA 70

−(∆1 − ∆2 )/(v1 − v2 ). This signifies an instability. If the system has no free energy, though (as is
the case in cold plasma), it cannot support instabilities in principle. This guarantees that the signs
of Vω,1 and Vω,2 in such a system are the same, so β1 β2 = |β|2 > 0. Then, δω1,2 are real and

δω1 − δω2 ≥ |β| (6.59)

at all δk. This shows that unless β = 0 (a degenerate case), δθ induces a nonvanishing “frequency
gap” between the curves δω1 (δk) and δω2 (δk). This constitutes “level repulsion”.

70
Lecture 7

Waves in warm fluid plasma

The cold-plasma model that was considered in the previous lecture misses thermal effects that can
be important. Here, we explore some of these effects semi-qualitatively within a basic fluid model. A
more accurate (kinetic) description will be presented in Part IV.

7.1 Introduction
Like in Lecture 2, let us start with the momentum equation
 
∂ ṽ s es 1 ∇Ps
+ (ṽ s · ∇) ṽ s = Ẽ + ṽ s × (B 0 + B̃) − + C s. (7.1)
∂t ms c ms ns
Because ∇Ps /ns is now retained, this equation must be complemented with equations for the density
ns and the pressure Ps . The density can be obtained from the continuity equation,
∂t ns + ∇ · (ns v s ) = 0, (7.2)
but handling the pressure is more complicated. A rigorous way to calculate Ps is to use the kinetic
approach (Part IV); but then the theory becomes more complicated and less transparent. Rigorous
fluid theories can be constructed asymptotically for some regimes (e.g., when plasma is strongly colli-
sional or not-so-warm), but they have limited applicability. Here, we adopt an alternative approach,
which is less rigorous but adequate qualitatively and has wider applicability.
First, consider a plasma that is in a global equilibrium. Then, as commonly done in thermody-
namics, one can assume the following simple model:
 γs
ns
Ps = P0s . (7.3)
n0s
Here, the constants P0s and n0s are the pressure and density of some fixed reference state and the
constant γs is called a polytropic index. This index depends on processes of interest and can be
derived from statistical physics. For example, isotropic processes correspond to γs = 1, as seen from
the equation of state for the (ideal) gas of species s, Ps = ns Ts , where Ts is the sth-species temperature.
Adiabatic dynamics corresponds to γs = (Ds + 2)/Ds [36], where Ds is the number of relevant degrees
of freedom. Which degrees of freedom are relevant can be guessed based on qualitative arguments or
by comparing with kinetic theory.
A natural generalization of Eq. (7.3) is the model in which plasma may evolve such that each fluid
element still remains in its own local equilibrium and thus conserves its own value of Ps /nγs s . This
corresponds to conservation of Ps /nγs s in the frame moving with the fluid velocity v s ; i.e.,
d
    
Ps ∂ Ps
0=
d t nγss ≡ ∂t + vs · ∇ nγss . (7.4)

71
LECTURE 7. WAVES IN WARM FLUID PLASMA 72

Below, we use Eq. (7.4) along with Eqs. (7.1) and (7.2) to study linear plasma waves.1

7.2 Nonmagnetized plasma

7.2.1 Basic equations
Let us consider nonmagnetized plasma first. We will assume the same model as in Lecture 2, except
for the following: (i) we will adopt the collisionless limit (νs → 0+) from the start; (ii) we will assume
that the plasma is homogeneous both in time and in space;2 and (iii) we will allow for nonzero Ps .
From Eq. (7.4), the latter satisfies the following linearized equation:
   
∂ Ps 1 ∂ P̃s γs Ps0 ∂ ñs 1 ∂ P̃s ∂ ñs
0≈ ≈ γs − γs +1 = γs +s − γs Ts0 . (7.5)
∂t nγs s ns0 ∂t ns0 ∂t ns0 ∂t ∂t

Then, the complete set of linearized material equations is as follows:

∂ ṽ s es ∇P̃s
= Ẽ − , (7.6a)
∂t ms ms ns0
∂ P̃s ∂ ñs
= γs ms vT2 s , (7.6b)
∂t ∂t
∂ ñs
= −n0s ∇ · ṽ s , (7.6c)
∂t
.
where vT s = T0s /ms is the unperturbed thermal speed of species s.
By differentiating Eq. (7.6a) with respect to t, one obtains

∂ 2 ṽ s es ∂ Ẽ ∇∂t P̃s es ∂ Ẽ γs vT2 s ∂∇ñs

2
= − = − . (7.7)
∂t ms ∂t ms ns0 ms ∂t ns0 ∂t
From the continuity equation, we obtain
∂∇ñs
= −n0s ∇ (∇ · ṽ s ) . (7.8)
∂t
Then, finally,

∂ 2 ṽ s es ∂ Ẽ
− γs vT2 s ∇ (∇ · ṽ s ) = . (7.9)
∂t2 ms ∂t

Correspondingly, the linear current j̃ s = es n0s ṽ s satisfies

 2 2
ωps

∂ 2 ∂ Ẽ
− γ s v T s ∇∇ j̃ s = . (7.10)
∂t2 4π ∂t

7.2.2 Dielectric tensor

It is seen from Eq. (7.10) that the spectral representations of the induced current and the electric field
are related by

ωp2
s = −i ω
−ω 2 1 + γs vT2 s kk j̃ (i)


Ẽ. (7.11)
4π
1 Keep in mind that the results presented below have limited applicability. As to be discussed in Part IV, some of
the waves in warm plasma are in fact heavily damped due to kinetic effects, particularly at large k.
2 Considering the general case requires cumbersome calculations, because warm plasma exhibits spatial dispersion.

72
LECTURE 7. WAVES IN WARM FLUID PLASMA 73

Let us consider projections of this equations on the axes perpendicular and parallel to k by applying
the corresponding projection matrices:
. kk . kk
Π⊥ = 1 − 2 , Πk = 2 . (7.12)
k k
(i) . (i) (i) . (i)
Since Π⊥ Πk = 0, one obtains that j̃ s⊥ = Π⊥ j̃ s and j̃ sk = Πk j̃ s satisfy

(i)
j̃ s⊥ =
i ω ωp2 Ẽ (i) iω ωp2
⊥, j̃ sk = Ẽ k . (7.13)
4π ω 2 4π ω 2 − γs k 2 vT2 s
Assuming that k is directed along the x axis, the corresponding susceptibility tensor is as follows:
ω2
 
− ω2 −γspk2 v2 0 0
Ts
ω2
 
χs (ω, k) = 
 0 − ωp2 0 
 (7.14)
ω2
0 0 − ωp2
and the dielectric tensor is
 
X k (ω, k) 0 0
(ω, k) = 1 + χs (ω, k) =  0 ⊥ (ω) 0 . (7.15)
s 0 0 ⊥ (ω)
The corresponding dispersion function is
 
k (ω, k) 0 0
D E (ω, k) =  0 ⊥ (ω) − N 2 0 , (7.16)
2
0 0 ⊥ (ω) − N
and the dispersion relation is
2
k (ω, k) ⊥ (ω) − N 2 = 0.

(7.17)
Just like in the case of cold nonmagnetized plasma (Lecture 2), there are two types of nonzero-
frequency waves in this case. One type corresponds to (Figs. 7.1 and 7.2)
N 2 = ⊥ (ω) = 0, i.e., ω 2 = ωp2 + k 2 c2 . (7.18)
These are the same transverse waves as in cold plasma (Sec. 2.2.4). Due to their transverse polar-
ization, the corresponding oscillation quiver velocities satisfy ∇ · ṽ s = 0; then, ∂t ñs = 0, and thus
∂t P̃s = 0, which is why these waves are independent of the plasma temperature.
The other type of waves predicted by Eq. (7.17) are longitudinal waves. Their dispersion relation is
X 
0 = k (ω, k) = 1 + χk (ω, k) s , (7.19)
s
2
  . ωps
χk (ω, k) s = − . (7.20)
ω 2 − γs k 2 vT2 s
In case of low-frequency oscillations, we expect oscillations to be isothermal; then γs = 1, so
2
ωps 1
χk,s (ω, k) ≈ ≡ 2 2 , (7.21)
k 2 vT2 s k λDs
.
where λDs = vT s /ωps is the Debye length. In contrast, high-frequency oscillations are expected to be
adiabatic and effectively one-dimensional (Ds = 1). Hence, γs = (Ds + 2)/Ds = 3, so we expect
2
ωps
χk,s (ω, k) = − . (7.22)
ω 2 − 3k 2 vT2 s

73
LECTURE 7. WAVES IN WARM FLUID PLASMA 74

N2
c2
2
3 υTi

c2
C2S

c2
3 υ2Te

ω
0 ω pi ω pe

Figure 7.1: The dispersion curves N 2 (k) for the three types of waves that can propagate in warm non-
magnetized plasma within the fluid approximation: blue – transverse electromagnetic waves, green –
electron plasma waves, red – the branch known as the ion acoustic wave (IAW) at ω  ωpi and as
the ion plasma wave (IPW) at ω & ωpi .

ω ω

kc kCS

3 kυTe ω pi
ω pe
kCS
3 kυTi

ω pi

k k
0 λ-1
De 0 λ-1
Di

Figure 7.2: Left – the same as Fig. 7.1 but in coordinates ω(k). Right – a close-up of the left figure
focusing on the IAW/IPW branch and extending to larger k.

74
LECTURE 7. WAVES IN WARM FLUID PLASMA 75

7.2.3 High-frequency oscillations: Langmuir waves

First, let us consider longitudinal waves at high (electron) frequencies, such that the ion contribution
is negligible. Then, Eq. (7.19) becomes
2
ωpe
0=1− , (7.23)
ω2 − 3k 2 vT2 e

or equivalently (Figs. 7.1 and 7.2),

ω 2 = ωpe
2
+ 3k 2 vT2 e . (7.24)

This is the dispersion relation of electron Langmuir waves, or electron plasma waves (EPW), in warm
plasma (cf. Sec. 2.2.3). Keep in mind that, by adopting γe = 3, we have restricted our model to the
high-frequency limit, ω  kvT e . Then, we can as well use
q
2 + 3k 2 v 2 ≈ ω
3k 2 vT2 e
ω= ωpe Te pe + (7.25)
2ωpe

(for ω > 0). Although small, the thermal corrections are important in that they make the group
velocity nonzero:

∂ω 3kvT2 e
vg = = . (7.26)
∂k ω
Notice also that the phase and group velocities are connected by the following relation:

v g · v p = 3vT2 e . (7.27)

7.2.4 Low-frequency oscillations: Debye shielding and ion sound

If all species are hot, then the dispersion relation becomes
X 1 1
0=1+ ≡1+ 2 2 . (7.28)
s
k 2 λ2Ds k λD

The solution to this, k = ±iλ−1

D (where the sign must be chosen such that it agrees with boundary
conditions), describes Debye shielding,

|Ẽ| ∝ |eikz | = e−ki z = e±z/λD . (7.29)

Another notable regime is when electrons are hot while ions are coldish. For simplicity, let us
consider a plasma with only one type of ions. Then, the corresponding dispersion relation is
2
ωpi 1
0=1− + 2 2 . (7.30)
ω2 − 3k 2 vT2 i k λDe

Its solution is readily found to be (Figs. 7.1 and 7.2)

2
ωpi k 2 λ2De
ω2 = 1 + 3k 2 vT2 i = ωpi
2
+ 3k 2 vT2 i . (7.31)
1+ k2 λ2De
1 + k 2 λ2De

In particular, at kλDe  1, Eq. (7.31) becomes

ω 2 ≈ CS2 k 2 , (7.32)

75
LECTURE 7. WAVES IN WARM FLUID PLASMA 76

Figure 7.3: Schematic of the electron-pressure oscillations that accompany oscillations of the cold-ion
density. Green are the ions; red are their Debye spheres, which contain electron pressure.

where CS , called the ion sound speed, is given by

. 2 2 Zi T0e + 3T0i
CS2 = ωpi λDe + 3vT2 i = . (7.33)
mi
.
(Here, Zi = |ei /ee | is the ion charge state, and we have used ni0 ei + ne0 ee = 0.) These waves are
known as ion acoustic waves (IAW), or ion sound waves. Notably, they exist even in the limit of zero
Ti0 . This is understood by the fact that each ion carries a Debye cloud of electrons. Even when there
is no pressure associated with ions per se, there is an electron pressure associated with each Debye
cloud. Oscillations of the ion density cause oscillations of this electron pressure. The latter creates a
restoring force density −∇Pe , which leads to sound-like oscillations (Fig. 7.3).
At k  λ−1De , Eq. (7.31) leads to

ω 2 ≈ ωpi
2
+ 3k 2 vT2 i . (7.34)

These are called ion plasma waves (IPW). The electron properties do not enter Eq. (7.34) explicitly
because sufficiently hot electrons do not contribute to k and are important for IPW only as a ho-
mogeneous neutralizing background. Also notably, the regime λ−1 −1
De  k  λDi corresponds to an
2 2
approximately constant frequency, ωp ≈ ωpi .

7.3 Magnetized plasma

To understand how the dispersion curves are affected by nonzero temperature in magnetized plasma,
let us start with the limit when the angle θ between the wave vector and the dc magnetic field is small.
Electron waves are not affected significantly in this limit, but at low frequencies, a new, ion-sound
branch appears. Because the magnetic Lorentz force is small at small θ, this branch will be similar to
that in Figs. 7.1 and 7.2 except where it crosses other branches, i.e., resonates with other waves. In
those regions, frequency gaps form much like it was discussed in Sec. 6.4.5. The corresponding mode
structure is shown in Fig. 7.4(a). As θ increases, the figure transforms continually into the one shown
in Fig. 7.4(b). For details see Refs. [3, 35] (Exercise 7.1).

Exercise 7.1: Explore the figures ω(k) in Ref. [35] and explain, qualitatively, how they relate
to the corresponding figures for cold plasma from Lecture 6. Plot the corresponding N (ω).

76
LECTURE 7. WAVES IN WARM FLUID PLASMA 77

Figure 7.4: From Ref. [37]. For ω(k) at even larger k, see Refs. [3, 35].

77
Problems for Part III

PIII.1 Methods of cold-plasma diagnostics

Here, you are asked to explore basic methods of CCS-plasma diagnostics that use electromagnetic
waves. Assume that: (i) the wave frequencies ω are large enough such that the ion response is negligi-
ble; (ii) the wavevectors k are parallel to the electron-density gradient, so the wave propagation can be
considered one-dimensional; and (iii) the inhomogeneity scales are much larger than the wavelength
λ, so the GO approximation can be used. Correspondingly, the phase shift that a wave gains by
propagating from location z1 to location z2 can be approximated by3
 z2
θ= dz k(z). (7.35)
z1

(a) Interferometry. — Assuming nonmagnetized plasma, calculate θ as a functional of the electron

density, ne (z). Show that, to the lowest order in ne , the phase shift acquired by the wave due
to the plasma is proportional to the space-averaged density. Can this method be applied “as is”
in magnetized plasma? If yes, what wave should one use? Estimate θ acquired by such a wave
with λ ∼ 1 µm in a typical tokamak.
(b) Reflectometry. — For the same type of waves, calculate the phase shift that a wave gains if
it starts at z = 0, reflects from a cutoff region, and returns back to z = 0. (Obviously, the
small-density approximation cannot be used here.) Express this shift θ(ω) as a functional of
z(ωp ) (this may require integration by parts), where z(ωp ) is a function inverse to ωp (z). Express
z(ωp ) as a functional of θ0 (ω) using the Abel transform, defined as follows [38]:
 x
g 7→ f : f (x) = d t (xg(t)
− t)α
, (7.36a)
0
 t 0

f 7→ g : g(t) =
sin(πα)
π
d x (t −f x)
(x)
1−α
f (0)
+ 1−α
t
. (7.36b)
0

Show that θ0 (ω) is the propagation time. Then, explain how the above result can be used to infer
n(z) from experimental measurements. What do you think are the limitations of this diagnostic?
(c) Faraday rotation. — Show that, to the lowest nonvanishing order in ω −1 , a linearly-polarized
wave propagating in plasma along a dc magnetic field experiences rotation of its polarization
angle ϑ at the rate

d ϑ = ωp2 |Ωe | . (7.37)

d z 2ω2 c
This is called the Faraday effect, which is commonly used to measure the magnetic-field strength
by measuring the rotation angle ∆ϑ experimentally.

3 Reflection points introduce additional order-one phase shifts [cf. Eq. (2.38)], but those can be ignored here.

78
PROBLEMS FOR PART III 79

PIII.2 Wave transformations in the ionosphere

Here, you are asked to study wave propagation in the Earth’s ionosphere.4 Assume that the ionospheric
plasma is collisionless, ions are immobile, and the Earth’s curvature is negligible.
(a) Assume the cold-plasma approximation and neglect the Earth’s magnetic field. At some altitude
(∼ 300 km), the electron density has the maximum value nmax = 106 cm−3 . Suppose there is
an antenna on the ground emitting radiation with frequency f = 12 MHz at some angle α with
respect to the vertical. At what α will this radiation be reflected back to the Earth?

Hint: Reflection occurs when the vertical group velocity becomes zero. How does the
horizontal wave number and the frequency evolve along the rays?

Now, consider the influence of the Earth’s magnetic field B 0 , assuming it is homogeneous and parallel
to the ground. Suppose k = (k⊥ , 0, kk ), where x is the vertical axis and the z axis is along B 0 .

(b) In magnetized electron plasma, waves can experience reflection at three different locations cor-
.
responding to three different values of X = ωp2 /ω 2 . Find these values X1,2,3 from Eq. (6.30)
.
as functions of Y = |Ωe |/ω and of Nk , assuming Nk 6= 0. Assuming also that 0 < Y < 1
and Nk2 6= 1, show that there is exactly one value of Nk2 , denoted N̄k2 (Y ), at which two of the
reflection points coincide. Calculate N̄k2 (Y ) and the corresponding value of X.
4 2
(c) Show that N⊥ satisfies aN⊥ + bN⊥ + c = 0, where a, b, and c depend on X. (The dependence
on Y and Nk is also assumed.) Find a(X) and outline how to find b(X) and c(X). Without
2
solving this equation, sketch N⊥ (X) at fixed Nk2 for Nk2 ≷ N̄k2 and Nk2 = N̄k2 . You may consider
it known (or show it yourself) that
 2 h
2 XY i
b − 4ac = Y 2 (Nk2 − 1)2 − 4(X − 1)Nk2 . (7.38)
1−Y2
2
Hint: Consider Nk = 0 first, for which case N⊥ (X) should be easy to find. (What are
the two modes in this regime?) Then, consider how the plot is modified for Nk 6= 0
2 2
by analyzing N⊥ (0), cutoffs, resonance(s), and the number of real roots for N⊥ .
(d) Now consider N⊥ in regions where it is real. Sketch N⊥ (X) corresponding to your sketches of
2
N⊥ (X) in part (c). (Remember to plot both N⊥ > 0 and N⊥ < 0.) Using these results, explain
the dependence of the field pattern on the launch angle α in Fig. 7.5. (Ignore the specific
numbers and focus on qualitative physics.)

PIII.3 MHD waves

In this problem, you are asked to derive the low-frequency modes of magnetized plasma from the
single-fluid ideal MHD model with nonzero temperature. Adopt the following model equations
∂ρm
+ ∇ · (ρm v) = 0, (7.39a)
∂t
∇B 2

∂ (B · ∇)B
ρm +v·∇ v = − − ∇P, (7.39b)
∂t 4π 8π
∂B
= ∇ × (v × B), (7.39c)
∂t
4 Similar effects are important in magnetically confined fusion plasmas, as to be discussed in Part IV.

79
PROBLEMS FOR PART III 80

Figure 7.5: The absolute value of the wave electric field |Ẽ(x)| (horizontal axis) versus the altitude x
(vertical axis, in km) for a standing wave launched from the ground (x = 0) at different angles α
(numbers on top) between k and the vertical. The figure is adapted from Ref. [39].

80
PROBLEMS FOR PART III 81

  
∂ P
+v·∇ = 0, (7.39d)
∂t ργm

where ρm is the mass density, P is pressure, and γ is a constant polytropic index. As usual, assume
that the background plasma is homogeneous and has no average velocity. Also assume homogeneous
stationary magnetic field B 0 . For simplicity, you may also adopt the coordinate system such that
B 0 = B0 ēz and k = k⊥ ēx + kk ēz . (Here, ēa is a unit vector along ath axis.)

(a) Linearize Eqs. (7.39) and derive a PDE for the fluid velocity; then use it to derive the dispersion
relation by adopting ∂t = −i ω and ∇ = i k. Show that this dispersion relation can be represented
in the form

(ω 2 − kk2 VA2 ) ω 4 − ω 2 k 2 (VA2 + VS2 ) + VS2 VA2 k 2 kk2 = 0,

 
(7.40)
.
where VS = (γP0 /ρm0 )1/2 and the index 0 denotes unperturbed quantities.

(b) Plot in polar coordinates the phase speed as a function of the angle between k and B 0 for all
branches. Which of these waves can propagate along B 0 ? Which of these waves can propagate
perpendicularly to B 0 ?
(c) Approximate ω 2 for VS  VA . What is the physical mechanism of each branch in this limit?

(d) Calculate the energy density of an Alfvén wave in a cold plasma with negligible kk in the limit
γA  1. (The answer is the same for both types of Alfvén waves.)

81
 Part IV

Waves in plasmas: kinetic theory

In this part of the course, we extend our previous models of plasma waves by including
kinetic effects.

82
Lecture 8

Introduction to kinetic theory of

plasma waves

Fluid theory used in the previous lectures is only a rough reduction of the complete (Klimontovich)
description that accounts for each individual particle in the plasma. The next best approach is the
kinetic approach, in which plasma is considered as a fluid in phase space. This lecture is intended as
an introduction into kinetic theory and its applications to modeling plasma waves.

8.1 Introduction
8.1.1 Distribution function
Suppose that the motion of a single particle is fully described by some set of phase-space variables Γ,
for example, (x, v) or (x, p). Then, one can define the particle phase-space density F (Γ) in space Γ,
independently for each given species (Box 8.1). Let us consider two different sets of such variables,
assuming that they are connected by some invertible function Q,

Q : Γ1 7→ Γ2 . (8.1)

First, suppose the phase-space density corresponding to a single particle with some coordinate Γ20 .
Such phase-space density is given by δ(Γ2 − Γ20 ). In terms of the particle coordinate Γ10 in the Γ1
space, the same function can be expressed as δ[Γ2 − Q(Γ10 )]. In the case of multiple particles, the
corresponding phase-space density in Γ2 must be averaged over all Γ10 , or in other words, integrated
over their phase-space density F 1 . This gives the following general rule for mapping the phase-space
density F 1 in Γ1 to the phase-space density F 2 in Γ2 :

F 2 (Γ2 ) = d Γ1 δ[Γ2 − Q(Γ1 )]F 1 (Γ1 ) = F 1 [Q−1 (Γ2 )] ∂Γ1
∂Γ2
, (8.2)

where the latter ratio denotes the Jacobian of the corresponding variable transformation and Q−1 is
the function inverse to Q.
Among all possible variable transformations, there are so-called canonical transformations, which
are special. They correspond to unit Jacobians, so the canonical phase-space densities (which we
denote with F as opposed to the general phase-space densities F ) are transformed simply as

F2 (Γ2 ) = F1 (Q−1 (Γ2 )). (8.3)

In this sense, the canonical phase-space density F is an invariant with respect to canonical transfor-
mations.

83
LECTURE 8. INTRODUCTION TO KINETIC THEORY OF PLASMA WAVES 84

Box 8.1: Kinetic modeling of quantum plasmas and general broadband waves

Here, we discuss only classical kinetic theory. For quantum particles, phase-space coordinates are
operators and there is no such thing as the phase-space density, so the theory has to be formulated
differently. For a system described by a state function |ψi, one starts by introducing the density
.
operator %̂ = |ψi hψ| and its Weyl symbol %, which is known as the Wigner function (or Wigner
tensor, if hx|ψi is a vector). The Wigner function satisfies the Moyal equation ∂t % = {{H, %}},
where H is the Weyl symbol of the particle Hamiltonian and {{... , ... }} are the so-called Moyal
brackets [40]. This equation is a generalization of the kinetic equation that is derived below.
(The classical distribution function is the coarse-grained limit of the Wigner function % up to a
constant factor.) The same formalism is applicable to classical waves [41, 42].

8.1.2 Liouville’s theorem

It is well known that the system evolution in time,
Qt : Γ0 7→ Γt , (8.4)
can be considered as a canonical transformation. (For example, see Ref. [12, Sec. 45]. Here, t is time
and Qt can be understood as the evolution operator.) This gives an equation describing the evolution
of the canonical phase-space density of any given species s:
Fs [Qt (Γ0 )] = F0s (Γ0 ). (8.5)
More explicitly, this can also be written as
Fs [t, x(t, x0 , p0 ), p(t, x0 , p0 )] = F0s (x0 , p0 ), (8.6)
where (x, p) are some canonical coordinates, (x0 , p0 ) are their initial values. By differentiating this
equality with respect to time, one obtains

0=
d F [t, x(t, x , p ), p(t, x , p )] = ∂Fs + d x · ∂Fs + d p · ∂Fs , (8.7)
dt s 0 0 0 0
∂t d t ∂x d t ∂p
which is known as Liouville’s theorem (Box 8.2).
In nonrelativistic plasma physics, it is customary to work with the particle phase-space density
f (t, x, v) in the (x, v) space instead of the canonical phase-space density F(t, x, p). The relation
between f and F is obtained using Eq. (8.2):
∂(x, p)
fs (t, x, v) = Fs [t, x, p(t, x, v)] . (8.8)
∂(x, v)
For nonrelativistic plasmas, which we consider below, one can take
es
p(t, x, v) = ms v − A(t, x), (8.9)
c
where A is the electromagnetic vector potential. The corresponding Jacobian is
∂(x, p)
= m3s , (8.10)
∂(x, v)
so fs and Fs differ only by a constant factor. Therefore, they satisfy the same equation,
∂fs F s ∂fs
+ v · ∇fs + · = 0, (8.11)
∂t ms ∂v
where F s = ms v̇ is the force.

84
LECTURE 8. INTRODUCTION TO KINETIC THEORY OF PLASMA WAVES 85

Box 8.2: Intuitive interpretation of Liouville’s theorem

Liouville’s theorem can also be interpreted as follows. Consider a small fluid element in phase
space. The shape of this element may become distorted with time but its volume d Γ is conserved,
because the transformation (8.4) is canonical. The number of particles inside this volume is
conserved too, by definition, so Fs d Γ = const. Then, Fs = const along the trajectory followed
by each given phase-space element.

8.2 Vlasov equation

8.2.1 Macroscopic fields and collision operator
In general, both fs and F s can be split into macroscopic parts, which we denote with the index “m”,
and microscopic fluctuations, which we denote with the prefix δ. Then, after averaging over fluctua-
tions, one obtains
 
∂fs,m F s,m ∂fs,m δF s ∂(δfs )
+ v · ∇fs,m + · =− · ≡ Cs , (8.12)
∂t ms ∂v ms ∂v
where Cs is called the collision operator. If Cs is large compared to the term (F s,m /ms ) · ∂v fs,m that
describes long-range interactions, then Eq. (8.12) is reduced to
∂fs,m
+ v · ∇fs,m ≈ Cs , (8.13)
∂t
which is called the Boltzmann equation. This equation adequately describes neutral gases, where
particles interact mainly in pairs, via short-range forces. In contrast, plasmas are, by definition,
ensembles where each particle interacts with many neighbors simultaneously (because plasmas as
ensembles where the number of particles in the Debye sphere is large). In this case, δF caused by
particle discreteness averages out almost to zero, so Cs is small. Below, we will mostly consider the
limit of ideal plasma, where the effect of the fluctuations is completely negligible. Then, one obtains
∂fs,m F s,m ∂fs,m
+ v · ∇fs,m + · = 0, (8.14)
∂t ms ∂v
or with F s,m explicitly written as the Lorentz force,
 
∂fs,m es 1 ∂fs,m
+ v · ∇fs,m + E s,m + v × B s,m · = 0. (8.15)
∂t ms c ∂v
This is known as the (collisionless) Vlasov equation. Below, we adopt Eq. (8.15) as our primary model
of plasma dynamics. The index m will be omitted for brevity.

8.2.2 Linearized Vlasov equation

As earlier, we will assume weak oscillating fields, so fs = n0s f0s + f˜s , where f˜s is linear in the wave field
and f0s is the unperturbed distribution. The coefficient n0s is introduced so that f0s be normalized
to unity,
 ∞
d v f0s (v) = 1. (8.16)
−∞

Then, f˜s satisfies

∂ f˜s ∂ f˜s
 
≡ Rs ,
es n0s 1 ∂f0s
+ v · ∇f˜s + v × Ωs · ≈− Ẽ + v × B̃ · (8.17)
∂t ∂v ms c ∂v

85
LECTURE 8. INTRODUCTION TO KINETIC THEORY OF PLASMA WAVES 86

.
where terms quadratic in the wave field have been dropped and Ωs = es B 0 /(ms c), as usual. (Unlike in
the fluid description that we studied earlier, the velocity is an independent variable here; thus, v · ∇f˜s
is a linear term and thus must be retained.) The first-order magnetic field that enters Eq. (8.17) can
be expressed through Ẽ using Faraday’s law:

B̃ = B̃ (f) + (ck̂/ω̂) × Ẽ, (8.18)

where B̃ (f) is independent of time. Hence,

  
Rs = R (f)
s −
es n0s ∂f0s
m ∂v
· Ẽ + v ×
k̂
ω̂
× Ẽ
s
 
(v · k̂)
= R (f)
e n
s 0s ∂f0s 1
s − Ẽ + k̂(v · Ẽ) − Ẽ
ms ∂vc ω̂ ω̂ c
   
·
= Rs −
(f) e n
s 0s ∂f0s v k̂ v k̂
b c
δcb 1 − + Ẽb , (8.19)
ms ∂vc ω̂ ω̂

where R (f) (f)

s is due to B̃ , so it is independent of Ẽ. This means that one can express f˜s as

f˜s = f˜(f) ˜(i)

s + fs , f˜(i) †
s = α̂s Ẽ, (8.20)
(f)
where f˜s in independent of Ẽ and α̂†s is a linear operator. (Because α̂†s Ẽ is a scalar, the coefficient
(i)
between f˜s and Ẽ is a covector, so it is introduced with † .) This yields the induced current density

j̃ (i)
s (t, x) = es d v vf˜(i)
s (t, x, v), (8.21)

whence the conductivity operator can be inferred:


σ̂ s = es d v v α̂†s . (8.22)

For simplicity, we will limit our consideration to homogeneous plasmas and to waves of the form
∝ e i k·x . Then, k̂ can be replaced with k. However, the time dependence has to be handled in a more
subtle manner, as will be discussed in Lecture 9.

8.3 Phase mixing

Before we proceed to the general kinetic description of plasma waves, let us discuss some paradigmatic
effects within the simple model of electrostatic interactions in nonmagnetized plasma. Because the
plasma is assumed homogeneous, the linearized Vlasov equation (8.17) can be simplified by application
of the spatial Fourier transform as follows:

∂ f˜s
+ i kvx f˜s = −
es n0s ∂f0s
Ẽ , (8.23)
∂t ms ∂vx

where f˜s and Ẽ are now independent of x but depend on k, and the x axis is chosen parallel to k.
This equation is similar, for example, to Eqs. (2.5) and (6.4) and can be solved in the same way:

s (v)e
−i kvx t
f˜s = f˜(f) + f˜(i)
s (t, v), (8.24)
 t
f˜(i)
s =−
es n0s ∂f0s (v)
ms ∂vx
d t0 e −i kv 0
x (t−t )
Ẽ(t0 ), (8.25)
0

(f)
where f˜s is determined by initial conditions but not by the field.

86
LECTURE 8. INTRODUCTION TO KINETIC THEORY OF PLASMA WAVES 87

For a delta-shaped field, Ẽ(t) = δ(t − 0), the induced perturbation is

f˜(i)
s (t, v) = −
ms ∂vx
e
es n0s ∂f0s (v) −i kvx t
. (8.26)

(This is understood as the Green’s function of the linearized Vlasov equation.) The corresponding

∞
induced current es −∞ d v vx f˜s (t, v) is, by definition, the conductivity in the (t, k) representation
(i)

[which is the Fourier image of the function Σ̄s (t, x) that we used earlier]:
 ∞
e2s n0s
Σ̄s,k (t) = −
ms
d v vx ∂f∂v e
0s (v) −i kv xt
. (8.27)
−∞ x

We now change the notation as follows:

 ∞
d vy d vz f0s (v) → f0s (vx ), (8.28)
−∞

∞
so f0s (v) is understood as the distribution of the velocity components parallel to k, and

∞ −∞
d v f0s (v) = 1
[as opposed to −∞ d v f0s (v) = 1 that we used previously]. Then,
2  ∞
ωps
Σ̄s,k (t) = −
4π
d v vf0s0 (v) e −i kvt . (8.29)
−∞

It is also instructive to express this function as follows:

2    ∞
ωps
Σ̄s,k (t) = −
4π −i t
1
∂k d v f0s0 (v) e −i kvt
−∞
2    ∞
ωps
=
4π −i t
1
∂k d v f0s (v)∂v (e −i kvt )
−∞
2   ∞
ωps

=
4π
∂k k d v f0s (v) e −i kvt
(8.30)
−∞
2
ωps
≡ ∂k [kF0s (−kt)] ,
4π
where F0s is the Fourier image of f0s . In the case when the background plasma is cold and stationary,
i.e., f0s (v) = δ(v), one has F0s = 1, so
2
ωps
Σ̄s,k (t) = , (8.31)
4π
in agreement with our earlier result (2.9). But the cold-plasma approximation is only a crude model.
In reality, f0s (v) is always a smooth function with some finite width vT s . Then, F0s has a characteristic
scale vT−1s , meaning F0s is small when its argument is much larger than vT−1s . (In fact, the Fourier
image of a smooth function decreases exponentially when the argument tends to infinity.) Therefore,
|Σ̄s,k (t)| → 0 on the time scale (kvT s )−1 .1 This effect is known as phase mixing, because it is caused
by destructive interference of the contributions to Σ̄s,k produced by f˜s at different velocities.
Note also that the decay of the plasma cumulative response, which is described by Σ̄s,k (t), does
not mean that oscillations of f˜s are attenuated. Once perturbed, the distribution function continues
to oscillate [Eq. (8.26)] until collisions come into play. In other words, the microscopic state of the
plasma can store information even after macroscopic currents have dissipated.2 One interesting effect
that results from this is the so-called plasma-wave echo [43].
1 In Sec. 11.1, we will explicitly derive Σ̄s,k for Maxwellian species.
2 For a mechanical analogy, see http://www.youtube.com/watch?v=8V6kc0PQa14.

87
Lecture 9

Eigenmodes in kinetic theory

Here, we discuss the concepts of eigenmodes in kinetic theory and dispersion relation in kinetic theory,
which are more subtle than those in fluid theory.

9.1 Case–van Kampen modes

Phase mixing can be interpreted as destructive interference of microscopic eigenmodes of the plasma.
Let us discuss these modes in detail. For simplicity, we will limit our consideration to the special
case when ions are immobile and the electron distribution f0e is isotropic, so it can be expressed as
.
a function of E = me v 2 /2Te ; i.e., f0e (v) = f¯(E). (The constant Te has units of energy and can be
considered as the electron temperature, but in general, f0e may be non-Maxwellian.) For simplicity,
we will also assume that f¯ is monotonically decreasing, so
me v ¯0 me v ¯0
∂v f0e = f (E) = − |f (E)|. (9.1)
Te Te
Then, Eq. (8.23) can be written as follows:

i ∂t f˜e = kvx f˜e + i eeTn0e |f¯0 (E)| vx Ẽ. (9.2)

s

Let us complement this equation with Ampere’s law. For electrostatic oscillations, it can be written
as ∂t Ẽ = −4π j̃x , or equivalently,

∂t Ẽ = −4πee d vx vx f˜e . (9.3)

By using a variable transformation f˜e → i αf˜e with an appropriate real constant α, one can bring this
set of equations to the form


i ∂t f˜e = kvx f˜e + vx Ẽ, i ∂t Ẽ = d vx vx f˜e . (9.4)

Let us discretize the velocity space into N chunks of the size ∆v centered√around vx = va ,
.
a = 1, 2, . . . , N . Let us also replace f˜e → (∆v)−1/2 f˜, and introduce ga = va ∆v. This brings
Eq. (9.4) to the Schrödinger form

i ∂t ψ = Hψ, (9.5)

88
LECTURE 9. EIGENMODES IN KINETIC THEORY 89

Box 9.1: Graph representation of kinetic Hamiltonians

Hermitian Hamiltonians are often represented as graphs. Graph’s nodes correspond to the com-
ponents of the state vector ψ and edges denote nonzero elements of the Hamiltonian matrix. In
particular, each diagonal element connects the corresponding node with itself, so is it represented
as a loop. The Hamiltonian (9.6) corresponds to a graph that is a “star with loops”: the node
that represents Ẽ is connected with each node representing f˜a , but no f˜a and f˜b are connected
directly with each other unless a = b. Such topology of Hamiltonian graphs is a signature feature
of collisionless-plasma models and mean-field theories in general.

where ψ = (f˜1 , f˜2 , . . . , f˜N , Ẽ)| is a column vector and (Box 9.1)
 
kv1 0 0 ... 0 g1
 0 kv2 0 . . . 0 g2 
 
 0 0 kv 3 . . . 0 g3 
H=  ... ... ... ... ...
. (9.6)
 ... 

 0 0 0 . . . kvN gN 
g1 g2 g3 . . . gN 0

Let us search for eigenmodes of this system in the form ψ ∝ e −i ωt . The corresponding frequencies ω
satisfy the following equation (Exercise 9.1):
N
X ga2
0 = D(ω) = ω − . (9.7)
a=1
ω − kva

Exercise 9.1: Derive Eq. (9.7) by proving the following equality for general ϑa and ga :
 
ϑ1 0 0 ... 0 g1
 0 ϑ2 0 ... 0 g2  ! N
N
|ga |2 Y
 
 0 0 ϑ3 ... 0 g3  X
det  = ϑ− ϑb .
 ... ... ... ... ... ...  ϑa
  a=1 b=1
 0 0 0 . . . ϑN gN 
g1∗ g2∗ g3∗ ∗
. . . gN ϑ

As an algebraic equation of order N + 1, Eq. (9.7) has N + 1 solutions for ω. All these solutions are
real because H is Hermitian (Exercise 9.2). The fact that Eq. (9.7) always has N + 1 real solutions
can also be seen graphically as shown in Fig. 9.1. The corresponding eigenvectors satisfy
ga
f˜a = Ẽ. (9.8)
ω − kva

Exercise 9.2: Show that if f¯(E) is nonmonotonic, then H is not Hermitian and that the shape
of D(ω) changes qualitatively such that complex roots in general become possible.

In the limit N → ∞, these modes are known as Case–van Kampen modes [44, 45].1 The mode
spectrum becomes continuous in this case, so any ω is an eigenfrequency. Strictly speaking, the concept
1 Here, we consider only a special case of Case–van Kampen modes that corresponds to monotonic f¯(E). One can

also generalize these modes by adding ion motion and background magnetic field.

89
LECTURE 9. EIGENMODES IN KINETIC THEORY 90

Figure 9.1: A graphical solution of Eq. (9.7) for N = 4. The solutions (black points) correspond to
the crossings of D of the horizontal axis. The vertical dashed lines correspond to ω = kva , the oblique
line represents the asymptotic D(ω) = ω.

of a dispersion relation becomes irrelevant then. In fact, for any bounded absolutely integrable function
u, one can find initial conditions such that Ẽ(t) = u(t) [46]. The only subtlety is that such initial
conditions are typically not analytic [because the eigenvectors (9.8) are not analytic] and thus are not
realized naturally. To study what happens for “natural” initial conditions, one should consider the
general initial-value problem. This will be discussed in Lecture 9.

9.2 Initial-value problem

(f)
As discussed in Lecture 8, the general solution of the Vlasov equation for each f˜s consists of f˜s ,
(i)
which is due to the free thermal motion, and the induced perturbation f˜s , which is a functional of
(f)
Ẽ. Correspondingly, the current density is j̃ = j̃ (f) + j̃ (i) , where j̃ (f) is determined by all f˜s and j̃ (i)
(i)
is determined by all f˜s . We have been ignoring j̃ (f) until now, but in general, this current density
is not negligible and determines the source term S in the wave equation [see Eq. (1.21)], which we
assume here in the form
c2
D̂ E Ẽ = S , D̂ E = (k̂k̂† − 1 k̂ 2 ) + ˆ. (9.9)
ω̂ 2

9.2.1 Equation for the field spectrum

Equation (9.9) is generally an integral equation of the form
 ∞  ∞
d t0 d x0 DE (t − t0 , x − x0 )Ẽ(t0 , x0 ) = S , (9.10)
t0 −∞

where the system is assumed homogeneous. It can be converted into a simpler, algebraic, equation
by applying the Fourier transform in x and the Laplace transform in time. Originally, the Laplace
transform of a given function F is defined as
 ∞
LF :
.
F (t) 7→ F̄ (s) = d t e −st F (t), (9.11)
0

90
LECTURE 9. EIGENMODES IN KINETIC THEORY 91

Box 9.2: Convolution theorem for the Laplace transform

The convolution theorem for the Laplace transform is valid only if the corresponding functions
are sufficiently well behaved. The dispersion operator D̂ E in the form (9.9) integrates the field,
so it is well behaved. In contrast, differential operators have singular kernels and require special
treatment. In particular,

LF 0 = −i ω F̄ (ω) − F (t0 ), LF 00 = −ω 2 F̄ (ω) + i ωF (t0 ) − F 0 (t0 )

(and so on for higher-order derivatives), as can be seen via taking the integral in Eq. (9.13) by
parts. This means that if D̂ is a polynomial of ∂t , then D̂ Ẽ = S becomes D̄(ω, k)Ē(ω, k) =
S̄ + ∆S̄ , where D̄(ω, k) and ∆S̄ are polynomials of ω and ∆S̄ is determined by F (t0 ), F 0 (t0 ), ...
In cold-plasma problems, it is convenient to work with the field equation in the form

4π i ω̂ (f) ω̂ 2
D̂ Ẽ = − j̃ , D̂ = D̂ E ,
c2 c2
(f)
because f˜s are time-independent in the absence of thermal motion and thus ω̂ j̃ (f) = 0. In
this case, the spectral representation of the field equation is simply D̄(ω, k)Ē(ω, k) = ∆S̄ . But
because this equation has the same form as Eq. (9.15) up to the definition of the dispersion
operator and the source term, the discussion of the initial-value problem in the main text applies
to such problems just as well.

where Re s must be large enough for the integral to converge. The inverse transform is
 a+i b
L−1 F̄ : F̄ (s) 7→ F (t) =
1
lim
2π i b→∞
d s e st F̄ (s), (9.12)
a−i b

where a is a real number that can be chosen arbitrarily as long as it is large enough such that all
singularities of F̄ remain to the left from the contour B. Here, we assume an equivalent but a slightly
modified definition:
 ∞
F̄ (ω) = d t e i ωt F (t), (9.13)
t0
 +∞+i a
F (t) =
1
2π
d ω e −i ωt F̄ (ω), (9.14)
−∞+i a

where the former integral is now taken from t0 instead of zero and a variable transformation s = −i ω
has been applied. The latter makes the Laplace transform look similar to the Fourier transform, except
for the following: (i) the integral in Eq. (9.13) is taken over half of the time axis; (ii) ω is generally
complex and must have a large enough imaginary part for the integral in Eq. (9.13) to converge; and
(iii) the integral in Eq. (9.14) is taken not necessarily along the real axis but parallel to it, specifically,
at a distance a such that all singularities of the integrand remain below the contour.
The Laplace transform in time and the Fourier transform in space convert the convolution integral
in Eq. (9.10) into a product (Box 9.2), so one obtains

E (ω, k)S (ω, k),

Ẽ(ω, k) = D −1 (9.15)
where the right-hand side is determined by the initial conditions. From here, the field in the (t, k)
space is obtained by applying the inverse Laplace transform:
 +∞+i a
Ẽ k (t) =
1
2π
d ω e −i ωt D−1
E (ω, k)S (ω, k). (9.16)
−∞+i a

91
LECTURE 9. EIGENMODES IN KINETIC THEORY 92

9.2.2 Dispersion relation and quasimodes

Let us analytically continue D E into the whole complex-ω plane (how to do this explicitly in practical
calculations will be discussed in Lecture 10) and consider bending the integration contour. Specifically,
instead of integrating along a straight line parallel to the real axis at large enough Im ω, let us shift
the contour downward. The integral will remain unaffected if the contour sticks to singular points of
the integrand. Now, let us adopt several assumptions.

Assumption 1: Let us assume that S , which is determined by the initial conditions,

is an entire function of ω (i.e., a function that is analytic everywhere possibly excluding
ω = ∞).

Then, the contour sticks only to singularities of D −1E . By definition of an inverse matrix, it follows
that such singularities are possible, in particular, when

det D E (ω, k) = 0. (9.17)

If det D E is analytic near such points, it can be Taylor-expanded. Then, the corresponding singular-
ities of D −1
E are poles. We denote them as ωq (k).

Assumption 2: Let us assume that D −1

E has no singularities other than poles.

Note now that, as we are shifting the contour further toward Im ω → −∞, the exponent under the
integral approaches zero (at t > 0), because

e −i ωt ∼ e (Im ω)t = e −|Im ω|t . (9.18)

Assumption 3: Let us assume that D −1 E S does not grow too rapidly (if at all) at
Im ω → −∞, so the smallness of the aforementioned exponent is enough to ensure that
the integral over the horizontal part of the contour vanishes.

Under these assumptions, the whole integral can be expressed as the sum over the contributions
of the aforementioned poles only, i.e.,2

Rq (k)e −i ωq (k)t ,
X
Ẽ k (t) = (9.19)
q

where the coefficients Rq are determined by the initial conditions, or more specifically, proportional
E (ω, k)S (ω, k) at ω = ωq (k). In principle, one can choose initial
to the corresponding residues of D −1
conditions such that all but one of these coefficients be zero; then, the resulting field is

Ẽ k (t) = Rq (k)e −i ωq (k)t . (9.20)

Because this solution has a well-defined complex frequency, ωq (k), it can be considered as an eigen-
mode. Also, the mode polarization Rq can in general be found as follows. By applying the Laplace
transform to Eq. (9.20), we obtain

Ẽ(ω, k) =
i Rq (k) . (9.21)
ω − ωq (k)

By substituting this into Eq. (9.15), we also obtain

D E (ω, k)Rq (k) = i [ωq (k) − ω]S (ω, k). (9.22)

assume that all poles of D −1 E are simple. For nth-order poles, additional terms like ∝ t e
2 Here, m −i ωq (k)t appear

in Eq. (9.19), with m = 1, 2, . . . (n − 1). This case will not be considered because it is not typical. In particular, the
presence of high-order zeros of det D E does not necessarily imply the presence of high-order poles in D −1 E .

92
LECTURE 9. EIGENMODES IN KINETIC THEORY 93

At ω = ωq (k), this gives

D E [ωq (k), k]Rq (k) = 0. (9.23)

.
This shows that the polarization vectors of complex-frequency eigenmodes, eq (k) = Rq (k)/Rq (k) can
be found as the zero eigenvectors of D E (ωq (k), k), much like in the case of real-frequency waves.
Finally, notice the following. If a plasma is described by (a finite number of) fluid equations,
then S is a polynomial in ω and D E is a rational matrix function of ω; i.e., each of its elements
is representable as p(ω)/q(ω), where p and q are polynomials (cf. Box 9.2). In this case, all the
assumptions in the above derivation are satisfied automatically. This means that the eigenmodes as
defined above are the same as those one would find by simply replacing ∇ with i k and ∂t with −i ω.
Such eigenmodes are the true modes of the system. If D E is not a rational function of ω, though,
the eigenmodes as defined here are not actually the true modes [unlike the Case–van Kampen modes
(Sec. 9.1)]; for example, they might not form a complete basis. However, such “quasimodes” are
relevant in that they correspond to smooth initial conditions and thus are typically representative of
plasma oscillations. Because of that, we will refer to det D E (ω, k) = 0 as “the” dispersion relation,
and we will refer to quasimodes as “the” plasma modes. However, “non-modal” effects are also, in
principle, possible.

93
Lecture 10

Dispersion properties of
nonmagnetized plasma

Here, we apply the results of the previous lectures to study the dispersion properties of nonmagnetized
plasma.

10.1 Dielectric properties

10.1.1 Susceptibility at Im ω > 0
A simple way to calculate the conductivity at Im ω > 0 is to calculate the plasma response to a
monochromatic field as discussed in Sec. 1.2.1. Then, one readily obtains

i
   
˜(i) es n0s ∂f0s v·k vb kc
fs = − δcb 1 − + Ẽb . (10.1)
ms (ω − k · v) ∂vc ω ω

The corresponding current density (8.21) can be written as follows:

2 
i ωps ∞  
k·v
 
j̃ (i)
s =−
4π
d v ω −vk · v δcb 1 −
ω
+
kc vb ∂f0s
ω ∂vc
Ẽb . (10.2)
−∞

Hence, the spectral conductivity at Im ω > 0 is

2 
i ωps ∞  
k·v
 
(σ s )ab (ω, k) = −
4π
d v ω −vka · v δcb 1 −
ω
+
kc vb ∂f0s
ω ∂vc
, (10.3)
−∞

and the corresponding susceptibility can be expressed as

2  ∞
ωps
   
k·v
(χs )ab (ω, k) =
ω −∞
d v
va
ω−k·v
δcb 1 −
ω
+
kc vb ∂f0s
ω ∂vc
. (10.4)

10.1.2 Susceptibility at any Im ω: Landau’s rule

In order to analytically continue these results to Im ω ≤ 0, let us start with proving two auxiliary
statements.

94
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 95

Lemma 1

Consider an integral of the form


.
IC (u) = d v vh(v)
−u
, (10.5)
C

where C is some contour in the complex-v space and h is an arbitrary function. Assuming
that h is absolutely integrable on C, it is easy to see that IC (u) is well-defined for all
u∈/ C, and so are all its derivatives. Therefore, IC (u) is analytic for all u ∈
/ C. 

Lemma 2

Suppose that u is initially in the region A1 that is above a given contour C. Consider
moving u to the region A2 that is on the other side of the contour C. When u crosses
C, the function IC (u) may become singular. But let us consider the integral IL (u) taken
over contour L that is the same as C except it is bended to remain below the pole at all v.
The is called a Landau contour. Then, according to Lemma 1, the function IL (u) remains
analytic. Because one also has IL (u) = IC (u) in the whole region A1 , the function IL (u)
represents the analytic continuation of IC (u) to the region A2 (and remember that the
analytic continuation is unique). By splitting the integral into the principal-value part and
the pole contribution, one finds that the expression


 0, u ∈ A1 ,
d v vh(v)
−u
= d v
h(v)
v−u
+ i πh(u) × 1, u ∈ C, (10.6)
L C 
2, u ∈ A2

represents the analytic continuation of IC (u). [Both terms in Eq. (10.6)are discontinuous


but their sum is continuous.] Note that unless u ∈ C, one can replace with , because
then the principal value of the integral coincides with the integral itself. Also note that
if the initial region A1 corresponds to the region below C, then i π in the above formula
must be replaced with −i π. 

By Lemma 2, the analytic continuation of Eq. (10.4)

2 
ωps
   
k·v
(χs )ab (ω, k) =
ω L
d v
va
ω−k·v
δcb 1 −
ω
+
kc vb ∂f0s
ω ∂vc
. (10.7)

This result may seem surprising in the following sense. The same formula for Im ω > 0 is read-
ily obtained from the linearized Vlasov equation if one simply adopts ∂t = −i ω. But for strictly
monochromatic oscillations of the distribution function, ∂t f˜s = −i ω f˜s is a precise equality irrespec-
tive of the sign of Im ω, so why should one expect a different result? The explanation is as follows.
Remember that f˜ consists of two terms: the field-induced term f˜s ∝ exp(−i ωt) and the term f˜s
(i) (f)

that is determined by free oscillations, which has the time dependence of the form f˜s ∝ exp(−i kvt).
(f)
(i) (f)
At Im ω > 0, f˜s eventually becomes much larger than f˜s , so the initial conditions do not matter.
(More precisely, the moment t0 at which the initial conditions are prescribed can be shifted to −∞,
(i) (i)
so f˜s = f˜s at any finite t.) But at Im ω < 0, f˜s is essential, and

∂t f˜s ≈ −i ω f˜(i)
s − i kv f s 6= −i ω fs .
˜(f) ˜ (10.8)

Although such f˜s is not monochromatic, integrals of this function (such as the current density) can
be monochromatic, which is how plasma supports the quasimodes that were introduced in Lecture 9.

95
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 96

10.1.3 General dielectric tensor

Finally, let us present the general expression for the dielectric tensor. Assuming the usual notation

. 1 X 2 . X 2
f0 (v) = 2 ω f0s (v), ωp2 = ωps , (10.9)
ωp s ps s
P
and using ab (ω, k) = δab + s [χab (ω, k)]s with Eq. (10.7), one obtains

ωp2
   
k·v
ab (ω, k) = δab +
ω
d v ω − k · v δcb 1 − ω + ω ∂v .
va kc vb ∂f0
(10.10)
L c

Because
  ∞ 
δab ∞
 
k · v ∂f0s
d v ω − k · v δcb 1 − ω ∂v = ω1
va
d v va ∂f0s
∂vb
=−
ω −∞
d v f0s = −
δab
ω
, (10.11)
L c −∞

one can also present this in the following alternative form, which will be useful below:

ωp2 ωp2
 
ab (ω, k) = 1 − 2 δab + 2
ω ω L
d v
kc va vb ∂f0
ω − k · v ∂vc
. (10.12)

10.2 Dielectric properties of isotropic plasma

10.2.1 Dielectric tensor
Let us now consider an important special case of isotropic plasma,1 that is, plasma where f0 (v) =
.
f¯0 (v), with v = (vx2 + vy2 + vz2 )1/2 . Then,

∂ ∂ q 2 va ¯0
f0 (v) = f¯00 (v) vx + vy2 + vz2 = f (v), (10.13)
∂va ∂va v 0
and it is easy to see that that  acquires the following diagonal form:
 
k 0 0
 =  0 ⊥ 0 , (10.14)
0 0 ⊥

assuming that the x axis is chosen along k. Correspondingly, the dispersion matrix is as follows:
 
k (ω, k) 0 0
D E (ω, k) =  0 ⊥ (ω, k) − N 2 0 , (10.15)
2
0 0 ⊥ (ω, k) − N
.
where N = kc/ω. This means that like in fluid plasma, waves can be of two types: transverse
electromagnetic waves and longitudinal electrostatic waves.

10.2.2 Transverse waves

The transverse waves satisfy

0 = ⊥ (ω, k) − (kc/ω)2 . (10.16)

1A more general case is addressed in a homework.

96
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 97

This is similar to the electromagnetic-wave dispersion that we studied earlier except ⊥ (ω, k) is some-
.
what different from its cold limit cold (ω) = 1 − ωp2 /ω 2 . From Eq. (10.12), one has

ωp2 ωp2
 
ab (ω, k) = 1 − 2 δab + 2
ω ω L
d v ωkv−akv
vb ∂f0s
, (10.17)
x ∂vx

so the thermal corrections can be estimated as follows:


ωp2 kv 2 ωp2 kvT ωp2 vT ωp2
⊥ (ω, k) − cold (ω) = 2
ω
d v ω − kv
y ∂f0s
∼ 2 < 2  2. (10.18)
L x ∂vx ω ω ω c ω

(Here, we have assumed that ω/k & c by analogy with waves in cold plasma, and we have also assumed
vT  c, because our theory is nonrelativistic to begin with.) Then,

⊥ (ω, k) ≈ 1 − ωp2 /ω 2 + O(vT2 /c2 ). (10.19)

This indicates that the electromagnetic waves in isotropic nonmagnetized plasma are not significantly
affected by (nonrelativistic) temperature; i.e., their dispersion relation is approximately the same as
in cold plasma,

ω 2 ≈ ωp2 + k 2 c2 . (10.20)

10.2.3 Longitudinal waves

The longitudinal waves satisfy
X
k (ω, k) = 0, k = 1 + χs,k , (10.21)
s

and
2 
ωps
   
χs,k (ω, k) =
ω
d v ω − kv δxx 1 − ω + ω ∂v .
vx kvx kvx ∂f0s
(10.22)
L x x

Let us integrate over vy and vz and change the notation as in Eq. (8.28). Then,
2 
ωps
χs,k (ω, k) =
ω
d v ω −v kv ∂f∂v0s , (10.23)
L

where v now denotes vx . Also notice that

v 1 (kv − ω) + ω 1 ω 1
= =− − 2 . (10.24)
ω − kv k ω − kv k k v − ω/k

Then,
 0  
f 0 (v)
 
d v vf0s (vx )
ω − kv
= dv 1 1
− − 2
ω
k k v − ω/k
0
f0s
ω
(v) = − 2
k L
d v 0s
v − ω/k
, (10.25)
L x L

∞
where we used that −∞ d v f0s 0
(v) = f0s (+∞) − f0s (−∞) = 0. This leads to the following expression
for the susceptibility (Exercise 10.1):
2  0
ωps
χs,k (ω, k) = − 2
k
d v vf−0s (v)
ω/k
, (10.26)
L

97
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 98

∞
Exercise 10.1: Show that Eq. (10.26) can as well be derived using σs (ω, k) = 0
d t e i ωt Σs,k (t),
with Σs,k taken from Eq. (8.29).

or more explicitly,

2
ωps ∞ 0 2
ωps  ω   0, Im ω > 0,
χk,s (ω, k) = − 2
k
d v v − ω/k − i π k|k| f0s0 k ×  1, Im ω = 0,
f0s (v)
(10.27)
−∞ 2, Im ω < 0.

Because v and the phase velocity ω/k can be comparable, the influence of kinetic effects on the
susceptibility can be strong, so let us discuss longitudinal waves in more detail. For simplicity, let us
focus on waves with small ωi and k (ω, k) that is smooth enough. Then, Eq. (10.21) can be simplified
as follows:

0 = k (ωr + i ωi , k) ≈ k (ωr , k) + i ωi ∂ω k (ωr , k). (10.28)

Because k on the right-hand side is evaluated at ωr , it can be calculated using Eq. (10.27) with
Im ω = 0:

k (ωr , k) = r (ωr , k) + i i (ωr , k), (10.29a)

ωp2 ∞
f00 (v)
.
r (ωr , k) = 1 −
k2
d v v − ω /k , (10.29b)
−∞ r
. ωp2 0 ω 
r
i (ωr , k) = −π f . (10.29c)
k2 0 k
For small enough i , one can further simplify the dispersion relation as follows:

r (ωr , k) + i [ωi ∂ω r (ωr , k) + i (ωr , k)] ≈ 0. (10.30)

The real and imaginary part of this complex equation give equations for ωr and ωi ; specifically,2

r (ωr , k) = 0, (10.31a)
i (ωr , k)
ωi = − . (10.31b)
∂ω r (ωr , k)

In Lecture 11, we will apply these results to study longitudinal waves in Maxwellian plasma.

10.3 Stability of electrostatic oscillations

As shown above, the dispersion relation for electrostatic oscillations in nonmagnetized plasma can be
written as

(ω) = 0, (10.32)

where the index k and the argument k are omitted for brevity and

ωp2 0
. X ωps
2
(ω) = 1 − 2
k
d v v f−0 (v)
ω/k
, f0 (v) =
ωp2
f0s (v). (10.33)
L s
2 Remember that this model holds only for smooth enough  (ω, k). Distributions with narrows beams can be
k
unstable even when i = 0; for example, see Problem PI.3.

98
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 99

At Im ω > 0, the Landau contour L in Eq. (10.33) can be replaced with the real axis:
 ∞
ωp2 0
(ω) = 1 − 2
k
d v v f−0 (v)
ω/k
, (10.34)
−∞

where the argument k is omitted for brevity. If f00 is absolutely integrable, the right-hand side is
analytic, which is proven like Lemma 1 in Lecture 10. At Im ω = 0, one has

ωp2 ∞ 0 ω2
d v v f−0 (v) − iπ
ω
p
(ω) = 1 − f0 . (10.35)
k2 −∞ ω/k k|k| 0 k

This function is also analytic provided that f0 is a physical distribution, which are always smooth on
the real axis.3 Thus, (ω) is analytic at Im ω ≥ 0. This property allows one to assess plasma stability
without actually solving Eq. (10.32), specifically, as follows.

10.3.1 Nyquist theorem

Consider the following integral over real ω:
 ∞ 0
. 1
N̄ =
2π i
d ω (ω)
(ω)
. (10.36)
−∞

At large enough ω, one has σ ∝ ω −1 (Appendix AI.1), so 0 (ω) ∝ ω −3 . Thus one can just as well
write Eq. (10.36) as
0
N̄ =
1
2π i
d ω (ω)
(ω)
, (10.37)
C

where C is the closed contour that includes the real axis R and a semi-circle of infinite radius at
Im ω > 0. By analyticity of  in the area encircled by C, the function 0 (ω)/(ω) can have singularities
within C only at ω = ωq that satisfy (ωq ) = 0. Also due to analyticity of , one has (ω) = αq (ω−ωq )nq
in the vicinity of these points, where αq is some constant, and nq is some positive integer. Then, the
integral over C can be expressed as a sum of residues at ωq :
0
d ω (ω)
X 1 (ω)
N̄ =
q
2π i Cq
nq −1
− ωq )
d ω nqααq (ω
X 1
=
q
2π i Cq (ω − ω )
q
n
q
q

d ω ω −1 ω
X nq
=
q
2π i Cq q
X
= nq . (10.38)
q

In the case of simple poles (nq = 1), the right-hand side equals the sum of zeros of  in the upper half
of the complex-ω plane, i.e., the number of unstable modes. One can also extend this statement to
general nq assuming the convention that each nq th-order zero counts as nq zeros.
Notice now that N̄ can also be calculated differently:
 
N̄ =
1
d ω
0 (ω)
=
1 d = 1
[ln (ω → +∞) − ln (ω → −∞)], (10.39)
2π i R (ω) 2π i (R)  2π i
3 In contrast, away from the real axis, f (v) is not necessarily analytic (cf. Problem PIV.2), so this argument is not
0
extendable to Im ω < 0.

99
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 100

where (R) is the projection of the real axis in the ω space to the  space. By definition of the complex
logarithm, ln  = ln || + i ϑ, where ϑ = arg . Also, |(ω → ±∞)| = 1, so
.

[i ϑ(ω → +∞) − i ϑ(ω → −∞)] =

1 ∆ϑ
N̄ = . (10.40)
2π i 2π
The expression on the right-hand side equals the number of times that (R) encircles the origin. In
combination with Eq. (10.38), we then arrive at the following theorem, known as the Nyquist theorem:

Theorem: The number of unstable modes equals the number of times the contour (R)
encircles the origin.

10.3.2 Single-peak distributions

The Nyquist theorem shows that single-peak distributions f0 cannot support unstable modes.4 This
is seen as follows. On the contour R, which is the real axis in the complex-ω plane, one has

ωp2 ∞ 0 ω2
d v v f−0 (v) − i π 2 f00
ω
p
(ω) = 1 − . (10.41)
k2 −∞ ω/k k k

Let us assume that, when projected to the complex- plane, this contour encircles the origin. For this
to occur, there must be some ω = ω∗ such that Re (ω∗ ) < 0 and Im (ω∗ ) = 0. Since ω∗ is real by
definition of R, this requires the following two conditions be satisfied simultaneously:

ωp2 ∞ 0
1−
k2
d v v −f0ω(v)/k < 0, (10.42a)
−∞ ∗
ωp2 0 ω 
∗
π 2 f0 = 0. (10.42b)
k k
Suppose that f0 (v) has only one peak, say, at some v = v∗ . Then, Eq. (10.42b) implies ω∗ /k = v∗ ,
so the inequality (10.42a) can be expressed as follows:

ωp2 ∞ 0

k2
d v vf0−(v)v > 1. (10.43)
−∞ ∗

But because the signs of f00 (v) and of (v − v∗ ) are opposite at all v, this integral is negative; hence,
the inequality (10.43) cannot be satisfied. This means that our assumption regarding the existence of
ω∗ is invalid, so (R) cannot encircle the origin. This means that by the Nyquist theorem, single-peak
distributions cannot support unstable modes.
4 In the figures below, which are taken from Ref. [1], H is the same as our .

100
LECTURE 10. DISPERSION PROPERTIES OF NONMAGNETIZED PLASMA 101

10.3.3 Double-peak distributions

Suppose now that f0 has two peaks. Then there are three extrema: a maximum v1 , a minimum v2 ,
and another maximum v3 . Then, much like in the previous case, there are generally three points
where (R) crosses the real axis in the complex- plane, namely at ωj = kvj . This implies a loop:

However, as long as v1 and v3 are not too far from each other, the loop is small and thus does not
encircle the origin. This means that by the Nyquist theorem, the appearance of a second peak does
not immediately lead to an instability. Having an instability requires the peaks to be sufficiently far
from each other.

101
Lecture 11

Electrostatic waves in isotropic

Maxwellian plasma

In this lecture, we apply the results of the previous lectures to study basic electrostatic waves in
isotropic Maxwellian plasma. (Because only such waves will be considered, the index k will be omitted.)
First, we present two alternative but equivalent calculations of the Maxwellian-species susceptibility.
Then, we discuss asymptotic approximations and, finally, we apply them to explicitly calculate an
approximate dispersion relation for Langmuir waves and ion acoustic waves in Maxwellian electron
plasma.

11.1 Susceptibility of Maxwellian plasma

11.1.1 Plasma dispersion function
Let us consider species s with the Maxwellian distribution

v2
 
1
f0s (v) = √ exp − 2 . (11.1)
2πvT s 2vT s

From Eq. (8.30), one obtains

2   ∞
ωps v2
 
Σ̄s,k (t) =
∂
4π ∂k
k dv √ 1
exp − 2 − i kvt
2vT s
−∞ 2πvT s
2   ∞  2
ωps

=
∂
4π ∂α
α d z √ exp − − i αz
1 z
2
−∞ 2π
2
ωps ∂
= [αI(α)], (11.2)
4π ∂α
.
where α = kvT s t and
 ∞
z2
 
. 1
I(α) = √ dz exp −i αz − 2
2π −∞
 ∞
2y i α
 
=√
1
π −∞
d y exp − √ − y 2
2
−α2 /2  ∞
e i
"  2 #
= √
π
d y exp − y + √ α
= e −α /2 .
2
(11.3)
−∞ 2

102
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 103

Σs,k (t)/Σs,k (0)

1.0

0.5

kυTs t
1 2 3 4 5

-0.5

Figure 11.1: Σ̄s,k (t) for Maxwellian plasma.

Using

(αe −α /2 ) = (1 − α2 )e −α /2 ,
∂ ∂ 2 2
[αI(α)] = (11.4)
∂α ∂α
one obtains
2 
ωps (kvT s t)2
 
2

Σ̄s,k (t) = 1 − (kvT s t) exp − . (11.5)
4π 2
−1
It is seen then that the phase mixing occurs on the time scale (kvT s ) , as anticipated (Fig. 11.1).
Also, using Eq. (11.5), one can readily calculate the conductivity in the spectral representation,
2  ∞
ωps
 
d t 1 − (kvT s t) exp i ωt − 2 (kvT s t) .
1
h i
2 2
σs (ω, k) = (11.6)
4π 0
It is also common to express this result as follows. Let us introduce the so-called plasma dispersion
function 1 (Fig. 11.2)
 ∞
z2 √
 
Z(ζ) = i dz exp i ζz − 4 = i πe −ζ 2 − 2S(ζ),
.
(11.7)
0

where S(ζ) is the Dawson function,

 ζ
1 −ζ 2 √
S(ζ) = e −ζ d y ey e
. 2 2
≡ π erfi(ζ). (11.8)
0 2
Then, σs can be expressed as (Exercise 11.1)
2
ωps
σs (ω, k) = i ζ 2 Z 0 (ζs ),
4πω s
.
ζs =
ω
√ . (11.9)
kvT s 2
The corresponding susceptibility is given by

4π i σs (ω, k) 2
ωps Z 0 (ζs )
χs (ω, k) = = − 2 ζs2 Z 0 (ζs ) = − 2 2 , (11.10)
ω ω 2k λDs
.
where λDs = vT s /ωps is the Debye length of species s. Because Z is an entire function of its argument,
the expression on the right-hand side of Eq. (11.10) is an entire function of ω. In particular, this means
that this expression can be used at any Im ω.
1 We assuming k > 0 for simplicity. For the general case and additional details, see Sec. 8.14 in Ref. [1]. Some
k
properties of the plasma dispersion function are also summarized in Appendix AIV.1.

103
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 104

Figure 11.2: Re Z(ζ) (left) and Im Z(ζ) (right) as functions of (Re ζ, Im ζ).

Exercise 11.1: Derive Eq. (11.9).

11.1.2 An alternative derivation using Landau’s rule

Now, let us derive an alternative representation of χs for Maxwellian species starting from Eq. (10.26).
First, note that
v2
   
0 1 v
f0s (v) = √ − 2 exp − 2 , (11.11)
2πvT s vT s 2vT s
so
 
d u uue− ω = − 2k21λ2 √1π e −z .
2 2 2
ωps −u /2
χs (ω, k) = 2 2 √
1
k vT s 2π
dz −2z
z − ζs
(11.12)
L kvT s Ds L

Note that
 
e −z = √1 de −z
2 2
!
√
1
π
dz −2z
z − ζs π
dz z −1 ζ dz
L L s

−z 2 d
 
= −√
1
π L
dz e dz z − ζ 1
s

2 d
 
=√
1
π L
dz e d ζ z − ζ
−z 1
s s

1 d
 
=√
π d ζs L
dz e −z 2 1
z − ζs
. (11.13)

Then, finally,

dz ze − ζ .
2
Z̃ 0 (ζs ) . 1 −z
χs (ω, k) = − 2 2 , Z̃(ζ) = √ (11.14)
2k λDs π L

104
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 105

This result is identical to Eq. (11.10) except the plasma dispersion function Z is replaced with Z̃.
Thus, Z̃ given by (11.14) is just an alternative representation of Z.

11.1.3 General susceptibility

The results of the previous sections are summarized as follows. Maxwellian species have the suscep-
tibility
Z 0 (ζs ) ω
χs (ω, k) = − , ζs = √ . (11.15)
2k 2 λ2Ds kvT s 2
Here, Z is the plasma dispersion function, which is given by Eq. (11.7) but can also be expressed as
 −z 2
Z(ζ) = √
1
π
d z ze − ζ . (11.16)
L

For real ζ, this gives

dz ze − ζ ,
∞ 2
−z √
π e −ζ .
1 2
Zr (ζ) = √ Zi (ζ) = (11.17)
π −∞

Then,
X  Z 0 (ζr ) 
r
r (ωr , k) = 1 − , (11.18a)
s
2k 2 λ2D s

√ X ζr e −ζ
2 
X  Z 0 (ζr )  
i
r

i (ωr , k) = − = π , (11.18b)
s
2k 2 λ2D s s
k 2 λ2D s
√
.
where ζr = ωr /(kvT 2). The index r in ζr is henceforth dropped for brevity.

11.2 Asymptotics
11.2.1 Warm species
Let us search for waves in plasma with “warm” species, i.e., such that ζ is large but finite. Then, due
to the exponential factor, the integral is mainly determined by the integrand at z  ζ. Then,
∞
1 1 1 X zn
=− ≈− . (11.19)
z−ζ ζ(1 − z/ζ) ζ n=0 ζ n

This leads to
∞ ∞ ∞ ∞
ζ −1 zn ζ −1 z 2n
dz e −z 2
dz e −z 2
X X
Zr (ζ) = − √ = − √ , (11.20)
π −∞ n=0
ζn π −∞ n=0
ζ 2n

where we used the fact that the integrals with odd powers of z are zero. Hence,
   ∞
1 1
Zr (ζ) ≈ − J(0) + 2 J(1) + . . . ,
ζ ζ
. 1
J(n) = √
π −∞
d z z 2n e −z .
2
(11.21)

This integral can be expressed through the gamma function. Alternatively, notice that

J(n) = (−1)n J¯(n) (1), (11.22)

105
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 106

(n)
where denotes the nthe-order derivative and
 ∞  ∞
¯
J(β)
. 1
=√
π
dz e −βz 2
=√
1
dz e −z 2 1
=√ . (11.23)
−∞ πβ −∞ β
Then, one readily obtains
¯ = 1,
J(0) = J(1) J(1) = −J¯0 (1) = 1/2, (11.24)

and so on. Hence,

 
1 1 1 1
Zr (ζ) = − 1 + 2 + ... ≈ − − 3. (11.25)
ζ 2ζ ζ 2ζ

Using Eq. (11.15), one then obtains the real part of the susceptibility in the form
 
1 1 3
χs,r (ωr , k) ≈ − 2 2 + 4
2k λDs ζs2 2ζs
2 2 2
ωps 2k vT s
 
3
=− 2 2 1+ 2
2k vT s ωr2 2ζs
2  2 2
ωps

3k vT s
=− 2 1+ (11.26)
ωr ωr2
and the imaginary part of the susceptibility in the form

√ ζs e −ζs
2

χs,i (ωr , k) ≈ π 2 2
k λDs
r 2
π ωr ωps
=
2 kvT s k vT2 s
2
e −ζs2
π ωr e −ζs
r 2

= . (11.27)
2 ωps (kλDs )3

11.2.2 Hot species

To approximate Z(ζ) for hot species (ζ  1), notice that
 ζ  ζ
S(ζ) = exp(−ζ 2 ) d y exp(y2 ) ≈ d y = ζ, (11.28)
0 0

so one readily obtains2

√ √ √
Z(ζ) ≈ ie −ζ Z 0 (ζ) ≈ −2i ζ e −ζ π − 2 ≈ −2(1 + i ζ π).
2 2
π − 2ζ, (11.29)

It is also instructive to notice that Re Z(ζ) = −2ζ for real ζ is seen from the following:

d z ze − ζ du e
2
∞ −z ∞ −(ζ+u)2
1 1
√ =√
π −∞ π −∞ u

= √
e −ζ 2 ∞
du e
−u2 −2ζu
e
π −∞ u
2 The second term on the right-hand side in Eq. (11.29) is much smaller than the first term and may, in fact, be
comparable to the corrections that we neglected in the Dawson function approximation. However, this term is more
important as it is the primary cause of damping, whereas corrections to the Dawson function would only slightly affect
the real part of the dispersion relation.

106
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 107

du e u
∞ −u2
1
≈√ (1 − 2ζu)
π −∞

du e
∞ 2 ∞
−u
=√
1
π u
2ζ
−√
π
d u e −u
2

−∞ −∞
= −2ζ. (11.30)

11.3 Waves
Now let us use the above results to derive the dispersion relations for Langmuir wave and ion acoustic
from kinetic theory.

11.3.1 Langmuir waves

For Langmuir waves, the electron contribution is much larger than the ion contribution. Assuming
that electrons are warm, one finds

π ωr e −ζe
2  2
3k 2 vT2 e
r
ωpe

r (ωr , k) ≈ 1 − 2 1 + ,  (ω
i r , k) ≈ . (11.31)
ωr ωr2 2 ωpe (kλDe )3
Equation (10.31a) leads to
2 
ωpe 3k 2 vT2 e

0 = r (ωr , k) ≈ 1 − 2 1 + , (11.32)
ωr ωr2


2
so one obtains ωr2 = ωpe + O k 2 vT2 e . The second term on the right-hand side is small, so to the
lowest (zeroth) order in the temperature, one has ωr2 ≈ ωpe 2
, as expected. Then, to the next (first)
order in the temperature, one has
2 
ωpe 3k 2 vT2 e

0=1− 2 1+ 2
. (11.33)
ωr ωpe
This leads to
3k 2 vT2 e
 
2 2 2
ωr = ωpe 1+ 2
= ωpe + 3k 2 vT2 e , (11.34)
ωpe
or in other words,
 
3
ωr ≈ ±ωpe 1 + k 2 λ2De . (11.35)
2
Equation (10.31b) leads to

π |ωr |3 e −ζe
r 2
ωi 1 i (ωr , k)
=− ≈− 3 3, (11.36)
|ωr | ωr ∂ω r (ωr , k) 2 2ωpe (kλDe )
where we have substituted
2  2
ωpe 2ωpe

∂r (ωr , k) ∂
≈ 1− 2 = 3 . (11.37)
∂ω ∂ω ωr ωr
2
We can also use the approximation |ωr | ≈ ωpe but ζe in e−ζe must be calculated more accurately:
2
ωr2 ωpe + 3k 2 vT2 e 1 3
ζe2 = = = 2+ , (11.38)
2k 2 vT2 e 2k 2 vT2 e 2κ 2

107
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 108

2.0
0.100
1.5
0.010

-ωi /ωpe
ωr /ωp,e

1.0
0.001
0.5 10-4

0.0 10-5
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
kλDe kλDe

Figure 11.3: ωr /ωpe (left) and −ωi /ωpe (right) as functions of kλDe for Langmuir waves in Maxwellian
electron plasma: red – numerical solution, blue – Eqs. (11.35) and (11.39).

.
where κ = kλDe . This leads to
r  
ωi π −3 1 3
≈− κ exp − 2 − . (11.39)
|ωr | 8 2κ 2
Also note that in application to Langmuir waves, our assumption that ζe  1 can be expressed simply
as κ  1, because ω 2 ≈ ωpe 2
. In this case, κ−3 exp(−κ−2 /2)  1, so ωi  ωr ; otherwise, our theory is
inapplicable. A comparison with the numerical solution of the dispersion relation is shown in Fig. 11.3.
Note that the fluid calculation of the Langmuir wave dispersion relation in Sec. 7 properly captured
ωr but not ωi . The kinetic calculation shows that in warm plasma, Langmuir waves have ωi < 0 and
therefore dissipate. This effect is known as Landau damping. It was predicted theoretically in Ref. [47],
and the first experimental observation was reported in Ref. [48]. The physical mechanism of Landau
damping will be discussed in Lecture 12.

11.3.2 Ion acoustic waves

Assuming that electrons are hot, one obtains

χe = −
1
Z 0
(ζ ) ≈
1
+
i ζ e √π . (11.40)
e
2k 2 λ2D,e k 2 λ2D,e k 2 λ2D,e

Assuming that electrons are hot, one obtains

1
χi = − Z 0 (ζi )
2k 2 λ2D,i
√
  
−2i e
1 −ζi2 1 3
≈− 2 2 π+ 2 1+ 2
2k λD,i ζi 2ζi
√
ie −ζi2
π ωp,i 2  
3
= 2 2 − 2 1+ 2
k λD,i ω 2ζi
2
ωp,i
≈− . (11.41)
ω2
Hence, the dielectric function in the form
2 √
1 ωp,i ζe π
r (ω, k) = 1 + − , i (ω, k) = . (11.42)
k 2 λ2D,e ω2 k 2 λ2D,e

108
LECTURE 11. ELECTROSTATIC WAVES IN ISOTROPIC MAXWELLIAN PLASMA 109

Equation (10.31a) leads to

2
1 ωp,i
1+ − = 0, (11.43)
k 2 λ2D,e ωr2

so one obtains
!−1
2
2 2 1 ωp,i k 2 λ2D,e k 2 Cs2
ωr = ωp,i 1+ 2 2 = = , (11.44)
k λD,e 1 + k 2 λ2D,e 1 + k 2 λ2D,e

where Cs is the ion sound speed given by

2
ωp,i ni e2i Te ei Te Zi Te
Cs2 = ωp,i
2
λ2D,e = 2
v 2
T,e = = ≡ . (11.45)
ωp,e ne e2e mi |ee | mi mi

In particular, let us consider the limit of small kλDe , when

2
∂r (ω, k) 2ωp,i 2 1
≈ ≈ . (11.46)
∂ω ω 3 ωr k λ2D,e
2

Then, Eq. (10.31b) leads to

p r r
π/2 ωr ωr 2 2 π ωr π Zi me
ωi = − 2 2 k λD,e = −ωr = −ωr  ωr . (11.47)
k λD,e kvT,e 2 8 kvT,e 8 mi

This shows that ions also experience Landau damping. Unlike for Langmuir waves, this damping is
not exponentially small at small kλD,e ; rather, it is small due to the smallness of me /mi . The reason
for this different scaling will become clear after we discuss the physical mechanism of Landau damping
in the next lecture.

109
Lecture 12

Landau damping and kinetic

instabilities

In this lecture, we discuss the physical mechanism of Landau damping and the associated nonlinear
effects. Although our discussion is limited to electrostatic interactions in nonmagnetized plasma, the
qualitative effects to be considered are relevant also in more general settings.

12.1 Passing and trapped particles

First, let us discuss the single-particle motion in a prescribed sinusoidal wave with a fixed ampli-
tude E0 . In the frame of reference traveling with the phase velocity ω/k, the wave potential is
.
stationary and the particle coordinate x̄ = x − (ω/k)t is governed by a nonlinear-pendulum equation
es
¨=
x̄ E0 sin(kx̄), (12.1)
ms

with equilibria at kx̄ = 2πn. (Here, n is an integer, and we will assume es E0 < 0 and k > 0, but the
signs can always be changed by a variable transformation x̄ → x̄ + π/k.) Let us consider the vicinity
of an equilibrium with, say, n = 0. Close enough to the equilibrium, Eq. (12.1) can be approximated
with the equation of a harmonic oscillator,

¨ + Ω2b0 x̄ = 0,
x̄ (12.2)

whose frequency, called the bounce frequency, is given by

s
. |es E0 |k
Ωb0 = . (12.3)
ms

More generally, one obtains from the energy conservation that

1 2
v̄ 2 − v [cos(kx̄) − 1] = v02 , (12.4)
2 t
or, equivalently,

v̄ 2 + vt2 sin2 (kx̄/2) = v02 , (12.5)

. . .
where v̄ = v − ω/k is the velocity in the moving frame, vt = 2Ωb0 /k, and v0 = |v̄(x̄ = 0)| is a constant
determined by the initial conditions. The trajectory with v0 = vt is the separatrix; it separates passing

110
LECTURE 12. LANDAU DAMPING AND KINETIC INSTABILITIES 111

ω
k

Figure 12.1: Trajectories of passing particles (non-shaded region) and trapped particles (shaded re-
gion). The arrows indicate the direction of the bounce motion. The dashed line corresponds to
v = ω/k.

(unbounded) trajectories from trapped (bounded) trajectories (Fig. 12.1). Passing trajectories, which
correspond to v0 > vt , extend from −∞ to +∞; particles on these trajectories have a nonzero average
velocity. In contrast, trapped trajectories, which have v0 < vt , are confined to a single wave period
and have zero average velocity in the moving frame, or the average velocity ω/k in the laboratory
frame.
The bounce frequency of trapped particles equals Ωb0 at v0 /vt  1 (even though Ωb0 is often called
“the” bounce frequency), but generally, it is found as follows. From Eq. (12.5), one obtains

d x̄ = ±qv2 − v2 sin2 (kx̄/2). (12.6)

dt 0 t

Then, the oscillation period T = d t can be expressed as follows:

 x
T = q
d x̄ =4 q
∗
d x̄ , (12.7)
v02 − vt2 sin2 (kx̄/2) 0 v02 − vt2 sin2 (kx̄/2)

where x∗ is the right stopping point. Let us adopt a new variable θ such that

cos θ d θ = cos(kx̄/2) d x̄.

vt kvt
sin θ = sin(kx̄/2), (12.8)
v0 2v0
Then, on x ∈ (0, x∗ ), one has
q
v02 − vt2 sin2 (kx̄/2)
d x̄ = kv q
2v0 cos θ
d θ = kv q
2
d θ. (12.9)
t 1 − sin2 (kx̄/2) t 1 − (v02 /vt2 ) sin2 θ

This leads to
4K(r) . v2
Tb = , r = 02 , (12.10)
Ωb0 vt
where K is the complete elliptic integral of the first kind. The corresponding bounce frequency is

. 2π πΩb0
Ωb = = . (12.11)
Tb 2K(r)

111
LECTURE 12. LANDAU DAMPING AND KINETIC INSTABILITIES 112

Ωb /Ωb0
1.0

0.8

0.6

0.4

0.2

υ20 /υ2t
0.0 0.2 0.4 0.6 0.8 1.0

Figure 12.2: Ωb /Ωb0 vs. v02 /vt2 : solid blue – exact formula (12.11); dashed black – asymptotic (12.12).

As seen in Fig. 12.2, Ωb ∼ Ωb0 for all trapped particles except those very close to the separatrix, where

Ωb (r → 1) π
≈ → 0. (12.12)
Ωb0 ln[16/(1 − r)]

12.2 Wave–particle energy exchange

Now, let us discuss the energy exchange between particles (electrons or ions) and the wave.

12.2.1 Direct Landau damping: wave dissipation

Particles with |v̄|  vt do not significantly participate in the resonant energy exchange with the wave,
so let us focus on trapped particles. Let us also split the trapped population into two groups: particles
that initially have v < ω/k and those that initially have v > ω/k. In Fig. 12.1, those are marked with
blue and red colors, respectively. At t > 0, the “blue” particles will, on average, go up in velocity;
thus, they will increase their kinetic energy in the laboratory frame at the expense of the electric-field
energy. In contrast, the “red” particles will on average decrease their energy and thus give energy to
the field. If the velocity distribution has a negative slope near the resonance [f00 (ω/k) < 0], then the
number of blue particles is larger than the number of red particles, so overall, the wave experiences
damping. This is the linear stage of Landau damping, which is described by the equations derived
(for the Maxwellian distribution) in Lecture 11.
If linear damping is strong enough, the wave dissipates before nonlinear effects become important.
Otherwise, Landau damping eventually enters a nonlinear stage [49, 50], when, loosely speaking, the
blue particles exchange places with the red ones. When that happens, the plasma starts to give energy
back to the field. But the system does not quite return to its original state after the bounce period,
because particles with different energies have different bounce periods. By the time 2π/Ωb0 , particles
with small v0 will have performed half-rotation around the trapping island but particles with larger
v0 will have not. This means that the absolute difference between the number of particles above and
below the resonance will be less than that in the initial state [Fig. 12.3(a)]. This difference is attenuated
further at the next half-rotation, and so on. Eventually, one ends up with a (quasi)stationary structure
where the density profile across the island is completely flattened [Fig. 12.4(a)], at least to the extent
that f000 (ω/k) is negligible. After that, individual particles continue to bounce back and forth, but the
macroscopic current density remains zero, so the electrostatic energy saturates at a constant level.1
1 The effect can be understood as phase mixing. It is not the same phase mixing as in the linear problem (Sec. 8.3),
but it is similar conceptually: each particle in the trapping island acts as an oscillator, and the macroscopic current is
determined by the average displacement of these oscillators from the equilibrium.

112
LECTURE 12. LANDAU DAMPING AND KINETIC INSTABILITIES 113

A A
(a) (b)
nonlinear stage
linear stage

nonlinear stage

linear stage

t t

Figure 12.3: A schematic of the evolution of the wave amplitude A(t) resulting from: (a) Landau
damping and (b) inverse Landau damping. The linear stage corresponds to A ∝ e ω t predicted by i

linear theory (Lecture 11). The dashed lines mark the asymptotic values at t → ∞.

f f
(a) (b)

υ υ
ω/k ω/k

Figure 12.4: Flattening of the distribution function near the resonance v = ω/k: dashed – initial
distribution, solid – distribution after saturation: (a) Landau damping – particles from the blue-
shaded region end up in the red-shaded region; (b) inverse Landau damping – particles from the
red-shaded region end up in the blue-shaded region.

These nonlinear effects limit the applicability of the linear Landau-damping theory presented in
Lecture 11. However, this theory can still be applicable at t & Ω−1 b0 if plasma is collisional. If the
collision rate νs exceeds Ωb0 , then the distribution function is kept close to Maxwellian at all times
and nonlinear flattening does not occur. That said, νs cannot be too large either. It should remain
small compared to kvT s , because otherwise collisional effects overshadow kinetic effects and the wave–
particle interaction ceases to be resonant, eliminating Landau damping. In summary then, for the
linear theory of collisionless Landau damping to apply at ωi . Ωb0 , one must have

Ωb0  νs  kvT s . (12.13)

12.2.2 Inverse Landau damping: wave amplification

Similar considerations apply when the slope of the distribution near the resonance is positive [f00 (ω/k) >
0]. In this case, Landau damping is replaced with inverse Landau damping, which causes wave amplifi-
cation [Fig. 12.3(b)]. This can result in instabilities, such as the bump-on-tail instability [Fig. 12.4(b)]
and the two-stream instability (Problem PIV.4). However, keep in mind that depending on the bound-
ary conditions, inverse Landau damping may produce a continuous spectrum of waves, in which case
the nonlinear saturation is different. (This will be discussed at the end of the course.) Also note
that unlike direct Landau damping, inverse Landau damping necessarily yields nonlinear structures
(Fig. 12.5). We will briefly describe such structures in the next section.

113
LECTURE 12. LANDAU DAMPING AND KINETIC INSTABILITIES 114

Figure 12.5: Results of particle-in-cell simulations illustrating the development of the two-stream
instability (from Ref. [51]).

12.2.3 Saturated states, BGK waves

Nonlinear waves produced as a result of Landau damping (direct or inverse) are close to so-called
Bernstein-Greene-Kruskal (BGK) waves [52], which are exact stationary solutions of the nonlinear
Vlasov equation. Such waves have zero linear damping rate but exhibit a nonlinear frequency shift
.
δω = ω − ωL [here ωL is the frequency that satisfies the linear dispersion relation (ωL , k) = 0], which
can be calculated as follows. For simplicity, suppose that trapped particles are well separated in the
velocity space from the bulk plasma (passing particles), which then can be described by real linear
(ω, k). For simplicity, assume that trapped particles are electrons. By treating them as external
charges with some density nt (x), one obtains from Gauss’s law that
k 2 (ω, k)ϕk = 4πee nt,k , (12.14)
.
where the index k denotes the corresponding Fourier harmonics. Using the notation Uk = ee ϕk for
the particle potential energy, one can also rewrite this as follows:
4πe2e nt,k ∂(ωL , k) ∂(ωL , k) 2ωp2 2δω
= (ω, k) ≈ (ω L , k) + δω = δω ≈ 3 δω ≈ ω . (12.15)
k 2 Uk ∂ω ∂ω ωL p

where we assumed  ≈ 1 − ωp2 /ω 2 . Assuming also that trapped particles are accumulated at bottoms
of the wave troughs (this corresponds to a delta-shaped spatial distribution), one has
nt,k n̄t n̄t
∼− 2 =− . (12.16)
k 2 Uk k U |ee kE|
(The minus is due to the fact that the density has a maximum where the potential energy has a min-
imum, and n̄t denotes the spatial average of nt , i.e., the number of trapped particles per wavelength.)
This shows that the frequency shift is negative and is given by
δω 4πe2e n̄t ω2
∼− = − t2 , (12.17)
ωp |ee kE| Ωb

114
LECTURE 12. LANDAU DAMPING AND KINETIC INSTABILITIES 115

where ωt is the plasma frequency associated with the density n̄t ; i.e.,
s
. 4πe2e n̄t
ωt = . (12.18)
me

Also notice the unusual scaling |δω| ∼ E −1 . This scaling changes if trapped particles
√ are spread out
across the island. In fact, if the distribution is smooth, one can show that |δω| ∼ E, which is indeed
seen in laser–plasma interactions. For further details on dispersion and adiabatic dynamics of BGK-
like waves, see, for example, Refs. [53,54]. In particular, note that due to the nonlinear frequency shift
δω = δω(|E|), BGK-like waves with trapped particles can be subject to nonlinear instabilities [55–58].

115
Lecture 13

Dispersion properties of magnetized

plasma

In this lecture, we extend our kinetic calculation of the plasma dielectric properties to three-dimensional
electromagnetic waves and magnetized plasma.

13.1 General dispersion operator from kinetic theory

Calculating the dielectric tensor of magnetized plasma requires a more general approach than the one
we have been using so far. Let us return to the linearized Vlasov equation (8.17) and consider its
characteristics, i.e., phase-space trajectories (x0 (t0 ), v 0 (t0 )) that particles would follow in the absence
of a wave:
d x0 = v 0 (t0 ),
d v0 =
1
F 0s [t0 , x0 (t0 ), v 0 (t0 )]. (13.1)
d t0 d t0 ms
(Here, F 0s is the zeroth-order Lorentz force, if any. The primes denote the fact that these trajectories
are different from the true ones, which account for both zeroth and first-order fields.) The full time
derivative of f˜s along such a trajectory is given by

d f˜ [t0 , x0 (t0 ), v0 (t0 )] = ∂ f˜s + v · ∇f˜ + F 0s · ∂ f˜s

!

d t0 s ∂t
s
ms ∂v 0 0 0 0 0 [t ,x (t ),v (t )]

= Rs [t0 , x0 (t0 ), v 0 (t0 )], (13.2)

where we substituted Eq. (8.17). Assuming the initial condition

x0 (t0 = t) = x, v 0 (t0 = t) = v, (13.3)

one then obtains a solution for f˜s at any given (t, x, v):
 t
f˜s (t, x, v) = f˜s [t, x0 (t), v 0 (t)] = f˜(0)
s + d t0 Rs [t0 , x0 (t0 ), v0 (t0 )], (13.4)
t0

(0)
where the integral depends on x and v through Eqs. (13.3) and f˜s is an integration “constant”.
Using Eq. (8.19) for Rs , one obtains
 t      
v · k̂
f˜(i)
s = − d t 0 es n0s ∂f0s
ms ∂vc
δ cb 1 −
ω̂
+
vb k̂ c
ω̂
Ẽb . (13.5)
t0 [t0 ,x0 (t0 ),v 0 (t0 )]

116
LECTURE 13. DISPERSION PROPERTIES OF MAGNETIZED PLASMA 117

From here, one can find the induced current density (8.21) and infer the conductivity (8.22); then D E
is readily obtained.

For the rest of this lecture, see the separate slides.

117
Lecture 14

Waves in magnetized plasma

In this lecture, we apply the results of Lecture 13 to explore kinetic waves in magnetized isotropic
Maxwellian plasma without flows.

14.1 Basic equations

Let us assume a homogeneous magnetized isotropic Maxwellian plasma without flows, with geometry
as in Lecture 6. As shown in Lecture 13, the dielectric tensor of such plasma can be written as1
∞
!
ωp2 X
e Yn ,
X
−λ
=1+ χs , χs = (14.1)
s
ω n=−∞
s

where the matrices Y n are given by

n2 In
 
−i n (In −
k⊥ nIn
An In0 ) An Bn
 λ Ω λ 
2
i k⊥ (I
   
 i n (In − In ) An
 0 n 0 0

Yn= In + 2λIn − 2λIn An n − In )Bn
, (14.2)
λ Ω
i k⊥ (I − I 0 )B

 
 k⊥ nIn 2(ω − nΩ) 
Bn − n n n In Bn
Ω λ Ω kk w 2

In = In (λ) are modified Bessel functions of the first kind, and

Z(ξn ) Z 0 (ξn ) 2 2
k⊥ w ω − nΩ
An = , Bn = − , λ= , ξn = , (14.3)
kk w 2kk 2Ω2 kk w

where Z is the plasma dispersion function introduced in Lecture 11 (Exercise 14.1).

Exercise 14.1: Identify the limit in which Eqs. (14.1)–(14.3) reproduce the cold-plasma dielec-
tric tensor (6.13).

As usual, the dispersion relation and the equation for the field polarization are

det D E (ω, k) = 0, D E (ω, k)h = 0, (14.4)

1 For an alternative representation of  that does not involve infinite series, see Ref. [59].

118
LECTURE 14. WAVES IN MAGNETIZED PLASMA 119

with the dispersion matrix given by

xx − Nk2
 
xy xz + N⊥ Nk
DE =  yx yy − N 2 yz . (14.5)
2
zx + N⊥ Nk zy zz − N⊥

Below, we will focus on waves propagating perpendicularly to the magnetic field (Nk → 0). For
parallel propagation (N⊥ → 0), see Problem PIV.7.

14.2 Perpendicular propagation: general considerations

At kk → 0, one has ξn → ∞, so Z(ξn ) ≈ −ξn−1 [Eq. (11.25)]. This leads to
 
1 kk w 1
An → − =− , (14.6a)
kk w ω − nΩ ω − nΩ
2
kk w 2

1 kk w
Bn → − =− → 0. (14.6b)
2kk ω − nΩ 2(ω − nΩ)2

However, (Yn )zz remains nonzero, because

2(ω − nΩ) 2(ω − nΩ) kk w 2 In

(Yn )zz = 2
In B n → − 2
In 2
= . (14.7)
kk w kk w 2(ω − nΩ) ω − nΩ

Then, the polarization equation (14.4) can be written as

  
xx xy 0 hx
 yx yy − N 2 0   hy  = 0, (14.8)
0 0 zz − N 2 hz

so there are two types of modes. Ordinary (O) modes, which are polarized along the z axis, satisfy

zz − N 2 = 0 (14.9)

and subsume the cold O wave as a special case. Extraordinary (X) modes, which are polarized in the
plane transverse to the z axis, satisfy
 
xx xy
det =0 (14.10)
yx yy − N 2

and subsume the cold X wave as a special case. The effects introduced by kinetic corrections for
O and X waves are similar, but Eq. (14.10) is somewhat richer than Eq. (14.9), so in this lecture we
will focus on X waves. For a more comprehensive discussion, see Refs. [60, 61].

14.3 Waves in the upper-hybrid frequency range

14.3.1 Electrostatic approximation
According to cold-plasma theory (Lecture 6), X waves can be adequately described within the electro-
static approximation at least at some frequencies, so let us assume that first. Within the electrostatic
approximation, the polarization vector h is parallel to k, and k is parallel to the x axis by our
convention. Then, from the x component of
  
xx xy hx
= 0, (14.11)
yx yy − N 2 hy

119
LECTURE 14. WAVES IN MAGNETIZED PLASMA 120

Exercise 14.2: For an x-polarized wave, the y component of Eq. (14.11) is yx (ω, k) = 0, which
is inconsistent with Eq. (14.12). Explain why Eq. (14.12) should be preferred over this equation.

one obtains, in the limit hy → 0, that (Exercise 14.2)

xx (ω, k) = 0. (14.12)

For clarity, let us focus on electron frequencies (of order ωuh ∼ Ωe or higher), where the ion
contribution is negligible. Then,  = 1 + χe ,
2 ∞ 2
ωpe
e −λe n In (λe )
X
(χe )xx ≈ An
ω n=−∞ λe
2 ∞ 2
ωpe
e −λe n In (λe ) 1
X
≈−
ω n=−∞ λe ω − nΩe
2 X∞ 2
ωpe
 
=−
ω n=1
e −λe n In (λe )
λe
1
ω − nΩe
+
1
ω + nΩe
2 X∞ 2
ωpe
=−
ω n=1
e −λe n In (λe )
λe
2ω
ω 2 − (nΩe )2
2 X∞
ωpe n2
=− 2
Ωe n=1
e −λe 2In (λe )
λe (ω/Ωe )2 − n2
, (14.13)

where we used In (λ) = I−n (λ). Let us introduce

2
. ωpe
. ω
q=
Ωe
, β2 = 2 ,
Ωe
αn (λ) =
λ
e .
. 2In (λ) −λ
(14.14)

Then, Eq. (14.12) leads to

∞
1 X n2 αn (λe )
2
= ≡ H(q, λe ), (14.15)
β n=1
q 2 − n2

with αn and H illustrated in Fig. 14.1. The modes described by Eq. (14.15) are known as (electrostatic)
electron Bernstein waves (EBW). Their frequencies can be approximately calculated as follows. Note
that
 n−1 r
λ 2 −3/2
αn (λ → 0) ∼ , αn (λ → ∞) ∼ λ , (14.16)
2 π

so there is always a mode with ω(λe → 0) → ωuh . Also, clearly, there are modes with ω → n|Ωe | at
both λe → 0 and λe → ∞. To connect these limits, let us consider EBW at β  1 (very underdense
plasma). In this case, H(q, λe ) has to be large to satisfy Eq. (14.15), so q must be close to some
integer n. Then, one can retain only the nth term in the sum in Eq. (14.15), so the latter becomes

n2 αn (λe )
1 − β2 ≈ 0. (14.17)
q 2 − n2

This leads to q 2 = n2 [1 + β 2 αn (λe )], or equivalently,

ω 2 = n2 [Ω2e + ωpe
2
αn (λe )]. (14.18)

120
LECTURE 14. WAVES IN MAGNETIZED PLASMA 121

α1 α2,3,4
1.0 (a) 0.10 (b)
0.8 0.08
0.6 0.06
0.4 0.04

0.2 0.02

λ λ
1 2 3 4 5 1 2 3 4 5

H
3
(c)
2

q
0.5 1.0 1.5 2.0 2.5 3.0 3.5
-1

-2

-3

Figure 14.1: Upper row: αn (λ) for various n. Lower row: H(q, λ) at λ = 0.2.

Thus, at n = 1, one has

ω(λe → 0) → ωuh , ω(λe → ∞) → |Ωe |, (14.19)

while at n > 1, one has ω → n|Ωe | at both λe → 0 and λe → ∞. The situation is somewhat different
in denser plasmas, as illustrated in Fig. 14.2.
Notably, all solutions of Eq. (14.15) are real. This means EBW as considered here do not dissipate.
The absence of dissipation is due to the fact that these waves propagate transversely to B 0 and
cannot resonantly interact with plasma particles. However, this ceases to be the case when relativistic
effects are taken into account [61]. Also, kk cannot be strictly zero in practice, and the ideal-plasma
approximation is inapplicable at very large k⊥ . In real plasmas, EBW dissipate efficiently, because they
have a small group velocity and thus take a long time to propagate through a plasma of a given length.
This makes these waves convenient vehicles for depositing energy into plasma, especially because EBW
can propagate in dense plasmas which electromagnetic waves cannot penetrate (Sec. 14.3.3). However,
note that EBW cannot be launched by antennas placed outside the plasma, because EBW cannot
propagate outside plasma. To understand how to launch these waves, electromagnetic effects have to
be considered. Qualitatively, these effects are understood as follows.

14.3.2 Electromagnetic dispersion

As discussed in Sec. 1.2.1, the electrostatic approximation generally holds when N 2  ab , which
in our case corresponds to large enough λe . At small λe , the waves are electromagnetic and can be
described as in Lecture 6 using the cold-plasma approximation. The only question then is how to
connect these two limits and, in particular, what happens near cyclotron resonances, where even small
thermal corrections matter. To answer this question, consider the following:

121
LECTURE 14. WAVES IN MAGNETIZED PLASMA 122

ω/Ωe  ω/Ωe 

5 (a) 5 (b)

4 4

3 3

2 2

1 1

λe λe
0.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0

2 2
Figure 14.2: The frequencies ω of the first four EBW in overdense plasma vs. λe = k⊥ vT e /Ω2e : left
– β = 3 (ωuh ≈ 3.16|Ωe |); right – β = 4 (ωuh ≈ 4.12|Ωe |). The dashed lines correspond to ω = n|Ωe |
with integer n and also to ω = ωuh .

• By studying asymptotics of Eq. (14.10) at λe → 0 or at λe → ∞ (or numerically), one can show

that there is an additional mode localized near each cyclotron resonance, ω ≈ n|Ωe |. These
modes are called Dnestrovskii–Kostomarov (DK) modes.

• The dispersion curve of the cold-plasma X wave crosses the resonances at ω = n|Ωe |, so with DK
modes taken into account, one can expect the new dispersion curves to exhibit avoided crossing
(Sec. 6.4.5) near the resonances. In other words, the X wave should continuously transform into
DK modes.
• Homogeneous Maxwellian plasma has no free energy, so the corresponding waves are stable. By
general theory of avoided crossing (Sec. 6.4.5), resonant coupling of two stable waves modifies
the corresponding dispersion curves that there remain two real frequencies for each real k.
These considerations are sufficient to plot the corresponding dispersion curves unambiguously. This
is illustrated in Fig. 14.3 for n = 1 in underdense plasma. There is no commonly accepted names for
the individual branches, but one can identify them locally based on the closest asymptotic. Then,
one can say that the electrostatic EBW continuously transforms into the cold X wave, and the latter
continuously transforms into the DK mode.

14.3.3 EBW application to plasma heating

Suppose one wants to heat a magnetically confined fusion plasma with an EBW. The antenna has to
be placed outside the plasma, so it launches a vacuum wave. The idea is to transform this vacuum
wave into an EBW adiabatically near the upper-hybrid resonance. Coupling with the DK mode is
diabatic (the corresponding frequency gap is small at nonrelativistic temperatures), so it is usually
ignored. However, delivering power to the upper-hybrid resonance can be tricky. This is seen from
the Clemmow–Mullaly–Allis (CMA) diagram, which shows the parameter ranges where a cold X wave
can propagate (Fig. 14.4; see also Problem PIII.2). If an X wave is launched from the low-B0 side
(|Ωe |/ω < 1, ωpe = 0), such wave is reflected at the low-density cutoff (R = 0). An X wave launched
from the high-B0 side (|Ωe |/ω > 1, ωpe = 0) can enter the plasma but cannot penetrate the dense
plasma core because of the high-density cutoff (L = 0).

122
LECTURE 14. WAVES IN MAGNETIZED PLASMA 123

ω/Ωe  N2
1.15 (a) 5 (b)

1.10 4

1.05 3

1.00 2

0.95 1

λ ω/Ωe 
0.2 0.4 0.6 0.8 1.0 0.6 0.8 1.0 1.2

Figure 14.3: Electromagnetic dispersion curves for β = 0.6, vT2 e /c2 = 0.1, and ω in the vicinity of ωuh :
(a) – in the (λ, ω/|Ωe |) space; (b) – same in the (ω/|Ωe |, N 2 ) space. On the left: the blue dashed
curve corresponds to the electrostatic approximation (14.15) for the EBW with n = 1; the red dashed
line corresponds to ω = |Ωe |. On both figures: the black dashed curves correspond to the cold-plasma
X wave.

To deposit power into the core then, one solution is to launch an X wave from the high-B0 side and
bounce it off the upper-hybrid resonance (S = 0 curve in Fig. 14.4). When the X wave approaches the
upper-hybrid resonance, the wave number k continues to grow (Exercise 14.3). Then, λe eventually
shifts to the right from the local maximum in the dispersion curve [Fig. 14.3(a)] and gradually turns
into the electrostatic EBW. The group velocity changes its sign then,
   2 2
∂ω ∂λe ∂ω k w ∂ω
k · vg = k · = k· = 2
< 0, (14.20)
∂k ∂k ∂λe Ω ∂λe

so the wave starts to propagate backwards. (This is called X–B conversion.) Now that the wave
energy is in the EBW, it is insensitive to the cutoff at L = 0, so it can penetrate the dense plasma
core, as desired.

Exercise 14.3: Using ray equations, explain why k continues to grow as the X wave is trans-
forming into an EBW.

An alternative solution is to launch an O wave at a special angle that ensures complete O–X
mode conversion at the location where ωp2 = ω 2 (Fig. 14.5). Then, the X wave propagates to the
upper-hybrid resonance and transforms into the EBW as usual. (This is called O–X–B conversion.)

14.4 Waves in the lower-hybrid frequency range

14.4.1 Electrostatic dispersion relation
In the lower-hybrid frequency range, one can expect ion Bernstein waves (IBW), with the dispersion
relation qualitatively similar to that of EBW. This is seen from the fact that the electrostatic dispersion

123
LECTURE 14. WAVES IN MAGNETIZED PLASMA 124

Figure 14.4: The CMA diagram for cold-plasma waves (copied from Ref. [1]). The X wave cannot
propagate in the area sandwiched between the curves R = 0 and S = 0, and it also cannot propagate
to the left from the curve L = 0.

124
LECTURE 14. WAVES IN MAGNETIZED PLASMA 125

N⟂ N⟂ N⟂
2 (a) N||2 < N
2 2 (b) N||2 = N
2 2 (c) N||2 > N
2

1 1 1

ω2p /ω2 ω2p /ω2 ω2p /ω2

0.5 1 0.5 1 0.5 1

-1 -1 -1

-2 -2 -2

Figure 14.5: The transverse refraction index for cold-plasma waves in a homogeneous magnetic field
vs. ωp2 /ω 2 . The three figures show different scenarios depending on how the initial Nk2 relates to the
optimum value N̄k2 = |Ωe |/(ω + |Ωe |). The black points mark cutoffs, and the vertical dashed lines
mark the location of the upper-hybrid resonance. See Problem PIII.2 for details.

relation is now
2 ∞ 2
ωpi
e −λi n In (λi )
X
0 = xx = 1 + (χe )xx + An (ζi ). (14.21)
ω n=−∞ λi

Because
λe w 2 Ω2 me
∼ e2 i2 ∼  1, (14.22)
λi w i Ωe mi
2
one can treat electrons as cold for all λi . mi /me , so (χe )xx ≈ ωpe /Ω2e [Eq. (14.13)]. Then, Eq. (14.21)
acquires the same form, up to coefficients, as the corresponding equation for electrostatic EBW:2
2 ∞
ωpi n2 In (λi )
e −λ
X
1+ 2 /Ω2 ω

i
An (ζi ) = 0. (14.23)
1 + ωpe e n=−∞
λi

14.4.2 IBW application to plasma heating

The lower-hybrid frequency range, where IBW appear, is particularly important for fusion appli-
cations. Like in the case of EBW, one can excite these waves using an external antenna, but the
conditions under which an external antenna can couple to lower-hybrid waves is more complicated.
In principle, one can, as with EBW, start at the high-B0 side, so the
p initial gyrofrequency Ωi,out be
p But the target region corresponds to ω ∼ ωlh ∼ |Ωi Ωe |. Because ω is fixed, this
not smaller than ω.
requires Ωi,out ∼ |Ωi Ωe |, i.e.,
r
Ωi,out mi
∼  1. (14.24)
Ωi me

This is impractical, so nonzero Nk is used instead, in which case the upper-hybrid resonance can be
made accessible from vacuum (almost). For further details, see Ref. [63].
2 If
ω  Ωi and λi  1, many terms in the sum (14.23) may be nonnegligible. Then, it may be more convenient to
use the expression that does not involve infinite series [59]. Notwithstanding large λi , the ion susceptibility may then
still be approximated well with the corresponding cold-plasma formula. For example, see Appendix B in Ref. [62].

125
Lecture 15

Collisionless dissipation in
magnetized plasma

In this lecture, we discuss mechanisms of collisionless dissipation of the wave power in magnetized
isotropic Maxwellian plasma without flows.

15.1 Single-particle picture

First, let us discuss collisionless wave absorption by individual particles. Consider a particle with
charge es that follows a given trajectory x(t). The power absorbed by this particle on average over
the field oscillations is given by
D E
P = es ẋ(t) · Ẽ[t, x(t)] . (15.1)

In the presence of a strong magnetic field B 0 = ēz B0 , the particle trajectory can be locally expressed as

x(t) = x0 + ēz vk t + x̃⊥ (t), (15.2)

where vk is the average velocity parallel to B 0 , and x̃⊥ is the transverse quiver displacement of the
particle from a field line that the particle follows on average, hx̃⊥ i = 0. The constant x0 can always
be made zero by redefining the origin. Then, assuming the notation

Ẽ(t, x) = Re E e −i ωt+i kk z+i k⊥ ·x⊥ ,

(15.3)

one can rewrite Eq. (15.1) as follows:

P = Re g(t) e −i (ω−kk vk )t ,
D E
(15.4)

where g is given by

⊥ (t) · E ⊥ + vk E z e
. ˙  i k ·x̃ (t)
g(t) = es x̃ ⊥ ⊥
. (15.5)

Because g is periodic in time with period 2π/Ωs , where Ωs is particle’s gyrofrequency, this function
can be represented as a Fourier series
∞
gn e i nΩs t
X
g(t) = (15.6)
n=−∞

126
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 127

with some coefficients gn . Then,

∞
e −i (ω−nΩ −k v )t
X D E
P = Re gn s k k
, (15.7)
n=−∞

so substantial energy exchange is possible when there exist an integer n such that
ω ≈ kk vk + nΩs . (15.8)
This is called resonant absorption at the nth harmonic.
Different particles have different vk ; however, assuming that the distribution of vk is centered
around zero (i.e., there are no average flows) and Ωs are much larger than kk ws , plasma as a whole
can effectively absorb the wave power only if
ω ≈ nΩs (15.9)
for some of its species s. Below, we show how this condition also flows from general kinetic theory,
and we also discuss how to explicitly calculate the wave absorption in Maxwellian plasma using results
from previous lectures.

15.2 Power absorption in Maxwellian plasma: basic formulas

As shown in Lecture 5, the wave power dissipated per unit volume is given by
ω †
Pabs = E A (ω, k)E , (15.10)
8π
where A is the anti-Hermitian part of the dielectric tensor and ω and k are the GO frequency and
wavevector, which are real by definition. For simplicity, let us assume that the plasma is collisionless,
nonrelativistic, isotropic, Maxwellian, and has no average flows. Then, as shown in Lecture 13,
∞
!
ωp2 X
e Yn ,
X
−λ
=1+ χs , χs = (15.11)
s
ω n=−∞
s

where the matrices Y n are given by

n2 In
 
−i n (In − In0 ) An
k⊥ nIn
An Bn
 λ Ω λ 
i k⊥ (I
  2  
i 0 n 0 0
 
Yn=  n (In − In ) A n In + 2λI n − 2λI n An n − In )Bn
. (15.12)
λ Ω
i k⊥ (I − I 0 )B

 
 k⊥ nIn 2(ω − nΩ) 
Bn − n n n In Bn
Ω λ Ω kk w 2
Here, In = In (λ) are modified Bessel functions of the first kind,
Z(ξn ) Z 0 (ξn ) 2 2
k⊥ w ω − nΩ
An = , Bn = − , λ= , ξn = , (15.13)
kk w 2kk 2Ω2 kk w
Z is the plasma dispersion function,
√
Z(ξ) = −2S(ξ) + i π e −ξ ,
2
(15.14)
and S is the Dawson function (Lecture 11). Because ω and k in Eq. (15.10) are real, only the imaginary
parts of An and Bn contribute to A . But ξn are also real, so Im An and Im Bn are determined entirely
by the second term in Eq. (15.14):
√
Im An =
kk w
e ,
π −ξn2
(15.15a)

127
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 128

√
1 d √ −ξn2 
πe ξn e −ξn = (ξn w) (Im An ).
π 2
Im Bn = − = (15.15b)
2kk d ξn kk

This leads to
√ ∞
!
π ωp2 X −ξn2 −λ
e e Ȳ n
X
A = (χs )A , (χs )A = , (15.16)
s
kk w ω n=−∞
s

where the matrices Ȳ n are given by

n2 In
 
−i n(In −
k⊥ w nIn
In0 ) ξn
λ Ω λ
i k⊥ w (I − I 0 )ξ
 
 2 
Ȳ n =  i n(In − In0 ) n
In + 2λIn − 2λIn0 . (15.17)
 
n n n
λ Ω
i k⊥ w (I − I 0 )ξ
 
 k w nI 
⊥ n
ξn − n n n 2In ξn2
Ω λ Ω
In a sufficiently cold plasma, ξn are large everywhere except at ω ≈ nΩs , where n is integer (n =
0, ±1, ±2, . . .). Then, collisionless dissipation is noticeable only at ω ≈ nΩs , in agreement with
Eq. (15.9). Below, we discuss qualitative physical mechanisms that contribute to Pabs described by
the above formulas.

15.3 Landau damping

Electrostatic waves polarized along B 0 are similar to the electrostatic waves in nonmagnetized plasma
(Problem PIV.7). Then, by analogy with Lecture 12, one can identify the E z -driven absorption at
the zeroth harmonic (n = 0) as Landau damping (LD). The corresponding absorption power density
is given by
ω
PLD = (A,zz )n=0 |E z |2 . (15.18)
8π
The corresponding power absorbed by species s is
ω
PLD,s = (χA,zz )s,n=0 |E z |2
8π
√
π ωp2 −ξ02 −λ

=
ω
8π kk w ω
e e 2I0 (λ)ξ02 |E z |2
s
ω|E z |2 ωp2  −λ
 
e I0 (λ) ξ0 e
 3 −ξ2

= √ 0 . (15.19)
4 π ω2 s

Note that PLD,s is maximized at λs → 0, in which case it is given by

ω|E z |2 ωp2
 
h(ξ0 ) = ξ03 e −ξ0 .
. 2
PLD,s = √ h(ξ0 ) , (15.20)
4 π ω2 s
1/2
In a cold plasma, ξ0 is large, so h(ξ0 ) is exponentially small; then because ξs ∝ ws−1 ∝ ms , the
1/2
electron contribution dominates. In hot plasma, ξ0 is small, so h(ξ0 ) ≈ ξ03 , and PLD,s ∝ ms ; in this
case, the ion contribution dominates.

15.4 Transit-time magnetic pumping

For electromagnetic waves, there is an additional dissipation mechanism that is similar to Landau
damping but determined by A,yy . It can be understood as follows. Linear waves with nonzero k⊥

128
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 129

kÞ = 0 kÞ ¹ 0

Figure 15.1: A schematic of magnetic field lines in magnetized plasma perturbed by an electromagnetic
wave: left k⊥ = 0, right – k⊥ 6= 0. The x axis is vertical, and the z axis is horizontal.

.
produce oscillations of the magnitude of the total field B = |B 0 + B̃| (Fig. 15.1):
q
B̃ = B̃⊥ 2 + (B + B̃)2 − B
0 0
q
≈ B02 + 2B 0 · B̃ − B0
c c
≈ B̃z = (k × Ẽ)z = k⊥ Ẽy . (15.21)
ω ω
For simplicity, suppose that ω  Ωs . Then, the oscillations of B produce an oscillating diamagnetic
2
force F̃ ≈ −µ ∂z B̃ on each particle along the z axis, where µ ≈ ms v⊥ /2B0 is particle’s magnetic
moment. In the complex form, one can write the ensemble-average of this force as

hF̃ is = −i kk hµis B̃z = −i kk

Ts c
k⊥ Ẽy , (15.22)
B0 ω

which can be viewed as an effective longitudinal electric field, Ẽeff,s = hF̃ is /es . Like the true longitu-
dinal electric field, Ẽeff,s produces Landau damping, which is known as transit-time magnetic pumping
(TTPM). Accordingly,

PTTMP,s ∼ PLD,eff,s
!
ω|Ẽeff,s |2 ωp2 3 −ξ2
= √ ξ e 0
4 π ω2 0
s
!
ω|Ẽy |2 c2 T 2 ωp2 ω 2 ω
= √ kk2 k⊥
4 π
2
e2 B02 ω 2 ω 2 kk2 w2 kk w
e −ξ02

s
!
2 2 ω2
ω|Ẽy | T 2 k⊥
= √
4 π
p ω
m 2 w 2 Ω2 ω 2 k k w
e −ξ0
2

s
!
ω|Ẽy |2 ωp2
λ 2 ξ0 e −ξ0
2
= √ . (15.23)
8 π ω
s

In particular, notice that TTMP is more efficient in high-β plasmas, because

!
|Ẽy |2 ωp2 ω −ξ2
PTTMP,s ∼ √
8 π
λ
ω kk w
e 0
s
ω|B̃z |2 T 4πn0 e2
 
= √ ξ0 e −ξ02
8 π mc2 Ω2 m s
2
 
ω|B̃z |
ξ0 e −ξ0
4πn0 T 2
= √
8 π B02 s

129
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 130

ω|B̃z |2
√ (βξ0 e −ξ0 )s .
2
= (15.24)
8 π
√
The maximum of this function is achieved at ξ0 ∼ 1, so PTTMP, max ∼ βω|B̃z |2 /(8 π).
A more rigorous expression for the TTMP comes from the general formula (15.10), where it
corresponds to the contribution of E y and n = 0:
ω
PTTMP = (A,yy )n=0 |E y |2 , (15.25)
8π
which leads to a result consistent with Eq. (15.23). That said, generally, one should also take into
account the “cross terms” Pyz,n=0 and Pzy,n=0 , which can be of the same order as Pyy,n=0 and
Pzz,n=0 [64]. In this sense, Landau damping and TTMP are not always separable.

15.5 Cyclotron damping

Another distinctive mechanism of collisionless dissipation is cyclotron damping (CD), which corre-
sponds to |n| ≥ 1. The basic features of cyclotron damping can be understood within a simple model
when Ẽz is negligible and λs is small. Then the power absorbed through CD by species s per unit
volume is given by (Problem PIV.9)
∞  n−1 ( )
2 2
√ ωps 2 |Ẽ+ | |Ẽ− |2
n λs
e +e
2
X
PCD,s ≈ π −ξn −ξ−n
. (15.26)
kk ws n=1 (n − 1)! 2 16π 16π
s

Note that the coefficients decrease rapidly with n, so the absorption at high-order resonances is
relatively weak. Also, efficient heating is possible only when (ξ±n )s are small enough, i.e., ω = ±nΩs ,
and the field is polarized such that its corresponding component Ẽ± = Ẽx ± i Ẽy does not vanish.
.
Like Landau damping, cyclotron heating can saturate through nonlinear effects. To understand
these effects qualitatively, let us consider a simple model in which a transverse wave propagates strictly
parallel to B 0 , i.e., k = kk ēz . The wave electric field Ẽ 0 in the frame moving at the phase velocity
v p = (ω/kk )ēz can be expressed through the wave fields in the laboratory frame using the well known
relativistic transformation [9]. Specifically, one finds that
vp
γp−1 Ẽ 0 = Ẽ + × B̃
c h
ω c i
= Ẽ + ēz × k × Ẽ
ck ω
= Ẽ + ēz × (ēz × Ẽ)
= Ẽ + ēz (ēz · Ẽ) − Ẽ(ēz · ēz )
= 0, (15.27)

where γp is the Lorentz factor associated with v p . Then, the particle energy in this frame is conserved,
0 2
(v⊥ ) + (vk0 )2 = const. (15.28)

(From now on, we assume that both v 0 and vp are nonrelativistic.) In the laboratory-frame variables,
this can be expressed as follows:

(v⊥ )2 + (vk − ω/kk )2 = const. (15.29)

This describes a (semi-)circle in the (vk , v⊥ ) plane shifted by ω/kk along vk axis (Fig. 15.2). Depending
on the initial phase at which a particle enters the wave, the particle may gain or lose energy from the
wave and thus can move up or down such a circle. This causes diffusion of the particle distribution

130
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 131

Figure 15.2: A schematic of the diffusion paths (sold curves) given by Eq. (15.29). The vertical red
line denotes the resonance (15.30). Also shown are isosurfaces of the Maxwellian distribution (dashed
curves). At sufficiently high ω/(kk ws ), the wave heats particles with large vk2 , and the majority of
2
those have low v⊥ (because the distribution function falls off as ∼ exp[−(vk2 + v⊥
2
)/ws2 ]), so on average,
particles are heated up. The shaded region corresponds to the trapping area.

in the velocity space along the circles (15.29), which are hence called diffusion paths. The diffusion
coefficient depends on the distance from the resonance (15.8) and is maximized at

vk = (ω − nΩs )/kk . (15.30)

To better understand how individual particles interact with the wave, let us differentiate Eq. (15.29)
with respect to time; then, one obtains
d vk 1 d v⊥2
 
=−
dt vk − ω/kk d t 2

=
kk
v⊥ ·
d v⊥
ω(1 − kvk /ω) dt
 
kk es 1 1
= v⊥ · Ẽ + v × B̃ + v × B 0 . (15.31)
ω(1 − kvk /ω) ms c c ⊥

Notice that
   
1 1
Ẽ + v × B̃ = Ẽ + v × (k × Ẽ)
c ⊥ ω ⊥
 
k(v · Ẽ) (k · v)
= Ẽ + − Ẽ
ω ω ⊥
 
kvk (v · Ẽ)
= Ẽ 1 − + k⊥
ω ω
 
k k vk
= Ẽ 1 − (15.32)
ω

(remember that k⊥ = 0 has been assumed) and

v ⊥ · (v × B 0 )⊥ = v ⊥ · (v ⊥ × B 0 )⊥ = (v ⊥ × v ⊥ ) · B 0 = 0. (15.33)

131
LECTURE 15. COLLISIONLESS DISSIPATION IN MAGNETIZED PLASMA 132

Hence, we obtain

d vk kk es

kvk

es kk v ⊥
= v⊥ · Ẽ 1 − = · Ẽ. (15.34)
dt ω(1 − kk vk /ω) ms ω ms ω

Suppose that the field is circularly polarized,

Ẽ = Re (ēx ∓ i ēy )E e −i ωt+i kz .

 
(15.35)

Then,

v ⊥ · Ẽ = |E | Re (v x ∓ i v y )e −i ωt+i kk z+i arg E

 
(15.36)
= v⊥ |E | sin[kk z − (ω ∓ Ωs )t + const], (15.37)

where we used that vx ∓ i vy ∝ e ±i Ωs t (cf. Sec. 6.1). Substituting this into Eq. (15.34) leads to

d 2z0 =
es (eff)
E sin(kk z 0 ), (15.38)
dt ms 0
where
. ω ∓ Ωs (eff) . k k v⊥
z0 = z − t + const, E0 = |E |, (15.39)
kk ω

and kk > 0 will be assumed for clarity. This is similar to Eq. (12.1) that we used to explore the
nonlinear saturation of Landau damping earlier; the only difference is that the resonance is now
(eff)
shifted by Ωs and E0 is replaced with E0 . This means that the nonlinear saturation mechanism for
cyclotron damping is similar to that of Landau damping, except that the trapping width (Fig. 15.2)
(eff)
is now of order (es E0 /ms kk )1/2 and the corresponding bounce time is
s s  
ms ms ω
T ∼ (eff)
= . (15.40)
|es E |kk |es E0 |kk kk v⊥
0

If v⊥ is small, reaching the nonlinear stage through cyclotron damping takes longer than at Landau
damping (for given kk es E0 /ms ). This is understood from the fact that a resonant particle with small
v⊥ travels primarily upward in Fig. 15.2, so it takes longer for vk to become nonresonant.
This picture can be significantly modified, though, when waves have a broad spectrum. A particle
interacting with multiple waves that have different phase velocities is not necessarily constrained to
a single diffusion path but can diffuse in wider regions of the velocity space. Then, the nonlinear
saturation of the wave absorption is governed by a very different equations. This regime will be
discussed, within a simpler model, in Lecture 16.

132
Lecture 16

Quasilinear theory

In this lecture, we discuss the nonlinear evolution of the particle distribution driven by self-consistent
waves with a sufficiently wide and dense spectrum. For simplicity, one-dimensional nonmagnetized
electron plasma will be assumed.

16.1 Introduction
16.1.1 One wave: nonlinearities due to trapped particles
The linear-wave approximation used in previous lectures implies

fs ≈ fs0 + fs1 , fs1 = O(Ẽ). (16.1)

As discussed in Lecture 12, this approximation holds in collisionless plasma only on time scales t .
Ω−1
b0 , where Ωb0 ∝ Ẽ
1/2
is the bounce frequency. At t & Ω−1 b0 , the distribution function acquires
structure with the characteristic velocity scale equal to the trapping-island size vt ∼ Ωb0 /k. Then,
near the resonance, ∂v fs scales as fs /vt ∼ Ẽ −1/2 . This makes the third term in the Vlasov equation

∂fs ∂fs es ∂fs

+v + Ẽ =0 (16.2)
∂t ∂x ms ∂v

scale as Ẽ∂v fs ∼ Ẽ 1/2 . Then, it cannot be balanced by the

Pother terms if fs contains only integer
powers of Ẽ. This means that the standard expansion f˜s = n f˜sn with f˜sn = O(Ẽ n ) is inapplicable
in this case, and the problem becomes complicated. Surprisingly, though, if particles interact with
many waves simultaneously, then the problem is simplified. This is understood as follows.

16.1.2 Two waves: Chirikov criterion

First, let us consider the case when particles interact with only two waves. Individually, each wave
would produce phase-space structures shown in Fig. 16.1. The two sets of trapping islands can be
made stationary if one considers them in the reference frames traveling with the corresponding phase
velocities, vp1 and vp2 . But in the case of two different phase velocities, there is no reference frame
where both wave fields are stationary simultaneously. Hence, two different regimes can be realized
depending on whether |vp1 − vp2 | is larger or smaller (comparable) than the island widths, namely,

vt1 + vt1 ≶ |vp1 − vp2 |, (16.3)

which is known as the Chirikov criterion. If the Chirikov criterion is not satisfied (vt1 + vt1 
|vp1 − vp2 |), then any given particle can be resonant to only one of the waves. In this case, each wave

133
LECTURE 16. QUASILINEAR THEORY 134

υ υ
(a) (b)

υ p1 υ p1
υ p2

υ p2

x x

Figure 16.1: A schematic of the trapping islands produced by two waves with different phase velocities,
vp1 and vp2 : (a) the Chirikov criterion is not satisfied (shaded are individual trapping islands); (b) the
Chirikov criterion is satisfied (shaded is the region of stochastic motion).

saturates more or less independently from the other one, i.e., as described in Lecture 12. But if the
Chirikov criterion is satisfied (vt1 + vt1 . |vp1 − vp2 |), then the islands overlap and trapped particles
no longer “belong” to a particular wave but are rather “shared” by the two waves. The trajectories
of such particles are generally stochastic and extend to the whole interval vp2 − vt2 . v . vp1 + vt1 .
Then, clearly, if the two waves saturate, they saturate together.

16.1.3 Many waves: statistical quasilinear approach

Now, suppose that there are not two but N  1 waves, which create a stochastic region in the velocity
space with the width ∆v ∼ N vt = O(Ẽ 0 ). Such broad-band spectra are naturally formed, for instance,
through bump-on-tail instabilities, with phase velocities in the range where f00 (v) > 0, assuming that
the mode spectrum is sufficiently dense.1 Then, Ẽ∂v fs ∼ Ẽfs /(N vt ) ∝ Ẽfs , so like in linear theory,
one can search for fs in the form (16.1). Even though the particle dynamics is complicated (stochastic)
in this case, a relatively simple statistical theory becomes possible if Ẽ is small enough. This theory
is known as quasilinear theory.
Below, we construct quasilinear theory assuming that the Chirikov criterion is satisfied, ions are
motionless, and plasma is not magnetized. (Adding magnetic field and allowing for multiple species
does not affect affect the physics qualitatively. For example, as discussed in Lecture 15, the equations
that describe cyclotron damping are similar to those describing Landau damping up to notation.) In
Sec. 16.2, we will describe the “standard” version of quasilinear theory, which is relatively easy to
derive but is oversimplified in some respects. A more rigorous, and more complete, version of this
theory will be outlined in Lecture 17.

16.2 Basic equations

In this section, we present a simplified version of quasilinear theory [1,65,66], which uses two additional
assumptions: (i) the plasma is homogeneous; (ii) the wave field evolves much faster than the linear
frequencies. (The second assumption is not always articulated in literature.) For simplicity, we will
also assume that the plasma is one-dimensional.
1 Whether the spectrum is dense or not depends on the boundary conditions. In unbounded plasma, any wavenumbers

are allowed, so the distribution of vp is continuous; then, the Chirikov criterion is automatically satisfied even at
vanishingly small amplitudes. In contrast, if the boundary conditions are periodic (as, for example, in a tokamak for
toroidal and poloidal modes), then vp are quantized and thus satisfying the Chirikov criterion requires finite amplitudes.

134
LECTURE 16. QUASILINEAR THEORY 135

16.2.1 Equation for the distribution function

Let us start with the Vlasov equation for the electron distribution
∂f ∂f ee ∂f
+v + Ẽ = 0. (16.4)
∂t ∂x me ∂v
(Because only electrons are considered, the index e in the distribution function is omitted.). Assuming
the Chirikov criterion is satisfied for many waves simultaneously, leading to stochastization of the
particle dynamics in a wide range of velocities ∆v = O(E 0 ), one can search for a solution in the form
f = f0 (t, v) + f˜(t, x, v), where f˜ is small. Then, Eq. (16.4) becomes

∂f0 ∂ f˜ ∂ f˜ ee ∂f0 ee ∂ f˜
+ +v + Ẽ + Ẽ = 0. (16.5)
∂t ∂t ∂x me ∂v me ∂v
By performing spatial averaging, one obtains the following equation for f0 :
* +
∂f0 ee ∂ f˜
+ Ẽ = 0. (16.6)
∂t me ∂v

The equation for f˜ is obtained by subtracting the equation for f0 from the original equation:
* +
∂ f˜ ∂ f˜ ee ∂f0 ee ∂ f˜ ee ∂ f˜
+v + Ẽ + Ẽ − Ẽ = 0. (16.7)
∂t ∂x me ∂v me ∂v me ∂v
| {z }
N

When the dynamics is regular, a small field can, in principle, make f˜ order-one over large enough
time. But here, we assume that the particle dynamics is stochastic, so correlations die out fast; then f˜
remains small at all times and the term N , which is nonlinear yet has zero average, can be omitted.
This is called the quasilinear approximation, and it leads to

∂ f˜ ∂ f˜ ee ∂f0
+v + Ẽ ≈ 0. (16.8)
∂t ∂x me ∂v
By applying the spatial Fourier transform to this equation, we also obtain

∂ f˜k (t, x, v)
+ i kv f˜k (t, x, v) +
ee ∂f0 (t, v)
Ẽk (t) = 0. (16.9)
∂t me ∂v
As discussed earlier (Sec. 8.3), the general solution of this equation can be written as
 t 0
f˜k (t, v) = f˜k (t0 , v) e −i kvt − e −i kvt d t0 e i kvt Ẽk (t0 ) ∂f0∂v
ee 0 (t , v)
. (16.10)
me t0

The function f0 (t0 , v) in the integrand is slow, whereas e i kvt and Ẽk (t0 ) are rapidly oscillating. Let
0

us express the latter as follows:

Ẽk (t) = Ẽk,0 e i θk (t) , (16.11)

where Ẽk,0 is a constant, and θk is a complex phase, so one can introduce a local complex frequency

ωk (t) = −dt θk (t) = −dt arg[Ẽk (t)] + idt ln |Ẽk (t)| ≡ ωk,r + i ωk,i .
.
(16.12)

Since the actual field is real, one has

 
Ẽ(t, x) =
∞
d k Ẽ ei kx = Ẽ ∗ (t, x) =
∞
d k Ẽ ∗ (t) e −i kx =  ∞
d k Ẽ ∗ e i kx .
k (t) k −k (t) (16.13)
−∞ 2π −∞ 2π −∞ 2π

135
LECTURE 16. QUASILINEAR THEORY 136

Box 16.1: Asymptotic representation of integrals of rapidly oscillating functions

.
t
Consider an integral of the form J (t) = t0 d τ e i ϑ(τ ) F (τ ), where F and ϑ0 are slow compared
with e i ϑ ; namely, ε = 1/(ϑ0 Tc )  1, where Tc is the characteristic time of F and ϑ0 . Then,
.


d τ iFϑ0(τ(τ)) d ed τ
t i ϑ(τ )
J (t) =
t0

t t
d F (τ )
 

i ϑ (τ ) e d e
F (τ ) i ϑ(τ )
= 0 − τ i ϑ(τ )
t0 t0 d| τ i{zϑ0 (τ ) }
O(εF )

i ϑ0 (t) e + const + O(εJ ).

F (t) i ϑ(t)
=

∗
Thus, Ẽ−k = Ẽk . This means that and ω−k = −ωk∗ = −ωk,r + i ωk,i , so

ω−k,r = −ωk,r , ω−k,i = ωk,i . (16.14)

Because ωk and f0 are assumed slow compared to e i θk (t) , one can represent f˜k as2 (Box 16.1)

f˜k (t, v) = g̃k (v) e −i kvt −

ee Ẽk (t) ∂f0 (t, v)
i me [kv − ωk (t)] ∂v . (16.15)

Assuming ωk,i > 0, the second term eventually dominates, so one obtains

ee Ẽk (t) ∂f0 (t, v)

f˜k (t, v) ≈ − . (16.16)
i me [kv − ωk (t)] ∂v

The actual distribution f˜(t, x, v) is obtained by taking the inverse Fourier transform of f˜k (t, v):
 
f˜(t, x, v) =
1 ∞
d k f˜k (t, v) e i kx = − mee
∞
dk Ẽk (t) e i kx ∂f0 (t, v)
. (16.17)
2π −∞ e −∞ 2π i [kv − ωk (t)] ∂v

Let us substitute this and the expression for Ẽ(t, x) into the equation for f0 :
* +
∂f0 ee ∂ f˜
=− Ẽ
∂t me ∂v
 ∞
d k Ẽk (t)e i kx ∂f0 (t, v)
* +
∂ e2
= Ẽ(t, x)
∂v m2e −∞ 2π i [kv − ωk (t)] ∂v
* 
d d k Ẽk (t)e i kx
+ !
∞ ∞
e2 k0 ∗
Ẽk0 (t)e
∂ −i k 0 x ∂f0 (t, v)
=
∂v m2e −∞ 2π −∞ 2π i [kv − ωk (t)] ∂v
 
d k 0 ∞ d k Ẽk∗0 (t)Ẽk (t) D i (k−k0 )x E ∂f0 (t, v)
!
∞
e2
=
∂
∂v m2e −∞ 2π −∞ 2π i [kv − ωk (t)]
e ∂v
 ∞ 
d k 0 ∞ d k Ẽk∗0 (t)Ẽk (t) 2π
!
∂ e2 0 ∂f0 (t, v)
= δ(k − k )
∂v m2e −∞ 2π −∞ 2π i [kv − ωk (t)] L ∂v
2 Although Eq. (16.15) is correct as the leading-order approximation, retaining corrections O(∂t f0 ) and O(∂t ωk ) is
generally necessary to keep the quasilinear model truly conservative. For Langmuir turbulence, this is not a big issue
somewhat accidentally. See the footnotes below, also see Lecture 17.

136
LECTURE 16. QUASILINEAR THEORY 137


dk
!
∞
∂ e2 |Ẽk (t)|2 ∂f0 (t, v)
= . (16.18)
∂v m2e −∞ 2πL i [kv − ωk (t)] ∂v

Here, we used that the plasma length L is assumed large enough, so

 L/2
e i (k−k )x 1
d x e i (k−k )x ≈ 2π
D 0
E 0
= δ(k − k 0 ). (16.19)
L −L/2 L
For ωk,i ≤ 0, the first term on the right-hand side of Eq. (16.15) may not be small compared to
the second term, but we will extrapolate the final result to this case using analytic continuation, as
usual. For this reason, the integration over k will, from now on, be done over the Landau contour L.

16.2.2 Diffusion coefficient

Hence, the equation for f0 is seen to be a diffusion equation,
 
∂f0 ∂ ∂f0 (t, v)
= D(t, v) , (16.20)
∂t ∂v ∂v
where

. e2
D(t, v) = 2
dk |Ẽk (t)|2
(16.21)
me L 2πL i [kv − ωk (t)]
serves as the diffusion coefficient. It is also common to express D as follows:

8πe2 1 |Ẽk (t)|2
D(t, v) =
m2e
d k i [kvU−k (t)
ω (t)]
,
.
Uk =
2πL 8π
, (16.22)
L k

and Uk is understood as the spectral density of the electric-field energy per unit volume, because
 ∞  ∞
d k0 ∗ d
 2 
Ẽk0 (t)e Ẽk (t)e
Ẽ 1 −i k 0 x k i kx
=
8π 8π −∞ 2π −∞ 2π
 ∞ 
=
1 d k0 ∞ d k Ẽ ∗ (t)Ẽ (t) 2π δ(k − k0 )
0 k
8π −∞ 2π −∞ 2π k L
 ∞
=
1 d k |Ẽ (t)|2
k
8π −∞ 2πL
 ∞
= d k Uk . (16.23)
−∞

Although Eq. (16.22) contains imaginary unit, the corresponding D is real, as easily seen from the
fact that Uk = U−k and ω−k = −ωk∗ . Furthermore, to the extent that the nonzero value of ωk,i is
negligible, one readily obtains, using Landau’s rule, that

. 8πe2
D(t, v) ≈ D̄(t, v) = Im d k kv − Uωk (t)
m2e L k (t) − i 0
2 2  ∞
δ(kv − ωk,r ) Uk d k.
16π e
= (16.24)
m2e 0

16.2.3 Field equations

To describe particles and fields self-consistently, the diffusion equation must be complemented with
equations for Uk . According to the equations that we introduced earlier, one has
d Uk =
1 1 d |Ẽk (t)|2
dt 2πL 8π dt
137
LECTURE 16. QUASILINEAR THEORY 138

1 1 d |Ẽk,0 e i θk (t) |2
=
2πL 8π dt
1 |Ẽk,0 e i Re θk (t) |2 de −2Im θk (t)
=
2πL 8π dt
1 |Ẽk,0 e i θk (t) 2
|
= 2ωk,i , (16.25)
2πL 8π
or in other words,3
d Uk = 2ωk,i Uk . (16.26)
dt
Since f˜ that we derived earlier is governed by the same equation as in stationary plasma, the frequen-
cies ωk (t) can be found using the GO dispersion relation:

4πe2
1−
me k 2
d v v∂−v fω0 (t,(t)/k
v)
= 0. (16.27)
L k

16.3 Properties and applications of quasilinear theory

16.3.1 Conservation laws
In summary, we have derived a closed set of equations that determine the self-consistent evolution of
the particle distribution and the field spectrum:
d Uk = 2ω U ,
 
∂f0 ∂ ∂f0 (t, v)
= D(t, v) , (16.28)
∂t ∂v ∂v dt k,i k

where D is given by Eq. (16.22) and ωk is found from Eq. (16.27). These equations conserve4 the total
number of particles:
d  ∞ d v f (t, v) =  ∞ d v ∂ D ∂f0 = 0,
 
(16.29)
d t −∞ 0
−∞ ∂v ∂v
the total momentum:
d  ∞ d v m vf (t, v) =  ∞ d v m v ∂ D ∂f0
 

d t −∞ e 0
−∞
e
∂v ∂v
 ∞
=− d v me D ∂f
∂v
0

−∞
 ∞  2
=− d v d k i (kv2U−kω ) 4πe
me
∂f0
∂v
−∞ L k
 ∞ 2 
= 2i d k Uk 4πe d v kv∂v−f0ω
−∞ me L k
 ∞ 
4πe2
= 2i d k kUk m 2
d v v −∂vωf0/k
−∞ ek L k
 ∞
= 2i d k kUk = 0 (16.30)
−∞
3 Strictly speaking, Eq. (16.26) should be the equation for the mode action I rather than for U . For Langmuir
k k
waves in homogeneous plasma, Ik ∝ Uk /ωk and ωk ≈ ωp = const, so the equations for Ik and Uk are usually not
distinguished. A more rigorous approach is described in Lecture 17.
4 As discussed in footnotes 2 and 3, the standard quasilinear theory presented here is oversimplified in two aspects.

It remains conservative because the two oversimplifications cancel each other.

138
LECTURE 16. QUASILINEAR THEORY 139

(where we substituted the dispersion relation and used the fact that Uk = U−k ), and the total energy:

d  ∞ d v me v2 f (t, v) +  ∞ d k U (t)
 

d t −∞ 2
0
−∞
k
 ∞ 2 ∞ 
= d v me2v ∂t f0 + d k U̇k
−∞ −∞
 ∞ 2
   ∞
= d v me2v ∂v∂
D
∂f0
∂v
+ d k 2ωk,i Uk
−∞ −∞
 ∞  ∞
=− d v me vD ∂f
∂v
0
+ d k 2ωk,i Uk
−∞ −∞
 ∞  2  ∞
=− d k d v i (kv2U−k ω ) 4πe v ∂f0
me ∂v
+ d k 2ωk,i Uk
−∞ L k −∞
 ∞
= 2i d k (ωk − i ωk,i )Uk
−∞
 ∞
= 2i d k ωk,r Uk
−∞
= 0. (16.31)

In the latter case, we used ω−k,r = −ωk,r , Uk = U−k , and also

 
4πe2 4πe2 − ωk + ωk ∂f0
me
d v kv −v ω ∂f0
∂v
=
me k L
d v kv kv − ωk ∂v
L k
2  
 
=
4πe
me k L
d v
∂f0
∂v
+
ωk
k L
d v
∂ v f0
kv − ωk
2 
 
= ωk 0 +
4πe
me k 2 L
d v
∂v f0
v − ωk /k
= ωk . (16.32)

Let us summarize the above equations concisely:

d  d v f (t, v) = 0, (16.33a)
dt 0

d

d t d v me vf0 (t, v) = 0,

(16.33b)

d 2 ∞ 

dt d v me2v f0 (t, v) + d k Uk (t) = 0. (16.33c)

−∞

Because f0 evolves due to Uk , this theory is not entirely linear. However, since the field spectrum is
determined by the local linear dispersion relation, this theory is called quasilinear.

16.3.2 Quasilinear evolution: broadband bump-on-tail instability

Let us return to discussing a one-dimensional broad-band bump-on-tail instability. Suppose the wave
spectrum is largest initially at some k. This produces flattening at velocities close to ωk,r /k. But then
the slope of the distribution increases at neighboring velocities. Due to this increase of f00 , the local ωi
increase too. This causes amplification of the wave field and increase of the diffusion coefficient, even
in regions where it was zero initially. Hence, the flattening proceeds until the distribution becomes
monotonic. This leads to formation of a “quasilinear plateau”, whose height is determined by the
particle conservation (Fig. 16.2).

139
LECTURE 16. QUASILINEAR THEORY 140

f f
(a) (b)

υ υ

Figure 16.2: Quasilinear evolution of the distribution with a bump on tail: (a) initial distribution;
(b) saturated distribution with a quasilinear plateau. The areas between the dashed curve and the
solid curve in (b) are equal as required by particle conservation.

Note that the formation of a quasilinear plateau indicates that the average momentum of the
resonant particles decreases. But the electrostatic field carries no momentum, so this momentum
change must be absorbed by the background distribution, which apparently experiences adiabatic
transformation while tail particles interact with the waves resonantly. This transformation can be
explained ad hoc by the difference between D and D̄ (16.24) [1]. A more systematic approach is
presented in Lecture 17.

140
Lecture 17

Dressed particles and

ponderomotive forces

In this lecture, we generalize quasilinear theory presented in Lecture 16. In particular, we introduce
the concept of dressed particles (also called oscillation centers) and ponderomotive effects.

17.1 Quasilinear theory for inhomogeneous plasma

A more accurate version of quasilinear theory, which properly describes the field evolution and allows
for plasma inhomogeneity, was proposed in Refs. [67, 68]. Those derivations are partly based on
heuristic arguments. Here, we briefly outline the derivation of the main equations using the same
operator approach that was used throughout the course. A generalized version of this calculation (not
limited to electrostatic waves in electron plasma) and a broader discussion can be found in Ref. [69].

17.1.1 Equation for the average distribution

As in the previous lecture, let us assume collisionless nonmagnetized electron plasma where the
Chirikov criterion is satisfied but the electric field is weak and the quasilinear approximation is ap-
plicable. For generality, we will allow plasma to be spatially inhomogeneous and not necessarily
one-dimensional. We will also work with the momentum distribution instead of the velocity distribu-
tion to make the equations more compact. Like in Sec. 16.2, we start with
∂ f˜
 
∂f0 ∂f0
+v· = −ee Ẽ · , (17.1a)
∂t ∂x ∂p
∂ f˜ ∂ f˜ ∂f0
+v· + ee Ẽ · = 0, (17.1b)
∂t ∂x ∂p
where h. . .i now is the statistical average. Using the same argument as in Sec. 16.2.1, let us neglect
the initial conditions in Eq. (17.1b). Then, using the propagator Ĝ of the linearized Vlasov equation,
the solution of Eq. (17.1b) can be written as
 
˜ ∂f0 ∂f0
f = −ee Ĝ Ẽ · ≡ −ee ĜÊ · , (17.2)
∂p ∂p
.
where Ê = Ẽ(t̂, x̂) is the electric field considered as an operator. [In the coordinate representation,
Ê a simply performs multiplication by Ẽa (t, x).] Accordingly, Eq. (17.1a) becomes
 
∂f0 ∂f0 ∂ ∂f0
+v· = D̂ab , (17.3)
∂t ∂x ∂pa ∂pb

141
LECTURE 17. DRESSED PARTICLES AND PONDEROMOTIVE FORCES 142

.
D̂ab = e2 hÊ a ĜÊ b i . (17.4)

It is straightforward to show that the Weyl symbol of D̂ab can be expressed through the Weyl symbol
of Ĝ as follows:

D(t, x, ω, k) = e2 d ω0 d k0 G(ω − ω0 , k − k0 ) hW E i(t, x, ω0 , k0 ). (17.5)

Here, hW E i is the average Wigner tensor of the electric field, i.e., the spectrum of the symmetrized
.
two-point correlation matrix Cab (t, x, τ, s) = hẼa (t + τ /2, x + s/2) Ẽb (t − τ /2, x − s/2)i:

hWE,ab i(t, x, ω, k) =
1
(2π)n+1
d τ d s Cab (t, x, τ, s) e −i k·s+i ωτ , (17.6)

.
which is also understood as the average Weyl symbol of Ŵ E = (2π)−(n+1) |Ẽi hẼ|. (Here, n is
the number of spatial dimensions.) Because the field is electrostatic, it is convenient to express
hW E i through the average Wigner function of the electrostatic potential, i.e., the spectrum of the
.
symmetrized two-point correlation function Cϕ (t, x, τ, s) = hϕ̃(t + τ /2, x + s/2) ϕ̃(t − τ /2, x − s/2)i:

hWϕ i(t, x, ω, k) =
1
(2π)n+1
d τ d s Cϕ (t, x, τ, s) e −i k·s+i ωτ , (17.7)

.
which is the average Weyl symbol of Ŵ ϕ = (2π)−(n+1) |ϕ̃i hϕ̃|. Because

Ŵ E = k̂ Ŵ ϕ k̂, (17.8)

one readily finds, using Eqs. (3.24) and (3.25), that

i k ∂hWϕ i − k ∂hWϕ i ,
 
hWE,ab i ≈ ka kb hWϕ i − b a (17.9)
2 ∂xa ∂xb

assuming the averaged quantities are smooth enough such that GO approximation is applicable.1
Similarly, because f0 is only weakly inhomogeneous, one can Weyl-expand D̂ like in Lecture 3, that
is, by using

D(t, x, ω, k) ≈ D(t, x, 0, 0) + ω ∂ω D(t, x, 0, 0) + (k · ∂k )D(t, x, 0, 0) (17.10)

and then applying the inverse Wigner–Weyl transform. After a tedious but straightforward calculation,
one arrives at the following diffusion equation:
 
∂F ∂H ∂F ∂H ∂F ∂ ∂F
+ · − · = D̄ab , (17.11)
∂t ∂p ∂x ∂x ∂p ∂pa ∂pb

D̄ab = πe 2
d ω d k ka kb hWϕ i(t, x, k · v, k), (17.12)

where nonlinearities beyond that of the second order in Ẽ have been neglected. (Note that Wϕ is
evaluated at ω = k · v, which indicates that diffusion is governed by resonant particles.) Here,

p2
 
1 ∂ ∂f0
F = f0 + Vab , H= + Φ. (17.13)
2 ∂pa ∂pb 2me
1 Strictly speaking, W can be delta-shaped [Eq. (17.17)], but we will be interested only in integrals of W . Con-
ϕ ϕ
tributions of higher-order derivatives of Wϕ to those integrals is small even for delta-shaped Wϕ when J and ωr are
smooth functions.

142
LECTURE 17. DRESSED PARTICLES AND PONDEROMOTIVE FORCES 143

The function H serves as an effective OC Hamiltonian, and Φ is known as the ponderomotive potential,
or more precisely, the ponderomotive energy of an OC. Specifically,
e2 ka kb hWϕ i(t, x, ω, k)
 
Vab =
∂
∂ϑ
d ω dk
ω−k·v+ϑ
, (17.14a)
ϑ=0
2
khWϕ i(t, x, ω, k)
Φ=
∂
∂p
dω dk e 2me (ω − k · v)
. (17.14b)

.
Also note that the OC velocity is not quite v = p/me but is rather given by

. ∂H p ∂Φ ∂Φ
V = = + ≡v+ . (17.15)
∂p me ∂p ∂p

(Because V − v = O(Ẽ 2 ), the velocities V and v are interchangeable in Eqs. (17.12) and (17.14)
within the accuracy of this theory.) Similar equations can be derived for plasma interaction with
electromagnetic waves, except the formulas for D̄ab , Vab , and Φ will be different.

17.1.2 Oscillation centers

The function F can be understood as the distribution of “oscillation centers” (OC). An OC is an
effective (“dressed”) particle whose coordinate coincides with the coordinate of the true particles
averaged over the high-frequency oscillations of the particle in the ambient wave field. Unlike a
true free particle, an OC experiences effective forces determined by the ponderomotive energy. For
example, in the case of a quasimonochromatic wave, Eq. (17.14b) reduces to simple formula [70]

e2 k 2 |ϕ̃c |2
Φ= , (17.16)
4me (k · v − ωr )2

where ϕ̃c is the complex amplitude of the actual potential ϕ̃ = Re (ϕ̃c e i θ ), θ is the GO phase, and
.
ωr = −∂t θ is the (real) GO frequency.
In the special case when ωr  k·v, Φ depends only on t and x and thus acts as an effective potential.
Such Φ which tends to expel particles from a high-frequency field. In more general settings, though
(for example when a particle interactions with a transverse waves in a strong magnetic field [71]), Φ
can be attractive as well. The concept of dressed particles is also widely known in atomic physics,
where the ponderomotive force −∇Φ is commonly called the dipole force [72].

17.2 Interaction with on-shell waves

In particular, let us consider interaction with “on-shell” waves, i.e., waves constrained by the dispersion
ω = ωr (t, x, k) in the (ω, k) space, so that

hWϕ i ∝ J(t, x, k) δ[ω − ωr (t, x, k)]. (17.17)

For such waves, the function J serves as the phase-space density of the wave action, if normalized
properly. One can show (Box 8.1) that it satisfies the so-called wave-kinetic equation (WKE) [41, 69]:
∂J ∂ωr ∂J ∂ωr ∂J
+ · − · = 2ωi J, (17.18)
∂t ∂k ∂x ∂x ∂k
which is a generalization (and correction) of Eq. (16.26).2 Then one has Ref. [67, 69]



∂t d p F + ∇ · d p V F = 0, (17.19a)
2 The WKE can be understood as the Vlasov equation for wave quanta. Alternatively, the Vlasov equation for plasma

particles can be considered as the classical (GO) limit of the WKE derived for particles as quantum waves.

143
LECTURE 17. DRESSED PARTICLES AND PONDEROMOTIVE FORCES 144

d p pF + d k kJ + ∇ · d p pV F + d k kvg J + ∇ d p ΦF = 0,



∂t (17.19b)






d p H0 F + d k ωr J + ∇ · d p H0 V F + d k ωr vg J + ∇ · d p V ΦF = 0,



∂t (17.19c)
. .
where H0 = p2 /(2me ) is the energy of a free OC and v g = ∂k ωr is the group velocity. Hence, equations
of quasilinear theory conserve the particle (OC) number, the total momentum, and the total energy:

d x d p F = const, (17.20a)
 
d x d p pF + d x d k kJ = const, (17.20b)
 
d x d p H0 F + d x d k ωr J = const. (17.20c)

The density of the energy–momentum density of the OC distribution is different from that of the
particle distribution precisely by the difference between the wave energy-momentum (Lecture 5) and
the field energy-momentum (which is why electrostatic waves can carry momentum while electrostatic
fields cannot):

(particle energy) + (field energy) = (OC energy) + (wave energy), (17.21a)

|
{z } |
{z } |
{z }
dv 1/2 me v 2 f0 d k Uk d k Uk

(particle momentum) + (field momentum) = (OC momentum) + (wave momentum) . (17.21b)

| {z } | {z } | {z }
d v me vf0 zero for electrostatic field dk P k

In particular, as F
diffuses, the OC momentum changes, but so does the wave momentum, and the
total momentum d v me vf0 remains conserved. Also note that in general, the ponderomotive force
on plasma is the sum of all the terms that are quadratic in Ẽ in Eq. (17.19b).

144
Appendices for Part IV

AIV.1 Plasma dispersion function

See Fig. 17.1.

145
APPENDICES FOR PART IV 146

Figure 17.1: Properties of the plasma dispersion function Z (copied from Ref. [73]).

146
Problems for Part IV

PIV.1 Wave propagation: initial-value problem

Consider one-dimensional propagation of a transverse electromagnetic wave in a cold stationary col-
lisionless nonmagnetized homogeneous electron plasma. As discussed in Lecture 2, such wave is
governed by the equation of the Klein–Gordon type,
∂t2 E − ∂x2 E + E = 0, (17.22)
where E is any of the transverse components of the electric field, t is measured in units ωp−1 , and x is
measured in units c/ωp . Here, you are asked to prove rigorously that signals governed by Eq. (17.22)
cannot propagate faster than at the speed of light.
(a) Using the Laplace transform in t and the Fourier transform in x, show that the general solution
of Eq. (17.22) can be written as follows:
 ∞
E(t, x) = d x G(t, x − x0 ) · E(x0 ). (17.23)
−∞

Here, the two-component vector

 
. E(0, x)
E(x) =
∂t E(0, x)
encodes the initial conditions, and the vector function

ds  ∞ dk est+ikx
 
. s
G(t, x) = (17.24)
B 2π i −∞ 2π 1 + k 2 + s2 1
serves as the Green’s function. (Here, B is the Bromwich contour, and the bracket is a two-
component vector.)
(b) Take the integral over k in Eq. (17.24). (This can be done analytically using the residue theorem.)
Prove that the remaining integral is identically zero at |x| > t > 0.

Hint: This is doable even without representing the remaining integral through ele-
mentary functions (which is also possible [74] but not required here).

PIV.2 Longitudinal waves in Lorentzian plasma

Consider a one-dimensionless electron plasma with a Lorentzian distribution,
∆ 1
f0 (v) = , (17.25)
π v 2 + ∆2
where ∆ is a constant.

147
PROBLEMS FOR PART IV 148

(a) Assuming k > 0, calculate the response function σk (t) and identify the characteristic time scale
of phase mixing of such plasma. Take the Laplace transform of σk (t) to obtain the spectral
representation of the conductivity, σ(ω, k).
(b) Using the result from part (a), calculate χ(ω, k).

(c) Show that the same result is obtained using Landau’s rule and

ωp2 0
χ(ω, k) = − 2
k
d v v f−0 (v)
ω/k
. (17.26)
L

Plot the corresponding contours for Im ω < 0, Im ω = 0, and Im ω > 0.

(d) Derive the exact dispersion relation of the longitudinal oscillations in such plasma. What is
their damping rate?

(e) Derive the general expression for σ(t, x) in terms of f0 (v). Plot the result for the Lorentzian
plasma. Explain the plot qualitatively.

PIV.3 Alfvén resonance

Consider a cold stationary electron–ion plasma with one type of ions and a background magnetic field
B 0 along the z axis. The plasma is homogeneous along the y and z axes, but the background density
and B0 depend on x. In this plasma, consider stationary waves with electric field Ẽ = E (x)e −i ωt+i kz z
with kz > 0 and 0 < ω/Ωi  1. Use that

S −iD 0
 
ω2
 =  iD S 0 , S ≈ Ωpi2  1, D ≈ − Ωω S  S . (17.27)
0 0 P i i

(a) Assume that P is large enough such that E z is negligible. Starting from Maxwell’s equations
.
and assuming the notation Nz = ckz /ω, show that E y satisfies

c d D
 2 2 2

+ S − Nz −
2
ω 2 d x2 S − Nz2 E y = 0. (17.28)

(b) Assume S = (1 + x/L)Nz2 , so that x = 0 corresponds to the so-called Alfvén resonance, where
S = Nz2 . Assume L > 0. Argue that Eq. (17.28) can be written as
d 2
x2 − 1
 
x 2+ E y = 0, (17.29)
dx 2α2

where x has been appropriately rescaled and α is a constant that you are asked to find. Plot
the GO dispersion curves kx (x) ≶ 0 corresponding to Eq. (17.29). Also find and plot the inverse
function, which has the form x(kx ) = x̄(kx ) ± ∆(kx ).
(c) In the GO limit, calculate the x-component of the group velocity at S − Nz2  D (for x > 1)
and S − Nz2  D (for −1 < x < 0) to determine the direction of the action flows along the
dispersion curves. Identify the branches at x → +∞ and at x → 0−. Describe what happens to
a wave launched toward the Alfvén resonance.

148
PROBLEMS FOR PART IV 149

.

(d) Consider g1 (kx ) = e i Θ(kx ) E y (x)e i kx x d x, where Θ(kx ) = x̄(kx ) d kx and x̄ is the same as in
.
problem (b). Take for granted (or prove it yourself) that g1 and g2 = −i sg1 − g10 satisfy
.

i ddk gg1 = si −s −i
    
g1 .
, s = α2 kx2 . (17.30)
x 2 g2

Assume that g1,2 ∝ e −i θ(kx ) , where ζ = θ0 (kx ) changes slowly with kx compared with θ. Derive
.
the two corresponding GO dispersion branches ζ1,2 (kx ) and explain how they are related to
x(kx ) in problem (b). By considering the “frequency gap” |ζ1 − ζ2 |, estimate the parameter that
determines the mode-conversion efficiency. Is energy deposition at the Alfvén resonance larger
at small α or at large α?

PIV.4 Two-stream instability in Lorentzian plasma

Consider a one-dimensionless electron plasma with a Lorentzian distribution with two streams.

(a) Derive the susceptibility corresponding to the electron distribution

∆ 1 ∆ 1
f0 (v) = + . (17.31)
2π (v − u)2 + ∆2 2π (v + u)2 + ∆2
√ √
(b) Show that the distribution has a single maximum if |u| < ∆/ 3 and two maxima if |u| > ∆/ 3.
(c) Plot characteristic Nyquist contours for this distribution for different u at fixed k. Argue that
the instability threshold can be found from (0, k) = 0. Solve this equation for k. How large
should |u| be for the plasma to be unstable? Can the distribution be double-peaked and yet
remain stable?

PIV.5 Weibel instability

In plasmas with anisotropic temperature, transverse waves can be subject to the so-called Weibel
instability. Here, you are asked to calculate this effect for nonmagnetized nonrelativistic collisionless
electron plasma with motionless ions using Eq. (10.12).

(a) Assume that f0 (v) is bi-Maxwellian with zero average velocity, namely,
!
1 vx2 vy2 vz2
f0 (v) = 3/2 2 exp − 2 − 2 − 2 , (17.32)
π w⊥ wk w⊥ w⊥ wk
2
where w⊥ = 2T⊥ /m and wk2 = 2Tk /m. Show that the dielectric tensor is diagonal. Then
calculate ⊥ explicitly and show that the dispersion relation of transverse waves can be written
as follows:
T⊥
ω 2 − k 2 c2 − ωp2 + ωp2 [1 + ζZ(ζ)] = 0. (17.33)
Tk

[If needed, the latter term can as well be represented in terms of Z 0 (ζ).]
(b) Briefly comment on how to recover the cold-wave dispersion relation (10.20) from Eq. (17.33).
(c) For the hot-plasma limit (ζ  1), show that Eq. (17.33) is approximately linear in ω and derive
the corresponding ω(k) explicitly. (For simplicity, assume k > 0.) Show that, at T⊥ > Tk , there
always exist k for which waves are unstable.

149
PROBLEMS FOR PART IV 150

(d) For the cold-plasma limit (ζ  1), assume that Im ζ > 0 and that Re ω is small (which can be
checked a posteriori ); then the plasma dispersion function can be approximated as3
1 1
Z(ζ) ≈ − − 3 .
ζ 2ζ

Assuming this asymptotic, show that Eq. (17.33) is biquadratic in ω (i.e., quadratic in ω 2 ).
Calculate ω(k) to the first order in the temperature and show that one of the modes is unstable.
Show that T⊥  Tk is required to justify the cold-plasma approximation for this mode.

(e) Compare your analytic calculations with numerical solution of Eq. (17.33).

PIV.6 Plasma susceptibility in magnetic field

Here, you are asked to derive one of the elements of the magnetized-plasma kinetic susceptibility
tensor using the following expression for the induced linear current density:

2   t−t0
ms ωps

s (t, x) = −e dp v d τ eiβ
j (i) i k·x−i ωt Ẽx U cos(φ + Ωs τ )
4π 0
 
∂f0s
+ Ẽy U sin(φ + Ωs τ ) + Ẽz − V cos(φ + Ωs τ ) , (17.34)
∂pk

where the axes are chosen such that ky = 0 and, accordingly,

k⊥ v⊥
β=− [sin(φ + Ωτ ) − sin φ] + (ω − kk vk )τ. (17.35)
Ω
Hint: For the remaining notation, see the slides from Lecture 13. In particular, re-
. .
member that Ωs = Ωs0 /γ is the relativistic gyrofrequency and Ωs0 = es B0 /(ms c) is the
nonrelativistic gyrofrequency. Also, f0s is the momentum distribution, not the velocity
distribution; the two are connected by the factor m3 , as explained in Sec. 8.1.2.
The derivation can be performed as explained on the slides, but you may also simplify
some of the steps, because (χs )xx is somewhat easier to calculate than the whole tensor.

(a) Show that Eq. (17.34) leads to

∞  ∞  ∞
ωp2 X
  2 2
(χs )xx =
ωΩs0 n=−∞ 0
d p⊥ 2πp⊥ d pk
Ωs
ω − kk vk − nΩs
n Jn (z)
z2
p⊥ U. (17.36)
−∞

(b) Assuming that f0s is isotropic, nonrelativistic, Maxwellian, and has zero average velocity, show
that
2 ∞
ωp0 n2 In (λs ) −λs
 
e An , An = k w . Z(ξn )
X
(χs )xx = (17.37)
ω n=−∞ λs k s

3 Note that this asymptotic is slightly different from the asymptotic of Z on the real axis derived in Sec. 11.2.1. See

Appendix AIV.1 for details.

150
PROBLEMS FOR PART IV 151

PIV.7 Kinetic waves propagating parallel to magnetic field

Here, you are asked to study waves propagating parallel to a static magnetic field in a static homoge-
neous anisotropic Maxwellian plasma (T⊥ 6= Tk ).
(a) Derive the limit of the dielectric tensor at k⊥ → 0. Show that it has the same form as in cold
plasma, except L, R, and P are replaced with, respectively,
X  ωp2  X  ωp2 2ω 
L̄, R̄ = 1 + A±1 , P̄ = 1 + B 0 . (17.38)
s
ω s s
ω kw2 s

Present the corresponding dispersion relation and identify the corresponding waves and their
polarizations.
(b) By retaining the lowest-order finite temperature corrections, show that the dispersion relation
for Alfvén waves can be written as
pk − p⊥
 
ω 2 = kk2 VA2 1 − 2 , (17.39)
B0 /(4π)

where pk and p⊥ are the parallel pressure and the perpendicular pressure, respectively. This
dispersion relation predicts the an instability, which is known as the firehose instability. When
does this instability occur? Qualitatively, what is its mechanism?
Hint: See slides for the formulas for An and Bn . Neglect the contribution from
the resonant pole but keep two terms when expanding the rest of the plasma disper-
sion function. Then simplify the result by adopting that ω  Ωi . However, when
expanding in ω, assume that the parameter kwk /ω is fixed and of order one.

PIV.8 Kinetic whistler waves

Here, you are asked to consider parallel propagation of high-frequency whistler waves (ω  Ωi ), for
which the ion susceptibility is negligible.

(a) Using Eqs. (17.38) with T⊥ = Tk , derive the whistler-wave dispersion relation. Unlike in Prob-
lem PIV.7, neglect thermal effects in Re R̄ but account for nonzero Im R̄.
(b) Present the dispersion relation for ωr . Show that in the regime when N 2  1 and ωr  |Ωe |,
√
the group velocity of these waves scales as vg ∝ ωr .

(c) Suppose that a whistler wave is excited by an antenna with fixed real frequency. Estimate the
distance that the wave propagates before it damps assuming that damping is weak.
Hint: The idea is the same as in Sec. 10.2.3 except: (i) the dispersion function is
different and (ii) complex is the wavevector rather than the frequency.

PIV.9 Cyclotron heating

(a) Consider an electromagnetic wave propagating approximately parallel to the static magnetic
field B 0 such that the longitudinal component of the electric field is negligible; however, allow
for small nonzero k⊥ . Show that the wave power dissipated per unit volume is approximately
given by Eq. (15.26).

151
PROBLEMS FOR PART IV 152

(b) Suppose that the static magnetic field is inhomogeneous, B0 ≈ B̄0 (1 + z/`0 ). Here, z is the
coordinate along the field (the wave propagates from z = −∞ to z = +∞), `0 is some large
enough characteristic length (you may assume that `0 > 0 for simplicity, but this is not essential),
and the constant B̄0 is the field strength at which the wave is in cyclotron resonance with some
minority ions s. Assume that the wave frequency ω is fixed and given. Also, assume that the
wave amplitude |Ẽ± | is given at z = 0 (and depends on z slowly enough such
that its value
outside the resonance region is unimportant). Calculate the total wave power d z PCD,s that
is dissipated through cyclotron damping on the minority ions per unit cross section of the wave
beam using two different methods (make sure that the results are consistent):
(i) using Eq. (15.26);
(ii) using Eq. (15.10) with the cold-plasma dielectric tensor (6.13).

Hint: See footnote 1 in Lecture 6 and also Problem PII.2(e).

(c) Propose an analytic estimate for the characteristic length of the absorption region aabs . Assum-
ing parameters typical for ion cyclotron heating in tokamaks, estimate aabs numerically.

152
Appendices

153
Appendix A

Abbreviations

This appendix summarizes the abbreviations used in the text.

Abbreviation Meaning

GO geometrical optics
CCS collisionless cold static
DK Dnestrovskii–Kostomarov
EBW electron Bernstein wave(s)
EPW electron plasma wave(s), a.k.a. (electron) Langmuir waves
IAW ion acoustic wave(s), a.k.a. ion sound waves
IBW ion Bernstein wave(s)
IPW ion plasma wave(s)
MHD magnetohydrodynamic(s)
OC oscillation center(s)
ODE ordinary differential equation
PDE partial differential equation
TE, TM transverse-electric, transverse-magnetic
WKB Wentzel–Kramers–Brillouin
WKE wave-kinetic equation
WWT Wigner–Weyl transform

154
Appendix B

Notation

This appendix summarizes the notations that are of general importance or often used in the main text.

Symbol Definition Explanation

.
... = ... definition
... · ... Eq. (C.5) scalar product of spatial or spacetime vectors
h...i average over fast oscillations
h...|...i scalar product of vectors in Hilbert space
h...| covector in Hilbert space (bra)
|...i vector in Hilbert space (ket)
.˜.. quantity of the first order in the wave amplitude
.ˆ.. operator
...∗ complex conjugate
...† Hermitian conjugate
...| transpose
...A anti-Hermitian part
...H Hermitian part
[..., ...] commutator

∂ partial derivative
∂a ∂/∂xa partial derivative with respect to a spatial coordinate
∂
α ∂/∂xα partial derivative with respect to a spacetime coordinate
principal-value integral

1 unity
1 unit matrix
1̂ unit operator
1̂ unit matrix operator

X Eq. (1.16) susceptibility in the coordinate representation

X̄ Box 1.1 susceptibility in the symmetrized coordinate representation
Λ eigenvalue of D H
Λ diagonal matrix with the eigenvalues of D H on the diagonal
Θ ∂kk ω photon inverse-mass tensor (times ~)
Σ Eq. (1.8) conductivity in the coordinate representation
Σ̄ Box 1.1 conductivity in the symmetrized coordinate representation

155
APPENDIX B. NOTATION 156

Ωs es B 0 /(ms c) gyrofrequency of species s

χ symbol of χ̂, subsumes the susceptibility in the spectral representation

χ̂ Eq. (1.15) susceptibility operator
 symbol of ˆ, subsumes the dielectric tensor
ˆ Eq. (1.17) dielectric operator
ε Eq. (3.5) geometrical-optics parameter
γ linear growth rate, also polytropic index
ϕ electrostatic potential
λ wavelength
λDs vT s /ωps Debye length of species s
2 2
λs k⊥ vT s /Ω2s roughly, squared ratio of the gyroradius and the transverse wavelength
π ≈ 3.14 ratio of a circle’s circumference to its diameter
θ eikonal phase, also angle between k an B 0
ρ charge density
σ symbol of σ̂, subsumes the conductivity in the spectral representation
σ̂ Eq. (1.5) conductivity operator
ω frequency
ω̄ −∂t θ local frequency of a quasimonochromatic wave
ω̂ i ∂t frequency operator
ωi Im ω imaginary part of the frequency
ωp Eq. (2.21) plasma frequency
ωps Eq. (2.6) plasma frequency of species s
ωr Re ω real part of the frequency

B, Ba magnetic field
B 0 , B0 background magnetic field
B , Ba magnetic field envelope
CS Eq. (7.33) ion sound speed
D Eq. (6.14b) element of  of CCS magnetized plasma
D, Dab symbol of a dispersion operator
D̂, D̂ab dispersion operator
Dϕ electrostatic dispersion operator
D̂ E Eq. (1.23) electromagnetic dispersion operator
E, Ea electric field
E , Ea electric field envelope
H

~ω photon energy or photon Hamiltonian
I dx I wave action
I Sec. 5.1 wave action density
Im imaginary part
J , Ja vg I action flux density
L Eq. (6.14d) element of  of CCS magnetized plasma
Lc characteristics spatial scale of an envelope and (or) of a medium
L Sec. 10.1.2 Landau contour
N ck/ω refractive index
N dimension of a wave field (N = 3 for the electric field)
P Eq. (6.14c) element of  of CCS magnetized plasma, also pressure
P χ̂E electric polarization
P kI wave momentum density
Pabs absorbed power
R Eq. (6.14d) element of  of CCS magnetized plasma

156
APPENDIX B. NOTATION 157

Q, Q Stern–Gerlach term
Re real part
S Eq. (6.14a) element of  of CCS magnetized plasma
S Eq. (5.17) Poynting vector
Tc characteristics temporal scale of an envelope and (or) of a medium
Ts temperature of species s
U ωI wave energy density
Vα ∂D/∂kα k-derivative of the symbol of a dispersion operator
W , W −1 Sec. 3.3 direct and inverse Wigner–Weyl transform
Zs es /e charge state of species s

arg argument (phase) of a complex number

c speed of light
c.c. complex conjugate
d differential
dτ d /d τ full derivative with respect to τ
e ≈ 2.72 Euler’s number
e |ee | elementary charge
ēa unit vector along ath axis
es charge of species s
h eigenvector of D H
~ Planck constant
i imaginary unit
j, ja current density
k, ka wavevector
k̄, k̄a ∇θ, ∂a θ local wavevector of a quasimonochromatic wave
k̂, k̂ a −i ∇, −i ∂a wavevector operator
k, kα (k0 , k) spacetime wavevector
k̂, k̂ α −i ∂α spacetime-wavevector operator
k0 −ω/c temporal component of the spacetime wavevector
ms mass of species s
n dimension of space, dim x
n n+1 dimension of spacetime, dim x
ns density of species s
x, xa spatial location
x̂, x̂a x, xa coordinate operator
x, xα {x0 , x} location in spacetime
x̂, x̂α {x0 , x} spacetime-coordinate operator
x0 ct temporal spacetime coordinate
vs fluid velocity of species s
vg ∂k ω group velocity
vp (k/k)(ω/k)
p phase velocity
vT s T0s /ms unperturbed thermal speed of species s
t time
tr √ trace
ws vT s 2 rescaled thermal speed of species s

157
Appendix C

Conventions

This appendix summarize the mathematical conventions assumed in the main text.

C.1 Numbers and matrices

For a given complex number z, its real part zr ≡ Re z and imaginary part zi ≡ Im z are defined as
. 1 . 1
zr = (z + z ∗ ) = zr∗ , zi = (z − z ∗ ) = zi∗ . (C.1)
2 2i
Accordingly,

z = zr + i zi , z ∗ = zr − i zi . (C.2)

Similarly, for any given matrix or operator M , its Hermitian part MH and anti-Hermitian part MA
are defined as
. 1 . 1
MH = (M + M † ) = MH† , MA = (M − M † ) = MA† . (C.3)
2 2i
Accordingly,

M = MH + i MA , M † = MH − i MA . (C.4)

C.2 Geometry
Euclidean or pseudo-Euclidean coordinates are assumed. The bold font is used for spatial coordinates
(x = {x1 , x2 , ...xn }, with any n ≥ 0), vectors (covectors), and matrices. The upper and lower indices
of these objects are marked with Latin indices and are interchangeable; for example, ka = k a , with
a = 1, 2, . . . n. For any two n-dimensional column vectors X and Y , one has
.
X · Y = X † Y = Xa∗ Y a , (C.5)

Here, † is a Hermitian conjugate (adjoint), or conjugate transpose, X † = X ∗| . The expression X † Y

is understood as a matrix product, with X † being a 1 × n matrix and Y being a n × 1 matrix.
Accordingly, XY † is a n × n matrix with elements X a Yb∗ .
For any matrix M̂ whose elements are operators, denoted M̂ ab , the adjoint M̂ † is the transpose
of the matrix with elements M̂ †ab ; i.e., (M̂ † )ab = M̂ †ba . For example, for the column-vector operator
k̂ = −i ∇, whose elements k̂ a = −i ∂a are Hermitian operators (here, ∂a = ∂/∂xa ), the adjoint is the
. .
row-vector operator k̂† = k̂| . Then, k̂† k̂ = k̂ 2 = −∇2 , while k̂k̂† is a matrix with elements k̂ a k̂ b .

158
APPENDIX C. CONVENTIONS 159

The spacetime coordinates and spacetime vectors (covectors) are denoted with the sans-serif font
in the invariant representation; but this font is used also for other objects. For example, x = {x0 , x},
.
and n = dim x = n + 1. The corresponding components are denoted with a regular font and Greek
indices (for example, xα ), which span from 0 to n. For the spacetime, Minkowski metric is assumed
in the form gαβ = g αβ = diag {−1, 1, 1, . . .}. For the objects in the Minkowski space, conventions
are similar to those above, except the indices are raised and lowered by the Minkowski metric. For
example, kα = gαβ k β , or equivalently, k α = g αβ kα , so in particular, k 0 = −k0 and k a = ka for a > 0.

159
Bibliography

[1] T. H. Stix, Waves in Plasmas (AIP, New York, 1992).

[2] E. R. Tracy, A. J. Brizard, A. S. Richardson, and A. N. Kaufman, Ray Tracing and Beyond:
Phase Space Methods in Plasma Wave Theory (Cambridge University Press, New York, 2014).
[3] D. G. Swanson, Plasma Waves (IOP, Philadelphia, 2003), 2nd ed.
[4] H. L. Pécseli, Waves and Oscillations in Plasmas (CRC Press, Boca Raton, 2013).
[5] M. Brambilla, Kinetic Theory of Plasma Waves: Homogeneous Plasmas (Clarendon Press, Ox-
ford, 1998).
[6] N. A. Krall and A. W. Trivelpiece, Principles of Plasma Physics (McGraw-Hill, New York, 1973).
[7] A. F. Aleksandrov, L. S. Bogdankevich, and A. A. Rukhadze, Principles of Plasma Electrody-
namics (Springer–Verlag, New York, 1984).
[8] E. M. Lifshitz and L. P. Pitaevskii, Physical Kinetics (Pergamon Press, New York, 1981).
[9] J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1999), 3rd edition.
[10] S. C. Wilks, J. M. Dawson, and W. B. Mori, Frequency up-conversion of electromagnetic radiation
with use of an overdense plasma, Phys. Rev. Lett. 61, 337 (1988).
[11] L. Tonks and I. Langmuir, Oscillations in ionized gases, Phys. Rev. 33, 195 (1929).
[12] L. D. Landau and E. M. Lifshitz, Mechanics (Butterworth–Heinemann, Oxford, 1976).
[13] N. A. Lopez and I. Y. Dodin, Restoring geometrical optics near caustics using sequenced meta-
plectic transforms, New J. Phys. 22, 083078 (2020).
[14] V. L. Ginzburg, The Propagation of Electromagnetic Waves in Plasmas (Pergamon, Oxford,
1970).
[15] I. Bialynicki-Birula, Photon wave function, in Progress in Optics, Vol. XXXVI (Elsevier, Ams-
terdam, 1996), edited by E. Wolf, p. 245; arXiv:quant-ph/0508202.
[16] S. W. McDonald, Phase-space representations of wave equations with applications to the eikonal
approximation for short-wavelength waves, Phys. Rep. 158, 337 (1988).
[17] I. Y. Dodin, D. E. Ruiz, K. Yanagihara, Y. Zhou, and S. Kubo, Quasioptical modeling of wave
beams with and without mode conversion. I. Basic theory, Phys. Plasmas 26, 072110 (2019).
[18] I. B. Bernstein, Geometric optics in space and time varying plasmas. I , Phys. Fluids 18, 320
(1975).
[19] Supplementary Material in Ref. [17].

160
[20] D. E. Ruiz, Geometric theory of waves and its applications to plasma physics, Ph.D. Thesis,
Princeton University (2017), arXiv:1708.05423.
[21] Yu. A. Kravtsov, L. A. Ostrovsky, and N. S. Stepanov, Geometrical optics of inhomogeneous and
nonstationary dispersive media, Proc. IEEE 62, 1492 (1974).

[22] M. Bornatici and Yu. A. Kravtsov, Comparative analysis of two formulations of geometrical
optics. The effective dielectric tensor , Plasma Phys. Control. Fusion 42, 255 (2000).
[23] M. Bornatici and O. Maj, Geometrical optics response tensors and the transport of the wave
energy density, Plasma Phys. Control. Fusion 45, 1511 (2003).
[24] G. B. Whitham, Linear and Nonlinear Waves (Wiley, New York, 1974).

[25] K. Y. Bliokh, A. Niv, V. Kleiner, and E. Hasman, Geometrodynamics of spinning light, Nature
Phot. 2, 748 (2008).
[26] R. G. Littlejohn and W. G. Flynn, Geometric phases in the asymptotic theory of coupled wave
equations, Phys. Rev. A 44, 5239 (1991).

[27] M. A. Oancea, J. Joudioux, I. Y. Dodin, D. E. Ruiz, C. F. Paganini, and L. Andersson, Gravi-

tational spin Hall effect of light, Phys. Rev. D 102, 024075 (2020).
[28] I. Y. Dodin and N. J. Fisch, Adiabatic nonlinear waves with trapped particles: III. Wave dynamics,
Phys. Plasmas 19, 012104 (2012), Appendix B.

[29] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Pergamon Press, New York,
1971).
[30] W. D. Hayes, Group velocity and nonlinear dispersive wave propagation, Proc. R. Soc. Lond. A
332, 199 (1973).
[31] D. E. Ruiz and I. Y. Dodin, On the correspondence between quantum and classical variational
principles, Phys. Lett. A 379, 2623 (2015).
[32] L. Landau, Zur Theorie der Energieubertragung. II , Phys. Z. Sowjetunion 2, 46 (1932).
[33] C. Zener, Non-adiabatic crossing of energy levels, Proc. R. Soc. London A 137, 696 (1932).
[34] E. R. Tracy and A. N. Kaufman, Metaplectic formulation of linear mode conversion, Phys. Rev.
E 48, 2196 (1993).
[35] T. E. Stringer, Low-frequency waves in an unbounded plasma, J. Nucl. Energy C 5, 89 (1963).
[36] L. D. Landau and E. M. Lifshitz, Statistical physics (Pergamon Press, New York, 1980), pt. 1.
[37] C. R. Sovinec, Note on kinetic Alfvén waves, UW-CPTC 09-2, https://cptc.wiscweb.wisc.edu/
wp-content/uploads/sites/327/2017/08/UW-CPTC 09-2 rev.pdf.
[38] E. W. Weisstein, Abel Transform, from MathWorld – A Wolfram Web Resource, https://
mathworld.wolfram.com/AbelTransform.html.
[39] B. Eliasson and K. Papadopoulos, HF wave propagation and induced ionospheric turbulence in
the magnetic equatorial region, J. Geophys. Res.-Space 121, 2727 (2016).

[40] J. E. Moyal, Quantum mechanics as a statistical theory, Proc. Cambridge Philosoph. Soc. 45, 99
(1949).
[41] S. W. McDonald and A. N. Kaufman, Weyl representation for electromagnetic waves: The wave
kinetic equation, Phys. Rev. A 32, 1708 (1985).
[42] http://www.princeton.edu/∼idodin/papers/APS-DPP-CPP.pdf.
[43] R. W. Gould, T. M. O’Neil, and J. H. Malmberg, Plasma wave echo, Phys. Rev. Lett. 19, 219
(1967).
[44] N. G. Van Kampen, On the theory of stationary waves in plasmas, Physica 21, 949 (1955).
[45] K. M. Case, Plasma oscillations, Ann. Phys. (N.Y.) 7, 349 (1959).
[46] E. C. Taylor, Landau solution of the plasma oscillation problem, Phys. Fluids 8, 2250 (1965).

[47] L. Landau, On the vibration of the electronic plasma, J. Phys. USSR 10, 25 (1946).
[48] J. H. Malmberg and C. B. Wharton, Collisionless damping of electrostatic plasma waves, Phys.
Rev. Lett. 13, 184 (1964).
[49] R. K. Mazitov, Damping of plasma waves, Zh. Priklad. Mekh. Tekh. Fiz. 1, 27 (1965).

[50] T. O’Neil, Collisionless damping of nonlinear plasma oscillations, Phys. Fluids 8, 2255 (1965).
[51] P. F. Schmit, I. Y. Dodin, and N. J. Fisch, Evolution of nonlinear waves in compressing plasma,
Phys. Plasmas 18, 042103 (2011).
[52] I. B. Bernstein, J. M. Greene, and M. D. Kruskal, Exact nonlinear plasma oscillations, Phys.
Rev. 108, 546 (1957).
[53] I. Y. Dodin, On variational methods in the physics of plasma waves, Fusion Sci. Tech. 65, 54
(2014).
[54] C. Liu and I. Y. Dodin, Nonlinear frequency shift of electrostatic waves in general collisionless
plasma: unifying theory of fluid and kinetic nonlinearities, Phys. Plasmas 22, 082117 (2015).

[55] W. L. Kruer, J. M. Dawson, and R. N. Sudan, Trapped-particle instability, Phys. Rev. Lett. 23,
838 (1969).
[56] I. Y. Dodin, P. F. Schmit, J. Rocks, and N. J. Fisch, Negative-mass instability in nonlinear plasma
waves, Phys. Rev. Lett. 110, 215006 (2013).

[57] S. Brunner, R. L. Berger, B. I. Cohen, L. Hausammann, and E. J. Valeo, Kinetic simulations

and reduced modeling of longitudinal sideband instabilities in non-linear electron plasma waves,
Phys. Plasmas 21, 102104 (2014).
[58] K. Hara, T. Chapman, J. W. Banks, S. Brunner, I. Joseph, R. L. Berger, and I. D. Boyd,
Quantitative study of the trapped particle bunching instability in Langmuir waves, Phys. Plasmas
22, 022104 (2015).
[59] H. Qin, C. K. Phillips, and R. C. Davidson, A new derivation of the plasma susceptibility tensor
for a hot magnetized plasma without infinite sums of products of Bessel functions, Phys. Plasmas
14, 092103 (2007).

[60] P. A. Robinson, Dispersion of electron Bernstein waves including weakly relativistic and electro-
magnetic effects. Part 1. Ordinary modes, J. Plasma Phys. 37, 435 (1987).
[61] P. A. Robinson, Dispersion of electron Bernstein waves including weakly relativistic and electro-
magnetic effects. Part 2. Extraordinary modes, J. Plasma Phys. 37, 449 (1987).
[62] I. Y. Dodin and A. V. Arefiev, Parametric decay of plasma waves near the upper-hybrid resonance,
Phys. Plasmas 24, 032119 (2017).
[63] R. I. Pinsker, Whistlers, helicons, and lower hybrid waves: The physics of radio frequency wave
propagation and absorption for current drive via Landau damping, Phys. Plasmas 22, 090901
(2015).

[64] M. Porkolab, On the physics of magnetosonic wave damping on electrons, AIP Conf. Proc. 244,
197 (1992).
[65] A. A. Vedenov, E. P. Velikhov, and R. Z. Sagdeev, Nonlinear oscillations of rarified plasma, Nucl.
Fusion 1, 82 (1961).

[66] W. E. Drummond and D. Pines, Non-linear stability of plasma oscillations, Nucl. Fusion Suppl.
3, 1049 (1962).
[67] R. L. Dewar, Oscillation center quasilinear theory, Phys. Fluids 16, 1102 (1973).
[68] S. W. McDonald, C. Grebogi, and A. N. Kaufman, Locally coupled evolution of wave and particle
distribution in general magnetoplasma geometry, Phys. Lett. A 111, 19 (1985).

[69] I. Y. Dodin, Quasilinear theory for inhomogeneous plasma, arXiv:2201.08562.

[70] J. R. Cary and A. N. Kaufman, Ponderomotive force and linear susceptibility in Vlasov plasma,
Phys. Rev. Lett. 39, 402 (1977).
[71] H. Motz and C. J. H. Watson, The radio-frequency confinement and acceleration of plasmas, Adv.
Electron. Electron Phys. 23, 153 (1967).
[72] J. Dalibard and C. Cohen-Tannoudji, Dressed-atom approach to atomic motion in laser light: the
dipole force revisited , J. Opt. Soc. Am. B 2, 1707 (1985).
[73] J. D. Huba, NRL Plasma Formulary (Naval Research Laboratory, Washington DC, 2002).

[74] https://en.wikipedia.org/wiki/Green%27s function.