Anda di halaman 1dari 161

Study on

Nonlinear Processes of Electron Beam Instabilities


via
Computer Simulations

Takayuki UMEDA

Department of Communications and Computer Engineering


Graduate School of Informatics
Kyoto University, Kyoto, JAPAN

Submitted: December 2003


Published: March 2004

Acknowledgements
I would like to express my sincere and hearty appreciation to Professor Hiroshi Matsumoto
for attracting me to space science and for giving me the opportunity to perform the computer
simulation study at Radio Science Center for Space and Atmosphere (RASC), Kyoto University.
I am also grateful to him for his stimulating supervision and constructive discussion throughout
the present study. I would like to express my deep appreciation to Professor Yoshiharu Omura
for his continual guidance, stimulating discussion, and many helpful suggestions throughout the
present study. I am also grateful to him for his critical reading of the manuscript and suitable
suggestions for revision. I would like to express my appreciation to Professor Kozo Hashimoto
for his useful discussion and helpful suggestions through research meetings. I am also grateful
to him for his careful reading of manuscript and constructive comments for revision.
I am deeply grateful to Dr. Hideyuki Usui for useful discussion and for his helpful assistance
in performing computer simulations. I am also grateful to Dr. Hirotsugu Kojima for his
variable comments in view of plasma wave observations and to Dr. Naoki Shinohara for his
encouragement. I would like to thank Messrs. Koichi Shin, Yoshikatsu Ueda and Tomohiko
Mitani for their helpful assistance in completing the present thesis. I wish to express my
appreciations to Professors Shoichiro Fukao, Toshitaka Tsuda, Toru Sato and all the stas of
RASC for useful comments through seminars.
I am grateful to Professor Peter H. Yoon at Maryland University for his useful lecture on the
nonlinear theory and for his guidance and discussion of the study in Chapter 5. I would like to
thank Dr. Taketoshi Miyake for useful discussion at scientic conferences. I am also grateful to
Dr. Tooru Sugiyama for discussion and helpful advice on programming. Thanks are also due to
discussion with my colleagues Keisuke Ninomiya, Toshihiro Sakakima, Toshiharu Hashimoto,
Teppei Ohyama and other students of the space plasma simulation group.
I am deeply grateful to Ms. Keiko Miwa for her oce administration and encouragement
throughout the present study. I am also indebted to Ms. Tomoko Kawamata for her help in
preparing materials concerning my grant and for her encouragement, and to Ms. Masae Asano
for maintenance of the computer system. I wish to thank Ms. Junko Kawamoto and Ms.
Michiyo Koyama for their oce work and encouragement.
The present work was supported by Japan Society for the Promotion of Science (JSPS),
research fellowship for young scientists and by Ministry of Education, Culture, Sports, Science and Technology, Grant-in-Aid for JSPS Fellows, 03821. The computer simulations were
performed on the KDK computer system at RASC.
Finally I would like to thank my parents Yoshiyuki, Fukuko, grand-parents Yoshiharu,
Kiyoko, and my brother Yoshihiro for their encouragement, hearty and continual support in
my private life.

ii

Abstract
Electron beam instabilities are one of the most fundamental processes in space plasmas as
well as in laboratory plasmas. Natural phenomena of wave-particle and wave-wave interactions
in regards to beam-plasma instabilities are often highly nonlinear including both coherent
and incoherent processes. In space plasmas, coherent nonlinear phenomena are observed as
electrostatic solitary waves and incoherent phenomena are observed as harmonics Langmuir
waves by recent spacecraft with a waveform receiver. Such nonlinear processes of electron beam
instabilities are too complicated for theoretical studies. In other to obtain better understanding
of both coherent and incoherent nonlinear processes of electron beam instabilities, we developed
new kinetic simulation codes of collisionless plasmas, particle code and Vlasov code, with a
higher-order accuracy.
We rst investigated formation process of electrostatic solitary waves observed by recent
spacecraft via one- and two-dimensional electrostatic particle simulations with open boundaries.
Since the previous simulations were performed in uniform periodic systems, wave-particle interaction of an electron beam instability was taking place uniformly in the systems. In the present
study, we inject a weak electron beam from an open boundary into the background plasma to
study spatial and temporal development of a bump-on-tail instability from a localized source.
In the open system, spatial structures of electron holes are determined by the distance from
the source of the electron beam. In an early phase of the simulation run, electron holes initially
uniform in the direction perpendicular to the magnetic eld become twisted through modulation by oblique electron beam modes to form two-dimensional structures. As the electron holes
propagate along the magnetic eld, they are aligned in the perpendicular direction through coalescence, forming one-dimensional structures in a distant region from the source. In a long time
evolution of the instability, ion dynamics becomes important in determining spatial structures
of electron holes. A lower hybrid mode is excited locally in the region close to the source of the
electron beam through coupling with electron holes at the same parallel phase velocity. The
lower hybrid mode modulates electron holes excited in later phases, resulting in formation of
modulated one-dimensional potentials. Since the perpendicular electric elds of electron holes
are carried by the electron holes at the drift velocity of the electron holes, they can be observed
even at a distant place from the source.
Secondly, we studied electromagnetic eld signature associated with electrostatic solitary
waves by performing two-dimensional electromagnetic particle simulations. Since an electron
hole is a coherent electrostatic potential structure, electron beam instabilities were conventionally studied by electrostatic particle simulations. However, the POLAR spacecraft and
FAST spacecraft observed electromagnetic eld signature associated with ESW. To study interaction between coherent electrostatic potentials and electromagnetic waves, we extend the
previous electrostatic particle model to an electromagnetic particle model. In the present two-

iii
dimensional simulations of a bump-on-tail type instability, electromagnetic eld components
are enhanced around two-dimensional electron holes. We found that the enhancement of electromagnetic elds is due to a current formed by electrons undergoing the E B0 drift, where
the electric eld is a perpendicular electrostatic eld at the edge of a two-dimensional electron
hole. An electromagnetic beam mode is excited by the current due to the drifting electrons
moving with the electron hole. The amplitude ratio of the electric eld to the magnetic eld
is estimated based on the present simulation result, and it is in agreement with those of the
POLAR and FAST observations.
Thirdly, we investigated generation of harmonics Langmuir waves. Nonlinear dispersion relations of harmonic Langmuir waves are derived based on a new weak turbulence theory. To
conrm the validity of the nonlinear theory we performed one-dimensional Vlasov simulation
of a weak beam instability. The present simulation showed generation of multiple harmonic
Langmuir modes up to 15th harmonics during the beam-plasma interaction. The frequencywavenumber spectrum obtained by taking the Fourier transformation of simulated electric eld
both in time and space shows an excellent agreement with the theoretical nonlinear dispersion
relations of harmonic Langmuir waves. The present result conrmed that harmonic Langmuir
waves are normal modes in plasmas. We also varied the wave spectrum of the initial perturbation to study eect of the initial wave spectrum. In all runs with dierent initial wave spectra,
including dierent power-law indices, white noise, and a monochromatic perturbation, the wave
spectrum of the Langmuir turbulence becomes power-law with a universal index of 5 6.

iv

Contents

Acknowledgments

Abstract

ii

List of Figures

vii

List of Tables

1 General Introduction

1.1

Electrostatic Solitary Waves (ESW) . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Nonlinear Evolution of Electron Beam Instabilities

. . . . . . . . . . . . . . . .

1.2.1

Bernstein-Greene-Kruskal(BGK) equilibrium . . . . . . . . . . . . . . . .

1.2.2

Generation of electron holes . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.3

Eects of beam density and background thermal velocity . . . . . . . . . 11

1.3

Multi-dimensional Electron Holes . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.4

Contribution of the Present Study . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Numerical Techniques for Particle Simulations

21

2.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.2

Electromagnetic Particle Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3

2.4

2.5

2.2.1

Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.2.2

Superparticle model

2.2.3

Grid assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.2.4

Time step chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

Poissons Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1

Charge density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.3.2

Electrostatic potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Electromagnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.1

Electric elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.4.2

Magnetic elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
v

CONTENTS

vi
2.5.1

Particle positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.5.2

Particle velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.6 Current Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32


2.6.1

Charge conservation method . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.6.2

Zigzag scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.6.3

Extention to second-order shape factor . . . . . . . . . . . . . . . . . . . 44

2.7 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.7.1

Field initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.7.2

Particle loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.8 Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.9 Particle Injection and Absorption at Bondaries . . . . . . . . . . . . . . . . . . . 49
2.10 Boundary Treatment for Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.10.1 Absorbing boundary conditions . . . . . . . . . . . . . . . . . . . . . . . 52
2.10.2 Improved masking method . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3 Formation of Electrostatic Solitary Waves in Open Systems

61

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 One-dimensional Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4 Two-dimensional Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.4.1

Linear analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.4.2

Spatial structures of electron holes . . . . . . . . . . . . . . . . . . . . . 70

3.4.3

Interaction between electron holes with lower hybrid waves . . . . . . . . 74

3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.5.1

Spatial prole of potentials along the magnetic eld . . . . . . . . . . . . 76

3.5.2

Eect of mass ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.5.3

Perpendicular scale of electron holes

. . . . . . . . . . . . . . . . . . . . 79

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4 Electromagnetic Signature Associated with Electrostatic Solitary Waves

83

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2 Simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.3.1

Electric and magnetic elds . . . . . . . . . . . . . . . . . . . . . . . . . 86

4.3.2

Electron E B drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3.3

Electromagnetic wave emission . . . . . . . . . . . . . . . . . . . . . . . 89

4.3.4

Eect of magnetization ratio . . . . . . . . . . . . . . . . . . . . . . . . . 90

4.3.5

Long-time nonlinear evolution . . . . . . . . . . . . . . . . . . . . . . . . 91

CONTENTS
4.4

4.5

vii

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.4.1

Waveforms of electromagnetic elds . . . . . . . . . . . . . . . . . . . . . 92

4.4.2

Amplitude of electromagnetic elds . . . . . . . . . . . . . . . . . . . . . 93

4.4.3

Contribution of ion E B drift . . . . . . . . . . . . . . . . . . . . . . . 96

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5 Nonlinear Theory and Vlasov Simulation of Harmonic Langmuir Waves

99

5.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5.2

Eigenmode Theory of Harmonics Langmuir Waves . . . . . . . . . . . . . . . . . 101

5.3

5.4

5.5

5.2.1

New weak turbulence theory . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.2.2

Nonlinear dispersion relation . . . . . . . . . . . . . . . . . . . . . . . . . 105

Vlasov Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111


5.3.1

One-dimensional Vlasov-Poisson solver . . . . . . . . . . . . . . . . . . . 111

5.3.2

Model and parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

Simulation Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114


5.4.1

Frequency-wavenumber characteristics . . . . . . . . . . . . . . . . . . . 114

5.4.2

Time evolution of each harmonic mode . . . . . . . . . . . . . . . . . . . 116

5.4.3

Time evolution of phase space . . . . . . . . . . . . . . . . . . . . . . . . 119

5.4.4

Power-low spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

5.4.5

Generation of harmonics from monochromatic perturbation . . . . . . . . 122

Summary and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

6 Concluding Remarks

127

A Numerical Techniques for Vlasov Simulations

131

A.1 Constrained Interpolation Prole (CIP) Scheme . . . . . . . . . . . . . . . . . . 131


A.1.1 Basic principle of the CIP scheme . . . . . . . . . . . . . . . . . . . . . . 131
A.1.2 Extension to higher-dimensions . . . . . . . . . . . . . . . . . . . . . . . 132
A.2 One-dimensional Vlasov-Poisson Solver . . . . . . . . . . . . . . . . . . . . . . . 133
A.2.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.2.2 Splitting method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A.2.3 Poissons equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
A.2.4 Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
A.2.5 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
References

139

Publication List

148

List of Figures
1.1 Broadband electrostatic noises observed by the GEOTAIL spacecraft.

. . . . .

1.2 GEOTAIL observation of electrostatic solitary waves. . . . . . . . . . . . . . . .

1.3 Nonlinear evolution of the electron two-stream instability in Run 1. . . . . . . .

1.4 Electrostatic potential and electron kinetic energy density at e t = 800 in Run 1.

1.5 Nonlinear evolution of the electron two-stream instability in Run 2. . . . . . . . 10


1.6 Nonlinear evolution of the electron two-stream instability in Run 3. . . . . . . . 10
1.7 Nonlinear evolution of the electron two-stream instability in Runs 4,5 and 6. . . 12
1.8 Nonlinear evolution of the electron bump-on-tail instability.

. . . . . . . . . . . 12

1.9 Phase-space electron distributions with dierent background thermal velocities.

13

1.10 Waveforms and velocity distribution functions corresponding to ESW and LW . 14


1.11 Initial electron velocity functions and linear dispersion relations.

. . . . . . . . 15

1.12 Parallel and perpendicular waveforms of ESW. . . . . . . . . . . . . . . . . . . . 16


1.13 Schematic illustration of a two-dimensional solitary potential structure. . . . . . 16
1.14 Electrostatic solitary waves observed by FAST in the auroral region.

. . . . . . 17

2.1 (a)Square-shaped superparticle and (b)triangular-shaped superparticle. . . . . . 24


2.2 Grid assignment of the three-dimensional system. . . . . . . . . . . . . . . . . . 25
2.3 Time step chart. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Vector relation in Buneman-Boris method.

. . . . . . . . . . . . . . . . . . . . 30

2.5 Decomposition of charge ux by Villasenor-Buneman method. . . . . . . . . . . 34


2.6 Decomposition of a charge ux by Esirkepovs method. . . . . . . . . . . . . . . 34
2.7 Particle trajectory of the two-dimensional zigzag scheme for i1 = i2 and j1 = j2 .

36

2.8 Particle trajectory of the two-dimensional zigzag scheme for i1 = i2 , j1 = j2 . . . . 36


2.9 Particle trajectory of the three-dimensional zigzag scheme. . . . . . . . . . . . . 40
2.10 Average CPU time used for the computation of current density for one time step. 42
2.11 Spatial proles of electrostatic potentials in the runs with ne1 = ne2 = 82 .

. . . 43

2.12 Spectral wave intensity |Bz | in the runs with ne1 = ne2 = 82 . . . . . . . . . . . . 43
2.13 Energy histories as derivations from initial energies.

. . . . . . . . . . . . . . . 47

2.14 Phase space for one-dimensional open system. . . . . . . . . . . . . . . . . . . . 51


2.15 Schematic illustration of particle injection.

. . . . . . . . . . . . . . . . . . . . 52

2.16 Shape of the masking function and procedure of the masking method . . . . . . 54
viii

LIST OF FIGURES

ix

2.17 Shape of the masking function for dierent r. . . . . . . . . . . . . . . . . . . . 55


2.18 Spatial and time proles of magnetic eld energy for light mode waves. . . . . . 56
2.19 Spatial and time proles of magnetic eld energy for whistler mode waves. . . . 56
2.20 Spatial and time proles of magnetic eld energy with a dierent masking factor. 57
2.21 Spatial and time proles of magnetic eld energy with improved masking method. 60
3.1

Schematic illustration of the initial condition. . . . . . . . . . . . . . . . . . . . 63

3.2

x vx phase diagrams at e t = 100, 200, 400 and 1000.

3.3

x vx phase diagram and the corresponding spatial prole of electric elds Ex .

3.4

k spectrum of |Ex | for x/e = 204.8 409.6, e t = 307.2 819.2.

3.5

Simulation results of counter-streaming electron beams.

3.6

Maximum growth rate and linear dispersion relation of electron beam modes.

3.7

Spatial proles of potentials at dierent times.

3.8

kx ky spectra of |Ex | and |Ey |.

. . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.9

Spatial proles of , Ex and Ey .

. . . . . . . . . . . . . . . . . . . . . . . . . . 73

. . . . . . . . . . . . . 66
67

. . . . . 68

. . . . . . . . . . . . . 69
. 70

. . . . . . . . . . . . . . . . . 71

3.10 Spatial proles of |Ey | along the magnetic eld and kx ky spectrum of |Ey |. . 74
2

3.11 kx and ky spectra of |Ey | for x/e = 128 256.

. . . . . . . . . . . . . 75

3.12 Schematic illustration on formation and evolution of potential structures. . . . . 77


3.13 kx ky spectra of |Ey | with dierent mass ratios. . . . . . . . . . . . . . . . . . 80
4.1

Schematic illustration of the simulation model.

. . . . . . . . . . . . . . . . . . 84

4.2

Velocity distribution of electrons assumed at the left boundary. . . . . . . . . . . 85

4.3

Electron phase-space distribution and spatial proles of Ex , Ey and . . . . . . 87

4.4

Spatial proles of (a)Ez , (b)Bx , (c)By and (d)Bz at e t = 128.

4.5

kx diagrams of Bx , By , Bz , Ex , Ey and Ez . . . . . . . . . . . . . . . . . . . 88

4.6

y
.
The spatial proles of Jz and ene E
B0

4.7

Long-time evolution of the electron beam instability.

4.8

Schematic illustration of magnetic eld and current density structures. . . . . . 93

4.9

Waveforms of Ex , Ey and By at y/e = 20, e t = 128.

. . . . . . . . . 87

. . . . . . . . . . . . . . . . . . . . . . . . 89

4.10 Potential model for analysis of amplitude ratio

|By |
.
|Ey |

. . . . . . . . . . . . . . . 92
. . . . . . . . . . . . . . 94
. . . . . . . . . . . . . . . 95

5.1

Dispersion relations and linear growth rates of harmonic Langmuir waves. . . . . 110

5.2

-kx diagram corresponding to the wave intensity|Ex (, kx )|2 . . . . . . . . . . . 114

5.3

Time evolution of the spectral wave energy density |Ex (t, kx )|2 .

5.4

Time histories of electric eld energies for the rst 6 harmonic modes.

5.5

Velocity space distribution functions and the corresponding x vx phase plots.

5.6

Continuation of Figure 6.5 for longer time periods. . . . . . . . . . . . . . . . . . 119

5.7

Spectra of wave electric eld energy versus wavenumber. . . . . . . . . . . . . . 121

5.8

Results of the with coherent initial perturbation with a single wave mode.

. . . . . . . . . 116
. . . . . 117
119

. . . 122

List of Tables
1.1 Simulation parameters for dierent runs. . . . . . . . . . . . . . . . . . . . . . .

2.1 Dierent cases of the particle positions at times t and t + t in the threedimensional zigzag scheme.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.1 Common simulation parameters for all runs. . . . . . . . . . . . . . . . . . . . . 65


3.2 The angles between the wave vector and the external magnetic eld, frequencies,
parallel and perpendicular wavenumbers of the excited lower hybrid modes for
dierent mass ratios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.1 Common parameters for simulation run. . . . . . . . . . . . . . . . . . . . . . . 86

Chapter 1
General Introduction
1.1

Electrostatic Solitary Waves (ESW)

The GEOTAIL spacecraft was launched in 1992 as the rst spacecraft of the International
Solar-Terrestrial Physics (ISTP) mission. The GEOTAIL Plasma Wave Instrument (PWI)
team made a series of plasma wave observations in the Earths magnetotail by the following
three sets of receivers onboard, Sweep Frequency Analyzer (SFA), Multi Channel Analyzer
(MCA) and Wave-Form Capture (WFC) [Matsumoto et al., 1994a].
Figure 1.1(a) shows a frequency spectrogram obtained by the SFA receiver on January 16,
1993. During the time period of 15:58 - 16:20 UT, there exists a strong wave emission extending
from a low frequency to the local electron plasma frequency. This emission is called Broadband
Electrostatic Noise (BEN) [Gurnett et al., 1976]. BEN emissions are commonly observed in
the plasma sheet boundary layer (PSBL) of the Earths magnetotail. BEN emissions usually
occur over a broad range of frequencies extending from about 10 Hz up to a few kHz which
corresponds the local plasma frequency e . There exists a quasi-upper cut-o around a few
hundreds of Hz, which is almost equal to the local electron cyclotron frequency e . BEN
emissions also have a clear lower cut-o around 10Hz. The lower cut-o frequency is almost
equal to the local lower hybrid resonance frequency omegaLHR . Similar broadband emissions
are also observed in various region of the Earths magnetosphere: cusp/cleft regions [Pottelette
et al., 1990], dayside auroral zone [Dubouloz et al., 1991], magnetosheath [Kojima et al., 1997],
and bow shock region [Matsumoto et al., 1997].
In spacecraft observations in 1970s, measurements of frequency spectra were mainstream.
Therefore it was believed that the BEN emission was a superposition of waves with a broadband
range of wave frequency. There were number of theoretical works regarding that BEN emissions
consist of normal mode waves unstable over a wide range of wave frequency [e.g., Schriver
and Ashour-Abdalla, 1987]. Some of linear dispersion relations for strongly unstable velocity
distribution functions consisting of electron and/or ion beams show a beam mode unstable over
a wide frequency range. However, such strong instabilities diuse the energetic electrons and
ions, and saturate as a single mode. Theoretical studies to explain BEN in terms of linear
analyses were not successful. It is noted, however, that Nishida et al. [1985] pointed out a
1

CHAPTER 1. GENERAL INTRODUCTION

Figure 1.1: (a) Frequency-time spectrogram of Broadband Electrostatic Noises (BEN) and
(b) the corresponding waveforms observed by the GEOTAIL spacecraft [after Matsumoto
et al., 1994b].

possibility that BEN emissions correspond to kinds of potential structures rather than noises
before the GEOTAIL waveform observation of BEN emissions.
The GEOTAIL PWI team provided a breakthrough by the WFC receiver onboard with
capability of recording waveforms of plasma waves. They revealed the real nature of BEN
emissions. During the event of BEN emissions as identied by the frequency spectra, the WFC
recorded a series of bipolar solitary waves (Figure 1.1). These impulsive waves are named
as Electrostatic Solitary Waves (ESW) after their characteristics [Matsumoto et al., 1994b;
Kojima et al., 1994]. Figure 1.1(b) shows the real waveform observed by one of the two sets
of the orthogonal dipole antennas (EU and EV ) [Matsumoto et al., 1994a]. From the computer
simulation, the GEOTAIL PWI team assumed that electrostatic solitary waves travel along
the ambient magnetic eld over a long distance at the velocity of an electron beam [Omura et

1.1. ELECTROSTATIC SOLITARY WAVES (ESW)

al., 1996]. They took the ambient magnetic eld as one of coordinates for the reference frame.
The electric elds EU and EV measured by the two sets of the orthogonal dipole antennas are
converted into components E|| and E , parallel and perpendicular respectively to the ambient
magnetic eld [Kojima et al., 1997].
Panels (a) and (b) in Figures 1.2 show waveforms of electrostatic solitary waves in the direction parallel to the ambient magnetic eld. Panels (b) and (e) show the corresponding
perpendicular electric elds of the solitary waves. As shown in the E|| E hodographs in Figure 1.2(c) and (f), these solitary waves are electrostatic in most cases, because the polarizations
of the bipolar pulses are mostly along the external static magnetic eld without any transverse
electric eld. The amplitude of electric eld of ESW in the PSBL region is very small, several
hundreds of V/m or smaller. The polarization of the bipolar pulses turns positive to negative
as well as negative to positive in the timescale of 150 msec, which corresponds to the electron
plasma oscillation period. This implies that ESW correspond to potential structures related to
the electron dynamics rather than the ions dynamics, i.e., positive potentials [Kojima et al.,
1997].
Since the GEOTAIL PWI team had experiences to observe bipolar waveforms of electric
elds in Particle-In-Cell (PIC) simulations of electron beam instabilities, the waveforms of
electrostatic solitary waves observed by the GEOTAIL spacecraft were familiar to them. From a
close similarity between observed waveforms and those found in PIC simulations, they assumed
that positive potentials are moving along the ambient magnetic eld as an explanation of the
bipolar waveforms of the electric eld [Matsumoto et al., 1994b]. When an isolated positive
potential crosses an antenna of the GEOTAIL spacecraft with a much higher velocity than the
velocity of the GEOTAIL spacecraft, a bipolar signature of electric eld corresponding to the
gradient of the potential can be recorded. From the observed waveforms of E|| and E , they
inferred that the positive potentials are uniform in the direction perpendicular to the ambient
magnetic eld [Kojima et al., 1997]. When the GEOTAIL spacecraft observes a potential with
a positive velocity, the polarization of E|| turns positive to negative. On the other hand, when
the GEOTAIL spacecraft observes a potential with a negative velocity, the polarization of E||
turns negative to positive.
Figure 1.2(g) shows a electron velocity distribution function obtained by the Low Energy Particle (LEP) instrument onboard the GEOTAIL spacecraft [Mukai et al., 1994]. The electron velocity distribution function is decomposed into the Maxwellian component and non-Maxwellian
components. The dashed line corresponds to the Maxwellian component, and the solid lines
show the non-Maxwellian components. The dash-dotted line shows the noise level. They found
enhancement of electron ux at high energy tails of the Maxwellian velocity distribution function, which can be regarded as diused electron beams after the saturation of electron beam
instabilities. The electron beams propagate with a velocity 10,000km/s. The densities of
beam electrons are 110% of the total electron density. In Figure 1.2(g), the electron beam

CHAPTER 1. GENERAL INTRODUCTION

Figure 1.2: Electric eld E|| and E for time period starting from (a-c) 14:14:50.539
UT and (d-f) 14:14:51.094 UT on September 16, 1993. (g) Reduced velocity distribution
functions. The solid line, dashed line and dot-dashed line correspond to beam component,
Maxwellian component and noise level, respectively [after Omura et al., 1999a].

1.2. NONLINEAR EVOLUTION OF ELECTRON BEAM INSTABILITIES

propagating in the negative direction have a slower velocity than that propagating in the positive direction. In this case, the waveform of the solitary wave changes with a longer timescale.
On the other hand, when the waveforms of the solitary waves change with a shorter timescale,
the velocity of the electron beam is higher, extending to 50,000km/s. The statistical analyses
of the GEOTAIL data [Omura et al., 1999a; Kojima et al., 1999c] are in good agreement with a
potential model based on the previous PIC simulations of an electron bump-on-tail instability
[Omura et al., 1996; Miyake et al. 1998], which is regarded as the most probable generation
mechanism of electrostatic solitary waves in the magnetotail.
Figure 1.2 also shows that the poralization of the solitary waves changes after 0.5 second,
implying that propagation direction of the solitary waves turns positive to negative with a
very short timescale. This result suggests that the solitary waves propagate in both tailward
and earthward directions in the magnetotail [Omura et al., 1999a; Kojima et al., 1999b]. The
statistical analyses of the GEOTAIL data also showed that in the neartail region, most of
solitary waves propagate in the earthward direction for XGSM > 30Re, where Re represents one Earth radius and GSM represents the Geocentric Solar Magnetospheric coordinate
system. On the other hand, most of solitary waves propagate in the tailward direction for
XGSM < 30Re. It is expected that acceleration regions of electrons are located at the nearEarth neutral line (XGSM = 20Re 40Re) and the distant neutral line (XGSM 140Re)
[Omura et al., 1999a,b; Kojima et al., 1999b], implying that generation of electrostatic solitary
waves is associated with the magnetic reconnection process.

1.2
1.2.1

Nonlinear Evolution of Electron Beam Instabilities


Bernstein-Greene-Kruskal(BGK) equilibrium

In early computer simulations such as water-bag model [e.g., Roberts and Berk, 1967; Berk et
al., 1970] and Particle-In-Cell (PIC) model [e.g., Mozer and Nielson, 1969] in the 1960s, it was
demonstrated that their appears solitary potential structures after the saturation of electron
beam instabilities. The potential structures form holes in the velocity-position phase space.
Since the solitary structures are positive potentials that trap electrons, they are called electron
holes. Figure 1.3 shows the temporal evolution of an electron two-stream instability with the
parameters in Table 1.1. The left panels show electron x vx space phase distribution function,
and the right panels show the corresponding waveforms of electric eld Ex normalized by
e Vt1 me /e, where e is the total electron plasma frequency, Vt1 is the initial thermal velocity
of the electron beam 1, and me and e are respectively the mass and the charge of an electron.
Velocities of electron holes formed in the saturation stage of an electron two-stream instability
are slightly dierent from each other. Therefore two electron holes coalesce with each other,
and merge into a larger and more intense electron hole. Formation and coalescence processes

CHAPTER 1. GENERAL INTRODUCTION

Figure 1.3: Nonlinear evolution of an electron two-stream instability with Vt1 = Vt2 = 1.0,
Vti = 0.125, Vd1 = 2.5, Vd2 = 2.5, Vdi = 2.5. Two electron beams have equal densities.
(left) Velocity-position phase-space electron distributions fe (x, vx ) at dierent times and
(right) the corresponding waveforms of the electric eld Ex normalized by e Vt1 me /e.
Velocity and distance are normalized based on the initial electron thermal velocity and
initial electron Debye length, respectively.

of electron holes were also observed in laboratory plasmas [e.g., Saeki et al., 1979]. However,
it was not clear that electron holes can really exist in space plasmas.
Such an electrostatic potential structure have been already described by an old theory proposed in 1957, in the year of the rst satellite Sputnik 1. In a paper entitled as Exact nonlinear
plasma oscillations, it was shown that equilibrium solutions to the time-independent VlasovPoisson equations could be constructed if the velocity distribution functions of electrons and
ions are expressed as functions of the sum of the electrostatic potential energy and the kinetic
energy, i.e., the Hamiltonian [Bernstein et al., 1957]. In this paper, it was also discussed that
there can exist every potential structures, such as bumps and dents, depending on the combination of trapped/untrapped electron and ion phase space distributions. The physical model
assumed in this paper is a purely one-dimensional system with an external static magnetic eld
along the system. The equilibrium solution in such a one-dimensional collision-less plasma has
generally been called a BGK mode after the three authors of the paper.
There were several theoretical works which showed that an electron hole corresponds to a

1.2. NONLINEAR EVOLUTION OF ELECTRON BEAM INSTABILITIES

Parameter

Run 1 Run 2 Run 3 Run 4 Run 5 Run 6

Drift velocity of electron beam 1

: Vd1

2.5

2.5

2.5

2.5

2.5

2.5

Thermal velocity of electron beam 1 : Vt1

1.0

1.0

1.0

1.0

1.0

1.0

: Vd2

-2.5

-2.5

-2.5

-2.5

-2.5

-2.5

Thermal velocity of electron beam 2 : Vt2

1.0

1.0

1.0

1.0

1.0

1.0

-2.5

-2.5

0.0

-2.5

-2.5

-2.5

0.125 0.0125

0.125

0.125

0.125

0.125

0.5

0.2

0.15

0.1

Drift velocity of electron beam 2


Drift velocity of ion beam

: Vdi

Thermal velocity of ion beam

: Vti

Density ratio of electron beam 1

: n1 /ni

0.5

0.5

Table 1.1: Simulation parameters for dierent runs.

BGK mode [e.g., Turikov, 1984; Krasovsky et al., 1997, 2003]. The kinetic description of the
BGK mode is derived from the Vlasov equation for the species s with /t = 0,
v
and Poissons equation,

qs fs
fs

= 0,
x
ms x v


2
1 
=

q
fs dv,
s
x2
o s

(1.1)

(1.2)

where is the electrostatic potential and the x-axis is taken along the magnetic eld. By taking
a frame of reference moving with an electron hole, we dene an energy variable
1
Ws ms v 2 + qs (x).
2

(1.3)

We can nd that an arbitrary function of the energy variable fs (x, v) = F (Ws ) can be at the
BGK equilibrium satisfying Eqs.(1.1) and (1.2). Denoting F (Ws ) as Fs+ and Fs respectively
for positive (v > 0) and negative (v < 0) velocities, we have the particle density expressed as
 0

ns (x) =
=


qs

fs (x, v)dv +



0

fs (x, v)dv

(1.4)

dWs .

(1.5)

Fs + Fs+

2ms (Ws qs )

The kinetic energy density is also expressed as


Us () =
=

1
2ms

1
ms v 2 fs dv
2

qs

(Fs + Fs+ ) 2ms (Ws qs )dWs .

(1.6)

Combining these expressions, we obtain a useful relation valid for the BGK equilibrium [Krasovsky
et al., 2003],
dUs
qs
= ns .
d
2

(1.7)

CHAPTER 1. GENERAL INTRODUCTION

Figure 1.4: Electrostatic potential and electron kinetic energy density Ue = 12 ne me v 2


at e t = 800 in Run 1. The potential and energy are normalized by Vt1 me /e and
2e Vt12 m2e /e2 , respectively. Distance is normalized based on the initial Debye length of
the major electrons.

Poissons equation (1.2) is rewritten as


2 d 
d2
=
Us .
dx2
0 d s

(1.8)

We multiply the above equation by d/dx and integrate over x to obtain



0 Ex2
= 2 (Us Us0 ).
2
s

(1.9)

The integration constant Us0 corresponds to an unperturbed kinetic energy density at innity
where Ex = 0. The equation derived above reveals that the electrostatic energy density is exactly twice as large as the perturbation of the kinetic energy at the BGK equilibrium [Krasovsky
et al., 2003].
In Figure 1.4, we show snapshots of electrostatic potential and electron kinetic energy
1
n m v 2 at e t = 800 of Run
2 e e
and 2e Vt12 m2e /e2 , respectively. By

density Ue =

1. The potential and energy are normalized

by Vt1 me /e

integrating Eq.(1.7) over and by assuming

e  ene , the kinetic energy density of trapped electrons by the BGK potential is roughly
estimated as
2(Ue Ue0 ) =

 0
0

e d  ene 0 ,

(1.10)

where 0 corresponds to the potential depth. The above relationship is also conrmed from
Figure 1.4.

1.2. NONLINEAR EVOLUTION OF ELECTRON BEAM INSTABILITIES

1.2.2

Generation of electron holes

It is suggested that there is a close similarity between observed waveforms of electrostatic solitary waves and waveforms of electron holes [Matsumoto et al., 1994b]. To study generation
mechanism of electrostatic solitary waves observed by the GEOTAIL spacecraft, the GEOTAIL PWI team performed a series of computer simulations of electron beam instabilities.
Eect of ion dynamics was indicated in the GEOTAIL observation [Kojima et al., 1994], while
simulations of electron beam instabilities were conventionally performed without ion dynamics.
Omura et al. [1994] performed a series of PIC simulations assuming two cold electron beams
and one ion beam as the initial condition. Electron beams are formed in various processes in
space plasmas, such as particle reection at shocks [e.g., Shimada and Hoshino, 2000], induction
electric eld due to variation of the geomagnetic eld [Omura et al., 2003], and parallel electric
eld of double layers [e.g., Singh, 2000a; Newman et al., 2001b]. Electron beams cause strong
electrostatic instabilities, which leads to diusion excitation of Langmuir waves, ion acoustic
waves or electron holes. In this section we review dierent nonlinear evolution of electron beam
instabilities depending on beam parameters. The simulations are performed with a new Vlasov
code described in Appendix A. We assumed a one-dimensional periodic system taken along an
ambient magnetic eld. Simulation parameters for dierent runs are listed in Table 1.1. We
used a mass ratio mi /me = 1600 close to the real mass ratio.
In a two-stream instability, a wave mode with the maximum growth rate grows to a saturation
level to form a series of large vortices (positive potentials) that can trap the whole population
of electrons. In a simulation run with two initially counter-streaming electron beams with
drift velocities +Vd and Vd and with equal densities, we found formation and coalescence of
trapped particle vortices moving with velocities close to zero (Run 1: Figure 1.3). Due to a
slight dierence between propagation velocities of the vortices, the electron vortices coalesce
with adjacent ones to form electron holes.
Ion dynamics becomes important in the formation process of electron holes. If the temperature of ions is cold enough compared with the electron temperature at the saturation stage, ion
Landau damping does not become eective to allow a stable propagation of ion acoustic waves.
In the case of Run 2, 12 mi Vti2 < 12 me Vte2 , nonlinear evolution of the electron beam instability is
almost the same as that of Run 1 at e t = 40, 80 and 160. In a long-time run up to i t  1,
however, the ion dynamics becomes eective, and electron holes decay into the ion acoustic
waves. At e t = 800, we found strong modulation of the ion phase space distribution as seen
in Figure 1.5b. Electron components around the ion drift velocity Vdi is also trapped by the ion
acoustic waves. In Figure 1.5b we found both electron holes and ion acoustic waves. However,
when the amplitude of the ion acoustic waves becomes much larger, the BGK equilibrium of
the electron holes is disturbed by the trapping of electrons by the ion acoustic waves.
Even if the temperature of ions is hot, the BGK equilibrium of the electron holes is disturbed

CHAPTER 1. GENERAL INTRODUCTION

10

Figure 1.5: Nonlinear evolution of an electron two-stream instability with Vt1 = Vt2 = 1.0,
Vti = 0.0125, Vd1 = 2.5, Vd1 = 2.5, Vdi = 2.5. Two electron beams have equal densities.
(a) Velocity-position phase-space electron distributions fe (x, vx ) at dierent times and (b)
the corresponding velocity-position phase-space ion distributions fi (x, vx ). Velocity and
distance are normalized based on the initial electron thermal velocity and initial electron
Debye length, respectively.

Figure 1.6: Nonlinear evolution of an electron two-stream instability with Vt1 = Vt2 = 1.0,
Vti = 0.125, Vd1 = 2.5, Vd1 = 2.5, Vdi = 0.0. Two electron beams have equal densities.
(a) Velocity-position phase-space electron distributions fe (x, vx ) at dierent times and (b)
the corresponding velocity-position phase-space ion distributions fi (x, vx ). Velocity and
distance are normalized based on the initial electron thermal velocity and initial electron
Debye length, respectively.

1.2. NONLINEAR EVOLUTION OF ELECTRON BEAM INSTABILITIES

11

by ion dynamics when ions are trapped by the negative potential wells formed by two positive
potentials corresponding to the adjacent electron holes. Let us think of a frame of reference
moving with the electron holes. If the drift kinetic energy of ions is smaller than the potential
energy of the well, the ions are trapped to destroy the potential well by their large inertia eect.
When the ions are in the rest frame of electron holes (Run 3: Figure 1.6), electron holes decay
into smaller holes to have certain drift velocities in the frame of ions. As a necessary condition
for stable propagation of electron holes, therefore, electron holes should have a high velocity in
the frame of ions.

1.2.3

Eects of beam density and background thermal velocity

There is no requirement that the two electron beams have equal densities. Assuming that one
electron beam is drifting against the other electron beam and the ion beam, we dene the
density ratio of the electron beam as R = n1 /ni , where n1 + n2 = ni . When the density of the
drifting electron beam becomes higher, electron holes are formed through long-time evolution
of the electron beam instability. As an extreme case, even when there is only a single electron
beam drifting against the ions, i.e., R = 100%, large electron holes can be generated after the
current driven instability called the Buneman instability [Buneman, 1959]. On the other hand,
when the density of the drifting electron beam is low, the beam instability leads to Langmuir
turbulence rather than to formation of electron holes. In the case of Vd /Vt2 = 20, where Vd
corresponds to the relative velocity between the two electron beam, Vd Vd1 Vd2 , it is found
that a necessary condition for formation of electron holes is R 30% [Omura et al., 1994]. In
the case of Vd /Vt2 = 5, the density ration R 15% is a necessary condition for formation of
electron holes as shown in Figure 1.7. A higher thermal velocity of the major electrons (Vt2 )
enables the electron beam instability to leads to formation of electron holes even with a lower
density ratio R.
In the cases of the two-stream instability and Buneman instability, the amplitude of electron
holes becomes very large, which corresponds to a potential depth that can trap whole population
of two electron beams or an electron beam in the Buneman instability. In these cases, the
potential energy of electron holes becomes larger than the thermal energy of electrons at the
saturation stage, e > me Vte2 . On the other hand, the amplitude of ESW (electron holes)
observed by the GEOTAIL spacecraft is of the order of 10100V/m, which corresponds to a
potential depth e 0.01me Vte2 . This implies that a small population of electrons are trapped
by the potential of electron holes. The observation contradicts with the generation mechanism
via the two-stream instability or Buneman instability, because the spatial scale and depth of
potentials in these instabilities are too large compared to the those of ESW in the PSBL region.
It is not necessary, however, for electron holes to be large enough to trap the whole population
of the two-streaming beam electrons as found in the two-stream instability. Formation of

CHAPTER 1. GENERAL INTRODUCTION

12

Figure 1.7: Nonlinear evolution of an electron two-stream instability with Vt1 = Vt2 = 1.0,
Vti = 0.125, Vd1 = 2.5, Vd1 = 2.5, Vdi = 2.5. Velocity-position phase-space electron
distributions fe (x, vx ) in (a) Run 4 (R = 0.2), (b) Run 5 (R = 0.15), (c) Run 6 (R = 0.1),
at dierent times, where R n1 /ni . Velocity and distance are normalized based on the
initial thermal velocity and initial Debye length of the major electrons, respectively.

Figure 1.8:

Nonlinear evolution of an electron bump-on-tail instability with

Vt1 = 1.0, Vt2 = 10.0, Vti = 0.125, Vd1 = 20.0, Vd2 = 0.0, Vdi = 0.0. The density ratio of beam electrons is R n1 /ni = 0.05. (left) Velocity-position phase-space electron
distributions fe (x, vx ) at dierent times and (right) the corresponding waveforms of the
electric eld Ex normalized by e Vt2 me /e. Velocity and distance are normalized based on
the initial thermal velocity and initial Debye length of the major electrons, respectively.

1.2. NONLINEAR EVOLUTION OF ELECTRON BEAM INSTABILITIES

13

Figure 1.9: Velocity-position phase-space electron distributions fe (x, vx ) at e t = 800


with dierent thermal velocities of the background electrons, Vt2 /Vd = 0.3, 0.2, 0.15, 0.1.
Vd = 20, Vt1 = 1.0, Vti = 0.125, Vdi = 0.0, R = 0.05. Velocity and distance are normalized
based on the initial thermal velocity and initial Debye length of the major electrons,
respectively.

electron holes is also found in a bump-on-tail instability induced by a weak electron beam
placed on a thermal tail of a warm Maxwellian velocity distribution function [Omura et al.,
1996]. Especially for a relatively weak beam forming a small bump on the high energy tail of the
velocity distribution function, electron holes with a small amplitude can be generated (Figure
1.8). Since the electron beam forms a positive gradient that give rise to a positive growth rate,
a coherent beam mode wave with a phase velocity slightly smaller than the drift velocity of the
electron beam grows to form a continuous sinusoidal potential structure. The beam electrons
are trapped in each potential well of the structure, and they are diused in the velocity phase
space through nonlinear trapping motion, forming a stable velocity distribution function. As
we found in the nonlinear evolution of the two-stream instability, the adjacent potentials with
trapped electrons coalesce with each other, gradually forming an isolated electron holes in a
long time evolution of the bump-on-tail instability.
In Figure 1.8 we assumed a weak electron beam with R = 5%. To show the eect of the
thermal velocity of the background major electrons Vt2 , we next decreased Vt2 from 10 to 2,
while the beam parameters Vd Vd1 Vd2 = 20, Vt1 = 1 and the ion parameters Vti = 0.125,
Vdi = 0.0 are common. As the thermal velocity of the background major electrons becomes
slower, the amplitude of electron holes becomes smaller. In the case of Vt2 /Vd > 0.2 we found
formation of electron holes. In the case of Vt2 /Vd < 0.2, on the other hand, the electron beam
instability develops to excite Langmuir waves with various phase velocities (Figure 1.9).
These simulation results are also conrmed by the GEOTAIL observation. Matsumoto et
al. [1999] showed a distinct dierence between velocity distribution functions of the thermal
electrons during observations of ESW and those during observations of Langmuir waves. In

CHAPTER 1. GENERAL INTRODUCTION

14

Figure 1.10: (a)Waveforms of electrostatic solitary waves and electron velocity distribution functions observed by the GEOTAIL spacecraft on 1993/6/24 at 15:38 UT.
(b)Waveforms of Langmuir waves and electron velocity distribution functions observed on
1993/6/24 at 15:43 UT. The left column shows the velocity distribution functions on the
Vx -Vy plane with the ambient magnetic eld B plotted as an arrow. The three-dimensional
velocity distribution functions are integrated over the velocity components perpendicular
to B, being reduced to the F (V|| ) as shown in the right column [after Matsumoto et al.,
1999].

regions where ESW are observed, the temperature of the thermal electrons is high. (Figure
1.10a). When the GEOTAIL spacecraft observes Langmuir waves, on the other hand, the
temperature of the thermal electrons is lower (Figure 1.10b).
When the thermal velocity of the background major electrons is fast, the linear dispersion
analysis in Figure 1.11 shows that the unstable mode is localized at the bump in the velocity
distribution function. In this case, the instability is a resistive kinetic instability. The velocity
distribution function is diused around a phase velocity slightly slower than the beam drift
velocity. On the other hand, when the thermal velocity of the background major electrons is
slow, unstable waves with various phase velocities are excited. In this case, the instability is a
reactive cold-beam instability. Beam electrons spread over a wide range in the velocity space.
Here, we note the terminology used in this thesis: The resistive instability is called bump-ontail instability, while cold two-stream instability or weak beam instability refers to the
reactive instability,
As shown in this section, nonlinear evolutions of electron beam instabilities are quite dierent
depending on the beam density and the temperatures of both beam and background electrons.
Electrostatic solitary waves can be regarded as a result of a coherent nonlinear process of

1.3. MULTI-DIMENSIONAL ELECTRON HOLES

15

Figure 1.11: Initial electron velocity functions for Vt2 /Vd = 0.1, 0.15, 0.2, 0.3 and the
corresponding linear dispersion relations, frequency (k) and growth rate (k). Vd = 20,
Vt1 = 1.0, Vti = 0.125, Vdi = 0.0, R = 0.05.

beam-plasma interactions, while generation of Langmuir waves is related to an incoherent


nonlinear process. Recent spacecraft observations and computer simulations have shown that
both coherent and incoherent nonlinear processes of beam-plasma interactions play important
roles on the generation of various plasma waves. In the present computer simulation study,
we will try to obtain further understanding of microscopic nonlinear beam-plasma interactions
driven by a weak electron beam.

1.3

Multi-dimensional Electron Holes

Currently electrostatic solitary waves are observed by recent spacecraft in various regions of
the Earths magnetosphere: plasma sheet boundary layer [Matsumoto et al., 1994; Franz et al.,
1998; Cattell et al., 1999], auroral region [Mozer et al., 1997; Ergun et al., 1998], bow shock
region [Matsumoto et al., 1997; Bale et al., 1998], magnetosheath [Kojima et al., 1997], solar
wind [Mangeney et al., 1999], and dayside magnetopause boundary [Matsumoto et al., 2003].
These solitary waves are electron holes which correspond to positive potentials. There also
exist ion holes which correspond to negative potentials [Kojima et al., 1999].
As reported by Kojima et al. [1997], the polarization of the ESW is parallel to the magnetic
eld. In most cases of the ESW, we nd bipolar variation of the parallel electric eld E|| , and we
nd little variation of the perpendicular electric eld as described in Section 1.1. This implies
that the electrostatic solitary waves observed by the GEOTAIL spacecraft in the magnetotail
have one-dimensional structures. Occasionally, however, there also exist electrostatic solitary
waves showing a unipolar variation of E as well as the bipolar variation of E|| as shown
in Figure 1.12. The maximum value of E coincides with the timing when the sign of E||
changes. To explain the observed waveforms of multi-dimensional ESW, the GEOTAIL PWI

CHAPTER 1. GENERAL INTRODUCTION

16

Figure 1.12: Waveforms of ESW observed

Figure 1.13: Schematic illustra-

by Geotail for 42 ms time period from (1)

tion of a two-dimensional solitary

12:47:39:520 UT and (2) 12:47:39:818 UT

potential structure and possible

on 940210. (a) Parallel component E|| ,

waveforms of E|| and E for all

(b) perpendicular component E , and (c)

dierent locations of the space-

hodograph in E|| -E plane [after Omura

craft and directions of the ESW

et al., 1999b].

[after Omura et al., 1999b].

team assumed a two-dimensional potential structure isolated in both parallel and perpendicular
directions relative to the ambient magnetic eld as schematically illustrated by Figure 1.13.
When the spacecraft sees one of edges of the potential structure, there appears a unipolar
waveform of E , either positive or negative depending on which edge of the potential structure
the spacecraft sees.
Compared with a two-dimensional simulation study of the bump-on-tail instability [Miyake
et al., 1998], they assumed that the two-dimensional structures of ESW represent a transient
process in the long time evolution of the bump-on-tail instability. In a two-dimensional system consisting of spatial coordinates parallel and perpendicular to the magnetic eld, there
exist uctuations with nite wavenumbers k|| and k . At the linear stage of the instability
the oblique modes (k = 0) as well as the parallel modes (k = 0) grow in time. At the
saturation of the instability, formation of two-dimensional potential structures is found in the
two-dimensional simulation. In the subsequent nonlinear evolution, the two-dimensional potentials coalesce with each other to form potential humps uniform in the direction perpendicular
to the ambient magnetic eld. Through the coalescence process, the two-dimensional structures are aligned in the perpendicular direction with perpendicular electric eld diminishing

1.3. MULTI-DIMENSIONAL ELECTRON HOLES

17

Figure 1.14: (a) The electric eld parallel to B0 . (b) The electric eld perpendicular to
B0 and in the spin plane of the satellite. (c) E along the spin axis of the satellite. (d)
A perturbation magnetic eld perpendicular to B0 . [after Ergun et al., 1998].

slowly. The two-dimensional characteristics totally disappear from the potential structures,
and there appear one-dimensional potential structures which are essentially the BGK modes.
The two-dimensional simulation result indicates that the source region of the instability, where
we can nd an unstable electron beam, is closer for the two-dimensional potentials than that
for the one-dimensional potentials. A preliminary study based on a statistical analysis of ESW
data observed by the Geotail spacecraft indicates have been conducted in order to identify the
acceleration region of electrons. More than 20% of ESW have distinct perpendicular electric
elds especially in two regions, around the near-tail neutral point XGSM = 25 75Re and
around the distant-tail neutral point XGSM = 175 225Re [Omura et al., 1999b]. The
result demonstrated that acceleration of electrons in the magnetotail is associated with the
magnetic reconnection process.
It is noted that such unipolar waveforms of the perpendicular electric elds of electrostatic
solitary waves are often observed in the auroral region (Figure 1.14). Three-dimensional structures of electrostatic solitary waves are observed by the POLAR spacecraft [Mozer et al., 1997]
and the FAST spacecraft [Ergun et al., 1998] with three sets of orthogonal dipole antennas.
Contrary to the GEOTAIL observation in the magnetotail, observations of electrostatic solitary
waves in the auroral region show the unipolar waveform of E in most cases, and there ex-

CHAPTER 1. GENERAL INTRODUCTION

18

ist little one-dimensional ESW. Multi-dimensional electrostatic potential structures of electron


holes and wave emission processes from such coherent potential structures are issues of recent
theoretical works [Vetoulis and Oppenheim, 2001; Newman et al., 2001a; Singh et al., 2001a;
Chen and Parks, 2002; Jovanovic et al., 2002] and computer simulations [Goldman et al., 1999;
Oppenheim et al., 1999, 2001; Muschietti et al., 1999, 2000; Miyake et al., 2001; Singh et al.,
2000b, 2001b, 2003; Umeda et al., 2002]. Meanwhile, the FAST and POLAR observation also
showed that there exist impulsive magnetic elds, and that these magnetic eld signatures are
associated with electrostatic solitary waves as shown in Figure 1.14. In this thesis we focus on
one- and two-dimensional potential structures of electrostatic solitary waves and associating
electromagnetic signatures.

1.4

Contribution of the Present Study

In Chapter 2, we give a guide of electromagnetic particle simulation code. The basic concept
of a particle code is not given in detail because it is found in the textbooks [e.g., Hockney
and Eastwood, 1981; Birdsall and Langdon, 1985; Matsumoto and Omura, 1985; Omura and
Matsumoto, 1993; Buneman, 1993]. Based on Kyoto university ElectroMagnetic Patricle cOde
(KEMPO) [Matsumoto and Omura, 1985; Omura and Matsumoto, 1993], we developed new
electrostatic and electromagnetic particle codes with the second-order shape-factor. Especially,
in the new electromagnetic code, we adopted a new charge conservation method with zigzag
particle trajectories [Umeda et al., 2003a] to avoid solving Poissons equation at every time
step. We also adopted an improved masking method for absorbing boundaries [Umeda et al.,
2001] to study spatial and temporal development of electron beam instabilities in open systems.
In Chapter 3, we study spatial and temporal development of an electron bump-on-tail instability in one- and two-dimensional open systems [Umeda et al., 2002]. The bump-on-tail
instability is considered to be the most probable generation mechanics of electrostatic solitary
waves (ESW) [Omura et al., 1994, 1996; Miyake et al., 1998], which was rst observed by the
GEOTAIL spacecraft in the Earths magnetotail [Matsumoto et al. 1994]. Simulations of electron beam instabilities are conventionally performed in uniform periodic systems, in which an
unstable velocity distribution is assumed to be uniform in the system. In the present study,
simulation systems are taken along an ambient magnetic eld. We inject a weak beam from a
boundary of the system into the background unperturbed plasma. The injected beam electrons
propagate along the ambient magnetic eld to set in the bump-on-tail instability. We rst
describe the present simulation model in detail. Secondly, we perform one-dimensional electrostatic particle simulation of weak beam injection into warm background electrons. Thirdly, we
extend the one-dimensional model to a two-dimensional model to include coupling of electrostatic solitary waves with obliquely propagating electrostatic modes such as lower hybrid waves
[Miyake et al., 2001]. In the two-dimensional model, beam electrons are placed uniformly in

1.4. CONTRIBUTION OF THE PRESENT STUDY

19

the direction perpendicular to the ambient magnetic eld. We nd that spatial potential structures of electron holes are determined by the distance from the source of the electron beam.
By performing a long simulation run, it is also found that electron holes are modulated in the
perpendicular direction by oblique electron beam modes in the timescale of electron dynamics,
and then by lower hybrid waves in the timescale of ion dynamics. We summarize development
of potential structures and generation of oblique electrostatic modes in terms of the distance
from the source and timescales of electrons and ions.
In Chapter 4, we study electromagnetic eld signatures associated with electrostatic solitary
waves [Umeda et al., 2004]. The POLAR spacecraft and FAST spacecraft observed that magnetic elds are accompanied by multi-dimensional electrostatic solitary waves [Mozer et al.,
1997; Ergun et al., 1998]. However, previous simulations and theoretical works were performed
with electrostatic approximation. To study interaction between coherent electrostatic potentials with electromagnetic elds, we extend the two-dimensional electrostatic particle model in
3 to an electromagnetic particle model. We nd that two-dimensional electron holes are accompanied by magnetic elds. The enhancement of the magnetic elds is due to a current formed
by electrons undergoing the E B0 drift, where the electric eld is a perpendicular electrostatic
eld at the edge of a two-dimensional electron hole. We also discuss possible emission process
of electromagnetic waves from the multi-dimensional electron holes propagating along the ambient magnetic eld. We next analyze waveforms of electromagnetic elds, and estimate the
amplitude ratio of the electromagnetic elds to the electrostatic elds. The order of amplitude
of the magnetic eld estimate based on the present simulation result is in agreement with the
POLAR and FAST observations.
In Chapter 5, we study an incoherent nonlinear process of electron beam instabilities, generation of harmonics Langmuir waves via theoretical and simulation studied [Yoon et al., 2003;
Gaelzer et al., 2003; Umeda et al., 2003b]. Harmonics Langmuir waves are observed in laboratory plasmas, in space plasmas and in computer simulations. Recently, dispersion relations of
harmonics Langmuir waves are derived by Yoon et al. [2003]. We rst described the generalized weak turbulence theory developed by Yoon [2000]. We then briey review the nonlinear
eigen-mode theory of harmonics Langmuir waves based on the new weak turbulence theory. To
conrm the validity of the nonlinear theory we have performed one-dimensional Vlasov simulation of a weak beam instability [Umeda et al., 2003b]. The results show an excellent agreement
between the theory and simulation, which indicates that harmonics Langmuir waves exist as
normal modes in plasmas. We also initiate the weak beam instability with dierent types of
initial wave spectra: white noise, power-law and monochromatic perturbations, to show that
the Langmuir turbulence has a universal power-law spectrum with index of 5 6.
In Chapter 6, we summarize the present study and give conclusions obtained by the present
computer simulations. Finally, we give suggestions for further studies.

20

CHAPTER 1. GENERAL INTRODUCTION

Chapter 2
Numerical Techniques for Particle Simulations
2.1

Introduction

In the space surrounding the Earth, electromagnetic elds accelerate plasma particles. Motions
of plasmas result in currents to change electromagnetic elds. Natural phenomena of waveparticle interactions observed in space plasmas are often highly nonlinear. Data of plasma
parameters obtained from spacecraft observations are limited in terms of space and time, because they are obtained only on the limited orbits of the spacecraft. It is dicult to distinguish
whether these data show a spatial variation or temporal variation. Meanwhile, theoretical studies cannot suciently explain the phenomena in the space plasmas because of the nonlinearity
and non-uniformity. In the late 1960s and early 1970s, computer simulation (or computer
experiment) has been developed as a third new approach in space plasma physics. The role of
computer simulation is to bridge gaps between the other two traditional approaches: theory
and experiment/observation. Based on a rapid advancement of supercomputer technology in
terms of its CPU speed and working area of memory, computer simulation is now widely recognized as an essential method rather than a powerful and useful tool to investigate nonlinear
wave-particle interaction in space plasmas.
Macroscopic physical processes of space plasmas are studied with MHD(Magneto-HydroDynamics) codes in which both electrons and ions are treated as a uid. Mesoscopic processes
are studied with hybrid codes in which electrons are treated as a uid while ions are treated
as individual particles. On the other hand, particle codes, in which both electrons and ions
are treated as individual particles, are used for study of microscopic physical processes such
as wave-particle interactions in space plasmas. In particle simulations, Maxwells equations
and the equations of charged particle motions are solved in self-consistent manner to follow the
individual nonlinear motions of a huge number of charged particles in their self-consistent elds.
Treating both electrons and ions as individual charged particles, one can solve microscopic
kinetics which occurs in the spatial scale of the electron Debye length and in the time scale
of the electron plasma period. The basic concepts of particle simulations are described in the
textbooks [e.g., Hockney and Eastwood, 1981; Birdsall and Langdon, 1985; Matsumoto and
Omura, 1985; Omura and Matsumoto, 1993; Buneman, 1993].
21

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

22

In the present study, we developed two- and three-dimensional particle code based on Kyoto
university ElectroMagnetic Patricle cOde (KEMPO) [Matsumoto and Omura, 1985; Omura,
1985; Omura and Matsumoto, 1993]. KEMPO has been developed and improved at RASC
of Kyoto University for more than two decades by a lot of scientists. The previous version
of KEMPO were developed with the rst-order shaped factor. as described in the textbooks
[Matsumoto and Omura, 1985; Omura and Matsumoto, 1993]. In this chapter we present
numerical techniques of three-dimensional electromagnetic particle code with the rst- and
second-order shape factors. In section 2.2, we briey give basic equations and concepts of
electromagnetic particle simulations. Sections 2.3-2.10 provide numerical techniques for solving
the basic equations of particle simulations and for an open boundary condition adopted in the
present study.

2.2
2.2.1

Electromagnetic Particle Code


Basic equations

Maxwells equations for the electric eld E (Ex , Ey , Ez ) and magnetic eld B (Bx , By , Bz )
are given as follows:
B = 0 J +

1 E
c2 t

B
E =
t

E =
0
B = 0

(2.1)
(2.2)
(2.3)
(2.4)

where J (Jx , Jy , Jz ), , 0 , 0 and c represent current density, charge density, magnetic permeability, dielectric constant and light speed, respectively. Eqs.(2.1) and (2.2) are solved to
advance electromagnetic elds in space and time. On the other hand, Eqs.(2.3) and (2.4) are
solved only for initialization of electromagnetic elds. These two equations are satised automatically, if Eq.(2.1) is solved correctly based on the current density J satisfying the continuity
equation of the charge density ,

= 0.
(2.5)
t
Neglecting relativistic eect, we have equations of charged particle motions as follows:
J +

dr
= v
dt
qn
dv
=
(E + v B)
dt
mn

(2.6)
(2.7)

where m, v, r and q represent particle mass, velocity, position and charge, respectively. The
subscript n represents the n-th particle.

2.2. ELECTROMAGNETIC PARTICLE CODE

23

The basic equations described above are written in the MKS unit system. In a simulation
however, magnetic permeability 0 and dielectric constant 0 can take arbitrary values, as far
as they satisfy the relation
0 0 =

1
.
c2

For simplicity, we adopt the following denition: 0 =

(2.8)
1
,
c2

0 = 1.

Electric and magnetic elds, current and charge densities are dened at spatial grid points,
while charged particles can take arbitrary positions. The current density and the charge density
are computed from the motion of particles. A charge and a charge ux of a particle is distributed
to the adjacent grid points with a xed assignment pattern, which is called PIC (Particle In
Cell) method. On the other hand, to compute velocities of charged particles, electric elds and
magnetic elds are interpolated from those at the adjacent grid points.
Discretization of space and time is necessary for numerical approach to and partial dierential
equations. The problem is how large we can choose grid spacing r and time step t keeping
the numerical stability. To avoid the numerical in stability, the CFL (Courant-Fredericks-Lewy)
or Courant condition should be satised. In the particle simulation, the CFL condition is

r
c D<
t

(2.9)

where D represents the spatial dimension of the simulation model. We note that particle
velocities are slower than the light speed c.

2.2.2

Superparticle model

In a particle simulation, all particle are treated as superparticles which have larger mass
and larger charge than those in the real space plasma. The superparticle model is necessary
for particle simulations, because it is impossible to treat huge number of particles such as
those in the real space plasma with the limited working area of computer memory. However,
we cannot avoid numerical noises such as enhanced thermal uctuations in the superparticle
model. Thermal uctuations are given by the thermal energy of one particle, i.e., 12 ms Vts2 . Here
Vt represents the thermal velocity for the s-th species (e.g., electron or ion). Since the mass of a
superparticle is much larger than that of a real plasma particle, numerical thermal uctuations
are much more enhanced than those in the real space plasma [Ueda et al., 1994].
In the present simulation code, we give the plasma frequency s and the charge-to-mass
ration qs /ms as input parameters for the s-th species. From these values, the mass of a particle
for the s-th species is given by
0 2s
.
(2.10)
(qs /ms )2 ns
Thus by using lots of superparticles per cell, i.e., by increasing the particle density ns , enhanced
ms =

thermal uctuations are suppressed.

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

24

Figure 2.1: (a)Square-shaped superparticle and (b)triangular-shaped superparticle.

The shape of the superparticle can be of any form. Based on the shape of the superparticle,
the charge and the charge ux of the superparticle are assigned to the adjacent grid points.
Normally the charge and the charge ux of the superparticle are assigned to each grid point
in such a way that the amount of the charge and the charge ux become proportional to the
area (or volume) of the superparticle in the territory of the grid point, which is called the areaweighting (or area-sharing) scheme [e.g., Morse and Neilson, 1969; Hockney and Eastwood,
1981; Birdsall and Langdon, 1985]. The assignment function determined by the shape of the
superparticle is called the shape factor. It is well known that the higher-order shape factor
can reduce the numerical noises.
The square-shaped superparticle with width x (Figure 2.1a) is widely used in particle
simulations. The square-shaped assignment function corresponds to the rst-order shape factor.
The one-dimensional rst-order shape factor is given by
Si () =

|xi |
x

for | xi |
for | xi | >

1
2
1
2

(2.11)

The actual computation of the rst-order shape factor for the n-th superparticle is realized by
the following procedure,
xn
xn
, Wi+1 =
i,
(2.12)
x
x
x
where i is the largest integer not greater than x
, which is given by the oor function,
Wi = i + 1

x
i = oor( x
).

In the present study we extended the previous square-shaped superparticle to the triangularshaped superparticle with width 2x (Figure 2.1b). The triangular-shaped assignment function
corresponds the second-order shape factor. The one-dimensional second-order shape factor is
given by

Si () =

xi 2
x

2
1 3
i|
|x
2 2
x

for | xi |

for | xi | >

for

1
2

1
2

< | xi |
3
2

3
2

(2.13)

2.2. ELECTROMAGNETIC PARTICLE CODE

25

Figure 2.2: Grid assignment of the three-dimensional system.

The actual computation of the second-order shape factor for the n-th superparticle is realized
by the following procedure,
1 1
xn
Wi1 = (
+ i)2 ,
2 2 x
where i is the integer nearest
i=

Wi =
x
,
x

3
xn
(
i)2 ,
4
x

1 1
xn
Wi+1 = ( +
i)2 ,
2 2 x

(2.14)

which is given by the anint function (in FORTRAN),

x
).
anint( x

2.2.3

Grid assignment

We dene full-integer grids [F ] at ix, jy, kz (i = 1, 2, ..., Nx , j = 1, 2, ..., Ny , k =


1, 2, ..., Nz ) and half-integer grids [H] at (i + 12 )x, (j + 12 )y, (k + 12 )z. For simplicity, we
express the location of the grid points with label (i, j, k) by [F F F ], and the grid points with
label (i + 12 , j + 12 , k + 12 ) by [HHH].
In the present three-dimensional particle code, we dene the charge density at [F F F ]. The
electrostatic potential is dened at the same grids. To realize centered deference form for the
spatial derivative in the charge continuity equation, the electric eld components Ex , Ey , Ez are
respectively dened at [HF F ], [F HF ], [F F H]. Components of the current density Jx , Jy , Jz are
respectively dened at the same grids of Ex , Ey , Ez for direct integration of the electric elds
in time. To realize centered deference form for the spatial derivative in Maxwells equations,
the magnetic elds component Bx , By , Bz are respectively dened at [F HH], [HF H], [HHF ].
The grid assignment of the three-dimensional system is illustrated in Figure 2.2.

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

26

Figure 2.3: Time step chart.

2.2.4

Time step chart

The time step chart adopted in the present code is based on that of KEMPO1 [Omura and
Matsumoto, 1993]. The quantities of elds and particles are advanced in time based on the
sequence shown in Figure 2.3. We dene a full-integer time nt and a half-integer time (n +
1
)t.
2

Basically the electric eld E at the full-integer time and the magnetic eld B at the

half-integer time are integrated in time by the leap-frog method. The particle positions r at
the the full-integer time and the velocities B at the half-integer time are also advanced by the
leap-frog method.
t

0. As the initial condition, we have r t , v t 2 , E t and B t 2 .


1. Advance the magnetic eld B from t
2. Compute B t by B t = (B t
t

t
2

to t +

t
2

t
2

t
2

to t +

t
2

using E t .

+ B t+ 2 )/2. Then advance the velocities of particles v from

using E t and B t .
t

3. Advance the positions of particles r from t to t + t using v t+ 2 .


4. Based on the charge conservation method, compute the current density J at t +
the positions r t and rt+t .
5. Advance the electric eld E from t to t + t using B t+

t
2

and J t+ 2 .

t
2

using

2.3. POISSONS EQUATION

2.3

27

Poissons Equation

This section provides the numerical techniques for solving Poissons equation,

2 = .

The total electric eld is given by the scalar potential and vector potential A as
A
E =
t
where the potentials are dened by

(2.15)

(2.16)

1 dv 
=
(2.17)
40 v r
0 J dv 
.
(2.18)
A=
4 v r
By the Coulomb gauge with A = 0, the scalar potential and the vector potential A are
uniquely determined to yield the induction electric eld given by

1 J 
0
ind
dv
E =
4 v r t
and the electrostatic eld given by
E es =

(2.19)

(2.20)

Solving Poissons equation, we can obtain spatial proles of scalar potentials and electrostatic
elds.

2.3.1

Charge density

We rst compute the charge density from the spatial distribution of particles. The spatial
prole of the charge density (i,j,k) is given by superposition of the charge of a particle assigned
to each particle,
(i,j,k) =

Np


qn Si (

n=1

xn
yn
zn
)Sj (
)Sk (
),
x
y
z

(2.21)

where Np represents the number of particles. With the rst-order weighting function Eq.(2.12),
the charge of one particle is assigned to
(i, j, k) = qn Wi Wj Wk ,

(i + 1, j, k) = qn Wi+1 Wj Wk ,

(i, j + 1, k) = qn Wi Wj+1 Wk ,

(i + 1, j + 1, k) = qn Wi+1 Wj+1 Wk ,

(i, j, k + 1) = qn Wi Wj Wk+1 ,

(i + 1, j, k + 1) = qn Wi+1 Wj Wk+1 ,

(i, j + 1, k + 1) = qn Wi Wj+1 Wk+1 , (i + 1, j + 1, k + 1) = qn Wi+1 Wj+1 Wk+1 ,


where i = oor(xn /x), j = oor(yn /y), k = oor(zn /z). The spatial prole of the charge
density is given by superposition of the charge contributed by each particle. With the secondorder weighting function Eq.(2.14), the prole of the charge density is obtained with a higher
accuracy. In this case, indices of the array are given by i = anint(xn /x), j = anint(yn /y),
k = anint(zn /z).

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

28

2.3.2

Electrostatic potential

The dierence form of Eq.(2.15) is given by


(i+1,j,k) 2(i,j,k) + (i1,j,k) (i,j+1,k) 2(i,j,k) + (i,j1,k)
1
+
(i,j,k) =
0
x2
y 2
(i,j,k+1) 2(i,j,k) + (i,j,k1)
+
z 2

(2.22)

There are a number of methods to solve the above equation. For the periodic boundary condition, we apply Fourier transformation to (i,j,k) to obtain (kl ,km ,kn ) ,
(kl ,km ,kn )
2
l:l
Nx
0, 1, 2, ..., N2z ,

where kl =

Ny Nz
Nx 



1
=
(i,j,k) e 1(ikl +jkm+kkn) ,
Nx , Ny , Nz i=1 j=1 k=1

= 0, 1, 2, ..., N2x , km =

2
m
Nx

: m = 0, 1, 2, ..., N2y , kn =

(2.23)
2
n
Nx

:n=

and Nx , Ny , Nz respectively represent the number of grid points in the x, y, z

directions. Eq.(2.22) is rewritten as


(Kx2 + Ky2 + Kz2 )(kl ,km ,kn ) = (kl ,km ,kn )
(2.24)




(kl ,km ,kn )
1
e 1(ikl +jkm+kkn) , (2.25)
(i,j,k) =
2
0 kl km kn Kx + Ky2 + Kz2
where Kx

sin(kl x/2)
,
x/2

Ky

sin(km y/2)
,
y/2

Kz

sin(kn z/2)
.
z/2

When the open boundary condition is imposed only in one direction, we apply the marching
method [Buneman, 1973] for solving Poissons equation. Poisson solver for a three-dimensional
system with open boundaries in multi-directions is still issue.

2.4
2.4.1

Electromagnetic Fields
Electric elds

The three-dimensional form of Eq. (2.1) can be rewritten as


Bz
Ex
By
Jx
= c2

t
y
z


Bz
Ey
2 Bx
= c

Jy
t
z
x


Bx
Ez
2 By
= c

Jz .
t
x
y

(2.26)
(2.27)
(2.28)

By using the grid assignment in Section 2.2.3, the centered dierence form of these equations
are given by
t+ t

t+t
t
2
= Ex,(i+
Ex,(i+
1
1
,j,k) tJx,(i+ 1 ,j,k)
,j,k)
2

(2.29)

2.4. ELECTROMAGNETIC FIELDS

+c2 t

29

t+ t

Bz,(i+
1
,j+ 1 ,k)
2

t+ t
Bz,(i+2 1 ,j 1 ,k)
2
2

t+ t
By,(i+2 1 ,j,k+ 1 )
2
2

t+ t

By,(i+2 1 ,j,k 1 )

2
2

t+ t

t+t
t
2
Ey,(i,j+
= Ey,(i,j+
1
1
,k) tJy,(i,j+ 1 ,k)
,k)
2

+c2 t

t+ t

Bx,(i,j+
1
,k+ 1 )
2

(2.30)

t+ t

2
Bx,(i,j+
1
,k 1 )
2

t+ t
t+ t
2
2

Bz,(i+ 1 ,j+ 1 ,k) Bz,(i 1 ,j+ 1 ,k)

2
2
2
2

t+ t

t+t
t
2
Ez,(i,j,k+
= Ez,(i,j,k+
1 tJ
1
)
)
z,(i,j,k+ 1 )

2
t+ t

B 2 1

+c2 t

t+ t

y,(i+ 2 ,j,k+ 21 )

2.4.2

(2.31)

By,(i2 1 ,j,k+ 1 )
2

t+ t

t+ t

2
2
Bx,(i,j
Bx,(i,j+
1
1
,k+ 1 )
,k+ 1 )
2

Magnetic elds

The three-dimensional form of Eq. (2.2) can be rewritten as


Ey Ez
Bx
=

t
z
y
By
Ez
Ex
=

t
x
z
Bz
Ex Ey
=

.
t
y
x

(2.32)
(2.33)
(2.34)

By using the grid assignment in Section 2.2.3, the centered dierence form of these equations
are given by
t+t
t
Bx,(i,j+
= Bx,(i,j+
1
1
,k+ 1 )
,k+ 1 )
2

2
2
t

E t+ 2

+t

t+ t

y,(i,j+ 21 ,k+1)

2
Ey,(i,j+
1
,k)
2

t+ t

t+ t

2
2
Ez,(i,j+1,k+
1 E
)
z,(i,j,k+ 1 )
2

Ex,(i+2 1 ,j,k+1) Ex,(i+2 1 ,j,k)

t+t
t
By,(i+
= By,(i+
1
1
,j,k+ 1 )
,j,k+ 1 )
2

2
2
t+ t

E 2

+t

t+ t

z,(i+1,j,k+ 21 )

2
Ez,(i,j,k+
1
)
2

t+ t

t+t
t
Bz,(i+
= Bz,(i+
1
1
,j+ 1 ,k)
,j+ 1 ,k)
2

+t

2
2
t
t+
t+ t

Ex,(i+2 1 ,j,k)
Ex,(i+
1
,j+1,k)

t+ t

t+ t
t+ t

2
2

Ey,(i+1,j+

1
,k)
y,(i,j+ 21 ,k)
2
.

(2.35)

(2.36)

(2.37)

30

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

Figure 2.4: Vector relation in Buneman-Boris method.

2.5
2.5.1

Equations of Motion
Particle positions

The dierence form of the equations of motion (2.6) in three-dimensions are given by
t+ t

xt+t
= xtn + vxn 2 t
n
t+ t
2

ynt+t = ynt + vyn

t+ t
2

znt+t = znt + vzn

2.5.2

(2.38)

(2.39)

t.

(2.40)

Particle velocities

The dierence form of the equation of motion (2.7) is given by


v t+

t
2

= v t

t
2

+ t

qs t
E + v t B t .
ms

However, since v t is not obtained, this equation is rewritten as

v

t+ t
2

= v

t t
2

qs t v t+
+ t
E +
ms

t
2

+ v t
2

t
2

(2.41)

(2.42)

By introducing the following to variables


t t
2

v
n = vn

t+ t
2

v+
n = vn

qs t t
E
ms
2
qs t t
,

E
ms
2
+

(2.43)
(2.44)

2.5. EQUATIONS OF MOTION

31

Eq.(2.42) can be rewritten as


v + v =

t
qs +
v + v B t .
ms
2

(2.45)

Taking the inner product of Eq.(2.45) with (v + + v ), we have


(v + )2 = (v )2 .

(2.46)

Thus the kinetic energy of a particle is conserved in the cyclotron motion. As shown in Figure
2.4, we have

t qs
|B|
=
2
2 ms

1
1
cos2
=
qs t 2
=
2
2
1 + ( ms 2 ) |B|2
1 + tan 2

t
v + v = 2v 0 cos2 B t .
2
2

(2.47)

tan

(2.48)
(2.49)

The actual computation is realized with the following four steps, which are called BunemanBoris method [Hockney and Eastwood, 1981; Birdsall and Langdon, 1985],

vxn

vyn
=

vzn
=

t t
vxn 2
t t
vyn 2
t t
vzn 2

= vxn
+
vxn
0

vyn
= vyn
+
0
vzn

vzn

vxn
= vxn
+
+
vyn

vyn

+
+
+

qs
E t t
ms x 2
qs
E t t
ms y 2
qs
E t t
ms z 2

qs

(vyn
Bzt
ms
qs
(v B t
ms zn x
qs

(vxn
Byt
ms
qs
1+( m

qs
1+( m

(2.50)

vzn
Byt ) t
2

vxn
Bzt ) t
2

t t
vyn Bx ) 2

qs
2
0
(vyn
Bzt
t 2
t 2 +B t 2 +B t 2 ms
B
)
(
)
x
y
z
2
qs
2
(v 0 B t
t 2
) (Bt 2 +Bt 2 +Bt 2 ) ms zn x

(2.51)

vzn
Byt ) t
2

0
vxn
Bzt ) t
2

qs
2
+

0
t
0
t t

vzn = vzn +
(v
B

v
B
)
2

xn
y
yn
x
qs t
2
2
2 ms
2
t
t
t
1+( m 2 ) (Bx +By +Bz )
s

t+ t

+ mqss Ext t
vxn 2 = vxn

t
t+

vyn

t+ t
2

vzn

+
= vyn
+
+
= vzn
+

s 2

qs
E t t
ms y 2
qs
E t t
ms z 2

(2.52)

(2.53)

Electric and magnetic elds in the above equations are those at the position of the particle.
Since the electromagnetic elds are dened at spatial grid points, we have to interpolate them
from the adjacent grid points to the position of the particle. To avoid a self-force, by which a
particle itself is accelerated, we have to use the same weighting in both assignment of densities
and computation of force. All components of the electromagnetic elds are dened at the
dierent grid points (see Section 2.2.3). As described in Section 2.3.1, we computed the charge
density based on the shape factor dened at the full-integer grids (i, j, k). Therefore we

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

32

rst relocate all the electromagnetic eld components to full-integer grids (i, j, k). Then eld
components at a position of a particle (xn , yn , zn ) are interpolated form those at the neighboring
grids:


xn
yn
zn
)Sj (
)Sk (
)Ex,(i,j,k),
x
y
z
i
j
k

xn
yn
zn
)Sj (
)Sk (
)Ey,(i,j,k),
Si (
x
y
z
i
j
k

xn
yn
zn
)Sj (
)Sk (
)Ez,(i,j,k),
Si (
x
y
z
i
j
k

xn
yn
zn
)Sj (
)Sk (
)Bx,(i,j,k),
Si (
x
y
z
i
j
k

xn
yn
zn
)Sj (
)Sk (
)By,(i,j,k),
Si (
x
y
z
i
j
k

xn
yn
zn
Si (
)Sj (
)Sk (
)Bz,(i,j,k).
x
y
z
i
j
k

Ex (xn , yn , zn ) =
Ey (xn , yn , zn ) =
Ez (xn , yn , zn ) =
Bx (xn , yn , zn ) =
By (xn , yn , zn ) =
Bz (xn , yn , zn ) =

2.6
2.6.1

Si (

(2.54)
(2.55)
(2.56)
(2.57)
(2.58)
(2.59)

Current Density
Charge conservation method

While the charge density is dened at full-integer grid points (i, j, k), the current densities Jx ,
Jy and Jz are respectively dened at (i + 12 , j, k), (i, j + 12 , k) and (i, j, k + 12 ) from the dierence
form of the continuity equation for charge
t+ t
2

t+t (i, j, k) t (i, j, k) Jx


0 =
+
t
t+ t
2

Jy

t+ t
2

(i, j + 12 , k) Jy
y

t+ t
2

(i + 12 , j, k) Jx
x

(i, j 12 , k)

t+ t
2

Jz

(i 12 , j, k)
t+ t
2

(i, j, k + 12 ) Jz
z

(2.60)
(i, j, k 12 )

The current densities Jx , Jy and Jz at each grid point are computed with dierent shapefactors, because they are dened at dierent spatial grid points. They are computed by the
superposition of a charge ux contributed by each particle,
Np

1
1
xn
yn
zn
Jx (i + , j, k) =
)Sj (
)Sk (
),
qn vxn Si+ 1 (
2 x
2
xyz n=1
y
z

(2.61)

Np

1
1
xn
yn
zn
)Sj+ 1 (
)Sk (
),
Jy (i, j + , k) =
qn vyn Si (
2
2
xyz n=1
x
y
z

(2.62)

Np

1
xn
yn
zn
1
Jz (i, j, k + ) =
qn vzn Si (
)Sj (
)Sk+ 1 (
).
2
2
xyz n=1
x
y
z

(2.63)

It is well known that the current densities obtained by the simple area-weighting scheme (2.61)(2.63) do not satisfy the continuity equation for charge exactly. In this case we have to solve

2.6. CURRENT DENSITY

33

Poissons equation for correction of electric elds [Hockney and Eastwood, 1981; Birdsall and
Langdon, 1985].
There are several numerical techniques for solving the continuity equation locally, which
allows us to avoid solving Poissons equation at every time step. They are called charge
conservation methods. As described by Eastwood [1991], a charge ux of a particle can be
computed from the start and end points of the particle movement, when both start and end
points are located in the same cell. When a particle moves across the cell meshes, the particle
movement is assigned to separate motions in each cell by the cell meshes. From superposition of
charge ux, charge conservation can be realized for any particle trajectories made up of straight
line segments between any start and end points.
In the Villasenor-Buneman method [Villasenor and Buneman, 1992], a particle trajectory
over one time step is assumed to be a straight line. Therefore all start and end points of
the particle trajectory segments are located along the straight line as shown in Figure 2.5.
Especially in the case of Figure 2.5, the original particle movement is described with movements
of three particles. The algorithm of the Villasenor-Buneman method is complex, because the
computation of cross-points of a particle trajectory and cell meshes is realized with several IF
statements.
In Esirkepovs method [Esirkepov, 2001], a charge ux is decomposed into movements of
four particles along the x and y axes in two-dimensions as shown in Figure 2.6a. In threedimensions, a charge ux is decomposed into movements of twelve particles along the x, y and
z axes as shown in Figure 2.6b. Each segment has its own weight as specied in Figure 2.6,
because the sum of charge ux segments is equal to qv. Since each particle moves only in one
dimension, algorithm of Esirkepovs method becomes easier and is realized without any IF
statements. The computation speed also becomes faster than that of the Villasenor-Buneman
method. Another advantage of Esirkepovs method is that it is easy to apply to higher-order
shape-factors.
In charge conservation methods, a particle trajectory over one time step is conventionally
assumed to be a straight line. However, when a particle moves from (x, y, z) to (x + vx t, y +
vy t, z + vz t) for one time step, the particle can take an arbitrary trajectory as long as it does
not move more than one grid spacing, x > vx t, y > vy t, z > vz t. In other words, a
particle trajectory needs not to be a straight line. Recently, a new charge conservation method
was developed by assuming that a particle trajectory is a zigzag line. [Tominaga, 1998; Umeda
et al., 2003]. The algorithm of the zigzag scheme is realized without an IF statement. The
number of segments which a charge ux is decomposed into is less than that of the two schemes.
Thus the computational speed of the zigzag scheme is faster than the other two schemes without
any substantial distortion of physics. In this section we describe the new charge conservation
method in detail. We also extend the rst-order scheme to the second-order scheme.

34

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

Figure 2.5: Decomposition of a charge ux by the Villasenor-Buneman method in


two-dimensions. (a)The original particle trajectory. (b)The charge ux is decomposed
into segments J1 , J2 and J3 at the sides of the cell square, where qv = J1 + J2 + J3 .
The dashed lines represent cell meshes. The number of segments is determined by the
number of cross-points between a particle trajectory and cell meshes.

Figure 2.6: Decomposition of a charge ux by Esirkepovs method. (a)In two-dimensions


a charge ux is decomposed into four segments along the x and y axes.

(b)In

three-dimensions a charge ux is decomposed into twelve segments along the x, y and


z axes. Each segment has its own weight as specied in the gure.

2.6. CURRENT DENSITY

2.6.2

35

Zigzag scheme

Two-dimensional scheme
In two-dimensions the dierence form of the continuity equation is written as
t+ t
2

t+ t
2

(i + 12 , j) Jx
x
t
t+t
(i, j)
(i, j)
=
.
t
Jx

(i 12 , j)

t+ t
2

Jy

t+ t
2

(i, j + 12 ) Jy
y

(i, j 12 )
(2.64)

Let us think of a particle with a charge q moving from (xt , y t) to (xt+t , y t+t ). We dene
x1 xt ,

y1 y t ,

(2.65)

x2 xt+t = xt + vxt+t/2 t, y2 y t+t = y t + vyt+t/2 t,


and
i1 oor(x1 /x), j1 oor(y1 /y),

(2.66)

i2 oor(x2 /x), j2 oor(y2 /y),

where i and j are the largest integer values not greater than x/x and y/y respectively, which
are given by the oor function. We also assume that the particle does not move more than
grid spacings x and y for one time step t, namely, vx t < x, vy t < y.
When a particle remains in the same cell square during its movement, a charge ux of a
particle can be computed from the start point (x1 , y1 ) and end point (x2 , y2) of the particle
movement. Charge conservation of the assigned current densities is realized with the procedure
described by Eastwood [1991],
Jx (i1 + 12 , j1 )

Jx (i1 + 12 , j1 + 1) =

1
F (1
xy x
1
F W ,
xy x y

Wy ), Jy (i1 , j1 + 12 )

Jy (i1 + 1, j1 + 12 ) =

1
F (1
xy y
1
F W ,
xy y x

Wx ),

(2.67)

where (Fx , Fy ) F represents charge ux given by


y1
x1
, Fy = q y2t
,
Fx = q x2t

(2.68)

and W represents the rst-order shape factor corresponding to the linear weighting function
dened at the midpoint between the start point (x1 , y1 ) and the end point (x2 , y2 )
Wx =

x1 +x2
2x

i1 , Wy =

y1 +y2
2y

j1 .

(2.69)

When a particle moves across cell meshes, we decompose the particle movement with a special
assignment pattern as shown in Figures 2.7 and 2.8. We have to think of the following four
cases depending on positions of the particle at time t and t + t: (a)i1 = i2 and j1 = j2 ,
(b)i1 = i2 and j1 = j2 , (c)i1 = i2 and j1 = j2 , and (d)i1 = i2 and j1 = j2 .

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

36

Figure

2.7:

Particle trajectory of the two-dimensional zigzag scheme for i1 = i2 and j1 = j2 (case


a). One particle moves from (x1 , y1 ) to ((i1 + 1)x, (j1 + 1)y) = (i2 x, j2 y), and
another particle moves from (i2 x, j2 y) to (x2 , y2 ) during the time from t to t + t.
The solid arrows represent particle trajectories. The dashed lines represent cell meshes.

Figure 2.8: Particle trajectory of the two-dimensional zigzag scheme for i1 = i2 , j1 = j2


2
(case b). One particle moves from (x1 , y1 ) to ((i1 + 1)x, y1 +y
2 ), and another particle
2
moves from (i2 x, y1 +y
2 ) to (x2 , y2 ) during the time from t to t + t. Here i1 + 1 = i2 .

The solid arrows represent particle trajectories. The dashed lines represent cell meshes.
By interchanging x and y, the case (b) is transformed to the case (c).

2.6. CURRENT DENSITY

37

(a) When i1 = i2 and j1 = j2 , the particle moves across two cell meshes. As shown in
Figure 2.7, we assume that the movement of the particle from (x1 , y1 ) to (x2 , y2 ) is described
as movements of two particles. One particle moves from (x1 , y1) to ((i1 + 1)x, (j1 + 1)y) =
(i2 x, j2 y), and another particle moves from (i2 x, j2 y) to (x2 , y2 ) during the time from t
to t + t. Therefore the particle trajectory becomes a zigzag line. If the particle trajectory
is assumed to be a straight line, the particle movement is decomposed into three segments (see
Figure 2.5). By assuming that the particle trajectory is such a zigzag line in Figure 2.7, the
cross-point of the particle trajectory and cell meshes can be determined independently of the
positions of the particle at t and t + t. Since we do not need to compute two cross-points of
the particle trajectory and cell meshes, computation becomes much easier and faster.
(b) When i1 = i2 and j1 = j2 , the particle moves to one of the adjacent cells in the x direction.
2
2
In this case one particle moves from (x1 , y1) to ((i1 + 1)x, y1 +y
) = (i2 x, y1 +y
), and another
2
2
2
particle moves from (i2 x, y1 +y
) to (x2 , y2 ) during the time from t to t + t. The particle
2

trajectory also becomes a zigzag line as shown in Figure 2.8.


(c) When i1 = i2 and j1 = j2 , the particle moves to one of the adjacent cells in the y direction.
2
2
We assume that one particle moves from (x1 , y1) to ( x1 +x
, (j1 + 1)x) = ( x1 +x
, j2 x), and
2
2
2
that another particle moves from ( x1 +x
, j2 x) to (x2 , y2) during the time from t to t + t.
2

Thus the particle trajectory becomes very similar to that in Case (b).
In both Cases (b) and (c), the number of segments is two, and is same as that of the case of
a straight line path. However, since we do not need to compute one cross-point of the particle
trajectory and a cell mesh, computation becomes easier.
(d) When i1 = i2 and j1 = j2 , the particle remains in the same cell square during its movement. In this case we do not need to decompose the particle movement into two segments,
and we can directly apply the procedure in Eq.(2.67) for computation of charge ux at each
grid point. However, we formally assume that the movement of a particle from (x1 , y1 ) to
(x2 , y2 ) is described as movements of two particles. That is, one particle moves from (x1 , y1) to
2 y1 +y2
2 y1 +y2
( x1 +x
, 2 ), and another particle moves from ( x1 +x
, 2 ) to (x2 , y2 ) during the time from t
2
2

to t + t.
Algorithm To reduce the four cases listed above, we introduce a relay point of particle
movement (xr , yr ) and the charge uxes F1 and F2 . That is, one particle moves from (x1 , y1 )
to (xr , yr ) and another moves from (xr , yr ) to (x2 , y2 ) during the time from t to t + t. In a
simple one-dimensional case, the relay point xr is given by

x1 +x2

xr =

, if i1 = i2 ,

max(i1 x, i2 x), if i1 = i2 .

(2.70)

38

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

In this paper we propose the following algorithm for computing the relay point (xr , yr ) without
any IF statement,






2
xr = min min(i1 x, i2 x) + x, max max(i1 x, i2 x), x1 +x
2
2
yr = min min(j1 y, j2 y) + y, max max(j1 y, j2 y), y1 +y
2

,
.

(2.71)

It is noted that computation speeds of the above two algorithms depend on CPU and compiler.
In general, computation of an algorithm with IF statements takes much time. However, the
algorithm in Eq.(2.70) is also useful on a computer system in which computation speed of an
IF statement is very fast.
By using (xr , yr ), a charge ux qv = q(vx , vy ) is decomposed into F1 = (Fx1 , Fy1 ) and
F2 = (Fx2 , Fy2 ) as follows:
y1
Fy1 = q yrt
,

x1
Fx1 = q xrt
,

yr
xr
Fx2 = q x2t
= qvx Fx1 , Fy2 = q y2t
= qvy Fy1 .

(2.72)

We next apply the algorithm in Eq.(2.67) for computation of charge ux at each grid point.
That is, we apply the shape-factors dened at the midpoints between (x1 , y1 ) and (xr , yr ), and
(xr , yr ) and (x2 , y2), respectively, to assign the segments F1 and F2 to the adjacent grid points.
This is because the particles exist at the midpoints at time t +

t
2

[Eastwood, 1991; Villasenor

and Buneman, 1991]. The rst-order shape-factors at the midpoints are dened by
Wx1
Wx2

x1 +xr
2x
xr +x2
2x

i1 , Wy1 =
i2 , Wy2 =

y1 +yr
2y
yr +y2
2y

j1 ,
j2 .

(2.73)

The segments of the charge ux assigned to 8 grid points are obtained by the following procedure.
Jx (i1 + 12 , j1 )

Jx (i1 + 12 , j1 + 1) =
Jx (i2 + 12 , j2 )

Jx (i2 + 12 , j2 + 1) =
Jy (i1 , j1 + 12 )

Jy (i1 + 1, j1 + 12 ) =
Jy (i2 , j2 + 12 )

Jy (i2 + 1, j2 + 12 ) =

1
F (1 Wy1 ),
xy x1
1
F W ,
xy x1 y1
1
F (1 Wy2 ),
xy x2
1
F W ,
xy x2 y2
1
F (1 Wx1 ),
xy y1

(2.74)

1
F W ,
xy y1 x1
1
F (1 Wx2 ),
xy y2
1
F W .
xy y2 x2

Finally we superpose the charge uxes contributed by each particle to obtain the total current
densities. We note that we can apply the simple area-weighting scheme (2.21) for computation
of Jz component in two-dimensions, because Jz is free from the charge continuity equation
(2.64).

2.6. CURRENT DENSITY

39

Three-dimensional scheme
In three-dimensions we have to think of eight cases depending on a particle movement whether
a particle moves across cell meshes or not. The eight cases can be reduced to four cases as
listed in Table 2.1. It is very easy to extend the two-dimensional zigzag scheme to the threedimensional one. Using the same procedure of the two-dimensional case. we can illustrate
particle trajectories for four cases in Figure 2.9. If the particle trajectory between the start
point (x1 , y1 , z1 ) and end point (x2 , y2 , z2 ) is a straight line, the number of segments varies 1 4
depending on the locations of the start and end points. By using the zigzag paths, the number
of segments is reduced to 2.

Table 2.1: Dierent cases of the particle positions at times t and t + t in the
three-dimensional zigzag scheme.

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

i1 = i2

j1 = j2

k1 = k2

Case (a)
Case (b)

Case (c)
Case (d)

In all cases a charge ux is also decomposed into two segments: F1 (Fx1 , Fy1 , Fz1 ) and
F2 (Fx2 , Fy2 , Fz2 ). By using a relay point (xr , yr , zr ) given by the algorithm in Eq.(2.71) or
Eq.(2.70) each segment of the charge ux is given by
y1
Fy1 = q yrt
,

x1
Fx1 = q xrt
,

z1
Fz1 = q zrt
,

yr
xr
zr
= qvx Fx1 , Fy2 = q y2t
= qvy Fy1 , Fz2 = q z2t
= qvz Fz1 .
Fx2 = q x2t

(2.75)

We assign each segment of the charge ux to the adjacent grid points using the following
rst-order shape-factors dened at the midpoints
Wx1 =
Wx2 =

x1 +xr
2
xr +x2
2

i1 , Wy1 =
i2 , Wy2 =

(xr ,yr ,zr )+(x1 ,y1 ,z1 )


2

y1 +yr
2
yr +y2
2

j1 , Wz1 =
j2 , Wz2 =

and

z1 +zr
2
zr +z2
2

(x2 ,y2 ,z2 )+(xr ,yr ,zr )


,
2

k1 ,
k2 .

(2.76)

The charge uxes F1 and F2 are assigned to 24 grid points as written in the following procedure.

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

40

Figure 2.9: Particle trajectory of the three-dimensional zigzag scheme with the second-order shape factor. (a)For i1 = i2 and j1 = j2 and k1 = k2 , one particle moves
from (x1 , y1 , z1 ) to ((i1 + 1)x, (j1 + 1)y, (k1 + 1)z, and another particle moves
from ((i1 + 1)x, (j1 + 1)y, (k1 + 1)z to (x2 , y2 , z2 ) during the time from t to
t + t.

(b)For i1 = i2 , j1 = j2 , k1 = k2 , one particle moves from (x1 , y1 , z1 ) to

z1 +z2
2
((i1 +1)x, (j1 +1)y, z1 +z
2 ). Another particle moves from ((i1 +1)x, (j1 +1)y,
2 )

to (x2 , y2 , z2 ). (c)For i1 = i2 , j1 = j2 , k1 = k2 , one particle moves from (x1 , y1 , z1 )


y1 +y2 z1 +z2
2 z1 +z2
to ((i1 + 1)x, y1 +y
2 ,
2 ). Another particle moves from ((i1 + 1)x,
2 ,
2 ) to

(x2 , y2 , z2 ). (d)For i1 = i2 and j1 = j2 and k1 = k2 , one particle moves from (x1 , y1 , z1 )


x1 +x2 y1 +y2 z1 +z2
2 y1 +y2 z1 +z2
to ( x1 +x
2 ,
2 ,
2 ). Another particle moves from ( 2 ,
2 ,
2 ) to (x2 , y2 , z2 ).

The dashed lines represent cell meshes.

2.6. CURRENT DENSITY

41

For l = 1, 2
Jx (il + 12 , jl , kl )

Jx (il + 12 , jl + 1, kl )

Jx (il +
Jx (il +

1
, j , kl + 1)
2 l
1
, j + 1, kl +
2 l

=
1) =

Jy (il , jl + 12 , kl )

1
, kl )
2

Jy (il , jl + 12 , kl + 1)

Jy (il + 1, jl +

Jy (il + 1, jl + 12 , kl + 1) =
Jz (il , jl , kl + 12 )

Jz (il + 1, jl , kl + 12 )

Jz (il , jl + 1, kl + 12 )
Jz (il + 1, jl + 1, kl +

=
1
)
2

1
F (1 Wyl )(1
xyz xl
1
F W (1 Wzl )
xyz xl yl
1
F (1 Wyl )Wzl
xyz xl
1
F W W
xyz xl yl zl

Wzl )

1
F (1 Wxl )(1
xyz yl
1
F W (1 Wzl )
xyz yl xl

Wzl )

1
F (1 Wxl )Wzl
xyz yl
1
F W W
xyz yl xl zl
1
F (1 Wxl )(1
xyz zl

(2.77)

Wyl )

1
F W (1 Wyl )
xyz zl xl
1
F (1 Wxl )Wyl
xyz zl
1
F W W .
xyz zl xl yl

Total current densities are obtained by superposition of the charge ux contributed by each
particle.

Test simulation runs


We performed test simulation runs using a new two-dimensional electromagnetic PIC codes.
We used the Villasenor-Buneman method and the present zigzag scheme for computation of
current densities. The simulation source codes were compiled with Intel Fortran Complier 7.0,
and the test runs were performed on a single Xeon 2.4GHz processor.
As a simulation on model, we set the external static magnetic eld in the x direction. The
electron cyclotron frequency is equal to the total electron plasma frequency e = e = 1.0. We
assumed two electron beams with the equal density and the equal thermal velocity Vt = 1.0.
One electron beam drifts against the other electron beam with drift velocity Vd = 4.0Vt along
the magnetic eld. Ions are assumed to be immobile background, and we used a reduced
speed of light c = 20Vt . We set that time step is e t = 0.025, and that grid spacing is
x = y = 1.0D , where D = Vt /e is Debye length. Two-dimensional simulation runs were
performed with 256 256 grid points. We varied number of superparticles per cell for both
electron beams, ne1 and ne2 , as a parameter for dierent simulation runs.
We rst measured average CPU time used for computation of current densities for one time
step. As shown in Figure 2.10, the computation speed of the two-dimensional zigzag scheme
is about 1.5 times as fast as that of the Villasenor-Buneman method. We also measured used
CPU time of three-dimensional scheme (not shown). In this case, computation speed of the
zigzag scheme is also about 1.5 times as fast as that of the Villasenor-Buneman method. We
note that the zigzag scheme equalizes the computation load balance of each processor in parallel

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

42

Figure 2.10: Average CPU time used for the computation of current density for one time
step. The experimental data of the present zigzag scheme and the Villasenor-Buneman
method are respectively plotted as circle and x-marks.

computation. Another advantage of the zigzag scheme is that the algorithm is much simple
and therefore easy for programming. It is more suitable for vector computation.
To analyze electrostatic elds, we plotted spatial proles of electrostatic potentials for the
two-dimensional runs with ne1 = ne2 = 82 in Figure 2.11. The proles of electrostatic potentials
are given by solving Poissons equation. In all runs we started with the same initial condition.
At the saturation stage (e t = 64), all runs give the same spatial proles of the electrostatic
potentials. At e t = 256, there is still little dierence of the spatial proles of electrostatic
potentials between the three methods. At e t = 1024, on the other hand, the dierence of
electrostatic potentials between the the three methods becomes more apparent. This dierence
is due to the accumulation of numerical error. The numerical error changes proles of background thermal noises. When we used fewer number of superparticles per cell, the dierence
of electrostatic potentials between the three methods appears in an earlier stage.
Next we analyzed electromagnetic elds. To compute spectrum of the enhanced thermal
uctuations, we performed Fourier transformation of Bz simulation data in both time and
space. The left panels of Figure 2.12 correspond to spectral wave intensity |cBz (, kx , ky )| at
ky e = 0.0245, and the right panels correspond to spectral wave intensity |cBz (, kx, ky )| at
kx e = 0.0245. As in the case of electrostatic elds, the results of the run with the zigzagscheme are very similar to those of the run with the Villasenor-Buneman method. The present
test simulation runs show that electrostatic elds as well as electromagnetic elds obtained
by the zigzag scheme are the same as those obtained by the conventional charge conservation
method.

2.6. CURRENT DENSITY

Figure 2.11: Spatial proles of electrostatic potentials in the two-dimensional runs with
ne1 = ne2 = 82 , for (a)the zigzag scheme and (c)the Villasenor-Buneman method.

Figure 2.12: Spectral wave intensity |cBz | in the runs with ne1 = ne2 = 82 , for
(a)the zigzag scheme and (c)the Villasenor-Buneman method. (left) kx diagram at
ky e = 0.0245 and (right) ky diagram at ky e = 0.0245. The intensity is normalized
by the thermal energy density of the electrons ne me Vt2 = (me e Vt /e)2 .

43

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

44

The numerical error between the three methods arises when a particle trajectory crosses
the cell meshes. When we consider the current density computed from a motion of only one
particle, the obtained value at each grid point is very dierent from each other. This is because
a segment of the charge ux is assigned with dierent weighting functions. However, when we
take average of Jx , Jy and Jz at adjacent grid points, the results respectively become qvx , qvy
and qvz , i.e., the same for the three methods. In other words, when distribution and motion of
particles in adjacent cells are almost uniform, the numerical error between the three methods
becomes almost zero. A wave mode with the minimum wave length is supported by two grid
spacings. Such a mode with a short wavelength is subject to the numerical error. However,
since wavelengths of electromagnetic modes are commonly much longer than two grid spacings,
they are not aected by the numerical error. By superposing a large number of particle motions,
distribution and motion of particles in adjacent cells can be regarded as uniform, and therefore
the numerical error of one particle does not become eective. The trajectory of each particle
does not aect the simulation result as conrmed by the present test simulations.

2.6.3

Extention to second-order shape factor

Decomposition of charge ux in one-dimensions


We apply the algorithm the zigzag scheme to the second-order shape factor Eq.(2.13). The
algorithm of the second-order scheme is the same as that of the rst-order scheme [Umeda et
al., 2003], while a trajectory of a particle movement is slightly modied. For simplicity, we rst
describe the one-dimensional scheme with the second-order shape factor. By using the same
procedure of the rst-order scheme, a trajectory of the particle movement for one time step is
decomposed into two segments depending on positions of the particle at time t and t + t.
When i1 = i2 , one particle moves from x1 to the midpoint (x1 + x2 )/2, and the other moves
from the midpoint to x2 during the time from t to t + t. When i1 = i2 , one particle moves
from x1 to the boundary between the territories of i1 and i2 , and the other moves to x2 during
the time from t to t + t. The relay point xr for the dierent cases is given by

(x1 +x2 )

xr =

, if i1 = i2 ,

2
(i1 +i2 )x
,
2

(2.78)

if i1 = i2 .

In the second-order scheme, the current densities at i1 12 , i1 + 12 , i2

1
2

and i2 +

1
2

are

assigned by keeping the following conditions.


Jx,(i1 + 1 ) Jx,(i1 1 ) =
2

Jx,(i2 + 1 ) Jx,(i2 1 ) =

q xr x1 xr +x1
( x
x t
q x2 xr x2 +xr
( x
x t

Therefore the current densities at i1


Jx,(i1 1 ) =
2

Jx,(i1 + 1 ) =
2

q xr x1
r +x1
(i1 x2x
x t
q xr x1 xr +x1
( 2x i1
x t

2i1 ), Jx,(i1 + 1 ) + Jx,(i1 1 ) =


2

2i2 ), Jx,(i2 + 1 ) + Jx,(i2 1 ) =

1
, i1
2

1
, i2
2

1
2

and i2 +

+ 12 ), Jx,(i2 1 ) =
2

+ 12 ), Jx,(i2 + 1 ) =
2

1
2

q xr x1
,
x t
q x2 xr
.
x t

(2.79)

are respectively given by

q x2 xr
2 +xr
(i2 x2x
x t
q x2 xr x2 +xr
( 2x i2
x t

+ 12 ),
+ 12 ).

(2.80)

2.6. CURRENT DENSITY

45

Three-dimensional scheme
The particle trajectories of the three-dimensional zigzag scheme with the second-order shape
factor are modied from those with the rst-order shape factor in Figure 2.9. (a) When i1 =
2 (j1 +j2 )y (k1 +k2 )z
i2 and j1 = j2 and k1 = k2 , one particle moves from (x1 , y1 , z1 ) to ( i1 +i
,
,
),
2
2
2

and another particle moves from ( (i1 +i22 )x , (j1 +j22 )y , (k1 +k22 )z ) to (x2 , y2 , z2 ) during the time
from t to t + t. (b) When i1 = i2 , j1 = j2 , k1 = k2 , one particle moves from (x1 , y1, z1 )
2
2
to ( (i1 +i22 )x , (j1 +j22 )y , z1 +z
), and another particle moves from ( (i1 +i22 )x , (j1 +j22 )y , z1 +z
) to
2
2

(c) When i1 = i2 , j1 = j2 , k1 = k2 , one

(x2 , y2 , z2 ) during the time from t to t + t.


particle moves from (x1 , y1 , z1 ) to
2 z1 +z2
( (i1 +i22 )x , y1 +y
, 2 )
2

2 z1 +z2
( (i1 +i22 )x , y1 +y
, 2 ),
2

and another particle moves from

to (x2 , y2 , z2 ). (d) When i1 = i2 and j1 = j2 and k1 = k2 , one particle

2 y1 +y2 z1 +z2
2 y1 +y2 z1 +z2
moves from (x1 , y1 , z1 ) to ( x1 +x
, 2 , 2 ), and another particle moves from ( x1 +x
, 2 , 2 )
2
2

to (x2 , y2, z2 ). The relay point of the particle movement between (x1 , y1 , z1 ) and (x2 , y2, z2 ), i.e.,
(xr , yr , zr ), for the dierent cases is given by

(y1 +y2 )

yr =
zr =

, if j1 = j2 ,

2
(j1 +j2 )y
,
2

(z1 +z2 )

(2.81)

if j1 = j2 .

, if k1 = k2 ,

2
(k1 +k2 )z
,
2

(2.82)

if k1 = k2 .

and Eq.(2.78).
To assign charge uxes to each grid point, we apply the area-weighting scheme. The shape
factor for assignment of charge uxes is dened at the midpoint of the particle movement,
because the particle exists at the midpoint at time t +

t
2

[Eastwood, 1991; Villasenor and

Buneman, 1992]. For simplicity we dene segments of the charge ux F1 = (Fx1 , Fy1 , Fz1 ) and
F2 = (Fx2 , Fy2 , Fz2 ) as
y1
x1
z1
Fx1 q xrt
, Fy1 q yrt
, Fz1 q zrt
,

(2.83)

yr
xr
zr
, Fy2 q y2t
, Fz2 q z2t
,
Fx2 q x2t

1 ,y1 ,z1 )
. We also dene x1 , y1 , z1 , x2 , y2 and z2 as
where F1 + F2 = qv = q (x2 ,y2 ,z2 )(x
t

x1
x2

xr +x1
2x
x2 +xr
2x

i1 , y1
i2 , y2

Then the shape factors at the midpoints


Wi1 1 =
Wj1 1 =
Wk1 1 =
Wi2 1 =
Wj2 1 =
Wk2 1 =

1
2
1
2
1
2
1
2
1
2
1
2

2
1

2
1

2
1
2
1
2
1
2

x1
y1
z1

x2
y2
z2

yr +y1
2y
y2 +yr
2y

j1 , z1
j2 , z2

(x1 ,y1 ,z1 )+(xr ,yr ,zr )


2

, Wi1 =
, Wj1 =
, Wk1 =
, Wi2 =
, Wj2 =
, Wk2 =

3
4
3
4
3
4
3
4
3
4
3
4

and

zr +z1
2z
z2 +zr
2z

k1 ,

(2.84)

k2 .

(xr ,yr ,zr )+(x2 ,y2 ,z2 )


2

x21 , Wi1 +1 =
y12, Wj1 +1 =
z12 , Wk1 +1 =
x22 , Wi2 +1 =
y22, Wj2 +1 =
z22 , Wk2 +1 =

1
2
1
2
1
2
1
2
1
2
1
2

2
1

2
1

2
1
2
1
2
1
2

+ x1
+ y1
+ z1

+ x2
+ y2
+ z2

are given by
,
,
,
,
,
.

(2.85)

46

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

The charge ux Fx 1 are assigned to the grid points in the territory of (i1 , j1 , k1 ) as written
in the following procedure.
Jx,(i1 1 ,j11,k1 1) =
2

Jx,(i1 1 ,j1,k1 1)
2

Jx,(i1 1 ,j1+1,k1 1) =
2

Jx,(i1 1 ,j11,k1 )

Jx,(i1 1 ,j1,k1 )

Jx,(i1 1 ,j1,k1 +1)

Jx,(i1 + 1 ,j1+1,k1 )

2
2
2
2

Jx,(i1 + 1 ,j11,k1 +1) =


2

Jx,(i1 1 ,j1+1,k1 +1) =


2

Jx,(i1 + 1 ,j11,k1 1) =
2

Jx,(i1 + 1 ,j1,k1 1)
2

Jx,(i1 + 1 ,j1+1,k1 1) =
2

Jx,(i1 + 1 ,j11,k1 )

Jx,(i1 + 1 ,j1,k1 )

Jx,(i1 1 ,j1+1,k1 )

2
2
2

Jx,(i1 1 ,j11,k1 +1) =


2

Jx,(i1 + 1 ,j1,k1 +1)


2

Jx,(i1 + 1 ,j1+1,k1 +1) =


2

1
F (1
xyz x1 2
1
F (1
xyz x1 2

x1 )Wj1 1 Wk1 1 ,

1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2
1
F (1
xyz x1 2

x1 )Wj1 +1 Wk1 1 ,

x1 )Wj1 Wk1 1 ,
x1 )Wj1 1 Wk1 ,
x1 )Wj1 Wk1 ,
x1 )Wj1 +1 Wk1 ,
x1 )Wj1 1 Wk1 +1 ,
x1 )Wj1 Wk1 +1 ,
x1 )Wj1 +1 Wk1 +1 ,
+ x1 )Wj1 1 Wk1 1 ,

(2.86)

+ x1 )Wj1 Wk1 1 ,
+ x1 )Wj1 +1 Wk1 ,
+ x1 )Wj1 1 Wk1 ,
+ x1 )Wj1 Wk1 ,
+ x1 )Wj1 +1 Wk1 ,
+ x1 )Wj1 1 Wk1 +1 ,
+ x1 )Wj1 Wk1 +1 ,
+ x1 )Wj1 +1 Wk1 +1 .

Similarly the other charge uxes are assigned to each grid point as follows:
Jy,(ii,j1 1 ,kk) =
2

Jy,(ii,j1 + 1 ,kk) =
2

Jz,(ii,jj,k1 1 ) =
2

Jz,(ii,jj,k1+ 1 ) =
2

1
F (1
xyz y1 2
1
F (1
xyz y1 2
1
F (1
xyz z1 2
1
F (1
xyz z1 2

y1 )Wkk Wii ,
+ y1 )Wkk Wii ,
z1 )Wii Wjj ,
+ z1 )Wii Wjj ,

for ii = i1 1, i1 , i1 + 1, jj = j1 1, j1 , j1 + 1, kk = k1 1, k1, k1 + 1,
Jx,(i2 1 ,jj,kk) =
2

Jx,(i2 + 1 ,jj,kk) =
2

Jy,(ii,j2 1 ,kk) =
2

Jy,(ii,j2 + 1 ,kk) =
2

Jz,(ii,jj,k2 1 ) =
2

Jz,(ii,jj,k2+ 1 ) =
2

1
F (1
xyz x2 2
1
F (1
xyz x2 2
1
F (1
xyz y2 2
1
F (1
xyz y2 2
1
F (1
xyz z2 2
1
F (1
xyz z2 2

x2 )Wjj Wkk ,
+ x2 )Wjj Wkk ,

(2.87)

y2 )Wkk Wii ,
+ y2 )Wkk Wii ,
z2 )Wii Wjj ,
+ z2 )Wii Wjj ,

for ii = i2 1, i2 , i2 + 1, jj = j2 1, 21 , j2 + 1, kk = k2 1, k2 , k2 + 1.
Total current densities are obtained by superposition of the charge uxes contributed by each
particle.

2.7. INITIALIZATION

47

Figure 2.13: Energy histories of test runs with (left)the rst-order shape factor and
(right)the second-order shape factor. The particle kinetic energy density, electric eld
energy density, magnetic eld energy density, and the total energy density are plotted as
derivations from initial values. The simulation parameters are the same as the test runs
in Section 2.6.2. All values are normalized by 12 e me Vte2 .

By using the second-order scheme for computation of both current densities and particle
velocities, the computation time becomes 2 times as long as that of the rst-order scheme in
two-dimensional simulations, and 3 times in three-dimensional simulations. However, with the
second-order scheme, the enhanced thermal uctuations with higher wavenumbers are much
more suppressed and the energy conservation are more improved. The numerical heating of
particles causes nonphysical increase in electric and magnetic eld energy densities as found
in the left panel of Figure 2.13. By applying the second-order shape factor for computation
of both current densities and particle velocities, such nonphysical increase in the eld energy
densities is suppressed as found in the right panel of Figure 2.13. The decrease in the kinetic
energy density and total energy density is due to relocation of elds described in Section 2.5.2.
The relocation of elds have an eect of numerical lter.

2.7
2.7.1

Initialization
Field initialization

As the initial condition, both electric eld and magnetic eld should satisfy Eqs.(2.3) and
(2.4). In the present simulation code we decompose the magnetic eld into electromagnetic
eld components and static ambient magnetic eld components,
B total = B em + B 0 .

(2.88)

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

48

We advance only the electromagnetic eld components B em by solving Maxwells equations


(2.1) and (2.2). The static ambient magnetic eld B 0 is only used to advance velocities of
particles.
As the initial condition, we dene
E em = 0 at t = 0

(2.89)

.
B em = 0 at t = t
2
Thus the initial electromagnetic elds are given by
= t=0

E t=0
B

t= t
2

(2.90)

= 0.

The ambient magnetic eld is determined so that B 0 satises Eq.(2.4). The initial electrostatic eld at t = 0 is determined by the spatial prole of the initial electrostatic potential
,
(2.91)

t=0
Ey,(i,j+
1
,k)

(2.92)

t=0
Ez,(i,j,k+
1
)
2

2.7.2

t=0
t=0
(i,j,k) (i+1,j,k)
x
t=0

t=0
(i,j,k)
(i,j+1,k)
=
y
t=0
(i,j,k) t=0
(i,j,k+1)
.
=
z

t=0
Ex,(i+
1
,j,k) =

(2.93)

Particle loading

The initial conditions on the particle distribution functions are very important in determining the later behavior of the system. The particle loading involves staring with prescribed
densities in space d0 (r) and in velocity f0 (v) and generating particle positions and velocities
t= t
2

(r t=0
n , vn

). In the present simulation code we used the method called inversion of cumula-

tive distribution functions [e.g., Birdsall and Langdon, 1985; Cartwright et al., 2000]. In this
t= t
2

methid, the initial densities are inverted to obtain (r t=0


n , vn

).

Supporse that we place particles in the one-dimensonal system to form a distribution function
h(x) from x = xmin to x = xmax . The cumulative distribution function is given by
x

H(x)  xxmin
max
xmin

h(x )dx
,
h(x )dx

(2.94)

where H(xmin ) = 0 and H(xmax ) = 1. By equating D(xm ) to a uniform distribution of numbers


Rm : 0 < Rm < 1, positions xm corresponding to the distribution h(x) are produced.
We assume that all particles have the Maxwellain velocity distrubitions given by

f (v) =

1
2Vt2

3
2

(v V d )2
exp
.
2Vt2

(2.95)

2.8. RENORMALIZATION

49

To obtain the velocities with the one-dimensional Maxwellain velocity distrubitions, the cumulative distribution function for the velocity
v

vmin

Vd )
exp (v 2V
2

vmin

Vd )
exp (v 2V
2

Rm = F (v)  vmax


2

dv 
dv 

(2.96)

is set equal to uniformly distributed numbers Rm , varying from 0 to 1.

2.8

Renormalization

We normalize input parameters so that they are converted to those normalized to the unit
system used in the simulation. To increase computational eciency, it is necessary to reduce
the number of operations involved in the dierence equations. In the present simulation code,
we adopted the cubic grid system, x = y = z = r. We normalize distance and time by
r and

t
,
2

respectively.

Let us dene two dierent unit systems. One is a unit system for real physical quantities
used as the input parameters (R-unit system). The other is a renormalized unit system used
in the simulation code (S-unit system). The grid spacing rR and time step tR in the R-unit
system can take arbitrary values as long as they satisfy the CFL condition. On the other hand,
in the S-system, they are xed as rS = 1, tS = 2. By assuming that x = rR and t = tR ,
2

all physical quantities are renormalized as follows:


1
x
x
R
1
t
t R

t
v
x
R
t2
E
x
R

distance

xS =

time

tS =

velocity

vS =

electric eld

ES =

magnetic eld

BS = tBR

current density

JS =

charge density

S =

energy density

S =

number density nS =
charge

qS =

mass

mS =

t3
J
x
R
2
t R
t4

x
2 R
D
x nR
t2
q
x
D R
t2
mR
x
D

where, D represents dimension of the system.

2.9

Particle Injection and Absorption at Bondaries

In particle simulations of space plasmas, both closed and open systems are used for studies
of nonlinear wave-particle interaction in space plasmas. Periodic boundary conditions and

50

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

reective boundary conditions are used to simulate micro-scale phenomena such as temporal
evolution of closed systems. It is easy to realize these boundary conditions. In a periodic
system, particles going over one boundary of the simulation system are injected from the other
boundary without any change of their velocity. That is,
new
old
vxn
= vxn
,

(2.97)

old
= xold
xnew
n
n + (xmin xmax ) sign(vxn ),

(2.98)

t
t
old
t
where xold
n = xn + vxn t, vxn = vxn . When a reective bondary condition is applied, the new

velocity of a particle is given by


new
old
= vxn
.
vxn

(2.99)

The new position is given by


= xold
xnew
n
n + 2xmin ,

(2.100)

= xold
xnew
n
n + 2xmax ,

(2.101)

at the left boundary, and

at the right bounday.


On the other hand, large-scale phenomena such as spatial evolution of localized perturbation
are studied in open systems rather than closed systems. Open systems are regarded as one
part of an innite space. Boundary conditions of open systems are called open or free
boundaries. Since we pick up a segment of real space plasmas as a simulation box, we need a
special numerical treatment for both particles and elds at the boundary between the simulation
box and the surrounding outer region. A requirement for open boundary conditions is to behave
as if there were no boundaries. In an open system, therefore, particles going over one boundary
of the system should not aect physical phenomena in the system any more. These particles are
absorbed (removed) at the boundary and we dont solve their motion, charge and current any
more. These particles are stocked and reused as new particles to be injected from boundaries
with new initial positions and velocities.
To inject particles with certain velocity distribution f (v) smoothly, let us assume an x vx
phase space of the one-dimensional system as shown in Figure 2.14. The particles in the left
triangle in the x vx phase space shown in Figure 2.14 come into the system for t. The
number of particles coming into the system for t with velocity v1 is gives by v1 f (v1 )t. This
indicates that particles coming into the system have a velocity distribution given by vf (v), i.e.
ux distribution. By assuming a normalized Maxwellian distribution


1
(v Vd )2

f (v) =
exp
,
2Vt2
2Vt

(2.102)

2.9. PARTICLE INJECTION AND ABSORPTION AT BONDARIES

51

Figure 2.14: The x vx phase space for the one-dimensional open system. Particles in
the left triangle come into the simulation system.

where Vd and Vt respectively represent drift velocity and thermal velocity, the number of particles coming into the system with positive velocity is given by
N + = n vx+ t,

(2.103)

where vx+ is the average velocity given by integration of ux distribution function over velocity,
vx+

 +

=
0

V2
Vt
uf (u)du = exp d 2
2Vt
2

Vd
+
1 + erf
2

V
d
2Vt



(2.104)

Here erf(x) is the error function given by


2
erf(x) =

 x

exp(t2 )dt.

(2.105)

The number of particles coming into the system with negative velocity is similarly given by
N = n vx t = nt

 0

uf (u)du.

(2.106)

To generate positions and velocities of injected particles, we adopt inversion of cumulative


distribution functions which is used for particle loading in Section 2.7.2. The cumulative distribution function of the ux vf (v) is given by
v

0 uf (u)du
F (v)  vmax
for v 0
uf (u)du
0
0
uf (u)du
for v < 0
 0v
vmin uf (u)du

(2.107)
(2.108)

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

52

Figure 2.15: Schematic illustration of particle injection [after Birdsall and Langdom, 1985].

By equating F (vm ) to a uniform distribution of numbers Rm : 0 < Rm < 1, velocities vm


corresponding to the distribution vf (v) are produced.
A particle injected at a boundary with a velocity vn is placed in the simulation system at
point xn = Rn vn t, where Rn is a uniform random number for each particle, tending to ll
the right triangle in the phase space of Figure 2.14. The procesure of the particle injection is
illustrated in Figure 2.15 [Birdsall and Langdom, 1985].

2.10

Boundary Treatment for Fields

2.10.1

Absorbing boundary conditions

In Section 2.9, numerical treatment for particles at open boundaries is described. In this section
we describe numerical treatment for elds.
It is easy to realize closed boundary conditions. A periodic boundary condition is realized
by setting that eld quantities at one boundary is equal to those at the other boundary,
HF (Nx + 1) = HF (1),
HH (0)

= HH (Nx ),

HH (Nx + 1) = HH (1),

(2.109)
(2.110)

where the subscripts F and H respectively represent full-integer grid and half-integer grid.
Reective boundary conditions are realized by setting that eld quantities at a boundary is
equal to those at the neighboring grid point as a mirror,
HH (0)

= HH (1),

HH (Nx + 1) = HH (Nx ).

(2.111)

2.10. BOUNDARY TREATMENT FOR FIELDS

53

On the other hand, at open boundaries, outgoing waves leave the system without non-physical
reection. We need to realize such a boundary condition called an absorbing boundary.
Algorithms of absorbing boundaries fall into two categories. In the algorithms in the rst
category [e.g., Lindman, 1975; Mur, 1981], eld quantities at boundaries are estimated by
assuming that outgoing electromagnetic waves have a certain phase velocity. Scalar wave
equations are solved to approximate boundary values, so that outgoing waves are not reected
at the boundaries. While there are many kinds of plasma waves which have dierent frequencies
and phase velocities, the algorithms in the rst category can absorb a single wave mode only.
The algorithms in the second category are to absorb outgoing waves using a resistive medium
such as a dielectric. In laboratory experiments, so called a wave absorber is located on the wall.
Absorbing boundaries with resistive mediums are useful, because they can suppress reection
of various plasma waves with dierent phase velocities at boundaries. However, it requires
additional working area of computer memory and processing time, because electromagnetic
waves are more absorbed in a resistive layer when the thickness of the layer becomes larger.
In particle simulations, the masking method (damping region method) [Tajima and Lee,
1981] is conventionally used as absorbing boundary conditions for electromagnetic elds. In
this method, outgoing waves are articially damped by simple masking of electromagnetic elds
in damping regions attached at both ends of the simulation box. Recently we introduced
two new parameters to control absorption of outgoing electromagnetic waves, and developed
an improved masking method [Umeda et al., 2001]. Using the new method, We can reduce the
size of damping regions substantially. In this section we describe the new masking method.

2.10.2

Improved masking method

Eective damping region


In the masking method, electromagnetic elds in damping regions are absorbed by multiplying
a masking function fM to the right hand side of the time dierence form of Maxwells equations
at each time step. In a one-dimensional system the procedure of the masking method is written
as
t

B t+ 2 (x)= B t 2 (x) t E t (x)




E t+t (x)= E t (x) t J

t+ t
2

(x) c2 B

t+ t
2

fM (x)

(x) fM (x)

(2.112)

The masking function and procedure of the masking method are schematically illustrated in
Figure 2.16. We use a parabolic masking function with fM = 1 in physical region and fM < 1
in damping regions, varying gradually from 1 to a smaller value (Figure 2.16). The masking
function is given by

fM (x) =

x /2
1 r |x|L
LD

for |x|
for |x| >

Lx
2
Lx
2

(2.113)

54

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

Figure 2.16: Shape of the masking function and procedure of the masking method

where Lx (= Nx x) represents the length of the physical region, and LD (= ND x) represents


the length of damping regions. We introduce a new parameter r : (0 < r 1) to change
the gradient of the function [Tominaga, 1998; Umeda et al., 2001]. The masking function fM
becomes 0 at x = (L/2 + LD /r). We can dene the length of an eective damping region
by LD /r as shown in Figure 2.17. In the conventional masking method [Tajima and Lee, 1981],
the masking function (2.113) is used with r = 1.0. Thus the masking parameter r is given by
r=

Length of real damping region


.
Length of eective damping region

(2.114)

To check the validity of the masking method, we performed simple test simulations. In a
one-dimensional system taken along the x axis, we attached the damping regions both ends of
the physical region. The number of grid points for physical region is Nx = 1000. We assume
a current source at the center of the simulation system. The external current J0 is driven in
the z direction to radiate electromagnetic waves propagating in both forward and backward
directions. The current J0 is given by

J0 =

AJ sin (2fJ t) for 0 fJ t 1


0

for

fJ t > 1

(2.115)

where AJ and fJ are the amplitude and frequency, respectively. The grid spacing x fJ /c =
0.002 and the time step t fJ = 0.0001 are the common parameters, where c is the speed
of light. As the initial condition, all electrons in the system form a Maxwellian velocity distribution. We use 1000 superparticles per cell. Ions are assumed to be uniform neutralizing
background with an innite mass in the simulation system.

2.10. BOUNDARY TREATMENT FOR FIELDS

55

Figure 2.17: Shape of the masking function (2.113) for dierent r. The dashed and solid
lines correspond to r = 1.0 and r = 0.5, respectively. Length of the eective damping
region is given by LD /r.

Firstly, we use the conventional method and change the length of the damping regions LD =
ND x to study the reection ratio of light mode in vacuum and whistler mode in plasmas with
e = 5.0, e = 5.0. In Figure 2.18, we plotted the wave density |cBy |2 for the light mode In
Figure 2.19, we plotted the wave energy density |cBy |2 for the whistler mode. The distance
is normalized by J = c/fJ . at each grid point and time step. Electromagnetic waves are
radiated at the center of the system and propagate to the boundaries. Although we cannot nd
substantial eect of the length of damping regions in the case of whistler mode as found in the
case of light mode, the results showed that a longer damping region gives better absorption.
In both cases, outgoing waves penetrate only 30% of the damping regions and are reected
near the boundary between the physical region and the damping region. This implies that a
large part (about 70%) of the damping region is not used for absorption of the waves. When
we use a large damping region, gradient of the masking function becomes small. Therefore,
the gradient of masking function is rather important than the size of damping regions. Since
change of function value indicates change of a medium where the waves are propagating, a
better masking function should be realized with a smaller gradient.
Next, we varying the masking parameter r to change the gradient of the masking function
(2.113), i.e. length of the eective damping regions. As described above, when we use large
masking factor, electromagnetic waves are reected near the boundary between the physical
region and the damping regions, and we cannot have good absorption. By varying r from 1.0 to

56

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

Figure 2.18: Spatial and time proles of magnetic eld energy, |cBy |2 for light mode
waves. (left)ND = 50, (middle)ND = 100, (right)ND = 500.

Figure 2.19: Spatial and time proles of magnetic eld energy, |cBy |2 for whistler mode
waves. (left)ND = 50, (middle)ND = 100, (right)ND = 500.

2.10. BOUNDARY TREATMENT FOR FIELDS

57

Figure 2.20: Spatial and time proles of magnetic eld energy, |cBy |2 with ND = 50.
(left)Light mode waves are absorbed with a masking factor r = 0.11. (right)Whistler
mode waves are absorbed with a masking factor r = 0.08.

smaller value, electromagnetic waves can penetrate the damping regions deeply, and reection
ratio becomes smaller. When we use too small masking factor, however, reection ratio becomes
larger because electromagnetic waves reach the boundary of the system with a nite amplitude
and are reected back into the system. Figure 2.20 shows the results of test runs with ND = 50.
The best absorption of the light mode is realized with r = 0.11, while the best absorption of
the whistler mode is realized with r = 0.08.
Let us consider that a nite amplitude wave with the following form is propagating in a
damping region,
BW (t) = BW 0 exp(i0 t)

(2.116)

From the computation (2.112), time variation of waves amplitude is expressed as


t+ t

t t
2

BW 2 BW
BW
=
t
t

= (i ir )BW

(2.117)

where is the masking factor (equivalent to fM ). From (2.116) and (2.117), we obtain the
damping ratio i as follows:
BW
t

t+ t
2

t t

BW 2
=
t
exp(i 02t ) exp(i 02t )
=
BW 0 (i0 t)
t
BW

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

58

0 t
1+
0 t
1
cos
i
sin
}BW 0 exp(i0 t)
t
2
t
2
0 t
1
cos
=
t
2

= {
i

(2.118)
(2.119)

By assuming t is small enough, the damping ratio normalized by the driving frequency 0 is
given by

i
1
=
0
0 t

(2.120)

Since the amplitude variation propagates at a group velocity, the spatial damping ratio ki
(k = kr + iki ) is given by
ki =

i
1
=
vg
vg t

(2.121)

We dene an eective spatial damping ratio in damping regions as


 LD
0

 1

ki dx = LD

ND r 2
r 2 x2
dx =
t
vg t
3vg x

(2.122)

t
represents a group velocity of waves in the simulation.
where we replace by 1 r 2 x2 . vg x
t
an eective group velocity. We have performed many test runs with various
We call vg x

combination of size of damping region ND , the masking parameter r and the eective velocity
t
vg x
. We found that the best masking parameter r0 satises the following relation [Umeda et

al., 2001],
ND r02
t  9 12.
vg x

(2.123)

This means that the amplitude of outgoing waves becomes 30 40dB when they reach the
edge of a damping region. Therefore nonphysical reection of outgoing waves takes place only
at the boundary between the physical region and the damping region.
Phase Retardation
We also introduce another masking factor in Eqs. (2.112). We can separate the right hand
side of the time dierence form of Maxwells equations into two parts. The rst part represents
original eld values, i.e., electromagnetic elds at the previous time step (at time t t
). The
2
second (at time t) represents the increments of the electromagnetic elds for one time step.
We use two masking parameters rd and rr , and apply the masking function (2.113) with the
dierent masking parameters as shown in the following form.
t

B t+ 2 (x) = fM (x, rd ) (B t 2 (x) t E t (x) fM (x, rr ))


t
t
E t+t (x) = fM (x, rd ) (E t (x) t J t+ 2 (x) c2 B t+ 2 (x) fM (x, rr ))

(2.124)

If rr = 0, i.e., fM (x, rr ) = 1.0, the above procedure is reduced to the method in Eq.(2.112).

2.10. BOUNDARY TREATMENT FOR FIELDS

59

We analyze eect of each masking factor. We replace both masking functions fM (x, rd ) and
fM (x, rr ) by factors and , as follows:
t+ t
B0 2
t+ t
E0 2

t E

t t
2

t J t c2 B t

=B
=E

t t
2

t+ t
2

t+ t
2

t+ t
B 0 2
t+ t
E 0 2

(2.125)

(2.126)

Eqs. (2.126) show that the amplitude (energy) of outgoing waves are gradually reduced in
the damping regions at each time step. The masking factor can be regarded as a damping
factor.
On the other hand, we cannot absorb outgoing waves when we apply assuming = 1. In
this case, we can obtain the wave equation (2.127) for light modes, as follows:

2E
t2
2B
t2

= ( c)2 2 E
= ( c)2 2 B

(0 < < 1)

(2.127)

Eqs. (2.127) shows that electromagnetic waves propagate slower than the light speed with
group velocity vg = c in damping regions. We also apply the masking factor of to the
current density J to mask motion of plasma particles in the damping regions. All particles
are treated as if they are moving with velocities v. In damping regions, therefore, not only
waves but also plasmas are moving slower than they move in a physical region. The masking
factor of can be regarded as a retarding factor.
In the new scheme, we can apply various combination of masking factors for both damping
and retarding parts varying rd and rr . The length of the eective damping region is 1/rd times
longer than that of a real damping region. A smaller damping factor corresponds to a longer
eective damping region. However, when we apply the masking function with too small r,
reection of outgoing waves becomes larger because outgoing waves reach the boundary of the
real damping region. By using larger retarding factor, electromagnetic waves propagate more
slowly in the damping regions. Reection of outgoing waves becomes smaller when we use a
combination of smaller rd and larger rr .
In most cases, the best retarding parameter is rr = 1.0. Because of the retarding eect,
however, wave length of outgoing waves gradually becomes shorter ( = 2c/) and wave
number of the waves becomes larger. When the eective wave number becomes larger than
the maximum wave number of the system (x/c > kmax = ), the waves cannot exist in
damping regions and are reected. We can summarize that the best retarding factor (the largest
rr ) is given by
rr =

1 x 2

1.0

ND
ND 1

for x
for x <

1 ( NNDD1 )2

1 ( NNDD1 )2

(2.128)

60

CHAPTER 2. NUMERICAL TECHNIQUES FOR PARTICLE SIMULATION

Figure 2.21: Spatial and time proles of magnetic eld energy, |cBy |2 with ND = 50.
(left)Light mode waves and (right)whistler mode waves are absorbed with a damping
factor r = 0.05 and a retarding damping factor r = 1.0.

It is noted that the retarding parameter must be rr 1.0 because electromagnetic waves cannot
propagate faster than the light speed. We compute eective spatial damping ratio of the new
scheme, as follows:
 LD
0

ki dx =

 LD
0

ND

t
c x

LD
1
dx =
ct
ct


rd
rr

2

 1
0

rd2 x2
dx
(1 rr2 x2 )
!

1
1 + rr x
|x
log |
2rr
1 rr x

ND
ND 1

(2.129)

where we replace and by 1 rd2 x2 and 1 rr2 x2 , respectively. We investigate the best
combination of rd and rr , and compute spatial damping ratio

 LD
0

ki dx. We also performed

many test runs and found that the best rd can be obtained by setting that

 LD
0

ki dx = 3.0 4.0.

In Figure 2.21, we applied the improved method with a damping factor r = 0.05 and a
retarding damping factor r = 1.0. For both light mode waves and whistler mode waves, we
obtain much better absorption than that in the left panels of Figures 2.18 and 2.19. Using the
improved masking method we can reduce the size of damping regions substantially.

Chapter 3
Formation of Electrostatic Solitary Waves in
Open Systems
3.1

Introduction

Electrostatic Solitary Waves (ESW) were rst observed by the GEOTAIL spacecraft in the
Earths magnetotail [Matsumoto et al., 1994]. Recently, ESW are observed in various regions
of the Earths magnetosphere. ESW are bipolar electric pulses longitudinal to the geomagnetic
eld. ESW can be modeled as electron phase-space density holes which are Bernstein-GreeneKruskal (BGK) modes, i.e., one-dimensional equilibrium solutions to the time-independent
Vlasov-Poisson equations [Bernstein et al., 1957; Krasovsky et al., 1997; Muschietti et al.,
1999]. The previous one- and two-dimensional simulation studies have conrmed that ESW
in the magnetotail are one-dimensional potentials generated through nonlinear evolution of an
electron bump-on-tail instability [Omura et al., 1994, 1996; Miyake et al., 1998]. In a long time
nonlinear evolution of the instability, electron holes coalesce with adjacent holes and merge into
larger, more intense and isolated holes.
In the auroral region, bipolar parallel electric elds accompanied by strong perpendicular
electric elds are observed by the POLAR spacecraft [Mozer et al., 1997] and FAST spacecraft
[e.g., Ergun et al., 1998]. This implies that electron holes in the auroral region have multidimensional structures. The recent two- and three-dimensional simulations of electron beam
instabilities have demonstrated that obliquely propagating electrostatic waves such as electrostatic whistler modes (oblique Langmuir modes) [Goldman et al., 1999; Oppenheim et al., 1999,
2001] and lower hybrid waves [Miyake et al., 2001; Singh et al., 2000b,2001b; Umeda et al.,
2002] are radiated from electron holes. Generation of these oblique electrostatic waves are also
described in the recent theoretical works [Vetoulis and Oppenheim, 2001; Newman et al., 2001a;
Singh et al., 2001a]. It is expected that multi-dimensional electron holes are formed through
interaction with these oblique electrostatic modes. Structure and stability of multi-dimensional
electron holes are also issues of recent theoretical works [e.g., Chen and Parks, 2002; Jovanovic
et al., 2002].
Simulations of electron beam instabilities were conventionally performed in uniform periodic
61

62

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

systems. In simulation runs with periodic boundary conditions, an unstable velocity distribution function is assumed to exist uniformly in space as an initial condition. In the real space
plasma, however, wave-particle interaction of electron beam instabilities does not necessarily
take place uniformly. Several numerical simulation were demonstrated that electron holes were
generated from a double layer [e.g., Singh and Schunk, 1984; Singh et al., 1987], shocks [e.g.,
Shimada and Hoshino, 2000] and magnetic reconnection [Drake et al., 2003]. Sources of electron beams are localized, when the electron beams result from acceleration by localized electric
elds such as a double layer [e.g., Singh, 2000a; Newman et al., 2001b] or an induction electric
eld [Omura et al., 2003]. Recently, stability, formation and coalescence processes of electron
holes were studied in realistic systems with open boundaries [Muschietti et al., 1999; Mandrake
et al., 2000; Omura et al., 2001; Umeda et al., 2002]. In the present study, we adopted such
a realistic simulation model where an electron beam is injected from a localized source into
a homogeneous background plasma to study spatial and temporal development of an electron
beam instability from the localized source. A similar simulation was performed by Mandrake et
al. [2000] with injection of a narrow electron beam into a non-uniform background plasma. In
the present study, we assume a wide electron beam with perpendicular dimension much longer
than both electron and ion gyro-radii. We focus on the interaction of electron holes with lower
hybrid waves that occurs under weak magnetic elds with e e > b .
In Section 3.2, we describe the model and parameters. Results of the one- and two-dimensional
simulations are respectively presented in section 3.3 and 3.4. Section 3.5 gives discussion, and
section 3.6 gives conclusion of this chapter.

3.2

Simulation Model

We modied the simulation code to develop one- and two-dimensional electrostatic particle
codes. The simulation system is taken in the x y plane. The ambient magnetic eld is taken
in the x direction. We assumed two-dimensional half-open system with periodic boundaries in
the y direction and open boundaries in the x direction.
As shown in Figure 3.1, we assume that the background plasma exist uniformly in the simulation box without beam electrons as the initial condition. In the region outside of the right
boundary, the velocity distribution function f is equal to that in the simulation box f0 . On
the other hand, in the region outside of the left boundary, we assume the bump-on-tail velocity
distribution function f+ . We assume that there is no perturbation in the regions outside of both
left and right boundaries. Ions with mass ratio mi /me = 100 are assumed to exist uniformly.
When a computer simulation is started, the electron beam is continuously injected from the end
of the system into the background homogeneous plasma. The injected electron beam and the
background electrons form a bump-on-tail velocity distribution function. The injected particles
(beam electrons, background electrons and ions) have ux velocity distribution functions given

3.2. SIMULATION MODEL

63

Figure 3.1: Schematic illustration of the initial condition. In the simulation box and the
region outside of the right boundary, electrons form a Maxwellian velocity distribution
function f0 = f . In the region outside of the left boundary, electrons form a bump-on-tail
velocity distribution function f+ . The local electron densities given by

f (v)dv are

assumed to be constant for all x. At the open boundaries, particles are injected with the
constant ux vf (v), while the exiting particles are removed.

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

64

by vx f (vx ) [Birdsall and Langdon, 1985]. The ux functions at the left and right boundaries are
given by vx f+ and vx f , respectively. In the present study, we inject particles with the constant
ux, while exiting particles are removed at the open boundaries as if there is no boundary.
In electrostatic simulations, we do not need a special numerical treatment of the open boundaries such as absorbing boundary conditions for electromagnetic waves used in electromagnetic
simulations [e.g., Umeda et al., 2001]. The electrostatic elds at the open boundaries are determined by the distribution of charged particles. We rst compute the charge density distribution
at the open boundaries. Then we solve Poissons equation to obtain electrostatic potential and
electrostatic eld. We used the marching method [Buneman, 1973] for solving Poissons
equation.
We assume that all electron and ion components have Maxwellian velocity distributions. The
initial electron velocity distribution in the simulation box (x Nd x) is given by


ne + nb
v2
exp x2 .
f (vx ) =
2Vte
2Vte

(3.1)

On the other hand, the electron velocity distribution in the region outside of the left boundary
(x < Nd x) is given by


ne
v2
nb
(vx Vd )2
exp x2 +
exp
.
f (vx ) =
2Vte
2Vtb2
2Vte
2Vtb

(3.2)

Here Vd , Vts and ns represent drift velocity, thermal velocity and density, respectively. Subscripts s = b, e, and i represent beam electrons, background electrons and ions, respectively.
We set up the initial electron density in the simulation box and in the regions outside of both
left and right boundaries constant. Since the electron beam is injected with a constant ux,
however, the charge neutrality is not necessarily satised in the present open system.
Since the charge neutrality is not enforced in the present open system, there exists a DC
component of parallel electric eld in the system. To suppress non-physical acceleration of the
background plasma, we subtract the DC electric eld from the parallel electric eld so that the
following condition is satised,


Ex,i,j = 0 (i = 1 Nx , j = 1 Ny ).

(3.3)

i,j

We performed a test run without cancellation of the DC electric eld. In this test run we
observed acceleration of the background electrons by the DC electric eld, which leads to nonphysical Langmuir oscillation of the background electrons. This is a limitation of the present
one-dimensional model. However, we also observed formation of electron holes in this test run.
The physical process of the electron beam instability is less aected by the DC electric eld,
because both beam electrons and background electrons are accelerated keeping the same relative
velocity between them. By subtracting the DC component, the simulations are performed in

3.3. ONE-DIMENSIONAL SIMULATION

Electron cyclotron frequency

65

e /e = 1.0

Drift velocity of electron beam Vd /Vt,e = 2.0


Thermal velocity of electron beam Vt,b /Vt,e = 0.1
Thermal energy ratio

Te /Ti = 64.0

Table 3.1: Common simulation parameters for all runs.

the rest frame of reference where the background plasma is free from the non-physical Langmuir
oscillation.
We used 1600 superparticles per cell respectively for electrons and ions. The density ratio
of the electron beam R is dened as R = nb /(nb + ne ). In the present study, we assume a
weak electron beam with R = 0.06. The common parameters for all simulation runs are listed
in Table 3.1. Assuming the electron plasma frequency e = 1.0 and the initial background
electron thermal velocity Vt,e = 1.0, we can normalize frequencies and velocities by e and
Vt,e , respectively. Length is normalized by the electron Debye length e = Vt,e /e . The
bump-on-tail velocity distribution function of the present study is based on the previous oneand two-dimensional studies [Omura et al., 1996; Miyake et al., 1998]. Electron bump-ontail velocity distribution functions are consistent with the recent statistical analyses of the
GEOTAIL data [Omura et al., 1999b; Kojima et al., 1999b]. In the particle measurement of
the GEOTAIL spacecraft, electron beams are observed as enhanced non-thermal uxes with
at diused velocity distributions that result in after saturation of electron beam instabilities.

3.3

One-dimensional Simulation

We present simulation results of a very long one-dimensional open system with Lx = 40960x
(x = 0.1e ), where Lx and x represent system length and grid spacing, respectively. We
show x vx phase diagrams at dierent times in Figure 3.2. A cold and weak electron beam is
injected from the left boundary into the warm background electrons to form the bump-on-tail
velocity distribution at x = 0. As the electron beam propagates from the left to the right, the
bump-on-tail instability evolves in space. Since the positive gradient of the velocity distribution
function is localized at the small bump, a coherent electron beam mode, whose phase velocity
is slightly smaller than the drift velocity of the bump electron beam, is excited. The beam
mode traps the beam electrons and diuses the bump to a at distribution, forming electron
holes in the x vx phase space. As the electron holes propagate, they coalesce with adjacent
holes to form larger and more isolated electron holes. The spatial scales of electron holes and
the distance between them become larger depending on the distance from the source of the
electron beam. As shown in Figure 3.3, at a distant place from the source of the electron beam
(x/e = 4096), we found larger and more isolated electron holes. We conrmed the formation

66

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

Figure 3.2: x vx phase diagrams at e t = 100, 200, 400 and 1000.

3.3. ONE-DIMENSIONAL SIMULATION

67

Figure 3.3: (a)A x vx phase diagram at e t = 4000 and (b)the corresponding spatial
prole of electric elds Ex . The amplitude of the electric eld is normalized by me Vt,e e /e.

and coalescence processes of electron holes in the one-dimensional open system.


In the present simulation model, the leading edge of the electron beam is associated with
the sudden turn-on of the electron beam. However, when electron holes penetrate into the
background plasma, they can propagate through the background plasma without signicant
disturbances of the background plasma. This implies that the electron holes are formed essentially by dynamics of trapped electrons rather than that of untrapped electrons with velocities
close to the beam drift velocity that pass through the electron holes. Thus the simulation result
is less aected by the sudden injection of the electron beam. Electron holes can propagate stably over a long distance through the background plasma, which is consistent with the statistical
analysis of the GEOTAIL data [e.g., Omura et al., 1999a; Kojima et al., 1999b]. It is possible
to observe electron holes at a more distant place from the source region where the instability
is taking place.
We compared the result of the open system with that of the periodic system [Omura et al.,
1996], and conrmed that time scale of evolution of the bump-on-tail instability in both systems
are the same. With conversion of time to position by x = Vd t, the temporal evolution of the
instability in the periodic system can be transformed to the spatial evolution of the instability
in the open system. This implies that the periodic system simulates a localized region of the
open system in a frame of reference moving with Vd .
We also performed the simulations with immobile ions (mi /me = ). However, we found
little dierence between the two cases. Even with cold ions, the electron holes do not lead to

68

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

Figure 3.4: A k spectrum of |Ex | for x/e = 204.8 409.6, e t = 307.2 819.2. The
amplitude is normalized by me Vt,e e /e.

nonlinear decay to ion acoustic waves such as observed in an electron two-stream instability
[Omura et al., 1994, 1996]. In Figure 3.4, we show an k spectrum obtained by taking
Fourier transformation of Ex data for x/e = 204.8 409.6, e t = 307.2 819.2. We found
clear enhancement of the beam mode, while the amplitudes of both Langmuir waves and ion
acoustic waves are very small. Since the electron holes propagate very fast, the electron holes
are separated from the background ions in the x vx phase space. Ions cannot be reected by
the positive potentials of electron holes.
In the magnetotail, the GEOTAIL spacecraft observed both tailward and earthward propagating ESW simultaneously [Omura et al., 1999a; Kojima et al., 1999a]. To demonstrate
the counter-streaming ESW, we performed another simulation run with counter-streaming two
electron beams in a short system (Lx = 8192x (x = 0.1e )). As shown in Figure 3.5, we
injected both forward and backward propagating electron beams from the left and right boundaries, respectively. As time elapses, we found formation of both forward and backward moving
electron holes. When the forward and backward traveling electron holes encounter, however,
they go through each other without coalescence. Since the relative velocity is much larger than
the trapping velocity of the electron holes, the electrons which form a potential of an electron
hole are not trapped by the other potential. Namely, there is little interaction between the
forward and backward propagating potentials.

3.4. TWO-DIMENSIONAL SIMULATION

69

Figure 3.5: Simulation results of counter-streaming electron beams. The x vx phase


diagrams at e t = 125, 250 and 500.

3.4
3.4.1

Two-dimensional Simulation
Linear analysis

In this section, we extend the previous one-dimensional simulations to a two-dimensional simulation with Lx Ly = 1024x 64y(x = y = e ). Before performing a two-dimensional
simulation, we numerically solved linear dispersion relations of the bump-on-tail instability. In
Figure 3.6a, we show the maximum growth rate of electrostatic waves as a function of wavenumbers kx and ky . We found that three modes are linearly unstable in this instability. (A)Electron
beam modes have the maximum growth rate /e = 0.1 at = 0 , where represents wave
normal angle. We also show the linear dispersion relation of the electron beam modes in Figure
3.6b. (B)A quasi-perpendicular mode is unstable at = 87 . The maximum growth rate of this
mode is /e = 0.01. Frequency of this mode is  LHR = 0.07e , LHR represents the lower
hybrid resonance frequency. (C)Oblique electrostatic modes are unstable for = 50 80 .
The maximum growth rate of the oblique modes is /e = 0.05 at = 75 . These modes are
unstable only in the presence of cold ions. Frequencies of these modes are LHR = 0.07e .

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

70

Figure 3.6: (a)Maximum growth rate /e as a function of kx and ky , and linear dispersion
relation of electron beam modes: (b)frequency and (c)growth rate with wave normal
angles of 0 , 15 , 30 , 45 and 60 , respectively.

3.4.2

Spatial structures of electron holes

To analyze spatial structures of electron holes, we show spatial proles of potentials at


2
dierent times in Figure 3.7. The potentials are normalized by me Vt,e
/e. The zero-level of

potentials corresponds to the spatial average. As the electron beam propagates along the external magnetic eld, the bump-on-tail instability develops in space as observed in the previous
one-dimensional simulation. The injected electron beam is uniform in the y direction. We
rst observed one-dimensional potentials in the leading edge of the electron beam (Figure 3.7a,
x/e = 85). Then the potentials excited by the instability are modulated in the y direction
(x/e = 45). According to the linear dispersion relation of the electron beam modes (Figure
3.6b), the parallel beam mode has the maximum growth rate, and oblique beam modes with
larger wave normal angle have smaller growth rates. Thus we observed the parallel beam
mode with ky = 0 at rst, and then we observed oblique beam mode with ky = ky .
As time elapses, amplitude of the oblique beam modes in the generation region of electron
holes (x/e = 64 128) becomes larger and larger. The electron holes excited in the later
phases are more modulated by the oblique beam modes. Spatial structures of these holes are
twisted as observed in Figure 3.7b and 3.7c. We also observed two-dimensional potentials isolated in both x and y directions infrequently, as at x/e = 130 in Figure 3.7d. The scales of x
and y axes of Figure 3.7d are the same. The ratio of the parallel to perpendicular characteristic
widths of the two-dimensional potential in Figure 7d is 0.5. In the present simulation, however,
existence of such two-dimensional potentials is infrequent, and the ratio of parallel to perpendicular characteristic widths of them is less than 0.5. Namely, the perpendicular dimension of
electron holes is much longer than the parallel dimension. In the leading edge of the electron

3.4. TWO-DIMENSIONAL SIMULATION

71

Figure 3.7: Spatial proles of potentials at dierent times. The potentials are normal2 /e.
ized by me Vt,e

beam, we found clear one-dimensional potential structures at all times. The one-dimensional
potentials can propagate without changing their characteristics. On the other hand, the twisted
potentials and two-dimensional potentials are aligned in the direction perpendicular to the external magnetic eld through the coalescence process [Miyake et al., 1998].
To show the dierence between potential structures in the leading edge of the electron beam
and those in the generation region of electron holes quantitatively, we analyzed Fourier spectra
of electric elds in these regions. We specied a region with width x/e = 64 from the leading
edge of the electron beam and computed kx ky spectra of |Ex | and |Ey | shown in Figure 3.8a.
The spectra are the time average at e t = 102.4, 204.8, 307.2, 409.6. The intensity is normalized
by me Vt,e e /e. We found enhancement of Ex components at ky  0, while the intensities of
Ey components are the same as that of the background thermal uctuations. We also show
kx ky spectra of |Ex | and |Ey | in the generation region of electron holes (x/e = 64 128)

72

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

Figure 3.8: (a)kx ky spectra of |Ex | and |Ey | in the leading edge of the electron beam.
We specied a region with width x/e = 64 from the leading edge of the electron beam.
(b)kx ky spectra of |Ex | and |Ey | for x/e = 64 128. The spectra are the time average
at e t = 102.4, 204.8, 307.2, 409.6. The amplitudes are normalized by me Vt,e e /e.

3.4. TWO-DIMENSIONAL SIMULATION

73

Figure 3.9: (a,d)Spatial proles of at e t = 1024.0 and 1228.8, respectively. (b,c)Spatial


proles of Ex and Ey for xe = 600 1024 at e t = 1024.0, respectively. The amplitude
2 /e and m V /e, respectively.
of potentials and electric elds are normalized by me Vt,e
e e t,e

in Figure 3.8b. The spectra are the time average at e t = 102.4, 204.8, 307.2, 409.6. We found
clear enhancement of |Ey | in Figure 3.8b. These spectra show that both parallel and oblique
electron beam modes exist in the generation region. The present result shows that there is a
spatial gap between the leading edge of the bipolar pulses (Ex ) and that of the oblique modes
(Ey ).
In Figure 3.9a and 3.9d we plotted spatial proles of potentials in later phases. In a region
far from the source of the electron beam (x/e = 700 1000) we found one-dimensional
electron holes. The one-dimensional potential structures are also indicated from the clear
bipolar waveforms of Ex in Figure 3.9b. However, these one-dimensional electron holes are
accompanied by Ey components as shown in Figure 3.9c. In a long time evolution of the bumpon-tail instability up to LHR t > 50, the intensities of Ey components become larger and larger

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

74

Figure 3.10: (a)Spatial proles of |Ey |2 along the magnetic eld: (A)Energy density of
|Ey |2 for ky e = 0 , and (B)energy density of |Ey |2 for ky e = 1.0 1.6. We
took time average of |Ey |2 for e t = 819.2 1024.0. (b)kx ky spectrum of |Ey | for
x/e = 128 256 at e t = 1024. The amplitudes of the electric elds are normalized by
me Vt,e e /e.

in a region close to the source of the electron beam.

3.4.3

Interaction between electron holes with lower hybrid waves

We show a spatial prole of |Ey |2 along the magnetic eld in Figure 3.10a. The energy density
of |Ey |2 for ky e = 0 averaged for e t = 819.2 1024.0 is shown by the black line
(A). We also plotted the energy density of |Ey |2 for ky e = 1.0 1.6 by the gray line (B).
We found strong enhancement of |Ey |2 with ky e = 1.0 1.6 for x/e = 150 200. We
specied a region x/e = 128 256 where Ey is enhanced, and computed a kx ky spectrum
of |Ey | at e t = 1024. In Figure 3.10b we found clear enhancement of a quasi-perpendicular
mode propagating at = 87 relative to the external magnetic eld. We found that the
region x/e = 150 200 is the source of the quasi-perpendicular mode. As in Figure 3.6,
this mode is linearly unstable. However, generation mechanism of this mode is not the linear
growth, because oblique electrostatic modes with  75 , whose growth rate is much larger
than that of the quasi-perpendicular mode, are not excited. In Figure 3.11 we show kx
and ky spectra of |Ey | for e t = 819.2 1228.8, x/e = 128 256. The plotted kx
and ky spectra are obtained by integration over ky and kx , respectively. We obtain the
frequency of the quasi-perpendicular modes as /e  0.08, which is close to the lower hybrid
resonance frequency (LHR /e  0.07). The kx spectrum shows that the parallel phase

3.4. TWO-DIMENSIONAL SIMULATION

75

Figure 3.11: (a) kx and (b) ky spectra of |Ey | for x/e = 128 256. These kx
and ky spectra are given by integration over all values of ky and kx , respectively. The
amplitude is normalized by me Vt,e e /e.

velocity of the quasi-perpendicular mode corresponds to the drift velocity of the electron holes
VEH = 0.8Vd = 1.6Vt,e . We conclude that a lower hybrid mode is excited through coupling of
the parallel phase velocity [Miyake et al., 2000]. The dispersion relation of the excited lower
hybrid mode at = 90 is given by
ky2 Vs2
+ 2i
=
2
2
1 + k y e
2

(3.4)

where Vs represents the ion sound speed [Lominadze, 1981]. In the present simulation LHR 
i . From Figures 3.10b and 3.11b the perpendicular wavenumber of the lower hybrid mode is

ky e  1.0. The perpendicular phase velocity of this mode is approximately equal to Vs / 2.

Therefore the propagation angle is given by  tan1 ( 2VEH /Vs ).


There are some recent theoretical works regarding emission process of electrostatic whistler
waves (oblique Langmuir waves) from the electron holes by bounce motions of the trapped
electrons [Vetoulis and Oppenheim, 2001] or vibration of the election holes [Newman et al.,
2001]. However, it is very dicult to compare the present simulation result with these works
because the parameters are very dierent. In the present study, since the frequency of the lower
hybrid mode is much smaller than the bounce frequency b , the emission process of the lower
hybrid mode is not consistent with that described in the former reference. However it is unclear
whether the present simulation result is consistent with the latter one. In the present study
the parallel phase velocity of the lower hybrid mode satises Cherenkovs condition, namely
= VEH kx . A theoretical work of this emission process is conducted by Singh et al. [2001].

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

76

3.5
3.5.1

Discussion
Spatial prole of potentials along the magnetic eld

In the very early phase of the simulation run (e t = 0 100), potential structures of electron
holes formed by the bump-on-tail instability are uniform in the direction perpendicular to the
external magnetic eld. As time elapses, amplitude of the oblique beam mode becomes larger
and larger in the generation region of electron holes. The excited oblique beam modes modulate
the background electron densities in the y direction. Potential structures initially uniform in
the y direction become twisted through modulation by the oblique beam mode excited later
than the parallel beam mode as schematically illustrated in Figure 3.12a. We also observed twodimensional electron holes isolated in both x and y directions infrequently (Figure 3.7d). As the
twisted electron holes and two-dimensional electron holes propagate farther from the generation
region, the electron holes are more and more isolated in the x direction through coalescence.
The perpendicular dimensions of two-dimensional electron holes also become longer and longer
through coalescence of the electron holes. Since the electrons trapped therein are mixed in the
y direction at the coalescence because of their cyclotron motions, the twisted potentials and
two-dimensional potentials are aligned in the y direction as schematically illustrated in Figure
3.12b. The perpendicular electric elds of the electron holes decay and potential structures
become one-dimensional. The decay process of the perpendicular electric elds through the
coalescence is described by Miyake et al. [1998, 2000].
One-dimensional potentials in the leading edge of the electron beam can propagate without
changing their characteristics. The linear dispersion relation of the electron beam modes in
Figure 6b shows that both parallel and oblique beam modes have the same parallel phase
velocity Vp / cos = VEH . However, group velocities of the oblique beam modes in the parallel
direction given by Vg cos = VEH cos2 are much slower than that of the electron holes (i.e.,
the parallel mode). Therefore, in the leading edge of the electron beam, potential structures
are free from the oblique beam modes propagating slower than the parallel beam mode in the
x direction.
In a long time evolution up to LHR t > 50, a strong lower hybrid mode is excited through
coupling with the electron holes [Miyake et al., 2000]. This wave-wave coupling process is
localized in a region close to the source of the electron beam, because the parallel group velocity
of the lower hybrid mode is much slower than the drift velocity of the electron holes. As shown
in Figure 3.10a, intensities of Ey components with ky e = 1 1.6 are especially large for
x/e = 150 200. The electron holes excited in the later phases are more modulated by the
lower hybrid mode in the y direction. As a result, the electron holes are accompanied by the
perpendicular electric elds to form modulated one-dimensional potentials as schematically
illustrated in Figure 3.7c. The perpendicular electric elds of electron holes are carried by

3.5. DISCUSSION

Figure 3.12: Schematic illustration on formation and evolution of potential structures


along the magnetic eld. (a)In the early phase of the simulation run (e t = 0 100),
potential structures are modulated by oblique electron beam modes. (b)In the later
phase (e t = 100 400), two-dimensional potentials and one-dimensional potentials
are respectively observed in the region close to the source of the electron beam and in
distant regions from the source. (c)For LHR t  1, modulated one-dimensional potentials
are formed by ion dynamics. (d)Schematic illustration of a modulated one-dimensional
potential.

77

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

78

the electron holes at the drift velocity of the electron holes. As the strong Ey component is
observed at x/e 1000 in Figure 3.12a, we can observe the perpendiculars electric eld of
electron holes even in distant regions from the source.

3.5.2

Eect of mass ratio

We performed additional three simulation runs with dierent mass ratios. The kx ky spectra
of |Ey | for x/e = 128 374 are plotted in Figure 3.13. Times of Figures 3.13b, 3.13c and
3.13d correspond to LHR t = 50.7. In the run with immobile ions (mi /me = ), simulation
results in early phases of the simulation run are almost the same as those with mobile ions
(mi /me = 100). In the later phase of this run, however, the lower hybrid mode is not enhanced
as in the kx ky spectrum of Figure 3.13a. From this result, we found that formation processes
of twisted electron holes and two-dimensional electron holes by the oblique beam modes are
due to electron dynamics.
In the run with lighter ions (mi /me = 25), we found excitation of the lower hybrid mode in
the earlier phase. The lower hybrid mode propagates at = 83 relative to B0 as shown in
Figure 3.13c. In the run with heavier ions (mi /me = 400), on the other hand, the lower hybrid
mode is not excited in the later phase (e t 800), and thus spatial structures of electron holes
are similar to those in the run with immobile ions. However, we performed a long run up to
LHR t > 50, and conrmed that a strong lower hybrid mode can be excited in the region close
to the source of the electron beam as shown in Figure 3.13d.
We listed the angles between the wave vectors and the external magnetic eld, frequencies , parallel and perpendicular wavenumbers kx , ky of the excited lower hybrid modes for
dierent mass ratios in Table 3.2. Eq.(3.4) shows that the wave normal angle and the parallel

wavenumber are approximately given by  tan1 ( 2VEH /Vs ) and kx  LHR /VEH , respec
tively. We obtain the perpendicular wavenumber as ky  2LHR /Vs . These equations are in
agreement with the parameters listed in Table 3.2. Thus the mass ratio is not essential in the
emission process of the lower hybrid mode, but it changes the time scale of growth of the lower
hybrid mode and its propagation angle. Even with much heavier ions, we expect that the lower

mi /me
25

/e kx e ky e

83 0.16

100

87

400

89 0.04

0.08

0.1

0.8

0.05

1.2

0.026 1.3

Table 3.2: The angles between the wave vector and the external magnetic eld, frequencies, parallel and perpendicular wavenumbers of the excited lower hybrid modes for
dierent mass ratios.

3.5. DISCUSSION

79

hybrid modes is excited. Since the parallel wavenumber of the lower hybrid mode becomes
much smaller with the real mass ratio, a much longer system in the x direction is needed for
observation of the lower hybrid mode in a more realistic system.

3.5.3

Perpendicular scale of electron holes

From the POLAR observation, Franz et al. [2000] estimated the perpendicular scale of electron
holes by assuming lx /ly = Ey /Ex , where lx and ly represent the parallel and perpendicular
characteristic widths of electron holes, respectively. In their estimation the ratio of the parallel
to perpendicular characteristic widths of two-dimensional electron holes in a plasma with a
parameter with e = e is approximately 0.7. In the present simulation runs, on the other
hand, spatial potential structures of electron holes are almost one-dimensional. Although twodimensional electron holes isolated in both x and y directions are formed through modulation
by the oblique beam modes, existence of such electron holes is infrequent. We note that the
perpendicular dimension of electron holes is aected by the system length in the y direction
because periodic boundary condition is applied in the y direction. To clarify the ratio of the
parallel to perpendicular characteristic widths of two-dimensional electron holes, we must use
a much longer system in the y direction. However, from the present simulation, the ratio of the
parallel to perpendicular characteristic widths of two-dimensional electron holes is expected to
be much less than 0.5.
Although potential structures are one-dimensional, the electron holes even in regions far from
the source are accompanied by perpendicular electric elds as in Figure 3.9. From Figures 3.9b
and 3.9c we obtain the amplitude ratio approximately Ey /Ex  0.7, which is consistent with
the POLAR observation [Franz et al., 2000]. In the region close to the source of the electron
beam, electron holes are strongly modulated in the y direction by the lower hybrid mode. The
perpendicular wavenumber of the lower hybrid mode is estimated as

ky 
k y e 

2i
LHR


Vs / 2
Vt,e me /mi

"
#
#
$

22e
.
2e + 2e

"
#
#
$

2e
2e + 2e
(3.5)

The parallel wavenumber of the beam mode is given by kx = e /Vd = 0.5/e . Thus the
perpendicular wavelength of the lower hybrid mode is shorter than the parallel characteristic
width of electron holes. The variation of the potentials in the perpendicular direction is smaller
than that in the parallel direction as found in Figure 3.9. One-dimensional potentials are
modulated in the perpendicular direction with a shorter wavelength as schematically illustrated
in Figure 3.12d.

80

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

Figure 3.13: kx ky spectra of |Ey | for x/e = 128 348 with dierent mass ratios:
(a)mi /me = at e t = 1228.8, (b)mi /me = 100 at e t = 716.8, (c)mi /me = 25 at
e t = 358.4, and (d)mi /me = 400 at e t = 1433.6. The amplitudes are normalized by
me Vt,e e /e. Times of (b),(c) and (d) correspond to LHR t = 50.7.

3.6. CONCLUSION

3.6

81

Conclusion

In the two-dimensional system with the open boundary, nonlinear evolution of the bump-ontail instability is dierent from that in the periodic system. Potential structures of electron
holes are similar everywhere in the periodic system. The coupling process takes place uniformly
in space, and the lower hybrid mode is excited uniformly. In the open system, on the other
hand, spatial structures of electron holes vary depending on the distance from the source of the
electron beam. The lower hybrid mode is excited locally in a region close to the source of the
electron beam. The spatial development of the instability cannot be transformed by taking a
frame of reference moving with Vd as observed in the previous one-dimensional system. In the
periodic system, the bump-on-tail instability saturates in an early phase, and then the lower
hybrid mode grows through the wave-wave coupling. However, since the electron holes, which
is energy source of the lower hybrid mode, begin to decay through their coalescence process
after the saturation, the growth of the lower hybrid mode also saturates in the early phase.
The amplitude of the lower hybrid mode is much smaller than that in the open system. In the
open system, on the other hand, the lower hybrid mode grows continuously in the region close
to the source of the electron beam, because the electron holes are continuously generated by
the constant ux of the electron beam.
In the early phase of the two-dimensional simulation runs, spatial structures of electron holes
are determined by the oblique electron beam modes. In the generation region of the electron
holes, electron holes initially uniform in the y direction are twisted through modulation by
the oblique beam modes. In regions far from the source of the electron beam, electron holes
become one-dimensional through the coalescence/decay process. In the later phase of the twodimensional simulation run, the wave-wave coupling process takes place uniformly in space in
the periodic system, and the lower hybrid mode is excited uniformly. In the open system, on
the other hand, the coupling process takes place in the localized region, because the parallel
group velocity of the lower hybrid mode is much slower than the drift velocity of the electron
holes. Electron holes are accompanied by the perpendicular electric elds through modulation
by the lower hybrid mode, resulting in formation of modulated one-dimensional potentials.
The modulated electron holes have unipolar waveforms of the perpendicular electric eld, as
observed by the FAST spacecraft in the auroral region [Ergun et al., 1998]. We also expect that
the perpendicular electric elds of electron holes can be observed even at a distant place from
the source, because they are carried by the electron holes at the drift velocity of the electron
holes.
In the open system, since both oblique electron beam mode and lower hybrid mode propagate
much slower than electron holes in the parallel direction, electron holes in the leading edge of
the electron beam are not aected by these oblique modes. On the other hand, electron holes
excited at the later times are more modulated by these oblique modes. In the present simulation

82

CHAPTER 3. FORMATION OF ESW IN OPEN SYSTEMS

study, we found that both perpendicular electric elds of electron holes and strong lower hybrid
modes are observed in the generation region of the electron holes. We can make use of these
characteristics to identify source regions of electron holes from waveform data of plasma wave
observations.
In the present study, we demonstrated excitation of oblique electrostatic modes via the wavewave coupling process for parameters of e e > b . There are also other instabilities of
electron holes under strong magnetic elds with e  e [Goldman et al., 1999; Oppenheim
et al., 1999] and weak magnetic elds with b > e [Muschietti et al., 2000], in which other
oblique modes with dierent parallel group velocities are excited. In the open system, parallel
group velocities of oblique modes are especially important because the dierence of parallel
group velocities results in a dierent wave-wave interaction region. Simulation studies of these
instabilities with open boundaries are left as future works.

Chapter 4
Electromagnetic Signature Associated with
Electrostatic Solitary Waves
4.1

Introduction

In Chapter 3, we have studied formation processes of one- and two-dimensional electron holes.
To study interaction between electron holes with electromagnetic elds, we extend the previous
two-dimensional electrostatic simulation to an electromagnetic particle simulation. We also
adopted the realistic simulation model used in Chapter 3 where an electron beam is injected
from a localized source.
Previous simulations and theoretical works of electron beam instabilities were performed with
electrostatic approximation, because electron holes are purely electrostatic potentials. Meanwhile, in the auroral region, bipolar parallel electric elds accompanied by strong perpendicular
electric elds which corresponding to multi-dimensional electron holes are observed by the POLAR spacecraft [Mozer et al., 1997] and FAST spacecraft [e.g., Ergun et al., 1998]. These
spacecraft also observed impulsive magnetic elds with electrostatic solitary waves. It is noted
that such electromagnetic signatures are not reported in the GEOTAIL observation. This chapter is the rst attempt to study electromagnetic eld signatures associated with electrostatic
solitary waves by performing electromagnetic simulations of electron beam instabilities. In
a potential structure of electrostatic solitary waves, electron density is spatially modulated.
Since the modulation of electron density propagates along the magnetic eld at a drift velocity
of an electron beam, there occurs temporal variation of a current. Therefore, it is possible
that electrostatic solitary waves become a source of electromagnetic emissions. We also discuss a possible mechanism of electromagnetic emissions from coherent electrostatic potential
structures.
Parameters of the present simulation are based on the GEOTAIL observation of electrostatic
solitary waves, and are in a weakly magnetized electron regime, e e > b , where b
represents the bounce frequency of the trapped electrons. Thus the parameters are dierent
from those in the auroral region. In discussion, electromagnetic signatures of electron holes
obtained by the present simulations are extended to a strongly magnetized electron regime.
83

84

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

In section 4.2, we present the two-dimensional simulation model. In section 4.3, we present
simulation results. Section 4.4 gives discussion, and section 4.5 gives conclusion of this chapter.

4.2

Simulation model

The simulation system is taken in the x y plane with Nx Ny = 1024 128 grid points. The
ambient magnetic eld is taken in the x direction. As schematically illustrated in Figure 4.1, we
assumed a two-dimensional half-open system with periodic boundaries in the y direction and
open boundaries in the x direction. To suppress non-physical reection of electromagnetic waves
at the open boundaries, we applied the improved masking method for an absorbing boundary
condition [Umeda et al., 2001]. We attached damping regions with Nd = 64 grid points at
both ends of the physical region. It is noted that a longer absorbing region gives better
absorption in such an absorbing boundary condition with a resistive layer. Due to limitation
of computation resource, Nd = Nx /4 Nx /20 grid points are commonly used.
The initial condition of the present two-dimensional simulation is the same as that of the
previous two-dimensional simulation in Chapter 3. We assume that background plasmas exist
uniformly in the simulation box without beam electrons. When a computer simulation is
started, beam electrons are continuously injected from the end of the left damping region
into the background homogeneous plasma. The injected beam electrons and the background
electrons form an unstable velocity distribution function shown in Figure 4.2. On the other
hand, at the end of the right damping region, we assume that the velocity distribution function
is equal to that in the physical region.
Number of superparticles per cell for beam electrons, background electrons and ions are
256, 1600 and 1600, respectively. The grid spacings x and y are equal to 0.5e . Due
the computation eciency we assume a reduced light speed. The common parameters for all
simulation runs are listed in Table 4.1. The present bump-on-tail type velocity distribution

Figure 4.1: Schematic illustration of the simulation model.

4.2. SIMULATION MODEL

85

Figure 4.2: Velocity distribution of electrons assumed at the left boundary. The solid
line shows the bump-on-tail velocity distribution with Vte /Vd = 0.5, Vtb /Vd = 0.125 and
R = 0.14.

is based on the previous one- and two-dimensional simulations [Omura et al., 1996; Miyake et
al., 1998; Umeda et al., 2002]. Electron bump-on-tail type velocity distribution functions are
consistent with the recent statistical analysis of the GEOTAIL data [Omura et al., 1999]. It is
noted that such a bump-on-tail type velocity distribution is also observed in a Vlasov simulation
of a double layer [Singh, 2000a]. As a parameter for dierent simulation runs, we varied the
electron magnetization ratio, i.e., the ratio of the electron plasma frequency to the electron
cyclotron frequency, as e /e = 1.0, 2.0, which satisfy the stable condition of electron holes
e > b [Miyake et al., 1998; Muschietti et al., 2000]. We again note that the present simulation
parameters are not in the strongly magnetized electron regime. In the present simulation, since
cyclotron motion of particles becomes very important, both electron and ion gyro-radii are
supported by more than 1 grid spacing.
The density ratio of the beam electrons R is dened as R = nb /(nb + ne ). To compare
the simulation with the previous electrostatic simulation, we rst used the same simulation
parameters as those in Chapter 3 (R = 0.06). However, since the amplitudes of excited electromagnetic elds are so small that the electromagnetic elds are disturbed by the enhanced
thermal uctuation of particle codes. In the present study, therefore, we assumed a stronger
electron beam with R = 0.14. To set the growth rate of an unstable beam mode comparable
to that of the previous electrostatic simulation, the thermal velocity of beam electrons is also
increased.
Assuming the electron plasma frequency e = 1.0 and the initial background electron thermal
velocity Vte = 1.0, we can normalize frequencies and velocities by e and Vte , respectively.

86

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

Electron cyclotron frequency

e /e 1.0

Drift velocity of beam electrons

Vd /Vte 2.0

Thermal velocity of beam electrons Vtb /Vte 0.25


Mass ratio

mi /me 100

Thermal energy ratio

Te /Ti 4.0

Light speed

c/Vte 10.0

Table 4.1: Common parameters for simulation run.

Distances are normalized by the electron Debye length e = Vte /e . In addition, electric elds
are normalized by e Vte me /e, and magnetic elds are normalized by e me /e. Potentials are
normalized by Vte2 me /e. Current densities are normalized by 0 2e Vte me /e.

4.3
4.3.1

Simulation Results
Electric and magnetic elds

To show spatial and temporal evolutions of the electron beam instability in the open system, we
rst plotted results of the run with e /e = 1.0. Figure 4.3a shows the x vx phase diagram at
e t = 256, which corresponds to the electron phase-space distribution

f (x, y, vx )dy. Figures

4.3b-d show the corresponding spatial proles of (b)Ex , (c)Ey and (d). It is noted that
evolution of the bump-on-tail instability in the present open system is very similar to that
in the previous three-dimensional simulations [Singh et al., 2000b,2001b; Singh, 2003]. When
we take a frame of reference moving with the electron beam in the open system, the spatial
evolution of the beam instability can be converted to the temporal evolution in the periodic
system [Umeda et al., 2002].
The electron beam instability takes place in a region close to the left boundary due to interaction between injected beam electrons and background electrons. Since an unstable beam
mode has the maximum growth rate in the direction parallel to the ambient magnetic eld,
purely one-dimensional electrostatic potentials are rst excited. As time elapses, oblique electron beam modes also grow, and they modulate background electron densities in the y direction.
Potential structures initially uniform in the y direction become twisted through modulation by
the oblique beam modes. As a result, potential structures in the region close to the source of
the electron beam become two-dimensional.
To study electromagnetic waves excited by the instability, we plotted snapshots of electromagnetic elds (a)Ez , (b)Bx B0 , (c)By and (d)Bz at e t = 256 in Figure 4.4. In the leading
edge of the electron beam, in which one-dimensional electron holes exist, there is little enhancement of magnetic elds. On the other hand, in regions where potential structures of electron

4.3. SIMULATION RESULTS

87

Figure 4.3: (a)Electron phase-space


distribution

f (x, y, vx )dy

and

Figure 4.4: Spatial proles of (a)Ez ,


(b)Bx , (c)By and (d)Bz at e t = 128.

the corresponding spatial proles of


(b)Ex , (c)Ey and (d) at e t = 128.

holes are two-dimensional, the magnetic eld components Bx , By and Bz are clearly enhanced.
This implies that the enhancement of the magnetic elds is caused by the perpendicular electric
eld Ey of the electron holes.
We computed frequency-wavenumber spectra of the magnetic eld components by performing
Fourier transformation in space (x, y) and time (t). The left panels of Figure 4.5 show the kx
diagrams of Bx B0 , By and Bz at ky e = 0.098, where ky e = 2/Ny y = 0.098. These
diagrams correspond to the spectral wave intensities obtained from the simulation data over
x/e = 128 256, y/e = 0 64, e t = 128 256. The kx diagrams show that the
magnetic eld components are enhanced along the beam mode line. These characteristics are
also found in the kx diagrams at ky e = 0.19 and 0.29. The result of Fourier analysis
implies that the excited magnetic elds are not normal modes.

4.3.2

Electron E B drift

We analyzed spatial proles of magnetic elds and current densities, and found that spatial
proles of Bx B0 , By and Jz almost satisfy Amperes law,
0 Jz =

By Bx

.
x
y

(4.1)

88

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

Figure 4.5: kx diagrams of Bx , By , Bz , Ex , Ey and Ez at ky e = 0.098. These


diagrams correspond to the spectral wave intensities obtained by Fourier transformation
of simulation data for x/e = 128 256, y/e = 0 64, and e t = 64 128.

This result means that most part of Bx and By are magnetostatic. To study generation mechanism of the current density Jz , we next compared the spatial prole of Jz in Figure 4.6a with
the spatial prole of Ey in Figure 4.3c. We found a good correlation between them. Perpendicular electric elds of electron holes correspond to electrostatic elds
 Ey . The ambient
y
magnetic eld exists in the x direction. Thus electrons can drift in the z direction, which corresponds to the direction perpendicular to both ambient magnetic eld B0 and perpendicular
electric eld Ey . We computed spatial proles of the current density Jz assuming that Jz is
induced by Ey B0 drift of electrons,
Jz = ene vz = ene

Ey
2
= 0 e Ey .
B0
e

(4.2)

As shown in Figure 4.6, we found that the obtained prole of Eq.(4.2) is in good agreement
with the spatial prole of Jz in the simulation. This implies that untrapped electrons as well
as trapped electrons of electron holes contribute to the enhancement of the current density Jz .

4.3. SIMULATION RESULTS

89

Figure 4.6: The spatial proles of (a)Jz and (b)Jz = ene By0 at e t = 128.
E

Thus two-dimensional potentials are accompanied by magnetic elds Bx and By . On the other
hand, since one-dimensional potentials are almost uniform in the y direction, they have no
substantial perpendicular electric eld Ey (i.e.,

 0). Both current density Jz and magnetic

elds Bx and By are little enhanced around one-dimensional potentials.

4.3.3

Electromagnetic wave emission

In the present simulation, we found that there exist current structures at edges of two-dimensional
electron holes in the perpendicular direction. There is a possibility of electromagnetic wave
emissions from the current. Electric elds can be separated into electrostatic elds E es and
electromagnetic elds E em ,
E es = ,

(4.3)

A
,
(4.4)
t
where A represents the magnetic vector potential dened by A = B with the Coulomb
E em =

gauge A = 0. In the two-dimensional system, Bx and By components contain magnetostatic


elds, while Ez component is purely electromagnetic. Since it is dicult to identify electromagnetic waves in the spatial prole of Ez in Figure 4a, we computed a frequency-wavenumber
spectrum of Ez by performing Fourier transformation in space (x, y) and time (t). The result
is plotted in the bottom-right panel of Figure 5. The spectra of Ex and Ey are also plotted for
comparison. The kx diagram of Ez clearly shows that electromagnetic waves are enhanced
along the beam mode line, as found in those of magnetic eld components. We also found
generation of electromagnetic waves at /e = 1.0 and /e = 1.8. However, these waves are
L- and R-modes enhanced by the thermal uctuations of the particle simulation. Since the
kx diagram at ky e = 0.098 is plotted, the frequencies of L- and R-modes become slightly
higher than their cut-o frequencies. The present result shows that electromagnetic waves are
excited from the moving current sources. The Ez component exists as an electromagnetic

90

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

beam mode. We note that the amplitude ratio of the electric eld Ez to the magnetic eld By
is equal to the phase velocity of the electromagnetic beam mode, |Ez |/|By |  Vp = 1.3Vte . On
the other hand, the amplitude ratio of the perpendicular electrostatic eld
to the magnetic
y
eld By is much larger than the light speed, |Ey |/|By |  30Vte .
As shown in Figure 4.5, the spectra of the electromagnetic beam mode are lower than the
electron plasma frequency. On the other hand, the spectra of the electron holes extend to the
electron plasma frequency, as shown in the kx diagram of Ex in Figure 5. The magnetic
vector potential is given by
2 A = 0 J .

(4.5)

Inserting Eq.(4) into the above equation and applying Fourier transformation, we obtain
(kx2 + ky2 )Az = 0 Jz 

2e
Ey .
c2 e

(4.6)

Thus the spectral wave intensity of the electromagnetic beam mode |Az (kx , ky )| is proportional
to |Ey (kx , ky )|/(kx2 + ky2 ). This means that the spectral wave intensity becomes smaller as the
wavenumber becomes larger such as a low-pass lter. Therefore, the spectral wave intensity
of the electromagnetic beam mode becomes smaller as the frequency becomes higher. The
wavelength of the electromagnetic beam mode becomes longer than the characteristic width of
electron holes.
The enhancement of the magnetic component Bz is also explained by Amperes law. An
electron hole is associated with a positive current, because the center of an electron hole is
positively charged [Krasovsky et al., 1997]. If an electron hole is uniform in the y direction,
obviously Bz becomes 0. On the other hand, if an electron hole has a nite dimension in the y
direction, Bz component becomes apparent at the edges of the two-dimensional electron hole.
Thus the spatial prole of Bz becomes very similar to that of Ey . The amplitude ratio of Ey
to Bz is |Ey |/|Bz |  20 30Vte , which is much larger than the light speed. We can separate
the electromagnetic component form the total Ey by
Eyem = Eytotal +

.
y

(4.7)

In this case, the amplitude ratio is equal to the phase velocity |Eyem |/|Bz | = Vp = 1.3Vte .

4.3.4

Eect of magnetization ratio

To demonstrate the validity of Eq.(4.2), we performed an additional run with a dierent ratio
of the electron plasma frequency to the electron cyclotron frequency e /e = 2.0. In this case
the amplitude of Jz becomes almost 0.8 times as large as that in the run with e /e = 1.0.
Since the amplitude of Ey component in the run with e /e = 2.0 becomes almost 1.6 times as
large as that in the run with e /e = 1.0, the simulation result is in agreement with Eq.(4.2).

4.3. SIMULATION RESULTS

91

The amplitudes of electromagnetic elds Bx B0 , By and Ez also become 0.8 times as large
as those in the run with e /e = 1.0, because these components are induced by the current
density Jz due to the Ey B0 drift of electrons.
We note that oblique electron beam modes with growth rates /e > 0.04 are excited in the
early stage of both runs. The growth rate becomes /e = 0.04 at  30 for e /e = 1.0,
and at  43 for e /e = 2.0. In the early stage (LHR t  1), since two-dimensional electron
holes are formed through modulation by oblique electron beam modes [Umeda et al., 2002],
the amplitude of Ey component is determined by the perpendicular wavelength of the oblique
electron beam modes. The ratio of the minimum perpendicular wavelength of oblique electron
beam modes in the run with e /e = 1.0 to that in the run with e /e = 2.0 is given by
tan 43/ tan 30  1.6. Therefore the amplitude of Ey in the run with e /e = 2.0 becomes 1.6
times as large as that in the run with e /e = 1.0. We also note that growth rates of oblique
beam modes do not change for a suciently high magnetization ratio (e /e  1.0). According
to the linear analysis, the wave-normal angle of the dominant beam mode with /e = 0.04
approaches to  54 (e /e 10.0), where we expect that the amplitude of Ey becomes 2.4
( tan 54 / tan 30 ) times as large as that in the run with e /e = 1.0.

4.3.5

Long-time nonlinear evolution

In Figures 4.3 and 4.4 we plotted results of an early stage, at e t = 256, in which both oneand two-dimensional electron holes exist. It is clearly demonstrated that electromagnetic elds
are accompanied by only two-dimensional electron holes. However, e t = 256 is a very short
timescale for evolution of electron holes and for eectiveness of ion dynamics. Therefore we
also show results of a later stage of the simulation run with e /e = 1.0. Figure 4.7a shows
the x vx phase diagram at e t = 1024. Figures 4.7b-h show the corresponding spatial proles
of (b), (c)
, (d)Jz , (e)Ex , (f)Ey , (g)Bx B0 and (h)By . At e t = 1024 the electron beam
y
has already reached the right boundary. Both electrostatic and electromagnetic waves are well
absorbed at the boundary. We can nd the bipolar signature of Ex corresponding to the electron
holes, while strong enhancement of Ey around x/e = 100 200 is also observed. In a later
stage of the simulation run upto LHR t  1, lower hybrid waves are excited due to ion dynamics
[Miyake et al., 2000; Singh et al., 2000b,2001b; Umeda et al., 2002]. From the perpendicular
electric eld of electron holes
, the current density Jz and magnetic elds Bx and By , we
y
can also conrm the formation of currents due to electron E B0 drift and enhancement of
electromagnetic elds even at the later stage. From the present simulation, we found that the
amplitude of electromagnetic waves is very small, |cB|  |E|. Since electrostatic waves are
dominant in the present simulations, the simulation results of electromagnetic simulations are
almost the same as those of the previous electrostatic simulations [Umeda et al., 2002]. We also
note that electron holes are stable for a long time in the weakly magnetized electron regime,

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

92

Figure 4.7: The electron phase-space distribution


spatial proles of (b),

(c)
y ,

f (x, y, vx )dy, and the corresponding

(d)Jz , (e)Ex , (f)Ey , (g)Bx B0 and (h)By at e t = 1024.

e e > b . However, in a long-time evolution of a strong magnetized system, it is found


that electron holes excited by a two-stream instability are unstable generating electrostatic
whistler waves [Goldman et al., 1999; Oppenheim et al., 1999, 2001].

4.4
4.4.1

Discussion
Waveforms of electromagnetic elds

Magnetic eld and current signatures associated with two-dimensional electron holes can be
modeled as a schematic illustration in Figure 4.8. Figure 4.8 a shows a case where an electron
hole is twisted in the y direction. On the other hand, Figure 4.8b shows a case where electron
holes have an edge in the y direction. To discuss waveforms of electrostatic and electromagnetic
waves, we specify one point in the y direction. In spacecraft observation, waveforms of plasma
waves are observed at a xed point. Waveforms of Ex , Ey and By shown in Figure 4.9 are those
at y/e = 20, e t = 128. These waves propagate toward right at the drift velocity of electron
holes as time elapses. Although the waveforms of Ey and By depend on the position y, By
component is strongly enhanced in the source region of electron holes, because Ey component
is also strongly enhanced by the oblique electron beam modes in the source region [Umeda et

4.4. DISCUSSION

93

Figure 4.8: Schematic illustration of magnetic eld and current density structures associated with two-dimensional electron holes. The current density is enhanced by the
E B drift of electrons, where E is perpendicular electric elds of electron holes
y ,
and B is the ambient magnetic eld B0 . (a)An electron hole is twisted in the y direction.
(b)Electron holes have an edge in the y direction.

al., 2002]. The waveforms of electrostatic and magnetic elds expected from Figure 4.8a are a
bipolar signature of Ey and a unipolar signature of By . However, the waveform of By in Figure
4.9 shows a sinusoidal signature. The present simulation box is very close to the source of the
electron beam, and thus the spatial distance between two electron holes is short, as one can
nd from the waveform of Ex in Figure 4.9. Therefore magnetic elds induced by electrons
undergoing the Ey B0 drift are superposed. When edges of neighboring two electron holes
have the opposite polarity as schematically illustrated in Figure 4.8b, magnetic elds associated
with the electron holes are superposed in such a way that the magnetic elds are reinforced,
which is observed as the strong bipolar Ey and unipolar By components at x/e = 220 in
Figure 4.9. On the other hand, when edges of neighboring two electron holes have the same
polarity, magnetic elds associated with the electron holes are superposed in such a way that
the magnetic elds cancel out each other.

4.4.2

Amplitude of electromagnetic elds

To discuss the amplitude ratio of electrostatic elds to magnetic elds, we performed a simple
one-dimensional analysis. We assume that a one-dimensional electron hole is tilted to the
direction of the ambient magnetic eld with angle as shown in Figure 4.10a. A potential of
an electron hole can be modeled as a Gaussian with width L and amplitude 0 [e.g., Krasovsky

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

94

Figure 4.9: Waveforms of Ex , Ey and By at y/e = 20, e t = 128.

et al., 1997; Muschietti et al., 1999],


x2
(x ) = 0 exp 2 .
2L


(4.8)

The perpendicular electric eld Ey is given by


Ey (x ) = Ex sin =

x sin
sin

(x ).
x
L2

(4.9)

The peak of the pulse is given at x = L. In the one-dimensional model, the magnetic
component By is given by


By (x ) = By cos = 0 cos

2 cos
Jz (x )dx = e 2
c e


Ey (x )dx =

2e sin 2
(x ).
2c2 e

(4.10)

In waveform observation, we use the maximum values of electric elds and magnetic elds to
obtain the amplitude ratio. Therefore, we take the maximum values of Eq.(4.9) and Eq.(4.10).
The amplitude ratio of Ey to By is given by
c2 e exp (0.5)
|Ey |
= 2
.
|By |
e L cos

(4.11)

When an electron hole has an edge in the perpendicular direction as in Figure 4.8b, we
assume that the potential is modeled as a two-dimensional Gaussian,

x2
y2
(x, y) = 0 exp 2 2 .
2Lx 2Ly

(4.12)

4.4. DISCUSSION

95

Figure 4.10: (a)Potential model for one-dimensional analysis. A Gaussian potential is


tilted to the direction of the ambient magnetic eld with angle . (b)Numerically computed amplitude ratio c2 e |By |/Lx 2e |Ey | versus Lx /e for dierent Ly /e .

The perpendicular electric eld Ey is given by


x2
y2
y
Ey (x, y) =
= 0 2 exp 2 2 .
y
Ly
2Lx 2Ly

(4.13)

To obtain the amplitude ratio of Ey to By , we solved the spatial prole of By numerically. We


took Fourier transformation of Ey and substituted the result into Eq.(4.6). The perpendicular
magnetic eld By is computed by
By (x, y) =

Az (x, y)
.
x

(4.14)

Figure 4.10b shows amplitude ratio c2 e |By |/Lx 2e |Ey | versus Lx /e for dierent Ly /e obtained by a numerical computation. For Lx  Ly the amplitude of By is given by |By | =
Lx 2e |Ey |/c2 e . For Lx Ly the amplitude of By is almost equal to |By |  0.2Lx 2e |Ey |/c2 e .
From the above analyses, the amplitude ratio of Ey to By is approximately given by
|Ey |
c2 e

.
|By |
Lx 2e

(4.15)

From the present simulation parameters, Lx  3e , e = e , the amplitude ratio is approximated as |Ey |/|By | 33Vte , which is in agreement with the simulation result. In the plasma
sheet boundary layer the parameters of electrostatic solitary waves are |E | 100V/m,
c 100Vte and Lx 10e . With a weak magnetization ratio e 0.1e , the amplitude of the
magnetic eld B in the plasma sheet boundary layer becomes of order of 0.01 0.1nT, which
may be too small to be measured by a waveform receiver.
The present simulations were performed in the weakly magnetized electron regime, e =
1, 2e , and therefore the magnetic eld is much weaker than the geomagnetic eld in the

96

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

auroral region, e 10e . However, the enhancement of magnetic elds associated with twodimensional electron holes is due to electron E B0 drift. The physical process is independent
of the magnetization ratio for e > b . Since we also used a typical two-dimensional Gaussian
potential model for estimation of the amplitude ratio of E to B , Eq.(17) is also applicable
to the strongly magnetized electron regime. In this paper we tried to estimate the amplitude
ratio of E to B in the auroral region based on the present simulation result. In consideration
with the parameters in the auroral region, |E | 100mV/m, c 100Vte and Lx e , the
amplitude of the magnetic eld B becomes of order of 0.01 1nT, which is in agreement with
those of the POLAR and FAST observations [Mozer et al., 1997; Ergun et al., 1998].

4.4.3

Contribution of ion E B drift

In the present simulations, we solved ion dynamics with a reduced mass ratio 100. However,
the enhancement of the current density Jz is not aected by ion dynamics, because the spatial
prole of the current density Jz can be obtained from Ey B0 drift of electrons. Electron
holes propagate in the direction parallel to the ambient magnetic eld with a velocity slightly
slower than the drift velocity of the electron beam Vd . The parallel dimension of an electron
hole is almost equal to 20e in the present simulation. Background electrons locally feel force
of a perpendicular electric eld of a two-dimensional electron hole for a time period of T =
20e /Vd = 10/e , which is almost equal to or longer than the electron cyclotron period 2/e .
Therefore, local untrapped electrons as well as trapped electrons of electron holes can lead to
E B0 drift in the z direction. On the other hand, the time period T = 10/e is much shorter
than the ion cyclotron period 2/i . Electron holes propagate away before local ions lead to
E B0 drift.

4.5

Conclusion

In conclusion, two-dimensional electron holes are accompanied by magnetic elds. On the


other hand, one-dimensional electron holes do not have any electromagnetic signatures. Twodimensional electron holes have perpendicular electrostatic elds E . Electrons lead to E B0
drift in the z direction, and the current density Jz is enhanced at edges of two-dimensional
electron holes in the perpendicular direction. Magnetic elds Bx and By are enhanced around
the current density Jz .
In the present simulation, the spatial distance between two electron holes is short because
the simulation box is very close to the source of the electron beam. Magnetic elds associated
with two-dimensional electron holes are superposed, and the waveform of the magnetic eld
becomes continuous. However, if the spatial distance between two electron holes is long enough,
it is possible to observe isolated magnetic elds associated with the electron holes such as those

4.5. CONCLUSION

97

observed by the FAST spacecraft in the auroral region [Ergun et al., 1998]. We also expect that
enhancement of magnetic eld components are frequently observed in source regions of electron
holes, which we can make use of to identify source regions of electron holes from waveform data
of plasma wave observations. The amplitude ratio of the electric eld to the magnetic eld
is estimated based on the present simulation result, and it is in agreement with those of the
POLAR and FAST observations.
The current structures induced by electrons undergoing E B0 drift are carried by electron
holes in the direction parallel to the ambient magnetic eld. An electromagnetic beam mode,
whose phase velocity and group velocity are equal to the drift velocity of the electron beam,
is excited by the moving current. The electromagnetic beam mode perturbs the background
plasma, and the magnetic eld components can be converted to a normal mode such as whistler
modes. We expect that the electromagnetic perturbation remains to be observed in distant regions, even when a multi-dimensional electron hole decays and the current structure disappears.
This process will be reported in a future paper.
In the present study, since we performed two-dimensional simulations, a current structure
induced by E B0 drift of electrons is not closed. In the real space plasma, we expect that
a current structure may form a loop along the edge of a multi-dimensional electron hole in the
perpendicular direction. Electromagnetic emissions from such a closed current structure will
be studied by future three-dimensional simulations.

98

CHAPTER 4. ELECTROMAGNETIC SIGNATURE ASSOCIATED WITH ESW

Chapter 5
Nonlinear Theory and Vlasov Simulation of
Harmonic Langmuir Waves
5.1

Introduction

The excitation of electrostatic multiple harmonic Langmuir waves has been observed in laboratory experiments since the 1960s [e.g., Apel, 1967, 1969], as well as in spacecraft observation
[e.g., Cairns, 1986] and in spaceborne rocket active experiment [e.g., Kellogg et al., 1986]. Such
a phenomenon was also observed in numerical simulations of the beam-plasma interactions [e.g.,
Joyce et al., 1971; Klimas, 1983, 1990; Schriver et al., 2000; Kasaba et al., 2001]. According
to these studies, harmonic Langmuir waves are excited at multiples of the plasma frequency
ne , n = 2, 3, 4, , and they all propagate with phase velocities roughly equal to the drift
velocity of the electron beam, /k V0 , where V0 is the average electron beam speed. This
implies that the wavelengths of the harmonic components, n , become shorter and shorter for
higher harmonics, n 2V0 /(ne ), or equivalently, the wave number of (n 1)-th harmonic
is given by k ne /V0 . Here we note that the rst harmonic mode corresponds to the L2
mode.
Available theories for the harmonic Langmuir wave generation can be categorized into two
classes. The rst class of theories, which were developed in the early 1970s, view the harmonics
as forced electrostatic perturbations generated by the interaction of trapped electrons and largeamplitude coherent Langmuir waves [e.g., Joyce et al., 1971; Klimas, 1983; ONeil et al., 1971,
1972]. Such theories are applicable for a suciently strong and cold beam, and for relatively
early nonlinear phase where the beam-plasma interaction is dominated by the formation of
coherent phase-space vortices. Another view, recently developed by Yoon [2000], is to consider
the harmonics as eigenmodes of nonlinear plasma system. This approach was prompted by
recent simulations [Schriver et al., 2000; Kasaba et al., 2001] which show that harmonic modes
persist even in late nonlinear phase when the coherent phase space structure is no longer
apparent, and when the plasma has entered a stage which can be genuinely characterized by
random phases.
Recently, the GEOTAIL spacecraft observed plasma waves at twoice of the plasma frequency
99

100

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

around the Earths electron foreshock, which are called 2fp emissions [Kasaba et al., 1997,
2000]. The simulations by Schriver et al. [2000] and Kasaba et al. [2001] were designed to
resolve only up to the rst harmonic (i.e., n = 1 and 2). These simulations clearly show on the
basis of simulated -k diagram, that the rst harmonic mode (n = 2) occupies a broad spectral
range which can be best described by a phenomenological dispersion relation,

kL2 e (2 + 3k 2 2e /2),

(5.1)

with a spectrum of k values largely centered around k 2e /V0 . Clearly, the broadband spectrum and the presistence of the harmonic mode in a fully turbulent stage cannot be discussed
solely on the basis of coherent nonlinear dynamics, which is why the alternative theory was
developed by Yoon [2000]. However, as with the simulations by Schriver et al. [2000] and
Kasaba et al. [2001], this theory is also restricted to the rst harmonic Langmuir mode only
[Yoon, 2000; Ziebell et al., 2001; Gaelzer et al., 2002].
At present, fully general theory which combines both aspects of turbulent (i.e., eigenmode
theory) and coherent nonlinear dynamics (forced perturbation theory) is not available. The
strengths and weaknesses of the two views are as follows: The coherent nonlinear theory has a
clearly identiable mechanism, i.e., particle trapping in the phase-space vortex, which accounts
for the generation of harmonic perturbations. On the other hand, it cannot be applied when
the system becomes turbulent, that is, when the spectrum broadens, and/or the phase-space
vortex is no longer characterized by a coherent wave. In contrast, the incoherent (or, turbulent)
eigenmode theory is applicable for a broadband spectrum, and it is capable of explaining the
persistence of harmonics over a long period. On the other hand, the theory does not have a
self-contained mechanism to generate the initial perturbations for the harmonic modes. Thus in
the eigenmode theory, an arbitrary level of initial harmonic mode has to be imposed [Gaelzer
et al., 2002, 2003]. In this respect, it can be envisaged that both coherent and incoherent
nonlinear theories have a role to play in a hypothetical complete theory.
The previous theory pertains only to the rst harmonic Langmuir mode (n = 2) [Yoon, 2000;
Ziebell et al., 2001; Gaelzer et al., 2002]. Obviously, it needs to be generalized to all higher
harmonics. Recently the nonlinear dispersion relations of all higher harmonics are derived by
Yoon et al. [2003]. To make a concrete comparison with the theoretical dispersion curves, we
performed Vlasov simulation of a weak-beam instability. In Section 5.2, we review the nonlinear
dispersion relations of harmonics Langmuir waves derived by Yoon et al. [2003]. In Section
5.3, we rst briey describe the one-dimensional electrostatic Vlasov code. We next describe
the model and parameters. Simulation results are presented in section 5.4. Section 5.5 gives
discussion and conclusion.

5.2. EIGENMODE THEORY OF HARMONICS LANGMUIR WAVES

5.2
5.2.1

101

Eigenmode Theory of Harmonics Langmuir Waves


New weak turbulence theory

The starting point of the nonlinear spectral balance equation is the self-consistent collisionless
Vlasov-Poisson equations. For simplicity, we assume one-dimensional system without magnetic
eld,

qs

Ex (x, t)
fs (x, vx , t) = 0,
+ vx
+
t
x ms
vx

1 

Ex (x, t) =
qs ns dvx fs (x, vx , t).
x
0 s

(5.2)
(5.3)

where fs stands for the particle distribution function for the s-th species normalized to 1. The
standard procedure is to separate the physical quantities into average and uctuating parts,
fs (x, vx , t) = Fs (vx , t) + fs (x, vx , t) and Ex (x, t) = Ex (x, t). In the weak turbulence theory
[Yoon, 2000], random-phase averages are employed. Note that the random-phase approximation
precludes coherent nonlinear eects. When these quantities are averaged over their phases, they
vanish. Inserting the above Vlasov-Poisson equations, we obtain

qs
qs

+
+ vx
+
Ex (x, t)
Fs (vx , t) +
Ex (x, t)
fs (x, vx , t) = 0, (5.4)
t ms
vx
t
x ms
vx

1 

Ex (x, t) =
qs ns dvx fs (x, vx , t). (5.5)
x
0 s

Averaging the above equations over the random phases of the uctuations, we obtain the
formal particle kinetic equation,
0=

qs
Fs
+
fs Ex ,
t
ms vx

(5.6)

where denotes ensemble averages over phases of the uctuations. Let us subtract the formal
particle kinetic equation from the original equation to obtain an equation for uctuation,

0=

Fs
qs
qs
+ vx
fs +
Ex
+
(fs Ex fs Ex ) .
t
x
ms
vx
ms vx

(5.7)

We assume that physical quantities are decomposed in the customary Fourier-Laplace transformation over the fast-time scale of the uctuation, while the amplitude of the spectra and
the average particle distribution evolve in slow-time scale,


fs (x, vx , t) =
Ex (x, t) =

dkd fs,k, (vx , t)eitikx ,

(5.8)

dkd Ex,k, (t)eitikx ,

(5.9)

Applying these transformations, the formal particle kinetic equation, the equation for uctuating eld and the equation for uctuating distribution are respectively expressed as follows:
qs

Fs (vx , t) =
t
ms vx

dkd

dk1 d1 fs,k, (vx , t)Ex,k1 ,1 (t) ei(+1 )ti(k+k1 )x ,

(5.10)

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

102

1 
ikEx,k, (t) =
qs ns dvx fs,k, (vx , t),
(5.11)
0 s

qs

i + ikvx +
fs,k, (vx , t) =
Ex,k, (t)
Fs (vx , t)
(5.12)
t
ms
vx


qs
dk1 d1
(fs,kk1,1 Ex,k1 ,1 fs,kk1,1 Ex,k1 ,1 ) .
+
ms
vx
To solve the above equations, we expand the perturbed particle distribution fs,k, ,
(1)

(2)

(3)

fs,k, = fs,k, + fs,k, + fs,k, + .

(5.13)

Inserting the above series expansion to the original equation (5.12), we obtain
qs
gk, Fs Ex,k, ,
ms 

qs
(1)
(1)
=
dk1 d1 gk, fs,kk1,1 Ex,k1 ,1 fs,kk1,1 Ex,k1 ,1 ,
ms 

qs
(2)
(2)
=
dk1 d1 gk, fs,kk1,1 Ex,k1 ,1 fs,kk1,1 Ex,k1 ,1 ,
ms

(1)

fs,k, =
(2)

fs,k,
(3)

fs,k,

(5.14)
(5.15)
(5.16)

..
.
where
gk,
Here

1
.
kvx + i0 vx

(5.17)

of Eq.(5.12) is 0, which means that the small-time scale of the uctuation is neglected

in the fast-time scale. Iterative solution for fs,k, is given by


fs,k, 

qs
gk, Fs Ex,k,
(5.18)
ms


q2
gk, gk2 ,2 Fs (Ex,k1 ,1 Ex,k2 ,2 Ex,k1,1 Ex,k2,2 )
+ s2 g
ms k1 +k2 =k 1 +2 =
+



qs3
g
gk, gkk1,1 gk3 ,3 Fs (Ex,k1 ,1 Ex,k2,2 Ex,k3,3
3
ms k1 +k2 +k3 =k 1 +2 +3 =

Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3 Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3 )


where


da1 da2 (a1 + a2 a).

(5.19)

a1 +a2 =a

We substitute fs,k, to Eq.(5.11),


kEx,k, 

2s Ex,k,

dvx

q2
+ s2
ms

qs 
dvx gFs
ms


(5.20)

gk, gk2 ,2 Fs (Ex,k1,1 Ex,k2,2 Ex,k1 ,1 Ex,k2 ,2 )

k1 +k2 =k 1 +2 =

dvx

gk, gkk1 ,1 gk3 ,3 Fs (Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3

k1 +k2 +k3 =k 1 +2 +3 =

Ex,k1 ,1 Ex,k2,2 Ex,k3 ,3 Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3 ) .

5.2. EIGENMODE THEORY OF HARMONICS LANGMUIR WAVES

103

The rst-order linear solution of the above equation is given by


(1)

0 = (k,) = 1 + (k,) ,

(5.21)

where
(1)
(k,)

2s 
k
Fs
dv
x
k2
kvx + i0 vx

(5.22)

For the second-order nonlinear terms, We assume that (k1 , 1 ) and (k2 , 2 ) are symmetrized.
The second-order nonlinear term becomes


qs
2s
dvx gk1 +k2 ,1 +2 (gk1 ,1 + gk2 ,2 ) Fs
s ms 2(k1 + k2 )

 qs
2s
1
(5.23)
=
dvx
1 + 2 (k1 + k2 )vx + i0
s ms 2k1 k2 (k1 + k2 )





Fs
Fs

k1

k2
k2
+ k1
.
vx 1 k1 vx + i0 vx
vx 2 k2 vx + i0 vx

(2)

(k1 ,1 |k2 ,2 )

For the third-order nonlinear terms, We assume that (k2 , 2) and (k3 , 3 ) are symmetrized. The
third-order nonlinear term becomes


qs2
2s
dvx gk1 +k2 +k3 ,1 +2 +3 gk2 +k3 ,2 +3 (gk2 ,2 + gk3 ,3 ) Fs
2
s ms 2(k1 + k2 + k3 )

 qs2
2s
1
=
dvx
2
1 + 2 + 3 (k1 + k2 + k3 )vx + i0
s ms 2k1 k2 k3 (k1 + k2 + k3 )


1
k1
(5.24)
vx 2 + 3 (k2 + k3 )vx + i0





Fs
Fs

k2

k3
+ k2
.
k3
vx 2 k2 vx + i0 vx
vx 3 k3 vx + i0 vx

(3)

(k1 ,1 |k2 ,2 |k3 ,3 )

Thus the result becomes


0 = (k,) Ex,k,

k1 +k2 =k 1 +2 =

(2)

(k1 ,1 |k2 ,2 ) (Ex,k1 ,1 Ex,k2 ,2 Ex,k1 ,1 Ex,k2 ,2 )

(3)

k1 +k2 +k3 =k 1 +2 +3 =

(k1 ,1 |k2 ,2 |k3 ,3 ) (Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3

Ex,k1 ,1 Ex,k2,2 Ex,k3,3 Ex,k1 ,1 Ex,k2 ,2 Ex,k3 ,3 ) .

(5.25)

Let us multiply Ex,k , to the above equation, and take the ensemble average of the resulting
equation,
2
(k + k  )( + )
0 = (k,) Ex,k,

2(k + k  )(

  (3)
+ )

k

k1 +k2 =k 1 +2 =

(2)

(k1 ,1 |k2 ,2 ) Ex,k1,1 Ex,k2 ,2 Ex,k ,

2
2
(k ,  |k  ,  |k,) Ex,k  ,  Ex,k, .

(5.26)

104

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

The above equation is obtained from the ensemble averages


Ex,k, = 0

(5.27)

2

Ex,k, Ex,k , = (k + k  )( + ) Ex,k,

(5.28)

Ex,k1,1 Ex,k2 ,2 Ex,k3,3 Ex,k,  (k1 + k2 + k2 + k  )(1 + 2 + 2 + )

(5.29)

2
2
Ex,k

(k1 + k2 )(1 + 2 ) Ex,k
1 ,1
3 ,3
2
2
Ex,k

+(k1 + k3 )(1 + 3 ) Ex,k
1 ,1
2 ,2

2
2
Ex,k
.
+ (k2 + k3 )(2 + 3 ) Ex,k
1 ,1
2 ,2

The last term of Eq.(5.26) is obtained from




k1 +k2 +k3 =k 1 +2 +3 =

(3)

2
2
(k1 ,1 |k2 ,2 |k3 ,3 ) (k)() (k1 + k2 )(1 + 2 ) Ex,k
Ex,k

1 ,1
3 ,3

2
2
+ (k1 + k3 )(1 + 3 ) Ex,k
Ex,k

1 ,1
2 ,2

= 2(k + k  )( +  )

  (3)
k

2
2
(k , |k , |k,) Ex,k
 ,  Ex,k, .

(5.30)

Let us consider solution of this equation,


(1)

(2)

Ex,k, = Ex,k, + Ex,k, + ,

(5.31)

(1)

= (k,) Ex,k, .

The second-order nonlinear term is given by


(2)

Ex,k, =

1
(k,)

(2)

k1 +k2 =k 1 +2 =

(1)

(1)

(1)

(1)

(k1 ,1 |k2 ,2 ) Ex,k1 ,1 Ex,k2 ,2 Ex,k1,1 Ex,k2,2 . (5.32)

The three-body correlation can be approximated as follows:


(2)

(1)

(1)

(1)

(2)

(1)

Ex,k1,1 Ex,k2,2 Ex,k,  Ex,k1,1 Ex,k2 ,2 Ex,k, + Ex,k1 ,1 Ex,k2 ,2 Ex,k,


(1)

(1)

(2)

+ Ex,k1,1 Ex,k2 ,2 Ex,k ,


(2)

2(k,|kk, )
(k , )

(1)

(1)

(Ex,k, )2 (Ex,kk, )2

(2)

2(k,|k, )
(k k  ,  )

(1)

(1)

(Ex,k, )2 (Ex,k, )2

(2)

2(k , |kk, )
(k, )

(1)

(1)

(Ex,k, )2 (Ex,kk, )2 ,

(5.33)

where we have deleted the subscripts of Ex , and (k, ) =  (k, ), (2) (k1 , 1 |k2 , 2 ) =
(2) (k1 , 1|k2 , 2 ).
We again consider the slow time perturbation in the linear term,


(k,+i ) Ex,k,
t

i (k,)
= (k,) +
Ex,k, ,
2 t

(5.34)

5.2. EIGENMODE THEORY OF HARMONICS LANGMUIR WAVES


1
2

where the factor

105

in front of the time derivative term stems from the fact that

Ex,k, = Ex,k, .
t
t

(5.35)

Using the following relation,


(2)

(2)

(k,|kk, ) = (k , |kk, )


(2)
(k,|k, )

(2)
(kk, |k, )

(5.36)
=

(2)
(k , |kk, )

(5.37)

we obtain the formal wave kinetic equation as




0 =

i (k,)
(k,) +
I(k, )
2 t

dk d

(2)

|(k , |kk, ) |2
(k,)

2
(2)
(k , |kk, )

(5.38)

I(k  , )I(k k  ,  )

I(k k  , )
I(k  , )
+
I(k, )
(k , )
(kk , )

dk  d  (k , |k , |k,)I(k, )I(k ,  ),


(3)

where I(k, ) Ex2 k, .

5.2.2

Nonlinear dispersion relation

We rst consider the linear dielectric function (k,) . For the fundamental Langmuir mode L1,
the dispersion equation is given by
0 = (k,) = 1

2e

2b
kVd


Z
(
)

Z (
),
2
2
2
2
2k Vte
2k Vtb
2kVte
2kVtb

(5.39)

where Z represents plasma dispersion function




ex
1
dx
Z() =
x
Z  () = 2(1 + Z()).
2

(5.40)
(5.41)

The analytical solution is well known,


2e
3k 2 Vte2
,
0 = (k,) = 1 2 1 +

2


3k 2 Vte2
L1
.
k = e 1 +
22e

(5.42)
(5.43)

For n-th harmonic Langmuir wave, we assume that the total spectral wave intensity is given
by the sum of individual wave intensity for each harmonic mode,
I(k, ) =


n

I Ln (k)( kLn ).

(5.44)

106

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Let us insert Eq.(5.38) into the real part of Eq.(5.26). We note that the third-order susceptibility
contributes very little to the dispersion relation, since it does not contain the inverse of the
linear dielectric function, as the second-order terms do. Thus we ignore the third-order nonlinear
terms in the discussion of the dispersion relations. The result is
0 =

Re(kn ,Ln ) I

(k) + 2Re

dkn

Ln |k k  , Ln Ln )
n
n ,
n

(kn kn , Ln Ln )

(2)

|(k

n ,n

(k

(5.45)

)2

(2)

(k

n,n

Ln

Ln |k  , Ln )
n ,
n

I Ln (kn )I Ln (kn )

|2



I Ln (kn )I Ln (kn ) .

Ln + Ln )
n +kn ,

Here, when two harmonics are adjacent to each other, the frequency dierence between the two
harmonics becomes e . In this case the second order term becomes large because the linear
dielectric function becomes very small,
(kn kn ,Ln Ln )

0 where n = n 1,

= 0 where n = n 1.

(5.46)

For the rst harmonic mode L2, keeping the nonlinear term with frequency dierence equal
to e , we have


0 = Re(k2 ,L2 ) I L2 (k2 ) + 4Re

dk1

(2)

(k1 ,L1 |k2 k1 ,L2 L1 )

2

I L1 (k1 )I L2 (k2 ).

(k2 k1 ,L2 )

(5.47)

We note that the last term of Eq.(5.44) does not contribute because the delta function is never
satised.
For the second harmonic mode L3, we have


0 = Re(k3 ,L3 ) I L3 (k3 ) + 4Re

dk2

(2)

(k2 ,L2 |k3 k2 ,L3 L2 )

2

I L2 (k2 )I L3 (k3 ).

(k3 k2 ,L3 L2 )

(5.48)

(1)

In this case the last term of Eq.(5.44) is also ignored because k3 ,L3 becomes nite.
For the third harmonic mode L4, we have


0 = Re(k4 ,L4 ) I L4 (k4 ) + 4Re

dk3

(2)

(k3 ,L3 |k4 k3 ,L4 L3 )


(k4 k3 ,L4 L3 )

2

I L3 (k3 )I L4 (k4 ).

(5.49)

We note that the coupling between the L2 mode with the L2 mode is ignored because frequency
dierence is not equal to e .
In general (for n 2), the (n 1)-th harmonic mode (i.e.,Ln) is generated through coupling
of L1 mode with L(n 1) mode. Therefore, we have


0 = Re (kn ,Ln ) + 4

dk(n1)

(5.50)

5.2. EIGENMODE THEORY OF HARMONICS LANGMUIR WAVES


%

(2)
(k
L(n1) |k k
Ln L(n1) )
n
(n1) ,
(n1) ,

(kn k(n1) ,Ln L(n1) )

107

)2

I L(n1) (k(n1) ) I Ln (kn ).

The second-order susceptibility can be approximated by


(2)

(k

(n1) ,

L(n1) |k

n k(n1) ,

Ln L(n1) )

2e
1
qe
me Ln L(n1) ( Ln L(n1) ) 2kn k(n1) (kn k(n1) )
2
2
(kn k(n1) )
k k(n1) (kn k(n1) ) kn k(n1)
+
n
Ln

L(n1)

kn k(n1) (kn k(n1) )2
+
(5.51)
Ln L(n1)


kn1
1
kn
qe

(kn k(n1) ) (5.52)



me 2n(n 1)2e n
n1


)2

(2)

(k

(n1) ,

L(n1) |k

n k(n1) ,

qe2 1 (n)
A
,
m2e 44e (kn ,k(n1) )

Ln L(n1) )

(5.53)

(n2 1)kn n(n 2)k(n1)

(n)

A(kn ,k(n1) ) =

n4 (n 1)4

2

(5.54)

where we ignored the thermal eect (Vt 0) and we approximated Ln ne [Yoon, 2000].
This lead to the following dispersion equation:
Re(k,Ln )

q2 1
= e2 4 Re
me e

(n)

dk(n1)

A(kn ,k(n1) ) I L(n1) (k(n1) )


(kn k(n1) ,Ln L(n1) )

(5.55)

From Eqs.(5.42) and (5.43), we approximate that


1
(kn k(n1) ,Ln L(n1) )

=



2e
3(kn kn1 )2 Vte2
1 Ln
1
+
( L(n1) )2
( Ln L(n1) )2

1

( Ln L(n1) )2
( Ln L(n1) )2 2e

2
3(kn kn1 )2 Vte
Ln
L(n1)
(
)2

2e
( Ln L(n1) )2 ( L1 )2

q2 1
2
0  1 Lne 2 e2 2
( )
me e

(5.56)

(n)

dk(n1)

A(kn ,k(n1) ) I L(n1) (k(n1) )


( Ln L(n1) )2 ( L1 )2

(5.57)

where L1 is the dispersion relation of the fundamental Langmuir mode (5.43). We further
approximate that
( Ln L(n1) )2 ( L1 )2 = ( Ln L(n1) + L1 )( Ln L(n1) L1)
 2e ( Ln L(n1) L1 ).

(5.58)

108

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Let us assume that the full dispersion relation can be expanded in a Taylor series in 2e ,
-

Ln

Ln -= |2e =0 +
2 + ,
2e -2 =0 e
Ln

(5.59)

where
Ln |2e =0 = e (n + (n) )

qe2
n2
(n)
dk(n1) A(kn ,k(n1) ) I L(n1) (k(n1) ).
(n) =
2(n2 1) m2e 4e

(5.60)

Here Te = 12 me Vte2 0 is assumed.


To obtain the thermal corrections to the nonlinear dispersion relation,
(n)
A(kn ,k(n1) ) I L(n1) (k(n1) )
qe2 1 
n2
dk(n1)
,
1=
2(n2 1) m2e 3e
Ln L(n1) L1

(5.61)

we take the derivative with respect to 2e . Ignoring 2e dependence of I L(n1) (k(n1) ), we have


qe2 1
n2
0 =
2(n2 1) m2e 3e

Ln (n) 2 kn  e
2 e

Ln kn 2
e
2e -2 =0

dkn1

L(n1) -kn1
2e

2e L(n1)

where we approximated kLn


kn1
n

L(n1) -

qe2 1
n2
2(n2 1) m2e 4e

2e =0

L(n1)

(kLn
kn1
n

kn1

2e

(n)

A(kn ,kn1 ) I L(n1) (kn1 )

2e =0

L1
(k
)2
n kn1 )

(5.62)

3
e (kn kn1 )2
2

(n)

dkn1 A(kn ,kn1 ) I L(n1) (kn1 )

2e =0

3
+ e (kn kn1 )2 ,

L1
(k
 kLn
ne . From the above equation we
n
n kn1 )

obtain
-

kLn
3
(n)
2
n
=
k
+
2
e
n
2e -2 =0
2
(n)
e

(n)

qe2
n2
=
2(n2 1) m2e 4e

2
kn1

(5.63)


(n)

dk(n1) A(kn ,k(n1) ) I L(n1) (k(n1) )

(5.64)

(n)
kn kn1 + (n) ,


with (1) = 0, (1) = 0. In general, therefore, we obtain the desired nonlinear dispersion relation
for harmonic Langmuir waves for n 2,

kLn = e n +

(n)

(n)
k

3
3k 2
+ k 2 2e + (n)
e .
2
k

(5.65)

5.2. EIGENMODE THEORY OF HARMONICS LANGMUIR WAVES

109

Eq.(5.65) is quite general but formal. In a time-dependent calculation of the wave kinetic
equation, Eq.(5.65) is solved at each moment in time by using the actual intensity I Ln (kn )
obtained on the basis of the wave kinetic equation. However, for simplicity, let us consider
specic models of the spectra of harmonic Langmuir waves. The physical situation is that an
electron beam interact with the thermal background plasma to excite primary Langmuir waves
(L1). The enhanced L1 mode is the source of the L2 mode, which in tern leads to L3 mode,
and so on. The primary Langmuir mode L1 is excited in a range of spectrum with average
wavenumber given by k0  e /V0 , where V0 is the average electron beam speed. In general,
(n 1)-th harmonic Langmuir mode Ln has a spectrum with average wavenumber located
at nk0  ne /V0 . On the basis of this consideration, let us model the spectra of harmonic
Langmuir modes by

Ln

In
(k nk0 )2
exp
(k) =
,
w
w2

(5.66)

where w represents the spread associated with the spectra (w  k0 ). Eq.(5.54) is rewritten as
A(k,k ) = n k 2 + n kk  + n k 2
(n)

(5.67)

(n + 2)
n4 (n 1)2
2(n + 1)(n 2)
=
n3 (n 1)3
(n 2)2
=
.
n2 (n 1)4

n =
n
n
By using the following properties:


(k  nk0 )2
1
exp
dk
w
w2



(k  nk0 )2
 1

dk k exp
w
w2




2
1

nk
)
(k
0
dk  k 2 exp
w
w2



(k  nk0 )2
 1
3
dk k exp
w
w2




2
1

nk
)
(k
0
dk  k 4 exp
w
w2


= 1,
= nk0 ,
w2
,
2
3
= n3 k03 + nk0 w 2,
2
= n2 k02 +

3
= n4 k04 + 3n2 k02 w 2 + nk0 w 4,
4

the quantity (n) is rewritten as


(n)

k =



qe2
n2
2
2
2
I

(n

1)
k
k
+
(n

1)

k
)
,
n1
n
n
0
n
0
2(n2 1) m2e 4e

(5.68)

where terms related to w is ignored. The quantity (n) is also rewritten as



(n)
k

(n)
k k0

n(n 1)
k0 (n 1)k .
2

(5.69)

110

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Figure 5.1:
k0 =

(a)Dispersion relations of harmonic Langmuir waves (5.70) with

2e +

2b /(Vd

Vtb ). (b)Linear growth rates of harmonic Langmuir waves (5.72).

The parameters nb /ne = 0.001, Vd = 5Vte and Vtb = 0.5Vte are used for the calculation.
(c)The ratio of the growth rate to the real frequency / versus k.

From Eqs.(5.68) and (5.69), the desired dispersion relations for harmonic Langmuir modes are
obtained,
%

kLn

= e

3
3
n + k 2 2e + (n 1)k0 (nk0 2k)2e ,
2
2

(5.70)

(n)

where we have ignored the small correlation term k .


Figure 5.1a shows the nonlinear dispersion relations for harmonic Langmuir waves (5.70). For
the calculation we used following parameters: nb /ne = 0.001, Vd = 5Vte and Vtb = 0.5Vte . The


average wavenumber k0 is given by k0 =

2e + 2b /(Vd Vtb ) rather than k0 =

2e + 2b /Vd ,

because the maximum positive derivative of the velocity distribution function occurs when
2f
= 0.
v2
2e + 2b .

In the calculation of the dispersion curves we also replaced 2e of Eq.(5.70) by


The straight dashed line shows the beam mode = k(Vd Vtb ). The beam mode is

not an actual eigenmode solution, but is shown here only as a reference, because the growth of
each harmonic is expected to occur only around the intersections between the harmonic mode
dispersion curves and the beam mode line.
Before we close, we discuss growth rates of harmonic modes. From Eq.(5.38) and Eq.(5.55),
we obtain
0 =


1 Re(k,) Ln
I (k) +
Im(kn ,Ln ) I Ln (k)
2
t
n

q2 1
e2 4 Im
me e

(5.71)

(n)

dk(n1)

A(kn ,k(n1) ) I L(n1) (k(n1) )I Ln (k)


(kn k(n1) ,Ln L(n1) )

We note that if we are only interested in the suciently early (i.e., quasi-linear) phase of
the harmonic mode generation process, then we may discuss the amplication of Ln mode by

5.3. VLASOV SIMULATION

111

considering its linear growth rate,


kLn 

Im(k, kLn )
Re (k,kLn )
kLn


8
nb
+
ne

kLn
kVte

Te
Tb

 32

3 

kLn

(kLn )2
exp 2 2
2k Vte

(5.72)

( Ln kV0 )2
(kLn kV0 ) exp k 2 2
,
2k Vtb

where
22e
Re(k, )

,

3
2e f -Im(k, ) = 2
.
k v -v=/k

(5.73)
(5.74)

Figure 5.1b shows the linear growth rates of harmonic Langmuir waves (5.72). We note that
the peak value of monotonically increases for increasing n, which conrms the ndings of the
previous Vlasov simulation [Klimas, 1983, 1990] that the initial growth rate for the harmonic
mode is higher than that of the fundamental by a factor roughly proportional to the harmonic
mode number. We also note that the growth of the harmonic mode occurs only in the vicinity
of the intersections between the beam mode line and the dispersion curves.
Since the maximum of the growth rate increases for increasing harmonic mode number, n, one
might wonder whether the original assumption,  , is violated for very high n. However,
since the frequency also increases as a function of n, the ratio, /, always remains small for
all n. To see this, we replot the result shown in a new format, in terms of the ratio /, in
Figure 5.1c. The numerical result reconrms the inequality,  , which validates the present
approximate analysis.

5.3
5.3.1

Vlasov Simulation
One-dimensional Vlasov-Poisson solver

Kinetic simulation codes are widely used for studies of nonlinear processes in plasmas. The
numerical methods for solving nonlinear Vlasov-Maxwell equations fall into two groups. One is
particle code, which follows trajectories of individual particles, described in Chapter 2. However, limitation of number of superparticles gives rise to the enhanced thermal uctuations
as described in Section 2.2.2. Another approach is to follow temporal evolution of velocity
distribution functions, which is called Vlasov simulation code. In Vlasov simulations, velocity
distribution functions are dened at grid points in position-velocity phase space. The advantage
of the Vlasov code is that we can suppress the thermal uctuations completely. On the other

112

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

hand, it requires substantial computer memory to realize three-dimensional velocity space and
two- or three-dimensional conguration space with sucient resolutions. Even with a recent
supercomputer systems it is still dicult (but not impossible) to perform a three-dimensional
(i.e., six-dimensional phase space) Vlasov simulation.
The generation of multiple harmonic Langmuir modes has also been studied with particle
simulations [e.g., Kasaba et al., 2001]. However, the generation of harmonics is restricted to
only the rst and second harmonics (n 3), because the generation of much higher harmonics
are disturbed by enhanced thermal uctuations. Therefore, in this chapter, we developed a
new one-dimensional electrostatic Vlasov code, in which enhanced thermal uctuations are
completelt suppressed. The new code solves two-dimensional x vx phase-space distribution
functions, and it is possible to perform simulations on a personal computer with a few gigabytes
of memory.
There are several numerical methods for integration of the velocity distribution function such
as Fourier-Hermite transformation [Joyce and Knorr, 1971], Fourier-Fourier transformation
[Klimas, 1983] and splitting methods [Cheng and Knorr, 1976; Gagn
e and Shoucri, 1977].
In the splitting method the velocity distribution function is advanced in time by shifting of
the function in both conguration and velocity spaces. Although the algorithm of the splitting
method is simple and easy for programming, this method requires a large number of grid points
in both conguration and velocity spaces to suppress numerical diusion. On the other hand,
the transformation methods can integrate the velocity distribution function stability without
numerical diusion, while the algorithm becomes very complex. Recently the Constrained
Interpolation Prole (CIP) scheme was developed by Yabe et al. [2001]. The CIP scheme is
a low diusion and stable numerical method, and can solve hyperbolic equations with high
accuracy. Since the CIP scheme conventionally used a cubic polynomial function, the scheme
was called Cubic Interpolated Propagation scheme [e.g., Yabe et al., 1991a,b]. The CIP scheme
was improved with dierent orders of polynomials to provide exact conservation of the total
density [e.g., Yabe et al., 2001]. Recently the CIP scheme with a cubic polynomial function
is applied to the splitting method for integration of the velocity distribution function [Utsumi
et al., 1998; Nakamura and Yabe, 1999]. By adopting the Constrained Interpolation Prole
(CIP) scheme for integration of velocity distribution functions, numerical stability is greatly
improved and numerical diusion becomes considerably small. Basic principle of the CIP
scheme is described in Appendix A.1, and detailed numerical techniques of the new Vlasov
code are presented in Appendix A.2.

5.3.2

Model and parameters

The simulation box is taken in the one-dimensioanl system along the ambient magnetic eld,
while the eect of the ambient magnetic eld is neglected in the present simulation. The velocity

5.3. VLASOV SIMULATION

113

distribution function f (x, vx ) is discretized over the xvx phase space with Nx Nv = 1024512
cells. Periodic boundary conditions are imposed at both ends of the simulation box. The
system length is Lx = 256e , where e = Vte /pe is the initial Debye length. The maximum
and minimum velocities considered in the simulation are respectively given by vmax = 5Vte
and vmin = 5Vte . Thus the grid spacing in the x and vx directions are respectively given by
x = 0.25e and v = 0.02Vte . The time step is t = 0.01pe .
In the present study we assume that a weak electron beam is drifting against the background
thermal electrons at the initial condition. Since the periodic boundary conditions are imposed,
both beam electrons and background electrons are assumed to exist uniformly. The density
ratio of the beam electrons to the background electrons is given by nb /ne = b /e = 0.001,
where b =

qe n2b
0 me

is the plasma frequency associated with the beam component. Here, of course,

qb = qe = e is the electron charge and mb = me is the electron mass. The initial thermal
and drift velocities associated with the beam electrons are respectively given by Vd = 5Vte and
Vtb = 0.5Vte . The ion dynamics are not considered in the present simulation. They are assumed
to form an immobile positively charged background.
In order to initiate the electron beam-plasma instability process, the seed perturbation must
be imposed at the outset in the Vlasov simulation. This is accomplished in our simulation
by the initial charge density (i.e., electron density) perturbation, which is modulated with a
charge-neutral function,
Nx /2
1 
n1 (x)/n0 =
A(ki ) cos(ki x + i ),
e i=1

(5.75)

where i = 2Ri with Ri being a uniformly distributed random number, 0 Ri < 1. The
dierence equation of the one-dimensional Poissons equation is given by
Ex (i) Ex (i 1)
(i) + (i 1)
.
=
x
20

(5.76)

By taking the Fourier transformation of the above dierence equation, we obtain that the
amplitude of each wave mode A(ki ) takes the following form [Umeda et al., 2003]:
A(ki ) =
Ki =

0 Ki |Ex (ki )|
,
cos[ki (x/2)]

sin[ki (x/2)]
.
ki (x/2)

(5.77)

In the present simulation the spectral shape of the initial noise is chosen to be in a power-law
form, although other choices can also be made,
log10 |Ex (ki )| = log10 ki
for
i = 1 Nx /2.

(5.78)

114

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Figure 5.2: An -kx diagram corresponding to the wave intensity, |Ex (, kx )|2 , over a
time period ranging from e t = 409.6 to e t = 512.0. The intensity is normalized by the
thermal energy of the background electrons ne me Vte2 . The solid lines show the linear and
nonlinear dispersion curves obtained from Eq.(5.70).

Finally, we note that the discretized wave number, ki , is related to the unit wave number
increment, k, by ki = ik where k = k1 = 2/Lx . Specically, the parameters of the
power-law spectrum presented in this chapter are = 4.0, = 2.5.

5.4
5.4.1

Simulation Result
Frequency-wavenumber characteristics

In Figure 5.2 we plot an kx diagram obtained by taking the Fourier transformation of


electric eld data in both space and time. This diagram shows |Ex (, kx )|2 over an interval
corresponding to the time period ranging from e t = 409.6 to e t = 512.0. The intensity is
normalized by the thermal energy density of the background electrons ne me Vte2 = (me e Vte )2 /e2 .
To compare the present simulation result with the theory, we have superposed the theoretical
dispersion curves, Eq.(5.70), on the kx diagram in Figure 5.2. In plotting this result, we
have carefully estimated the actual thermal velocity of the background electrons, average drift
velocity, and the thermal velocity of the beam electrons after the saturation of the instability

5.4. SIMULATION RESULT

115

(e t = 409.6). This is because the theoretical dispersion relation (5.70) depends on the particle
parameters, Vd , Vte , and Vtb . We could have simply used the initial values for these quantities,
Vte = 0.2Vd , Vtb = 0.1Vd , and still would have obtained a fairly reasonable agreement. However,
since both the background and beam electrons undergo slight heating, while the average beam
speed decreases during the wave growth and saturation process, we have decided to improve
upon the theoretical dispersion curve by computing the average values for these quantities
during the time period of the simulation. We nd that the numerically computed quantities
correspond to Vte = 0.2Vd , Vtb = 0.15Vd , and Vd = 0.96Vd . This implies a decrease in average
drift speed and a modest heating in the beam electrons. Making use of these rened electron
speeds, we nd an excellent agreement between the nonlinear dispersion relation (5.70) and the
present Vlasov simulation result.
Graphical representation of the simulation results in kx diagram format, as we do here,
was not done in early simulations which reported Langmuir harmonic generation by beamplasma instability [e.g., ,Klimas, 1983, 1990]. The kx diagram is very informative, since
without such a plot, one cannot fully characterize the true nature of the harmonic modes. In
the absence of such an information, early interpretation of the harmonic generation tended to
refer to these modes as a quasi beam mode,  kx Vd . In the recent simulations by Schriver
et al. [2000] and Kasaba et al. [2001], on the other hand, one can nd similar representations
of the simulation result such as Figure 5.2. However, these works were limited to the rst
harmonic (n = 2) only. In contrast, Figure 5.2 shows not only the rst harmonic but all higher
harmonics up to n = 12, which has not been reported. The kx diagram clearly shows that
the harmonic Langmuir waves are not excited along the beam-mode line,
= k x Vd ,
but are excited in kx space, dictated according to the nonlinear dispersion relation [see
Eq.(5.70)],
= kLn
.
x
Note that the intensity of higher harmonics decreases as the harmonic mode number increases.
The high harmonic modes possess very low intensity levels. As such, it would be very dicult
to simulate these modes by means of the present particle-in-cell simulations owing to the high
noise by the enhanced thermal uctuations. In this regard, the Vlasov simulation is an ideal
tool to study the phenomenon of Langmuir turbulence involving the harmonic generation.
According to the kx diagram in Figure , the higher harmonics have incrementally broader
ranges of wave numbers over which the modes are excited. This result is also in agreement
with the theoretical growth rate property, Eq.(5.72), of the harmonic Langmuir modes (see
Figure 5.1b). However, we should also point out that the agreement between the theory and
simulation becomes better for the lower harmonics. As the harmonic mode number increases,
on the other hand, the theoretical curves and the excited wave intensities begin to deviate albeit

116

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Figure 5.3: Time evolution of the spectral wave energy density |Ex (t, kx )|2 normalized by
the thermal energy of the background electrons ne me Vte2 .

only slightly. Specically, the theoretical dispersion relations can be described as parabolas,
whereas the excited wave intensities can be characterized more or less as short straight lines.
This is not too surprising, since the theoretical curves are the result of an approximate analysis.

5.4.2

Time evolution of each harmonic mode

Figure 5.3 shows the time evolution of the spectral wave energy density |Ex (t, kx )|2 normalized
with respect to the same thermal energy density of the background electrons as before. According to Figure 5.3, appreciable enhancements of the electric eld energy occur in the vicinities of
wave numbers corresponding to k1 Vd /e = 1.2, k2 Vd /e = 2.4, k3 Vd /e = 3.6, k4 Vd /e = 4.7,
k5 Vd /e = 5.9, k6 Vd /e = 7.1, and so on. When compared with Figure 5.2, one may note that
these values correspond to the central wave numbers for L1, L2, L3, . . . modes. Consequently,
k1 , k2 , k3 , . . . correspond to the wave number at which the peak intensities of L1, L2, L3,
. . . modes take place.
It should be noted that, at rst sight, the emergence of enhanced harmonic mode structure
does not seem to take place until the system has evolved to e t 200, or so. However, this is
only because the gray-scale plot does not show very small intensities clearly. The subsequent
gure will show that the harmonic modes do indeed emerge very early on, in agreement with

5.4. SIMULATION RESULT

117

Figure 5.4: Time histories of electric eld energies for the rst 6 harmonic modes,
|Ex (kn )|2 (n = 1, 2, 6), where k1 Vd /e = 1.2, k2 Vd /e = 2.4, k3 Vd /e = 3.6,
k4 Vd /e = 4.7, k5 Vd /e = 5.9, and k6 Vd /e = 7.1. The normalization is as before,
i.e., with respect to the thermal energy of the background electrons, ne me Vte2 . Straight
lines represent the linear growth rate for each harmonic

the result of the previous Vlasov simulations [Klimas, 1983, 1990] where it is reported that
the harmonics appear as early as a few plasma oscillation periods. In spite of the fact that
physical parameters as well as the numerical algorithms adopted in our study are distinct from
those simulations [Klimas, 1983, 1990] it is remarkable that these works are in good overall
agreement.
To understand the onset of the harmonics in more detail, we specically chose the rst
6 harmonic modes (with wave number corresponding to k1 , k2 , . . . , k6 ) and plotted the time
history of the wave energy density evolution for each mode. The result is displayed in Figure 5.4.
It can be seen that the fundamental Langmuir (n = 1) mode grows in an exponential fashion at
the start of the simulation run. The linear theoretical prediction for L1 mode evolution is shown
by the straight line. As one can see, the agreement is excellent. To improve upon the comparison
between the theory and simulation, the theoretical growth rate was computed on the basis of the
exact dispersion equation (5.39), rather than on the approximate analytical expression (5.72).
We found that although the analytical formula does a good job of an overall description of
the instability property, it nevertheless tends to overemphasize the actual growth rate, because

118

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Eq.(5.72) is obtained by using several approximations. Therefore, we have decided to resort


to the full solution of Eq.(5.39), which can be done for L1 mode. Thus, the maximum linear
growth rate for L1 mode was determined by numerically solving Eq.(5.39). The subsequent
growth rates for higher-harmonic modes were determined by simply multiplying the harmonic
mode number to the maximum growth rate of the L1 mode. The result is a good overall
agreement with the theory, although for harmonic modes corresponding to n 3 the very
early-time behavior of the modes is obscured by the sea of initial noise.
Note that for a brief period following the start of the simulation, the wave modes in kn (n 2)
initially decrease sharply. This occurs for a period of several tens of plasma oscillations, and is
consistent with the similar behavior reported by Klimas [1983, 1990]. Since the wave energy at
a given kn is shared by the fundamental Langmuir mode as well as all the higher harmonics,
the initial damping can be attributed to the Landau damping of L1 mode at those high kn
values. During the early time (e t less than 50 or so), the harmonic modes have not reached
suciently high intensity to exhibit themselves above the L1 mode noise level.
At some later time, around e t 50 or so, the rst harmonic mode (n = 2) begins to
emerge from the background of initial noise. It is dicult to determine the precise initial
level of L2 mode, since the mode emerges from the ambient background noise. However, once
the mode exhibits itself above the noise, the subsequent amplication follows the theoretical
prediction very well, as indicated by the straight line. Similar behavior can be discerned for
higher harmonics, k3 , k4 , , etc.
On the basis of Figure 5.4 alone, it is dicult to ascertain whether the onset of harmonic
modes takes place after some time has elapsed following the initial excitation of L1 mode,
or whether they are all simultaneously triggered at the outset. However, on the basis of
the excellent agreement between the theoretical quasi-linear growth rate prediction and the
dynamical behavior of the harmonic modes as seen in the simulation (Figure 5.4), it is reasonable
to theorize that the harmonic modes were all present at time t 0, but were hidden beneath
the noise, and become apparent only after some time.
The generalized weak turbulence theory, as presented by Gaelzer et al. [2003], assumes small
but nite levels of harmonic modes at the outset. Although it is not possible to compute
the initial noise level within the context of such a theory in a self-consistent manner, it is
nevertheless plausible that the presence of L1 mode, however low its intensity may be, leads to
the forced nonlinear perturbation which is rich in higher harmonic content. Such a concept was
already quantitatively demonstrated by ONeil et al. [1971, 1972] in the context of coherent
nonlinear theory. Thus, it is possible that a similar mechanism will provide the necessary
seed perturbation level in the case of broadband turbulence theory, in close analogy with the
coherent nonlinear theory [ONeil et al., 1971, 1972], although no such theory is available at
the present time. In any case, it might be possible to explain the existence of the harmonic
content at the outset naturally, by such a higher-order nonlinear theory.

5.4. SIMULATION RESULT

Figure 5.5: The velocity space distribution functions

Nx1

dx f (x, vx ) at

119

Figure 5.6: Continuation of Figure


5.5 for longer time periods.

dierent times and the corresponding


x vx phase plots.

5.4.3

Time evolution of phase space

We now show the time evolution of the distribution function in Figures 5.5 and 5.6. The set of


panels on the left show the average velocity distribution functions Nx1 dx f (x, vx ) at dierent
times, while the panels on the right show the corresponding x vx phase-space diagrams. At
e t = 204.8, one can observe that the phase-space structure is modulated with the characteristic
wavelength corresponding to the dominant k1 wave mode. The ne structure, already evident at
e t = 307.2, indicates the excitation of smaller-scale modes, i.e., the harmonics. At e t = 409.6,
one can clearly see that nonlinear vortices undergo multiple folding upon itself. The largestscale vortices are, of course, those corresponding to the dominant k1 mode, but within the
major vortices, one can discern the emergence of ne structure, which was already apparent
at earlier times, but has now become obvious. Again, the existence of these ne structure is
intimately related to the harmonic modes, as already noted.
In Figure 5.6 we continue with the display of phase-space evolution for longer time periods.
From the phase-space plots shown in Figure 5.6, it can be said that the system has now entered
a stage which can be characterized as purely turbulent. The time evolution of the averaged
velocity space distribution is in agreement with the customary quasi-linear plateau formation.

120

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

According to Figures 5.5 and 5.6, the full quasi-linear saturation of the beam-plasma instability
is achieved near the end of the simulation run, namely, at e t = 819.2.

5.4.4

Power-low spectrum

The potential signicance of this process in understanding the turbulence in plasmas has been
already noted in early studied on harmonic Langmuir wave generation [Apel, 1967, 1969; ONeil,
1971]. As we will show below, the spectrum of harmonic Langmuir wave intensities form a quasi
power-law feature in the saturation stage. One of the ways to characterize a turbulence (in any
medium) is to consider the quasi-stationary power-law spectrum. The reason why the power-law
spectrum is of interest is that such a spectrum implies a quasi scale-free structure associated
with the turbulence. A good example of the scale-free structure is the well-known fractals.
The power-law index characterizes the scaling property of the turbulence. For this reason, to
obtain a scale-invariant Langmuir turbulence theory has been an ongoing eort [Gaelzer et al.,
2003]. According to our theory and simulation, however, the power-law spectrum seems to be
a natural byproduct of harmonic generation.
In Figure 5.7 we plot the normalized wave electric eld energy |Ex (kx )|2 /ne me Vte2 versus
normalized wavenumber kx Vd /e . The wave spectrum is plotted at intervals of e t = 204.8,
307.2, 409.6, and 819.2. The straight line indicates the power-law prole associated with initial
wave intensity, which is given by a formula equivalent to Eq.(5.78). At e t = 204.8, the
enhancement of k1 mode is visible, while other wave modes are also seen to be emerging from
the noise level. At e t = 307.2, the enhancements of the harmonic modes at k2 , k3 , etc., are
more clearly visible, at least up to n = 5. Note that the wave spectrum at this time can already
be characterized by a power-law. At e t = 409.6, the individual peaks are not so apparent
because of the overlap of harmonic spectra. Moreover, the spectral shape deviates somewhat
from the pure power-law. However, this is only a transient behavior, and at the nal time,
e t = 819.2, it can be seen that the spectrum has settled down to a quasi power-law with index
5 to 6.
Note that the fundamental mode, L1, is characterized by a clear peak which does not fall along
the simple power-law prole. We have thus excluded this L1-mode peak in the interpretation
of the power-law distribution with an index equal to 5 to 6. We believe that the
enhanced peak associated with the fundamental L1 mode is owing to the coherent particle
trapping saturation mechanism, which tends to lead to higher saturation level than predicted
by simple quasi-linear theory.
In the present simulations, we found that the high-end of the kx spectrum, namely, those
spectral range corresponding to kx Ve /e > 10, or so, exhibits a slight steepening of the powerlaw distribution. This occurred in all our simulation runs regardless of the initial conditions.
Thus, we believe that it is owing to the grid-scale dissipation in the numerical scheme. This

5.4. SIMULATION RESULT

121

Figure 5.7: Spectra of the normalized wave electric eld energy |Ex (kx )|2 /ne me Vte2 versus
kx Vd /e , at e t = 204.8, 307.2, 409.6, and 819.2. The dashed line represents the initial
prole of the wave intensity. Note that the nal spectrum is roughly characterized by the
initial power-law index of 5.

behavior is not shown in Figure 5.7, since we restricted ourselves to the range, 101 < kx Ve /e <
10. However, in the next example, such a behavior will be clearly shown. In any case, it is
interesting to note that such a feature is often present in many turbulent systems, such as in
neutral uid turbulence, where it is known there as the dissipation range behavior. It is also
interesting to note that in many real turbulent systems, the spectra often possess a prominent
peak corresponding to low-frequency and large-scale coherent structure. The spectrum of the
harmonic Langmuir wave possesses these features.
The overall power-law index of roughly 5 to 6 is also seen in the early beam-plasma
experiment by Apel [1967, 1969]. In the present simulation such a power-law index emerges
quite naturally. One might note, however, that the initial wave spectrum which we implemented
at the start is also of a power-law type with the similar power-law index close to 5 or 6.
Thus, one might naturally wonder whether the nal result somehow reects what was imposed
at the outset. In order to see how robust the nal outcome is, we have performed several runs

122

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Figure 5.8: This gure shows the case of coherent initial perturbation with a single wave
mode at kx = k1 . (a)Time evolution of spectral wave energy density |Ex (t, kx )|2 versus
time e t. (b)Time history of rst six harmonics mode intensities. (c)The evolution of
spectral wave intensity |Ex (kx )|2 versus kx , at e t = 0.0 and 819.2. (d)x vx phase-space
diagram at e t = 409.6.

with dierent proles associated with the initial noise, including dierent power-law indices,
white noise, and a monochromatic perturbation. The end result invariably seems to be a quasi
power-law with an approximate 5 to 6 index, although the detailed structure of the nal
spectrum (at e t = 819.2) is dierent from run to run.

5.4.5

Generation of harmonics from monochromatic perturbation

As a concrete demonstration of this nding, we show in Figure 5.8, the case of a monochromatic
perturbation. In this run, we initiated the instability by imposing a sinusoidal perturbation with
a single wave mode with kx = k1 . Panel (a) shows the time evolution of spectral wave energy
density |Ex (t, kx )|2 versus time e t, in the same format as Figure 5.3. Panel (b) displays the
time history of rst six harmonics mode intensities in the same format as Figure 5.4. In panel

5.4. SIMULATION RESULT

123

(c) the plot of wave spectrum |Ex (kx )|2 is shown versus kx which corresponds to e t = 819.2.
The initial wave spectrum is also shown for comparison. Note that although at t = 0, there
should only be a single mode at kx = k1 , the small but nite ( 1035 ) noise level is owing to
the roundo error inherent with any digital computation. We also plotted an xvx phase-space
diagram at e t = 409.6 in panel (d).
Note from Figure 5.8b, that because there are practically no wave power at wavenumbers
other than kx = k1 , the sea of initial perturbation which obscures the very early-time behavior associated with the harmonic modes is absent in the present case. As a result, one can
clearly identify each harmonic mode as they begin to grow from small noise level. As straight
lines which represent the linear theory demonstrate, the very early-time behavior of harmonic
growths is in excellent agreement with the theory. However, since the coherent initial input is
in violation of the assumption of quasi-linear theory which is applicable only for a broadband
spectrum, the subsequent dynamics is very dierent from the prediction of the simple quasilinear theory. Indeed, as panel (b) shows, the intensities associated with the harmonics show
amplitude modulations, which must be due to coherent nonlinear physical eects. Such a modulation was also present to some degree in the case of power-law initial uctuations, as Figure
5.4 shows. However, the present Figure 5.8b demonstrates much more prominent amplitude
modulations.
The nal spectrum shown in Figure 5.8c is of course very dierent when compared with Figure
5.7, since Figure 5.8c is comprised of discrete multiple peaks. This shows that the system retains
the memory of the initial condition. However, the striking fact is that if we connect the peaks of
the individual spikes (minus the enhanced fundamental peak), we again obtain a rough estimate
of power-law-like dependence with index 5 or 6 in the range 100 < kx Vd /e < 101 .
It is noteworthy that this run corresponds to a coherent initial input, and yet, it leads
to a qualitatively similar result in comparison with the case of broadband incoherent initial
noise. Consequently, the present example may shed some light on the alternative mechanism of
harmonic generation, i.e., the coherent nonlinear theory rst proposed in the early 1970s. In our
simulation, it is found that in the relatively early phase of the simulation run (e t = 0 100),
when the harmonic growth is still largely dictated by linear theory, the spectrum of the electric
eld is characterized by a steep power-law with an approximate 15 index. However, in the nal
stage, the saturated spectrum features 5 to 6 index. To understand the time evolution of
spectral index, we focus on the nonlinear trapping dynamics. Note that although the primary
L1 mode grows according to the linear theory, it is clear that in the present example, the
saturation is achieved by nonlinear trapping, which occurs around e t = 400. The phase-space
diagram in Figure 5.8d clearly shows this, which leads to the diusion of the electron beam
over the velocity range specied by
L1 (k1 )
L1 (k1 )
VT < vx <
+ VT ,
k1
k1

(5.79)

124

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

where VT = 2T /k1 and T is given by


/

T =

k1 |qe ||Ex (k1 )|


.
me

(5.80)

Coincident with the trapping, around e t 200 and beyond, it is seen that the harmonic mode
amplitudes rapidly increase with the eective amplication rates well above those predicted by
quasi-linear theory (this behavior is also seen in the case of incoherent run, Figure 5.4, to some
extent). A possible mechanism for such a rapid growth of the harmonics beyond e t 200 is
the eects due to the nonlinear trapping process, which creates a phase-space vortex rich in
harmonic content. As a consequence, the modes with wavenumbers, 2k1 , 3k1 , 4k1 , 5k1 , and so
on, experience additional enhanced amplication, thus yielding the eventual power-law index
of 5 6.

5.5

Summary and Discussion

The results we have discussed thus far indicate the importance of the harmonic generation
as a new way of approaching the Langmuir turbulence. The present Vlasov simulation of
multiple harmonic Langmuir modes (up to 15th harmonics), generalizes the previously available
simulations which were restricted to the second harmonic only. The frequency-wavenumber
spectrum obtained by taking the Fourier transformation of electric eld data both in time
and space shows an excellent agreement with the theoretical nonlinear dispersion relations for
harmonic Langmuir waves. The quasi power-law spectrum associated with the saturated wave
energy shows that it is natural for the Langmuir turbulence involving the harmonics to form a
power-law distribution in wave energy spectrum. According to the theory generalized by Yoon
et al. [2003] and Gaelzer et al. [2003], the evolution equation for the spectral wave energy
density for a harmonic Langmuir mode, In ILn (k) = |ELn (k)|2 , is governed by an equation
of the form

In
(5.81)
= n In .
t
As Figure 5.4 shows, the theoretical amplication rate agrees well with the simulation, at least
during the early phase of the evolution. Thus, the above equation appears to describe the time
evolution of the harmonic modes quite satisfactorily in an overall sense.
However, the above equation requires a nite level of harmonics at the outset, but such a
level becomes a free parameter in the weak turbulence theory [Yoon et al., 2003; Gaelzer et
al., 2003]. Moreover, the nal outcome of the solution highly depends on the initial condition,
since the equation is essentially linear. For this reason, it is needed to manipulate the initial
condition very carefully in order to reproduce the saturation wave spectrum which is consistent
with the laboratory experiment [Apel, 1967, 1969] as well as with the active rocket experiment
[Kellogg et al., 1986].

5.5. SUMMARY AND DISCUSSION

125

In the simulation, on the other hand, regardless of the form of the initial wave level, the
system seems to readjusts itself in two crucial ways. First, it damps out the unwanted noises
early on in the simulation such that, even if we start the run with a wrong form of initial
wave spectrum, the system self-regulates to create the true modes. This is evident from the
initial rapid Landau damping of the modes seen in Figure 5.4, for instance. Then, once the
true modes are generated at relatively early time, the higher harmonics grow with higher linear
growth rates through some mode-couplings, (as evidenced by the amplitude modulations seen in
Figure 5.8), which the incoherent turbulence theory may not be able to explain, if they involve
coherent nonlinear dynamics. Even with a single mode perturbation, the beam-plasma system
nally produces a quasi power-law spectrum with an apparently universal spectral index of 5
to 6.
Therefore, one of the issues from the standpoint of theory is to determine the initial level
of the harmonic uctuations in a self-consistent manner, or equivalently, by including coherent
theory. We also have to include the mode-coupling terms on the right-hand, which is ignored
in Eq.(5.71), so as to limit the unchecked growth of the high-harmonic modes at the nonlinear
stage (as seen in Figure 5.4),

In
= n In +
an,m In Im .
t
m

(5.82)

The physical origin of the source can be attributed to either the generalized spontaneous uctuations, or higher-order nonlinear correction terms which may lead to the forced nonlinear
perturbation rich in harmonic content. The mode-coupling terms may involve particle trapping
and coherent nonlinear dynamics, which the generalized weak turbulence theory cannot handle
yet. However, the theory and simulation show remarkable agreements on many aspects, such
as the k diagram obtained from the simulation data and theoretical dispersion curves, and
the initial growth property. Although the incoherent turbulence theory excludes the trapping
eects at the outset, it is remarkable that the prediction of the nonlinear dispersion relation
for the harmonics and the early-time growth behavior are in remarkable agreement with the
simulation, even for a coherent initial input.

126

CHAPTER 5. THEORY AND SIMULATION OF HARMONIC LANGMUIR WAVES

Chapter 6
Concluding Remarks
In this thesis, we have studied coherent and incoherent nonlinear processes of electron beam
instabilities by performing computer simulations. One of the most important coherent nonlinear
processes of beam-plasma interactions in space plasmas is formation of electrostatic solitary
waves observed by the GEOTAIL spacecraft and other recent spacecraft. Incoherent nonlinear
processes are observed as harmonics Langmuir waves.
To obtain better understanding of nonlinear beam-plasma interactions in space plasmas, we
have developed a new particle code in Chapter 2. By applying the second-order shape factor,
the energy conservation is improved substantially. In the present study, we also developed
two new numerical schemes. One is charge conservation method for solving the continuity
equation for charge without solving Poissons equation at every time step. The computation
speed of the present scheme is faster than the conventional charge conservation method. The
other numerical scheme is an improved masking method for open boundary conditions. The
improved scheme can reduce the size of damping regions substantially. By using these new
schemes, computation eciency is much increased.
In Chapter 3 we studied formation processes of one- and two-dimensional electrostatic solitary
waves/electron holes in realistic simulation models with open boundaries. We injected a weak
electron beam from a boundary of the simulation system into the background plasma to induce
the electron bump-on-tail instability. The instability develops in space as the electron beam
propagates along the ambient magnetic eld. Since electron beam modes have the maximum
growth rate in the direction parallel to the ambient magnetic eld, we rst found formation
of one-dimensional electron holes. In a timescale of electron dynamics, oblique electron beam
modes also grow in a region close to the source of the electron beam. Electron holes initially
uniform in the y direction are twisted through modulation by the oblique beam modes. As the
two-dimensional electron holes propagate along the ambient magnetic eld, they are aligned
in the perpendicular direction through coalescence. Electron holes become one-dimensional in
regions far from the source of the electron beam. In a timescale of ion dynamics, we found
generation of lower hybrid waves in the generation region of the electron holes. Electron holes
are accompanied by perpendicular electric elds through modulation by the lower hybrid waves.
We found that the modulated electron holes have unipolar waveforms of the perpendicular
127

128

CHAPTER 6. CONCLUDING REMARKS

electric eld, as observed by the recent spacecraft, GEOTAIL, POLAR, FAST and WIND.
The present simulations were performed under the condition for stability of electron holes,
e e > b , in which coupling of electron holes with lower hybrid waves becomes dominant.
We need to perform further simulations with dierent amplitudes of the ambient magnetic eld
or dierent temperatures of the background electrons to study coupling of electron holes with
other oblique modes such as whistler waves.
In Chapter 4 we studied electromagnetic signature associated with electrostatic solitary
waves. Spatial and temporal development of electrostatic potentials is almost the same as
that in the previous electrostatic simulation, while we found clear enhancement of magnetic
eld components around two-dimensional electron holes. At the perpendicular edge of twodimensional electron holes, electrons undergo E B drift, where the electric eld is a perpendicular electrostatic eld of the electron holes and the magnetic eld is the ambient magnetic
eld. We found that a current is induced by the electron E B drift, and that magnetic elds
are enhanced around the current. Therefore, one-dimensional electron holes without perpendicular electric elds do not have any electromagnetic signatures. We also estimated the amplitude
of the magnetic eld based on the present simulation result. The amplitude ratio of the electric
eld to the magnetic eld of the present analysis is in agreement with those of the POLAR
and FAST observations. The current structures propagate along the ambient magnetic eld,
because electrons undergoing E B0 drift are carried by electron holes in the direction parallel
to the ambient magnetic eld. We conrmed that the moving current perturbs the background
plasma to excite electromagnetic waves. We need to perform further computer simulations
and theoretical analyses to investigate conversion eciency of electrostatic to electromagnetic
energy.
In Chapter 5 the generation of harmonics Langmuir waves has been studied rst by nonlinear
weak turbulence theory and then by Vlasov simulations. We briey reviewed generalized weak
turbulence theory, and derived nonlinear dispersion relations of harmonics Langmuir waves
by solving the Vlasov-Poisson equations including nonlinear wave-wave coupling terms. To
conrm the validity of the nonlinear theory we then performed one-dimensional electrostatic
Vlasov simulations of a weak-beam instability with dierent types of initial wave spectra. We
rst initiated the beam instability with a power-law wave spectrum. In this run we found the
generation of multiple harmonic Langmuir modes up to 15th harmonics during the beam-plasma
interaction. Frequency-wavenumber spectrum computed by performing Fourier transformation
of simulated electric eld in space and time shows an excellent agreement with the theoretical
nonlinear dispersion relations of harmonic Langmuir waves. Time histories of spectral energy
densities for each harmonic mode also showed that a higher harmonic mode has a larger growth
rate, which is also in a good agreement with the nonlinear theory. We next initiated the beam
instability with dierent power-law indices, white noise, and a monochromatic perturbation.
In all runs, the saturated wave spectra become power-law with a universal index of 5 6.

CHAPTER 6. CONCLUDING REMARKS

129

Although the weak turbulence theory is based on incoherent nonlinear theory, harmonics modes
are also excited from a monochromatic perturbation, which implies that coherent nonlinear
dynamics, i.e., nonlinear trapping process, as well as incoherent nonlinear dynamics plays an
important role in the generation of harmonic Langmuir waves.
In summary, we have developed two dierent kinetic simulation codes, particle code and
Vlasov code, with a higher-order scheme, and performed computer simulations of electron beam
instabilities to study both coherent and incoherent nonlinear processes. Natural phenomena of
beam-plasma interactions are highly nonlinear, and it is practically impossible to describe its
long-time evolution by analytic theories. In this thesis we successfully revealed the generation
processes of electrostatic solitary waves and harmonics Langmuir waves performing computer
simulations. The success of the present study implies that computer simulations/experiments
are now essential approaches for studies of nonlinear physics. A beam-plasma interaction is one
of the most fundamental microscopic processes in plasmas, and is studied by kinetic simulations.
On the other hand, macroscopic and mesoscopic processes, such as magnetic reconnection and
shocks, are studied with uid codes, in which electron kinetics is neglected by treating electrons
as a uid. However, the large-scale processes result in geomagnetic eld variation, and such a
change in the geomagnetic eld produces induction electric elds that can accelerate charged
particles to generate electron and ion beams. This suggests that we need to study macroscopic
and mesoscopic processes with kinetic simulation codes, because beam-plasma interactions play
important roles in the large-scale processes. Rapid advancement of supercomputer technology
will enable us to perform large-scale kinetic simulations. Now we are developing a prototype of
the Space Simulator using Earth Simulator which is currently the fastest supercomputer in the
world. The Space Simulator can be regarded as a virtual magnetosphere or a numerical chamber
in which we can reproduce various physical processes in the Earths magnetosphere or virtually
perform space experiments. The Space Simulator will be developed with supercomputers in a
future generation to forecast the geospace environment. The particle code and Vlasov code used
in the present study will be parts of the system. We hope that the present study of microscopic
instabilities would be a rst step toward the development of the Space Simulator. As a next
step we will perform computer simulations with non-uniform background plasmas, and then
study coupling of microscopic processes with macro or mesoscopic processes by performing
large-scale kinetic simulations. We also hope that numerical techniques for kinetic simulations
of collisionless plasmas presented in this thesis provide a useful basis for particle codes and
Vlasov codes.

130

CHAPTER 6. CONCLUDING REMARKS

Appendix A
Numerical Techniques for Vlasov Simulations
A.1
A.1.1

Constrained Interpolation Prole (CIP) Scheme


Basic principle of the CIP scheme

In this section the basic principle of the CIP scheme is briey reviewed by using a simple
one-dimensional hyperbolic equation,
f
f
+v
= 0.
(A.1)
t
x
If the velocity v is constant, the theoretical solution of Eq.(A.1) is obtained by shifting a prole
as f (x, t + t) = f (x vt, t). Similarly, even if the velocity is not constant, an approximate
solution is obtained by assuming that t is small enough, as
f (xi , t + t)  f (xi vt, t)

(A.2)

where xi represents the spatial grid point. By assuming that velocity v is independent of x,
dierential of Eq.(A.1) is given by
(x f )
(x f )
+v
=0
t
x
where x f f /x. It is noted that the Vlasov equation falls into this condition.

(A.3)

If two values f and x f are given at two neighboring grid points, the prole between the two
points can be described by a cubic polynomial function such as
F (x) = A1i (x xi )3 + A2i (x xi )2 + A3i (x xi ) + A4i ,

(A.4)

and dierential of Eq.(A.4) is given by


F (x)/x = 3A1i (x xi )2 + 2A2i (x xi ) + A3i .

(A.5)

When xxi 0, A1i , A2i , A3i and A4i are determined from the given values of these functions
at both ends,
F (xi ) = A4i = fi ,

(A.6)

F (xi )/x = A3i = x fi ,


F (xi+1 ) = x3 A1i + x2 A2i + xA3i + A4i = fi+1 ,
F (xi+1 )/x = 3x2 A1i + 2xA2i + A3i = x fi+1 ,
131

(A.7)
(A.8)
(A.9)

132

APPENDIX A. NUMERICAL TECHNIQUES FOR VLASOV SIMULATIONS

where xi+1 xi = x. The value A1i is given by (A.9)x -(A.8)2 as


A1i =

x fi + x fi+1
fi fi+1
+2
.
2
x
x3

(A.10)

The value A2i is given by (A.8)3(A.9)x as


A2i =

fi fi+1
2x fi + x fi+1
3
.
x
x2

(A.11)

Similarly when xxi < 0, the values of the functions at both ends are given by Eqs.(A.6),(A.7)
and
F (xi1 ) = x3 A1i + x2 A2i xA3i + A4i = fi1 ,
F (xi1 )/x = 3x2 A1i 2xA2i + A3i = x fi1 .

(A.12)
(A.13)

The value A1i is given by (A.12)2+(A.13)x as


A1i =

x fi + x fi1
fi fi1
2
.
2
x
x3

(A.14)

The value A2i is given by (A.12)3+(A.13)x as


A2i =

2x fi + x fi1
fi fi1
.
3
x
x2

(A.15)

As a result of the above simultaneous equations, the propagating prole is given by


fit+t F (xi + ) = A1i 3 + A2i 2 + x fit + fit ,

(A.16)

x fit+t (F/x)x=xi + = 3A1i 2 + 2A2i + x fit ,

(A.17)

where vt. The values Ai1 and Ai2 are summarized as follows:
x fit + x fist
fit fist
+
sign()

2
,
x2
x3
fit fist
2x fit + x fist
3
= sign()
,
x
x2

A1i =

(A.18)

A2i

(A.19)

where is = i + sign() and sign() = 1 returns sign of .

A.1.2

Extension to higher-dimensions

In a two-dimensional case

f
f
f
+ vx
+ vy
=0
t
x
y

(A.20)

where vx and vy are respectively velocities in the x and y directions, the solution to Eq.(A.20)
splits into the following fractional steps:
f  (xi , yj )

= f t (xi vx t, yj ),

f t+t (xi , yj ) = f  (xi , yj vy t).

(A.21)

A.2. ONE-DIMENSIONAL VLASOV-POISSON SOLVER

133

Here we have to pay attention to the solution, because the CIP scheme needs information on
the spatial derivatives. To solve the second equation of (A.21), we need f  and y f  . However,
y f  is yet unknown because Eqs.(A.16)-(A.19) give f  and x f  in solving the rst equation of
(A.21). To estimate y f  the following equation is solved additionally,

(y f ) = (vx x f ).
t
y

(A.22)

There are several methods to solve this equation. In the present study we applied the centered
dierence scheme as follows:

t+ t
t
2
vx,(i,j+ 1 ) x f(i,j+ 1 ) vx,(i,j 1 ) x f(i,j 1 )
(A.23)
2
2
2
2
x

t+ t
t
2
t

vx,(i,j+1) x f(i,j+1) vx,(i,j1) x f(i,j1)


= y f(i,j)
2x


t 
t

t

t
= y f(i,j)

vx,(i,j+1) x f(i,j+1)
+ x f(i,j+1)
vx,(i,j1) x f(i,j1)
+ x f(i,j1)
.
4x


t
= y f(i,j)

y f(i,j)

Therefore in the second equation of (A.21) the following equation is solved additionally,

t+ t
t
2
vy,(i+ 1 ,j) y f(i+ 1 ,j) vy,(i 1 ,j)y f(i 1 ,j)
(A.24)
2
2
2
2
y


t 
t+t
t+t



= x f(i,j)

vy,(i+1,j) y f(i+1,j)
+ y f(i+1,j)
vy,(i1,j) y f(i1,j)
+ y f(i1,j)
.
4y

t+t

= x f(i,j)

x f(i,j)

In a three-dimensional case
f
f
f
f
+ vx
+ vy
+ vz
=0
t
x
y
z

(A.25)

where vz is velocity in the z direction, the solution to Eq.(A.25) similarly splits into the following
three steps:
f  (xi , yj , zk )

= f t (xi vx t, yj , zk ),

f  (xi , yj , zk )

= f  (xi , yj vy t, zk ),

(A.26)

f t+t (xi , yj , zk ) = f  (xi , yj , zk vz t).

A.2
A.2.1

One-dimensional Vlasov-Poisson Solver


Basic equations

In one-dimensions the Vlasov equation without static magnetic eld is given by


qs
fs
fs
fs
+ vx
+
Ex
= 0,
t
x
ms vx

(A.27)

where the subscript s represents particle species (e.g., electron and ion). The prole of electric
eld Ex is given by solving Poissons equation,

1 
Ex (x)
=
qs
fs (x, vx )dvx .
x
0 s

(A.28)

134

APPENDIX A. NUMERICAL TECHNIQUES FOR VLASOV SIMULATIONS

The velocity distribution functions fs are dened at grid points in the two-dimensional x
vx phase space with Nx Nvx ,s . We can dene the velocity grid spacing vs for each particle
species as

vmax,s vmin,s
,
(A.29)
Nvx ,s 1
are respectively upper and lower cuto velocities, and f (vmax,s ) =
vs =

where vmax,s and vmin,s

f (vmin,s )  f (). Thus the velocity grid points vx,(s,j) are dened for each particle species.
In the present study we specify plasma frequency of s , charge-to-mass ratio qs /ms drift velocity Vd,s and thermal velocity Vt,s of each species. The normalized initial velocity distribution
function of each species gs (vx,(s,j) ) g(s,j) is given by
g(s,j) =

(vx,(s,j) Vd,s )2
1
exp
.
2
2Vt,s
2Vt,s

(A.30)

Therefore we initiate fs (xi , v(s,j) ) f(s,i,j) as


f(s,i,j) = i g(s,j),

(A.31)

where i represents initial perturbation in the x direction. It is noted that we have to give
an initial perturbation that seeds an instability. If there is no initial perturbation, the
distribution function becomes uniform in the x direction and the electric eld becomes Ex = 0
at all times. In other words, no instability takes place in the system. To apply the CIP scheme
we also give derivatives of the initial distribution function, x f and vx f as
x f(s,i,j) = x i g(s,j),
vx,(s,j) Vd,s
f(s,i,j).
vx f(s,i,j) =
2
Vt,s

(A.32)
(A.33)

In the present code the initial perturbation i and its derivative x i are given by
Nx /2

i = 1 +
i =

A(km ) cos(km xi + m ),

(A.34)

km A(km ) sin(km xi + m ).

(A.35)

m=1
Nx /2

m=1

Here we assume that the amplitude A(km ) is small enough ( 1).

A.2.2

Splitting method

In the present study we adopted the time-advance algorithm called splitting method proposed
by Cheng and Knorr [1976]. The Vlasov equation splits into the following three steps, and the
numerical procedure is represented by
fs (xi , vx,(s,j)) = fst (xi vx,(s,j)t/2, vx,(s,j)),
qs t+ t
fs (xi , vx,(s,j)) = fs (xi , vx,(s,j)
E 2 t),
ms x,i
fst+t (xi , vx,(s,j)) = fs (xi vx,(s,j)t/2, vx,(s,j)).

(A.36)
(A.37)
(A.38)

A.2. ONE-DIMENSIONAL VLASOV-POISSON SOLVER

135

With the CIP scheme the computation of the rst step (A.36) and third step (A.38) are
respectively rewritten as
t
t
t
t
A1(s,i,j) = (x f(s,i,j)
+ x f(s,is,j)
) sign(
vx,(s,j)) 2(f(s,i,j)
f(s,is,j)
),

(A.39)

t
t
t
t
vx,(s,j)) (2x f(s,i,j)
+ x f(s,is,j)
) 3(f(s,i,j)
f(s,is,j)
),
A2(s,i,j) = sign(

(A.40)


3
2
t
t
= A1(s,i,j) vx,(s,j)
+ A2(s,i,j) vx,(s,j)
x f(s,i,j)
vx,(s,j) + f(s,i,j)
,
f(s,i,j)

(A.41)


2
t
= 3A1(s,i,j) vx,(s,j)
2A2(s,i,j)vx,(s,j) + x f(s,i,j)
,
x f(s,i,j)

(A.42)


t
vx f(s,i,j)
= vx f(s,i,j)
(A.43)


1
t

t

vx,(s,j+1) x f(s,i,j+1)
+ x f(s,i,j+1)
vx,(s,j1) x f(s,i,j1)
+ x f(s,i,j1)
,
4

and




C1(s,i,j) = (x f(s,i,j)
+ x f(s,is,j)
) sign(
vx,(s,j)) 2(f(s,i,j)
f(s,is,j)
),

(A.44)





vx,(s,j)) (2x f(s,i,j)
+ x f(s,is,j)
) 3(f(s,i,j)
f(s,is,j)
),
C2(s,i,j) = sign(

(A.45)

t+t
3
2


= C1(s,i,j) vx,(s,j)
+ C2(s,i,j)vx,(s,j)
x f(s,i,j)
vx,(s,j) + f(s,i,j)
,
f(s,i,j)

(A.46)

t+t
2

= 3A1(s,i,j) vx,(s,j)
2A2(s,i,j)vx,j + x f(s,i,j)
,
x f(s,i,j)

(A.47)

t+t

= vx f(s,i,j)
,
(A.48)
vx f(s,i,j)


1

t+t

t+t
+ x f(s,i,j+1)
vx,(s,j1) x f(s,i,j1)
+ x f(s,i,j1)
,
vx,(s,j+1) x f(s,i,j+1)
4

where is = i sign(
vx,(s,j)). Here we rescaled x f and vx f by x fs x fs x and vx fs
vx fs vs , respectively, and vx,(s,j) vx,(s,j)t/2x to reduce the number of computation.
Similarly The second step of the splitting method (A.37) is rewritten as




B1(s,i,j) = (x f(s,i,j)
+ x f(s,i,js)
) + sign(Ex,(s,i) ) 2(f(s,i,j)
f(s,i,js)
),

(A.49)





+ x f(s,i,js)
) 3(f(s,i,j)
f(s,i,js)
),
B2(s,i,j) = sign(Ex,(s,i) ) (2x f(s,i,j)

(A.50)


3
2


x,(s,i)
Ex,(s,i) + f(s,i,j)
= B1(s,i,j) Ex,(s,i)
+ B2(s,i,j) E
+ vx f(s,i,j)
,
f(s,i,j)


2

= 3B1(s,i,j) E
vx f(s,i,j)
x,(s,i) + 2B2(s,i,j) Ex,(s,i) + x f(s,i,j) ,

(A.51)
(A.52)



= x f(s,i,j)
(A.53)
x f(s,i,j)



1





+ E
,
x,(s,i+1) vx f(s,i+1,j) + vx f(s,i+1,j) Ex,(s,i1) vx f(s,i1,j) + vx f(s,i1,j)
4

where js = j + sign(Ex,(s,i) ). To reduce the number of computation, we rescaled Ex by Ex,(s,i)


qs /ms Ex,i t/vs .

A.2.3

Poissons equation

To solve Poissons equation, we rst compute spatial prole of charge density at time t + t
.
2
However, as shown in Eqs.(A.36)-(A.36), the distribution function at t + t
is not given. In the
2

136

APPENDIX A. NUMERICAL TECHNIQUES FOR VLASOV SIMULATIONS

is computed from the the distribution function


splitting method, the charge density at t + t
2
f  as
t+ t
2

0


s


 qs ns,i
2s

vs
f(s,i,j)
=
.
qs /ms
0
s
j

(A.54)

In the present code both electric eld Ex and charge density are dened at full-integer grids
i = 1, 2, ..., Nx . Therefore electric eld is computed from with the following procedure,
Ex,i = Ex,i1 +

A.2.4

i + i1
x.
20

(A.55)

Renormalization

As described in Section 2.8, we also renormalize input parameters (R-unit system) to quantities
in the simulation code (S-unit system) to reduce the number of operations. We normalize
distance and time by r and t , respectively. By assuming that x = rR and t = tR , all
2

physical quantities are renormalized as follows:

A.2.5

distance

xS =

time

tS =

velocity

vS =

electric eld

ES =

charge density

S =

energy density

S =

1
x
x
R
1
t
t R

t
v
x
R
t2
E
x
R
t2 R
t4

x
2 R

Boundary conditions

The distribution functions at both ends of the system fi=0 and fi=Nx +1 are dened (computed)
depending on boundary conditions. On the other hand, values of the distribution functions at
the boundaries in the velocity direction are given by fj=0 = fj=1 , fj=Nvx +1 = fj=Nvx , because
we assume f (vmax,s ) = f (vmin,s )  f ()  0.
Periodic boundary
In periodic systems, values of the distribution function at the boundaries are given by
fi=0 = fi=Nx ,
fi=Nx +1 = fi=1 .

(A.56)
(A.57)

Open boundary
When vx is positive, the distribution function at the left boundary is given from Eqs.(A.16)(A.19)
by setting i = Nx + 1 and is = Nx . When vx is negative, the distribution function at the right
boundary is similarly given form Eqs.(A.16)(A.19) by setting i = 0 and is = 1. We note

A.2. ONE-DIMENSIONAL VLASOV-POISSON SOLVER

137

that the propagating proles are completely absorbed at the boundaries without any physical
reection. Since the hyperbolic equation (A.1) is a one-way wave equation, computed values at
the boundaries satisfy absorbing boundary conditions such as Murs method [1981] and Lindmans method [1975]. On the other hand, we have to input the distribution function at the left
boundary for vx < 0 and and that at the right boundary for vx > 0.
Inow boundary. We can input arbitrary velocity distribution functions with vx > 0 at
x = 0, and with vx < 0 at x = Nx x. The inow boundary condition is used to inject plasma
ow such as electron and/or ion beams from the boundary.
Outow boundary. Velocity distribution functions evolve in time, and the initial distribution function is perturbed. Thus the distribution functions at the open boundaries are also
perturbed by physical phenomena in the system. To give feedback of the perturbation in the
system to distribution functions inputed at the boundaries, we assume
fi=0,Nx +1
 0,
x

(A.58)

(x fi=0,Nx +1 )  0.
x

(A.59)

One numerical scheme is described below,

fib =
x fib =

fi=1

at ib = 0

fi=Nx

at ib = Nx + 1

x fi=1

at ib = 0

x fi=Nx

at ib = Nx + 1

(A.60)
.

(A.61)

Another scheme is described by Newman et al. [2001]. A velocity distribution function at the
outow boundary is adjusted to equal the distribution function in a narrow region near the
boundary.

138

APPENDIX A. NUMERICAL TECHNIQUES FOR VLASOV SIMULATIONS

References
[1] Apel, J. R., Harmonic generation and turbulencelike spectrum in a beam-plasma interaction,
Phys. Rev. Lett., 19, 744-747, 1967.
[2] Apel, J. R., Nonlinear eects and turbulent behavior in a beam-plasma instability, Phys.
Fluids, 12, 640-648, 1969.
[3] Bale, S. D., P. J. Kellogg, D. E. Larson, R. P. Lin, K. Goetz, and R. P. Lepping, Bipolar
electrostatic structures in the shock transition region: Evidence of electron phase space
holes, Geophys. Res. Lett., 25, 2929-2932, 1998.
[4] Berk, H. L., C. E. Nielsen, and K. V. Roberts, Phase space hydrodynamics of equivalent
nonlinear systems: Experimental and computational observations, Phys. Fluids, 13, 980995, 1970.
[5] Berenger, J.-P., A perfectly matched layer for the absorption of electromagnetic waves, J.
Comput. Phys., 114, 185-200, 1994.
[6] Bernstein, I. B., J. M. Greene, and M. D. Kruskal, Exact nonlinear plasma oscillations,
Phys. Rev., 108, 546-550, 1957.
[7] Birdsall, C. K., and A. B. Langdon, Plasma Physics via Computer Simulation, McGraw-Hill,
New York, 1985.
[8] Buneman, O., Dissipation of currents in ionized media, Phys. Rev., 115, 503-517, 1959.
[9] Buneman, O., Inversion of the Helmholtz (or Laplace-Poisson) operator for slab geometry,
J. Comput. Phys., 12, 124-130, 1973.
[10] Buneman, O., TRISTAN: The 3-d electromagnetic particle code, in Computer Space
Plasma Physics, edited by H. Matsumoto and Y. Omura, pp.67-84, Terra Scientic, Tokyo,
1993.
[11] Cairns, I. H., New waves at multiples of the plasma frequency upstream of the Earths
bow shock, J. Geophys. Res., 91, 2975-2988, 1986.
139

REFERENCES

140

[12] Cartwright, K. L., J. P. Verboncoeur, and C. K. Birdsall, Loading and injection of


Maxwellian distributions in particle simulations, J. Comput. Phys., 162, 483-513, 2000.
[13] Cattell, C. A., J. Dombeck, J. R. Wygant, M. K. Hudson, Comparisons of Polar satellite
observations of solitary wave velocities in the plasma sheet boundary and the high altitude
cusp to those in the auroral zone, Geophys. Res. Lett., 26, 425428, 1999.
[14] Chen, L.-J., and G. K. Parks, BGK electron solitary waves in 3D magnetized plasma,
Geophys. Res. Lett., 29, 1331, doi:10.1029/2001GL013385, 2002.
[15] Cheng, C. Z., and G. Knorr, The integration of the Vlasov equation in conguration space,
J. Comput. Phys., 22, 330-351, 1976.
[16] Drake, J. F., M. Swisdak, C. Cattell, M. A. Shay, B. N. Rogers, and A. Zeiler, Formation
of electron holes and particle energization during magnetic reconnection, Science, 299, 873877, 2003.
[17] Dubouloz, N., R. Pottelette, M. Malingre, G. Holmgren, and P. A. Lindqvist, Detailed
analysis of of broadband electrostatic noise in the dayside auroral zone, J. Geophys. Res.,
96, 3565-3579, 1991.
[18] Eastwood, J. W., The virtual particle electromagnetic particle-mesh method, Comput.
Phys. Commun., 64, 252-266, 1991.
[19] Eastwood, J. W., W. Arter, N. J. Brealey and R. W. Hockney, Body-tted electromagnetic
PIC software for use on parallel computers Comput. Phys. Commun., 87, 155-178, 1995.
[20] Ergun, R. E., C. W. Carlson, J. P. McFadden, F. S. Mozer, L. Muschietti, I. Roth, and R.
Strangeway, Debye-scale plasma structures associated with magnetic-eld-aligned electric
eld, Phys. Rev. Lett., 81, 826-829, 1998.
[21] Esirkepov, T. Z., Exact charge conservation scheme for particle-in-cell simulation with an
arbitrary form-factor, Comput. Phys. Commun., 135, 144-153, 2001.
[22] Franz, J. R., P. M. Kintner, C. E. Seyler, J. S. Pickett, and J. D. Scudder, On the perpendicular scale of electron phase-space holes, Geophys. Res. Lett., 27, 169-172, 2000.
[23] Gaelzer, R., L. F. Ziebell, and P. H. Yoon, Generation of harmonic Langmuir mode by
beam-plasma instability, Phys. Plasmas, 9, 96-110, 2002.
[24] Gaelzer, R., P. H. Yoon, T. Umeda, Y. Omura, and H. Matsumoto, Harmonic Langmuir
waves. II. Turbulence spectrum, Phys. Plasmas, 10, 373-381, 2003.

REFERENCES

141

[25] Gagne, R., and M. Shoucri, The splitting scheme for the numerical solution of a onedimensional Vlasov equation, J. Comput. Phys., 24, 445-449, 1977.
[26] Goldman, M. V., M. M. Oppenheim, and D. L. Newman, Nonlinear two-stream instabilities
as an explanation for auroral bipolar wave structures, Geophys. Res. Lett., 26, 1821-1824,
1999.
[27] Gurnett, D. A., L. A. Frank, and R. P. Lepping, Plasma waves in the distant magnetotail,
J. Geophys. Res., 81, 6059-6071, 1976.
[28] Hockney, R. W., and J. W. Eastwood, Computer simulation using particles, McGraw-Hill,
New York, 1981.
[29] Jovanovic, D., P. K. Shukla, L. Steno, and F. Pegoraro, Nonlinear model for electron phase-space holes in magnetized space plasmas, J. Geophys. Res., 107, 1110,
doi:10.1029/2001JA900180, 2002.
[30] Joyce, G., and G. Knorr, Numerical integration methods of the Vlasov equation, J. Comput. Phys., 8, 53-63, 1971a.
[31] Joyce, G., G. Knorr, and T. Burns, Nonlinear behavior of the one-dimensional weak beam
plasma system, Phys. Fluids, 14, 797-800, 1971b.
[32] Kasaba, Y., H. Matsumoto, and R. R. Anderson, Geotail observation of 2f p emission
around the terrestrial electron foreshock, Adv. Space Res., 20, 699-702, 1997.
[33] Kasaba, Y., H. Matsumoto, Y. Omura, R. R. Anderson, T. Mukai, Y. Saito, T. Yamamoto,
and S. Kokubun, Statistical studies of plasma waves and backstreaming electrons in the
terrestrial electron foreshock observed by Geotail, J. Geophys. Res., 105, 79-103, 2000.
[34] Kasaba, Y., H. Matsumoto, and Y. Omura, One- and two-dimensional simulations of
electron beam instability: Generation of electrostatic and electromagnetic 2f p waves, J.
Geophys. Res., 106, 18693-18711, 2001.
[35] Kellogg, P. J., S. J. Monson, W. Bernstein, and B. A. Whalen, Observations of waves
generated by electron beams in the ionosphere, J. Geophys. Res., 91, 12065-12078, 1986.
[36] Klimas, A. J., Numerical method based on the Fourier-Fourier transform approach for
modeling 1-D electron plasma evolution, J. Comput. Phys., 50, 270-306, 1983a.
[37] Klimas, A. J., A mechanism for plasma waves at the harmonics of the plasma frequency
in the electron foreshock boundary, J. Geophys. Res., 88, 9081-9091, 1983b.

REFERENCES

142

[38] Klimas, A. J., Trapping saturation of the bump-on-tail instability and electrostatic harmonic excitation in Earths foreshock, J. Geophys. Res., 95, 14905-14924, 1990.
[39] Kojima, H., H. Matsumoto, T. Miyatake, I. Nagano, A. Fujita, L. A. Frank, T. Mukai, W.
R. Parterson, and R. R. Anderson, Relation between electrostatic solitary waves and hot
plasma ow in the plasma sheet boundary layer: Geotail observations, Geophys. Res. Lett.,
21, 2919-2922, 1994.
[40] Kojima, H, H. Matsumoto, S. Chikuba, S. Horiyama, M. Ashour-Abdalla, and R. R.
Anderson, Geotail waveform observations of broadband/narrowband electrostatic noise in
the distant tail, J. Geophys. Res., 102, 14439-14456, 1997.
[41] Kojima, H., H. Matsumoto, and Y. Omura, Electrostatic solitary waves observed in the
geomagnetic tail and other regions, Adv. Space Res., 23, 1689-1697, 1999a.
[42] Kojima, H., K. Ohtsuka, H. Matsumoto, Y. Omura, R. R. Anderson, Y. Saito, T. Mukai,
S. Kokubun, and T. Yamamoto, Plasma waves in slow-mode shocks observed by Geotail
spacecraft, Adv. Space Res., 24, 51-54, 1999b.
[43] Kojima, H., Y. Omura, H. Matsumoto, K. Miyaguti, and T. Mukai, Characteristics of
electrostatic solitary waves observed in the plasma sheet boundary: Statistical analysis,
Nonlinear Processes in Geophysics, 6, 179-186, 1999c.
[44] Krasovsky, V. L., H. Matsumoto, and Y. Omura, Bernstein-Greene-Kruskal analysis of
electrostatic solitary waves observed with Geotail, J. Geophys. Res., 102, 22131-22139,
1997.
[45] Krasovsky, V. L., H. Matsumoto, and Y. Omura, Electrostatic solitary waves as collective
charges in a magnetospheric plasma: Physical structure and properties of Bernstein-GreeneKruskal (BGK) solitons, J. Geophys. Res., 108, 1117, doi:10.1029/2001JA000277, 2003.
[46] Lindman, E. L., Free-space boundary conditions for the time dependent wave equation, J.
Comput. Phys., 18, 66-78, 1975.
[47] Lominadze, D. G., Cyclotron waves in plasma, Pergamon Press, Oxford, 1981.
[48] Mandrake. L., P. L. Pritchett, and F. V. Coroniti, Electron beam generated solitary structures in a nonuniform plasma system, Geophys. Res. Lett., 27, 2869-2872, 2000.
[49] Mangeney, A., C. Salem, C. Lacombe, J. -L. Bougeret, C. Perche, R. Manning, P. J.
Kellogg, K. Goetz, S. J. Monson, and J. -M. Bosqued, WIND observations of coherent
electrostatic waves in the solar wind, Ann. Geophysicae, 17, 307-320, 1999.

REFERENCES

143

[50] Matsumoto, H., and Y. Omura, Particle simulation of electromagnetic waves and its application to space plasmas, in Computer Simulation of Space Plasmas, edited by H. Matsumoto
and T. Sato, pp.43-102, Terra Sci., Tokyo, 1985.
[51] Matsumoto, H., I. Nagano, R. R. Anderson, H. Kojima, K. Hashimoto, M. Tsutsui, T.
Okada, I. Kimura, Y. Omura, and M. Okada, Plasma wave observations with Geotail spacecraft, J. Geomag. Geoelectr., 46, 59-95, 1994a.
[52] Matsumoto, H., H. Kojima, T. Miyatake, Y. Omura, M. Okada, I. Nagano, and M. Tsutsui,
Electrostatic solitary waves(ESW) in the magnetotail: BEN wave forms observed by Geotail,
Geophys. Res. Lett., 21, 2915-2918, 1994b.
[53] Matsumoto, H., H. Kojima, Y. Kasaba, T. Miyake, R. R. Anderson, and T. Mukai, Plasma
waves in the upstream and bow shock regions observed by GEOTAIL, Adv. Space Res., 20,
683-693, 1997.
[54] Matsumoto, H., L. A. Frank, Y. Omura, H. Kojima, W. R. Paterson, M. Tsutsui, R.
R. Anderson, S. Horiyama, S. Kokubun, and T. Yamamoto, Generation Mechanism of
ESW based on GEOTAIL Plasma Wave Observation, Plasma Observation and Particle
Simulation, Geophys. Res. Lett., 26, 421-424, 1999.
[55] Matsumoto, H., X. H. Deng, H. Kojima, and R. R. Anderson, Observation of Electrostatic Solitary Waves associated with reconnection on the dayside magnetopause boundary,
Geophys. Res. Lett., 30, 1326, doi:10.1029/2002GL016319, 2003.
[56] Miyake, T., Y. Omura, H. Matsumoto, and H. Kojima, Two-dimensional computer simulations of electrostatic solitary waves observed by Geotail spacecraft, J. Geophys. Res., 103,
11841-11850, 1998.
[57] Miyake, T., Y. Omura, and H. Matsumoto, Electrostatic particle simulations of solitary
waves in the auroral region, J. Geophys. Res., 105, 23239-23249, 2000.
[58] Morse, R. L. and C. W. Neilson, Numerical simulation of a warm two-beam plasma, Phys.
Fluids, 12, 2418-2425, 1969.
[59] Mozer, F.S., R. Ergun, M. Temerin, C. Cattell, J. Dombeck, and J. Wygant, New features
of time domain electric-eld structures in the auroral acceleration region, Phys, Rev. Lett.,
79, 1281, 1997.
[60] Mukai, T., S. Machida, Y. Saito, M. Hirahara, T. Terasawa, N. Kaya, T. Obara, M. Ejiri,
and A. Nishida, The low energy particle (LEP) experiment onboard the Geotail satellite,
J. Geomag. Geoelectr., 46, 669-692, 1994.

REFERENCES

144

[61] Mur, G., Absorbing boundary conditions for the nite-dierence approximation of the
time-domain electromagnetic-eld equation, IEEE Trans. Electromagnetic. Compat., EMC23, 377-382, 1981.
[62] Muschietti, L., R. E. Ergun, I. Roth, and C. W. Carlson, Phase-space electron holes along
magnetic eld lines, Geophys. Res. Lett., 26, 1093-1096, 1999.
[63] Muschietti, L., I. Roth, C. W. Carlson, and R. E. Ergun, Transverse instability of magnetized electron holes, Phys. Rev. Lett., 85, 94-97, 2000.
[64] Nakamura, T., and T. Yabe, Cubic interpolated propagation scheme for solving the hyperdimensional Vlasov-Poisson equation in phase space, Comput. Phys. Commun., 120, 122154, 1999.
[65] Newman, D. L., M. V. Goldman, M. Spector, and F. Perez, Dynamics and instability of
electron phase-space tubes, Phys. Rev. Lett., 86, 1239-1243, 2001a.
[66] Newman, D. L., M. V. Goldman, R. E. Ergun, and A. Mangeney, Formation of double
layers and electron holes in a current-driven space plasma Phys. Rev. Lett., 87, 255001, doi:
10.1103/PhysRevLett.87.255001, 2001b.
[67] Nishida, A., T. Hada, K. A. Anderson, R. R. Anderson, S. J. Bame, and E. W. Hones Jr.,
Broadband electrostatic noise in the magnetotail: Its relation to plasma sheet dynamics, J.
Geophys. Res., 90, 4453-4460, 1985.
[68] Omura, Y., and H. Matsumoto, KEMPO1: Technical guide to one-dimensional electromagnetic particle code, in Computer Space Plasma Physics, edited by H. Matsumoto and
Y. Omura, pp.21-65, Terra Scientic, Tokyo, 1993.
[69] Omura, Y., H. Kojima, and H. Matsumoto, Computer simulation of electrostatic solitary
waves: A nonlinear model of broadband electrostatic noise, Geophys. Res. Lett., 21, 29232926, 1994.
[70] Omura, Y., H. Matsumoto, T. Miyake, and H. Kojima, Electron beam instabilities as
generation mechanism of electrostatic solitary waves in the magnetotail, J. Geophys. Res.,
101, 2685-2697, 1996.
[71] Omura, Y., H. Kojima, N. Miki, T. Mukai, H. Matsumoto, and R. Anderson, Electrostatic
solitary waves carried by diused electron beams observed by the Geotail spacecraft, J.
Geophys. Res., 104, 14627-14637, 1999a.
[72] Omura, Y., H. Kojima, N. Miki, and H. Matsumoto, Two-dimensional electrostatic solitary
waves observed by Geotail in the magnetotail, Adv. Space Res., 24, 55-58, 1999b.

REFERENCES

145

[73] Omura, Y., H. Kojima, T. Umeda, and H. Matsumoto, Observational evidence of dissipative small scale processes: Geotail spacecraft observation and simulation of electrostatic
solitary waves, Astrophys. Space Sci., 277, 45-57, 2001.
[74] Omura, Y., W. J. Heikkila, T. Umeda, K. Ninomiya, and H. Matsumoto, Particle simulation of plasma response to an applied electric eld parallel to magnetic eld lines, J.
Geophys.Res., 108, 1197, doi:10.1029/2002JA009573, 2003.
[75] ONeil, T. M., J. H. Winfrey, and J. H. Malmberg, Nonlinear interaction of a small cold
beam and a plasma, Phys. Fluids, 14, 1204-1212, 1971.
[76] ONeil, T. M., and J. H. Winfrey, Nonlinear interaction of a small cold beam and a plasma.
Part II, Phys. Fluids, 15, 1514-1522, 1972.
[77] Oppenheim, M. M., D. L. Newmann, and M. V. Goldman, Evolution of electron phasespace holes in a 2-D magnetized plasma, Phys. Rev. Lett., 83, 2344-2347, 1999.
[78] Oppenheim, M. M., G. Vetoulis, D. L. Newmann, and M. V. Goldman, Evolution of
electron phase-space holes in 3D, Geophys. Res. Lett., 28, 1891-1894, 2001.
[79] Pottelette, R., M. Malingre, N. Dubouloz, B. Aparicio, R. Lundin, G. Holmgren, and G.
Marklund, High-frequency waves in the cusp/cleft regions, J. Geophys. Res., 95, 5957-5971,
1990.
[80] Roberts, K. V. and H. L. Berk, Nonlinear evolution of a two-stream instability, Phys. Rev.
Lett., 19, 297-300, 1967.
[81] Saeki, K., P. Michelsen, H. L. Pecseli, and J. Juul Rasmussen, Formation and coalescence
of electron solitary holes, Phys. Rev. Lett., 43, 501-504, 1979.
[82] Schriver, D., and M. Ashour-Abdalla, Generation of high-frequency broadband electrostatic noise: The role of cold electrons, J. Geophys. Res., 92, 5807-5819, 1987.
[83] Schriver, D., M. Ashour-Abdalla, V. Sotnikov, P. Hellinger, V. Fiala, and A. Mangeney,
Excitation of electron acoustic waves near the electron plasma frequency and at twice the
plasma frequency, J. Geophys. Res., 105, 12919-12927, 2000.
[84] Singh, N., Electron holes as a common feature of double-layer-driven plasma waves, Geophys. Res. Lett., 27, 927-930, 2000a.
[85] Singh, N., S. M. Loo, B. E. Wells, and C. Deverpalli, Three-dimensional structure of
electron holes driven by an electron beam, Geophys. Res. Lett., 27, 2469-2472, 2000b.

REFERENCES

146

[86] Singh, N., S. M. Loo, and B. E. Wells, Electron hole as an antenna radiating plasma waves,
Geophys. Res. Lett., 28, 1371-1374, 2001a.
[87] Singh, N., S. M. Loo, and B. E. Wells, Electron hole structure and its stability depending
on plasma magnetization, J. Geophys. Res., 106, 21183-21198, 2001b.
[88] Singh, N., Space-time evolution of electron-beam driven electron holes and their eects on
the plasma, Nonlinear Proc. Geophys., 10, 53-63, 2003.
[89] Tajima, T., and Y. C. Lee, Absorbing boundary condition and Budden turning point
technique for electromagnetic plasma simulation, J. Comput. Phys., 42, 406-413, 1981.
[90] Tominaga, T., Study on eciency increasing of electromagnetic particle codes, Master
Thesis (in Japanese), Dep. Electr. Commun. Eng., Kyoto Univ., Kyoto, Japan, 1999.
[91] Turikov, V. A., Electron plase space holes as localized BGK solutions, Phys. Scr., 30,
73-77, 1984.
[92] Ueda, H., Y. Omura , H. Matsumoto, and T. Okuzawa, A study of the numerical heating
in electrostatic particle simulations, Comput. Phys. Commun., 79, 249-259, 1994.
[93] Umeda, T., Y. Omura, and H. Matsumoto, An improved masking method for absorbing
boundaries in electromagnetic particle simulations, Comput. Phys. Commun., 137, 286-299,
2001.
[94] Umeda, T., Y. Omura, H. Matsumoto, and H. Usui, Formation of electrostatic solitary
waves in space plasmas: Particle simulations with open boundary conditions, J. Geophys.
Res., 107, 1449, doi:10.1029/2001JA000286, 2002.
[95] Umeda, T., Y. Omura, P. H. Yoon, R. Gaelzer, and H. Matsumoto, Harmonic Langmuir
waves. III. Vlasov simulation, Phys. Plasmas, 10, 382-391, 2003a.
[96] Umeda, T., Y. Omura, T. Tominaga, and H. Matsumoto, A new charge conservation
method in electromagnetic particle simulations, Comput. Phys. Commun., 156, 73-85,
2003b.
[97] Umeda, T., Y. Omura, and H. Matsumoto, Two-dimensional particle simulation of electromagnetic eld signature associated with electrostatic solitary waves, J. Geophys. Res.,
109, A02207, doi:10.1029/2003JA010000, 2004.
[98] Utsumi, T., T. Kunugi, and J. Koga, A numerical method for solving the one-dimensional
Vlasov-Poisson equation in phase space, Comput. Phys. Commun., 108, 159-179, 1998.

REFERENCES

147

[99] Vetoulis, G., and M. Oppenheim, Electrostatic mode excitation in electron holes due to
wave bounce resonance, Phys. Rev. Lett., 86, 1235-1238, 2001.
[100] Villasenor, J., and O, Buneman, Rigorous charge conservation for local electromagnetic
eld solvers, Comput. Phys. Commun., 69, 306-316, 1992.
[101] Yabe, T., and T. Aoki, A universal solver for hyperbolic equations by cubic interpolated
propagation, I. One-dimensional solver, Comput. Phys. Commun., 66, 219-232, 1991a.
[102] Yabe, T., T. Ishikawa, P. Y. Wang, T. Aoki, Y. Kadota, and R. Ikeda, A universal solver
for hyperbolic equations by cubic interpolated propagation, II. Two- and three-dimensional
solver, Comput. Phys. Commun., 66, 233-242, 1991b.
[103] Yabe, T., F. Xiao, and T. Utsumi, The constrained interpolation prole method for
multiphase analysis, J. Comput. Phys., 169, 556-593, 2001.
[104] Yoon, P. H., Generalized weak turbulence theory, Phys. Plasmas, 7, 4858-4871, 2000.
[105] Yoon, P. H., R. Gaelzer, T. Umeda, Y. Omura, and H. Matsumoto, Harmonic Langmuir
waves. I. Nonlinear dispersion relation, Phys. Plasmas, 10, 364-372, 2003.
[106] Ziebell, L. F., R. Gaelzer, and P. H. Yoon, Nonlinear development of weak beam-plsama
instability, Phys. Plasmas, 8, 3982-3995, 2001.

Publication List
Refereed Journal Articles
1. Omura, Y., H. Kojima, T. Umeda, and H. Matsumoto, Observational evidence of dissipative small scale processes: Geotail spacecraft observation and simulation of electrostatic
solitary waves, Astrophysics and Space Science, Vol.277, No.1, pp.45-57, 2001.
2. Umeda, T., T., Y. Omura, and H. Matsumoto, An improved masking method for absorbing boundaries in electromagnetic particle simulations, Computer Physics Communications, Vol.137, No.2, pp.286-299, 2001.
3. Umeda, T., Y. Omura, H. Matsumoto, and H. Usui, Formation of electrostatic solitary
waves in space plasmas: Particle simulations with open boundary conditions, Journal of
Geophysical Research, Vol.107, No.A12, 1449, doi:10.1029/2001JA000286, 2002.
4. Yoon, P. H., R. Gaelzer, T. Umeda, Y. Omura, and H. Matsumoto, Harmonic Langmuir
waves. I. Nonlinear dispersion relation, Physics of Plasmas, Vol.10, No.2, pp.364-372,
2003.
5. Gaelzer, R., P. H. Yoon, T. Umeda, Y. Omura, and H. Matsumoto, Harmonic Langmuir
waves. II. Turbulence spectrum, Physics of Plasmas, Vol.10, No.2, pp.373-381, 2003.
6. Umeda, T., Y. Omura, P. H. Yoon, R. Gaelzer, and H. Matsumoto, Harmonic Langmuir
waves. III. Vlasov simulation, Physics of Plasmas, Vol.10, No.2, pp.382-391, 2003.
7. Omura, Y., T. Umeda, and H. Matsumoto, Simulation of electron beam instabilities and
nonlinear potential structures, in Space Plasma Physics, edited by J. Buchner, C. T. Dum,
M. Scholer, Springer-Verlag Berlin Heidelberg, pp.79-92, 2003.
8. Omura, Y., W. J. Heikkila, T. Umeda, K. Ninomiya, and H. Matsumoto, Particle simulation of plasma response to an applied electric eld parallel to magnetic eld lines, Journal
of Geophysical Research, Vol.108, No.A5, 1197, doi:10.1029/2002JA009573, 2003.
9. Umeda, T., Y. Omura, T. Tominaga, and H. Matsumoto, A new charge conservation
method in electromagnetic particle-in-cell simulations, Computer Physics Communications, Vol.56, No.1, pp. 73-85, 2003.
148

PUBLICATION LIST

149

10. Umeda, T., Y. Omura, and H. Matsumoto, Two-dimensional particle simulation of electromagnetic eld signature associated with electrostatic solitary waves, Journal of Geophysical Research, Vol.109, No.A2, A02207, doi:10.1029/2003JA010000, 2004.

Oral Presentations
1. Umeda, T., Y. Omura, H Usui, and H. Matsumoto, Two-dimensional electromagnetic
simulations of electrostatic solitary waves in open systems (in Japanese), 2000 Japan
Earth and Planetary Science Joint Meeting, Tokyo, June, 2000.
2. Umeda, T., Y. Omura, H Usui, and H. Matsumoto, Electromagnetic particle simulation
of electrostatic solitary waves in a two-dimensional open system (in Japanese), 108th Society of Geomagnetism and Earth, Planetary and Space Science (SGEPSS) Fall Meeting,
Tokyo, November, 2000.
3. Umeda, T., Y. Omura, H Usui, and H. Matsumoto, Two-dimensional particle simulations
of electromagnetic emission from electrostatic solitary waves (in Japanese), 2001 Japan
Earth and Planetary Science Joint Meeting, Tokyo, June, 2001.
4. Umeda, T., Y. Omura, H. Matsumoto, and H Usui, Electromagnetic particle simulation
of electron beam instabilities in a two-dimensional open system, 2001 Asia-Pacific Radio
Science Conference (AP-RASC), Tokyo, Japan, August, 2001.
5. Umeda, T., Y. Omura, and H. Matsumoto, An improved masking method for absorbing
boundaries in electromagnetic particle simulations, 6th International School/Symposium
for Space Plasma Simulations (ISSS-6), Garching, Germany, September, 2001 (Invited).
6. Umeda, T., Y. Omura, H. Matsumoto, and H Usui, Particle simulations of electromagnetic emissions from electrostatic solitary waves (in Japanese), 110th Society of Geomagnetism and Earth, Planetary and Space Science (SGEPSS) Fall Meeting, Fukuoka,
November, 2001.
7. Umeda, T., Y. Omura, H. Matsumoto, and H Usui, Formation of electrostatic solitary
waves in a two-dimensional open system (in Japanese), 2002 Japan Earth and Planetary
Science Joint Meeting, Tokyo, May, 2002.
8. Umeda, T., Y. Omura, H. Matsumoto, and H. Usui, Formation of electrostatic solitary
waves: Two-dimensional particle simulation with open boundary conditions, 2002 Western Pacific Geophysics Meeting (WPGM), Wellington, New Zealand, July, 2002.

PUBLICATION LIST

150

9. Umeda, T., Y. Omura, and H. Matsumoto, Particle simulations of electromagnetic emissions from electrostatic solitary waves, 27th General Assembly of the International Union
of Radio Science (URSI), Maastricht, Netherlands, August, 2002.
10. Umeda, T., P. H. Yoon, Y. Omura, and H. Matsumoto, Vlasov simulation of harmonic
Langmuir waves (in Japanese), 112th Society of Geomagnetism and Earth, Planetary and
Space Science (SGEPSS) Fall Meeting, Chofu, November, 2002.
11. Umeda, T., Y. Omura, and H. Matsumoto, Particle simulations of electrostatic solitary
waves in space plasmas, 7th International Workshop on the Interrelationship between
Plasma Experiments in Laboratory and Space (IPELS), Whitesh, Montana, USA, June,
2003 (Invited).
12. Umeda, T., Y. Omura, and H. Matsumoto, Particle simulations of electromagnetic emissions from electrostatic solitary waves, 23rd General Assembly of the International Union
of Geodesy and Geophysics (IUGG), Sapporo, Japan, July, 2003.
13. Umeda, T., Y. Omura, and H. Matsumoto, Particle simulations of electromagnetic emissions from electrostatic solitary waves (in Japanese), 114th Society of Geomagnetism and
Earth, Planetary and Space Science (SGEPSS) Fall Meeting, Toyama, November, 2003.