NUNC COCNOSCO EX PARTE

TRENT UNIVERSITY
LIBRARY
PLASMA PHYSICS
 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/plasmaphysicsOOOOtane
McGraw-Hill Physical and Quantum Electronics Series

Hubert Heffner and A. E. Siegman, Consulting Editors

Beam: Electronics of Solids

Collin: Foundations for Microwave Engineering
Elliott: Electromagnetics
Johnson: Field and Wave Electrodynamics
Kirstein, Kino, and Waters: Space-charge Flow
Louisell: Radiation and Noise in Quantum Electronics
Moll: Physics of Semiconductors
Papas: Theory of Electromagnetic Wave Propagation
Siegman: Microwave Solid-state Masers
Smith, Janak, and Adler: Electronic Conduction in Solids
Smith and Sorokin: The Laser
Tanenbaum: Plasma Physics
White: Basic Quantum Mechanics
PLASMA PHYSICS
B. Samuel Tanenbaum
Associate Professor of Engineering
Case Institute of Technology

McGraw-Hill Book Company

New York St. Louis San Francisco
Toronto London Sydney
 , \ Z. 5

PLASMA PHYSICS
Copyright © 1967 by McGraw-Hill, Inc. All
Rights Reserved. Printed in the United States
of America. No part of this publication may be
reproduced, stored in a retrieval system, or
transmitted, in any form or by any means,
electronic, mechanical, photocopying, record¬
ing, or otherwise, without the prior written per¬
mission of the publisher.
Library of Congress Catalog Card Number 67-22971

62812
12 34567890 MAM M 7432106987
 Dedicated to my parents
Rena A. Tanenbaum and Harry M. Tanenbaum
and to my wife
Carol Binder Tanenbaum
 PREFACE

This text on plasma physics is written for students at the senior or first-year
graduate level. It is an extended version of notes for a first-year graduate
course in plasma dynamics that has been offered for many years at Case. The
book (like the course for which it was developed) is intended to serve both as
an introduction to the theory of plasmas for students planning to do further
work in the subject and as a survey for students with other research interests
who want to have some knowledge of plasmas.
The book has been written with the realization that a first course in
plasma physics may attract students with a wide variety of backgrounds.
Students taking the course at Case have included physicists, chemists, mathe¬
maticians, astronomers, mechanical engineers, and electrical engineers. For
this reason every effort has been made to make the book self-contained by
including brief developments of all the mathematics, electromagnetic theory,
fluid mechanics, and kinetic theory used in the text. In addition the material
is arranged so that, for the most part, the more elementary topics (both
conceptually and mathematically) are presented in the first half of the text,
and the more advanced topics are left for the final chapters. Hence an
undergraduate or a graduate student with a weak background in mathematics
should be able to read the first three chapters with no trouble, as well as a good
number of the more advanced topics in the final three chapters and in the
Appendix.
The topics included in the book clearly fail to cover all areas of plasma
physics. Aside from the first two introductory sections there is almost no
discussion of the experimental aspects of the subject. Moreover, there are
some important theoretical topics, like radiation and atomic processes in a
plasma, which are scarcely mentioned, and others, like plasma instabilities and
magnetohydrodynamics, which are covered very briefly. I have attempted,
however, to cover a broad selection of topics with particular emphasis on the
various models that are used to predict the behavior of plasmas. In addition
I have tried to introduce the reader to much of the common terminology and to
most of the “well-known results” in plasma physics. In this connection I
have attempted throughout to use derivations which do not require too much
mathematical sophistication and which are complete enough for easy recogni¬
tion of all the assumptions and approximations used.
The idea of developing this material into a text was suggested to me by
Professor 0. K. Mawardi, who heads the plasma research program at Case.
During the past several years, Professor Mawardi has developed a full sequence
of plasma-physics courses, which now includes one senior course, three regular
graduate courses, and a number of special-topic courses. His suggestions
concerning topics to include in the text were very valuable, as was his constant
encouragement.
Many of my other colleagues at Case have helped me in preparing this
text, either by reading critically portions of the manuscript or by helping me
to learn those areas of plasma physics where my own background was weak.
Those who were of particular help include Professors R. E. Collin, E. Reshotko,
W. B. Johnson, R. B. Block, E. J. Morgan, R. T. Compton, E. D. Thompson,
and L. S. Taylor. A number of other people have also suggested changes
which improved the text. I would especially like to thank Drs. D. A. Meskan,
S. P. Heims, L. H. Holway, Jr., G. A. Otteni, and E. L. Walker and several
anonymous reviewers chosen by the publisher. In addition, I am indebted
for many helpful comments to all the students who used earlier versions of this
material as course notes.
R. S. Hirsh, who kindly agreed to read through the complete manuscript,
corrected many small errors and made numerous suggestions which, I believe,
make the text easier for the reader to follow. His assistance is greatly
appreciated.
The number of books dealing with plasma physics and related topics has
grown rapidly in recent years. A list of many of these texts is included in the
Bibliography at the end of this book. Of these references, the ones which were
most useful to me are the fine texts “Physics of Fully Ionized Gases,” by
Spitzer, “Elements of Plasma Physics,” by Gartenhaus, and “Foundations of
Plasma Dynamics,” by Holt and Haskell; the excellent chapter by Mintzer in
“Mathematics of Physics and Chemistry,” vol. 2, edited by Margenau and
Murphy; and the very readable nonmathematical surveys “Project Sherwood:
The U.S. Program in Controlled Fusion,” by A. S. Bishop, and “Elementary
Plasma Physics,” by L. A. Arzimovich.
A number of additional acknowledgments are due here. First, Professor
R. E. Bolz, head of the Engineering Division, both by his encouragement and
by his kindness in providing a relatively light teaching load, made it possible
for me to complete this manuscript in a fairly short time. During this time
my work was supported in part by research grants to the plasma group from
the National Science Foundation and from the National Aeronautics and
Space Administration. Next, I am greatly indebted to Mrs. J. L. Griffen, who
did a remarkably accurate job of typing all the material involved in preparing
this book. Finally, I must express my gratitude to Professor D. Mintzer,
who has been teacher, fellow researcher, and friend for many years.
Despite the surprisingly large amount of work required, I have found the
preparation of this text to be a remarkable educational experience. I hope
that the students who use the book will find it useful and will learn from the
reading a sizable fraction of what I learned from the writing.

B. Samuel Tanenbaum
 CONTENTS

PREFACE

1 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS 1

1.1 Introduction 2
1.2 Applications of Plasma Physics 14
1.3 Review of Vector Analysis and Electromagnetic Theory 30
1.4 Motion of a Charge in Uniform Fields 36
1.5 Motion in a Nonuniform B Field 43
1.6 Adiabatic Invariants; Applications; Time-varying Fields 52

2 MAGNETOIONIC THEORY 61

2.1 The Langevin Equation 62

2.2 Electron Conductivity and Mobility 65
2.3 Conductivity Tensor and Dielectric Tensor 69
2.4 Wave Propagation with Bo = 0 72
2.5 The Effect of Collisions 82
2.6 Resonances and Reflection Points with B0 ^ 0 86
2.7 Wave Propagation Parallel or Perpendicular to Bo 92
2.8 Alternative Descriptions of Dispersion Relations 98
3 CONTINUUM EQUATIONS FOR A PLASMA 103

3.1 Derivation of Continuum Equations 104

3.2 Treatment of Plasma as a Mixture 109
3.3 Diffusion 113
3.4 Wave Propagation in a Warm Plasma 122
3.5 Longitudinal Waves in a Fully Ionized Gas ' 129
3.6 Transverse Waves in the Direction of Bo 134
3.7 Linearized Magnetohydrodynamic Equations 138
3.8 Magnetic Pressure, Viscosity, Reynolds Number, and Diffusivity 145
3.9 MHD Waves 150
3.10 The Nonlinear MHD Equations 155

4 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA 159

4.1 Distribution Function and the Boltzmann Equation 160

4.2 Transport Equations 166
4.3 The BV Equation and the Relaxation Model 172
4.4 Application to Longitudinal Electron Waves 177
4.5 Landau Damping 181
4.6 Landau Damping of Transverse Waves 188
4.7 The Two-stream Instability 198
4.8 The Debye Potential Problem 203
4.9 Plasma Sheaths 210

5 THE BOLTZMANN EQUATION FOR A PLASMA 223

5.1 Two-body Collisions and the Boltzmann Collision Term 224

5.2 The H Theorem 230
5.3 Maximum Entropy and the Maxwellian Distribution Function 235
5.4 Lowest-order Transport Equations for a Gas Mixture 240
5.5 Calculation of Collision Frequencies 246
5.6 The Electron Runaway Effect 254
5.7 The Thirteen-moment Equations for an Electron Gas 260
5.8 Calculation of Transport Coefficients 267
5.9 Distribution Function for a Strong E Field 273
5.10 The Fokker-Planck Equation 282

6 THE BBGKY EQUATIONS FOR A PLASMA 291

6.1 The Liouville Equation 292

6.2 The Equation for /(1) 295
6.3 The BV Equation with Self-consistent Field 299
6.4 Corrections to the BV Equation 303
6.5 Formal Perturbation Technique for Weak Correlations 309
MATHEMATICAL APPENDIX 315

A.l Delta Function and Fourier Transform Theory 316

A.2 Evaluation of Scattering Angle x 321
A.3 Derivation of Boltzmann Collision Term 326
A.4 Calculation of Cross Sections 330
A.5 The Chapman-Enskog Technique 335
A.6 Evaluation of (df/dt)c for Electron-Neutral Collisions 340
A.7 Important Formulas and Definitions 347

BIBLIOGRAPHY 351

INDEX 355
chapter one

MOTION OF
A CHARGED
PARTICLE IN
ELECTROMAGNETIC
FIELDS
chapter one

1.1 INTRODUCTION

Definition of a plasma

In this text we are concerned with the behavior of plasmas. The term
plasma, first applied by Langmuir in 1928 to the ionized gas in an
electric discharge, is now used to describe a large class of basically
neutral mixtures containing some charged particles, which might be
positive or negative ions, free electrons, or holes (vacancies in the other¬
wise filled electron band of a solid that behave like electrons with a
positive charge). Although the most actively studied plasmas are still
ionized gases, some electrolytic liquids and some solids can also be
classified as plasmas.
Three basic features characterize plasmas and distinguish them
from ordinary solids, liquids, or gases. The most notable feature is
that some, at least, of the particles are charged and therefore interact
with each other in accordance with the coulomb-force law. This
coulomb force (which is proportional to R~2, with R the distance
between the particles) falls off very slowly with distance compared with
most other interparticle forces, f Hence, in a plasma every charged
particle interacts simultaneously with many of its neighbors, giving
the plasma a cohesiveness which is often compared to that of a mold or
jelly-

Debye shielding

A second, related feature of a plasma is that the charged particles tend

to rearrange themselves in such a way as to effectively shield any
electrostatic fields due either to a surface at some nonzero potential or
to a charge within the plasma. This rearrangement of the charged
particles (which is analyzed in detail in Secs. 4.8 and 4.9) effectively
cancels out any electrostatic fields within a distance on the order of a
length \D) which is known as the Debye length. It is shown in Sec. 4.8

f The force between most neutral molecules in a gas, for example, is large when
the particle centers are within one diameter of each other but very nearly zero
when they are farther apart.
 1.1 INTRODUCTION 3

that

/ T\1/2
\d = 69.0 ( — j meters (1.1.1)

where T = temperature, °K
ne — electron number density, particles/m3
From the plot of \D shown in Fig. 1.1 it can be seen that the Debye
length is less than 1 cm for laboratory plasmas of all types, although
it rapidly becomes very large for gases with very low electron number
densities (or extremely high temperatures).
The Debye shielding effect, while characteristic of all plasmas, does
not occur in every mixture containing charged particles. Fora mixture
which can be treated with classical equations there are two necessary
conditions for shielding to occur. The first and most obvious require¬
ment is that the physical dimensions of the system be large compared to
\d, since otherwise there is simply 7iot enough room for the shielding to
occur. In addition, there must be enough electrons within a distance \d
from the disturbance to produce the shielding; hence the average dis¬
tance between electrons must be small compared with the Debye length,

6 8 10 12 14 16 18 20 22 24 26 28

log]0 ne, meter'3

Fig. 1.1 The Debye length (in meters) plotted as a function of electron number
density and temperature.
4 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

or, equivalently, the number of electrons Nd in a sphere of radius Ad

must be much greater than 1. From Eq. (1.1.1) we note that

y3/2
Nd = |nc7rXz>3 = 1.37 X 106 ^ (1.1.2)
7le

A plot of Nd is shown in Fig. 1.2, where it can be seen that Nd is

large for hot or rarefied gases and small for dense or cool material.

Quantum-mechanical effect on temperature

From Fig. 1.2 it is clear that solids do not satisfy the plasma shielding
criterion ND » 1 if we use the usual lattice temperature (about 300°Iv
or less) in Eq. (1.1.2). However, because of the high electron number
densities in a solid, quantum-mechanical effects become important and
require the temperature of the electrons to be as much as three or four
orders of magnitude higher than the lattice temperature, thus bringing
some solids into the range where Nd exceeds 1.
To see how this comes about we note that, from the uncertainty

o
bo
o

Fig. 1.2 The number of particles in a Debye sphere plotted as a function of

electron number density and temperature.
 1.1 INTRODUCTION 5

'principle of quantum mechanics,

ApAx>~ (1.1.3)

where Ap and Ax are the uncertainties in the momentum and position of

an electron and h is Planck’s constant. This condition shows that as
the particles get packed closer and closer together (so that Ax is small),
the iqomentum must get very large, since p must be at least of the
same order as Ap. The electron temperature (as we shall see in Sec. 4.2)
for electrons having no net drift velocity is given by

m(v2) _ (p2)
(1.1.4)
3k 3 niK

where m = electron mass

(■v2) = average of square of electron velocity
k = Boltzmann’s constant
From Eq. (1.1.4) it is clear that the electron temperature is relatively
high whenever the momentum is forced to be large by the uncertainty
principle.
To get a crude numerical estimate of this effect, we can let Ax be
of the order of the average distance between electrons, ne~113, so that,
from Eq. (1.1.3),

hne113
Ap > (1.1.5)
47r

In addition, if p2 is of the same order as (Ap)2, then, from Eqs. (1.1.4)

and (1.1.5), we have

h2n 2/3
T > -:0 ‘ - = 7.4 X 10_16ne2/3 (1.1.6)
487r2mK

Equation (1.1.6) provides a lower bound for the electron tempera¬

ture and is not of great significance for gaseous plasmas, where ne is
seldom greater than 1021 particles/m3. However, for a solid there are
two factors which make this minimum value for T quite high: (1) the
number densities are very large (ne may be 1028 particles/m3); and (2)
in the solid, because of the periodic structure of the lattice, the electrons
to a good approximation act as if they were free but with a mass that
is lighter than m. The value of this effective mass depends on the
particular material, but for some solids it may be as little as 0.01m.
With this light mass and high number density T can reach 104°K, a
6 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

temperature which the lattice itself could not attain without melting
or vaporizing. This value for the electron temperature, as can be seen
in Fig. 1.2, raises the value of Nd to values comparable to, or greater
than, 1; hence at least some shielding effects can occur.

The cor criterion

Plasmas are sometimes defined to include all media that have some
charged particles and satisfy the shielding criteria, Nd > 1 and \d <
smallest linear dimension of the plasma. However, this definition is
really a little too broad, as can be seen by a simple example. From
Fig. 1.2 note that a gas with a temperature of 1000°Iv and with 106
electrons/m3 has a very large value for ND■ If nn, the number density
of neutral molecules, is also relatively low (as in the interstellar gas),
this medium should be treated as a plasma. However, if this is a
relatively dense gas (where nn is 15 or 20 orders of magnitude larger
than ne), then the effect of the electrons is negligible, and the medium
can be treated as a neutral gas.
The crucial difference between these two cases is that re„, the
average time an electron travels between collisions with neutral mole¬
cules, is large (perhaps days) for the rarefied gas and exceedingly small
(perhaps 10~u sec) for the dense one. Therefore, the electrons can
have a relatively independent behavior in the former case, having a
separate drift velocity and temperature, for example, from that of the
neutrals, while in the latter case there are so many electron-neutral
collisions per second that the electrons are always forced to be in com¬
plete equilibrium with the neutrals.
This shows up clearly in a calculation of the dc. electrical con¬
ductivity ao of a weakly ionized gas, where (in Sec. 2.2) we find that
do = nfi^Ten/m = 2.8 X 10~8ncTen. Thus, in the example we were just
considering, the interstellar gas is a very good conductor, while the
dense gas is a very poor one.
From the analysis of collision frequencies in Chap. 5 we find that,
for a gas

4 X 1015
(1.1.7)
nnTe1/2

so that for ionized gas in the laboratory ren is typically between 10-5
and 10-9 sec. In a solid, where electrons suffer collisions with vibrating
lattice ions, with impurities, and with lattice imperfections, t is much
shorter than this unless one works with very pure crystals and at a
low lattice temperature (where lattice vibrations are greatly reduced).
However, at temperatures of a few degrees Kelvin, using very pure
samples of bismuth, values of r in excess of 10-9 sec are possible.
 1.1 INTRODUCTION 7

Although these values for r may seem short, one must remember that
plasma experiments often involve waves at microwave (or higher) fre¬
quencies, where the period of an oscillation is less than 10-10 sec; hence
the time between collisions is still large compared to the time during
which the plasma parameters are changing. For this reason, when
dealing with wave phenomena, if to is the frequency in radians per
second, then the product cor is large when plasma effects are important
and small when there are too many collisions for the electrons to be
uncoupled from the rest of the system.
Naturally occurring plasmas
Throughout- the bulk of this text we shall be concerned with two closely
related goals: (1) the development of the various sets of equations that
are used to study the behavior of plasmas, together with some discussion
of the limitations and approximations involved in each approach; and
(2) a few solutions to the plasma equations, chosen to illustrate either
important physical properties of the plasma or useful mathematical
methods for dealing with the equations.
In view of these goals we shall not devote much space to the many
actual and potential applications of plasma physics. However, it is
important to describe some of these briefly at least, since it is these
applications which motivate (and finance) most of the research in the
subject.
To start, our interest in plasmas results from the fact that most-
of the universe, with a few exceptions like the earth, is a plasma. First,
there are the stars, which because of their high temperatures are almost
completely ionized. Next there is the interstellar gas, which, despite
its low number density, still exhibits some plasma properties. Finally,
there are the numerous special plasmas that show up in a wide variety
of geophysical and ast-rophysical problems. These include the iono¬
sphere, the region of partly ionized gas that extends from about 50 km
above the earth to several earth radii;f the Van Allen belts of energetic
ionized particles trapped at high altitudes by the earth’s magnetic
field; the beams of plasma which erupt from the sun during magnetic
storms and which are related to the spectacular aurora borealis seen in
the earth’s upper atmosphere; the steady stream of plasma known as
the solar wind, which greatly distorts the earth’s magnetic field at
heights greater than two or three earth radii; and the spectacular gas
discharges created by a lightning stroke.
Solid and liquid plasmas
A great many solid, liquid, and gaseous plasmas, man-made as well as
natural, are present on earth. Among the solids alone, for example,
f One earth radius is about 6370 km.
8 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

there are several kinds of plasma. These can be divided into two
classes, compensated plasmas, consisting of electrons and holes, and
uncompensated plasmas, where the numbers of mobile negative and
positive charges are not equal and overall charge neutrality is main¬
tained by heavy positive or negative ions. Compensated plasmas
include metals like iron and tungsten, semimetals (so-called because
the number density of electrons and holes is much lower than for the
metals) like bismuth and antimony, and semiconductors. Uncom¬
pensated plasmas include metals like copper and sodium and semi¬
metals or semiconductors that are doped by adding impurity atoms
which release electrons or holes, depending on whether the impurity
atom has more or fewer electrons than the regular atoms of the lattice.
In addition to these “natural” solid-state plasmas, i.e., plasmas
which exist in the crystals as they are grown, it is also possible to
create plasmas in a semiconductor by applying an external electric field
strong enough to cause avalanche breakdown. This process occurs when
the electrons gain enough energy from the applied field to dislodge other
electrons when they collide with lattice atoms, so that a high electron
number density is quickly created.
Examples of liquid plasmas include the liquid metals, like mercury,
which was used in the early experimental work in magnetohydro¬
dynamics, and electrolytic solutions of all types. (The Debye shielding
theory, in fact, was first developed for electrolytes.)

The Saha equation

The gaseous plasmas, finally, are the ones which have been most inten¬
sively studied. Any neutral molecule will gradually lose its electrons
if enough energy is supplied to it. The energy required for single
ionization of some typical atoms is shown in Table 1.1. This ionization

TABLE 1.1 IONIZATION POTENTIALS

Element Ionization energy Ui for first electron, ev

Argon 15.68
Cesium 3.87
Hydrogen 13.53
Helium 24.46
Mercury 10.39
Potassium 4.32
Sodium 5.12
Oxygen 13.55
 1.1 INTRODUCTION 9

energy is given in electron volts, where 1 ev = 1.G02 X 10~19 joule.

One can easily find that kT = 1 ev when T = 11600°K; hence tempera¬
tures ina hot plasma are often given in electron volts, kilo electron volts
(103 ev), or million electron volts (106 ev), but they can be converted
to degrees Kelvin by multiplying the number of electron volts by the
factor 1.16 X 104.
For a gas in equilibrium at some temperature T one can calculate
the percent of the atoms that are ionized by using the Maxwell-
Boltzmann distribution law, derived in Sec. 5.3. This law states that if
na and nb are the number density of particles having energies of Ua
and Ub, respectively, then the ratio r = na/nb is given by

r = — = g- e-ui‘T U in joules; T in °Iv

Mb Qb
U, T in ev (1.1.8)

where U = Ua — Ub and g„ is the degeneracy factor giving the number of

states with the energy Up. For a system having only two energy levels
Ua and Ub, if we let a be the percent of all the particles that are in
the higher energy state a, then clearly

We now consider two special cases. First, if ga = gb, note that,

from Eq. (1.1.8), r increases with increasing temperature from 0 (when
T = 0) to 1 (when T —* °o). Hence a plot of a(T) has the form shown
in Fig. 1.3, where a increases gradually from 0 to 50 percent. (Higher
ratios of na to nb, which would occur if enough energy were forced into

a(%)

100
- 75
50
25

-3-2-1 0 1 2 3 T/U

Fig. 1.3 Percent of particles in an excited state as a function of

temperature when ga = gb-
10 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

this system and no higher energy states were available, correspond to

negative temperatures; however, these will not concern us here.)
The second case, which is the one that applies in the ionization
problem, occurs when the upper state is much more degenerate than
the lower, so that ga » gb, and, from Ecp (1.1.8),

r = Ce~UIT C» 1 (1.1.10)

From this equation we easily find that r = 1 (or a = 50 percent) when

T = Ti/2) where

U
T1/2 (1.1.11)
In C

Thus, as shown in Fig. 1.4, 50 percent of the atoms are in the upper
energy level when the temperature (in electron volts) is still well belotv
the excitation energy U.
Another interesting feature in this second case is that the percent
of particles in the excited state changes from nearly 0 to nearly 100
over a narrow temperature range, much as in an ordinary phase transition
from a solid to a liquid or from a liquid to a gas. A good measure of this
is the temperature difference AT that would exist between a = 0 and
a = 100 percent if the curve of a(T) were a straight line with the slope
of the true a(T) curve at T1/2. (Note the dotted curve in Fig. 1.4.)
From Fig. 1.4 it is easily seen that

Hence, from Eqs. (1.1.9) to (1.1.12) one easily finds that if C is assumed

Fig. 1.4. Percent of parti¬

cles in an excited state as a
function of temperature
with C = ga/gb » 1.
 1.1 INTRODUCTION 11

to be a slowly varying functionf of T,

Ar-w (LU3)
This result shows that if In C is sufficiently large, that is, ga » gb, the
curve of a(T) looks like a step function, with the transition from state
6 to state a occurring at 7\/2.
For the ionization problem that we are concerned with here, state a
is the ion-electron pair, state b is the neutral atom, and U is the ioniza¬
tion potential. Values for ga and gb are computed from quantum-
mechanical arguments^ and, in general, are quite complicated. How¬
ever, if the internal degrees of freedom of the particles are neglected,
one finds, to a good approximation, that

2-ir menT\31
C = g- = ni-
gb \ h2
2.405 X 1021T3/2nr1 T in °K

3.00 X 10"T^nr1 T in ev (1.1.14)

with rii the number of ions per cubic meter. When this expression is
used in Eq. (1.1.10), the result, known as the Saha equation, is clearly

r = - = 3 X 1027 T3<2ni-1e-uiT (1.1.15)

Plots of the percent of ionization of hydrogen as a function of

temperature are shown in Fig. 1.5 for values of nt = Hi + n„ ranging
from n( = 1025 particles/m3 down to nt = 1016 particles/m3. The first
of these, at room temperature, is the number density at a little below
atmospheric pressure, while the latter is a typical number density in a
high vacuum. Since C is large, significant degrees of ionization are
achieved for temperatures that are well below the ionization energy of
13.5 ev. In addition, since C is proportional to nr1, as we go to lower
number densities, the values of both rl\/2 and AT decrease significantly,
as expected from Eqs. (1.1.11) and (1.1.13).
From the figure it can be seen that high ion number densities are

t If the variation in C witli temperature is not neglected, one finds that AT is

given by the expression in Eq. (1.1.13) divided by the factor 1 + e, where e, which
is generally very small, is given by

Tm / dC\
6 [In (In C)]irm
C In C\df)Tut
J See, for example, W. B. Thompson, “An Introduction to Plasma Physics,”
pp. 17-20, Addison-Wesley Publishing Company, Inc., Reading, Mass., 1962.
12 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

Temperature, ev

Fig. 1.5 Percent ionization njnt for hydrogen as a function of tem¬

perature for several number densities int = ne + n„ in particles per
cubic meter).

possible at equilibrium in hydrogen (or other gases with comparable

ionization potentials) only at temperatures greater than a few thousand
degrees Kelvin. While so high a temperature is not easily attained in
an ordinary furnace, it can be achieved by electrical heating of a high-
density gas in a high-pressure-constricted electric arc, and for short
times in a shock tube, or in the flow field around a space vehicle during
reentry into the earth’s atmosphere. It is also evident from Eq. (1.1.11)
that the temperature required to obtain a high degree of ionization is
almostf directly proportional to U. Hence in a gas like cesium vapor,
where the ionization potential is less than one-third that of hydrogen,
a high percent of ionization can be obtained at temperatures below
1000°K. For this reason, although cesium vapor is highly corrosive
and difficidt to work with, it is used quite often for studies where a
steady homogeneous plasma is desired.

Nonequilibrium plasmas
Thus far we have discussed only those gaseous plasmas produced by
raising the temperature to a point where the equilibrium percent ioniza¬
tion is reasonably high. Although this accounts for many astrophysical
f The effect is a little more complicated than Eq. (1.1.11) suggests, since In C
is a slowly varying function of temperature.
 1.1 INTRODUCTION 13

plasmas, it is a relatively uncommon method for obtaining gaseous

plasmas in the laboratory. The more usual technique involves some
nonequilibrium process by which the degree of ionization is raised far
above its equilibrium value.
The most common example of this is the gas discharge, in which a
voltage applied across a gas, under suitable conditions, accelerates free
electrons to energies high enough to ionize other atoms in a collision,
thus rapidly creating an “avalanche” of electrons. Examples of gas
discharges include lightning strokes, sparks, commercial neon tubes,
arc discharges used for welding, mercury rectifiers, and various types of
laboratory dark, arc, and glow discharges. A notable feature of gas
discharges is that the electron temperature is almost always much
higher than the temperature of the heavy particles. This is due to two
factors, both of which are examined in later sections: (1) the energy to
produce the discharge is provided by the applied electric field which
heats light electrons in a plasma much more efficiently than heavy
ions; and (2) the energy transfer from the electrons to the heavy particles
is very slow, since the change in the kinetic energy of two colliding
particles is always very small if the collision partners are a light fast-
object and a heavy slow one. (Think of trying to speed up a slowly
moving bowling ball by hitting it with a ping-pong ball.)
Thus, as opposed to the condition of thermal equilibrium, where
the degree of ionization can be found from the Saha equation, in a gas
discharge we have an energy source which preferentially heats the elec¬
trons to a temperature on the order of 1 ev, while the ions and neutrals
are usually not too much above room temperature. Although these
figures are typical ones, they certainly do not cover all discharges. In
electric-welding arcs, electron temperatures of several electron volts are
produced, while the heavy-particle temperatures are about \ ev, and
still higher temperatures are obtained in plasma torches, i.e., plasma
jets produced by forcing a gas through an internal arc.
In addition to gas discharges there are nonequilibrium plasmas,
which are produced by radiation incident upon a gas. Perhaps the
most common example of this is the photoionization process that occurs
when an atom is ionized by absorbing a photon, whose energy hf equals
the ionization potential. Since the ionization potential for gases is
typically about 15 ev, the frequency range for photoionization is around

_ 15ev _ 3 6 x 1Q15 (1.1.16)

J h
which is in the ultraviolet band of frequencies, f Ultraviolet radiation
from the sun produces the relatively high degree of ionization in the
f These bands are more commonly classified by the wavelength \ of the^radi¬
ation, where X = c/f. Clearly, for this example, X = 8.4 X 10“8 m = 840 A.
14 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

ionosphere during the daylight hours, and when this radiation source is
“turned off” at sunset, the electron number density falls off gradually
to a much lower equilibrium value.
Lower-frequency optical or radio-frequency (r-f) radiation cannot
produce a plasma by photoionization, since each photon’s energy is
too small. However, a focused laser beam or a high-power r-f signal
has such a high electric-field strength that it can ionize a gas and produce
a plasma in the same manner as in an ordinary gas discharge. Ioniza¬
tion is also produced by x-rays or gamma rays, which have still higher
frequencies than ultraviolet. Although these high-energy sources are
not usually used to produce laboratory plasmas, their effect is utilized
in the ionization chamber, a device which determines the intensity of an
x-ray beam by measuring the number of ions created as the beam passes
through a gas in the chamber. In summary, there are a great many
ways to create a gaseous plasma. We have touched briefly on some of
the methods, and space prevents us from discussing still others, e.g.,
ionization produced by surface effects. For a full discussion of the
processes which produce gaseous plasmas the reader should consult one
of the many books dealing solely with this subject.f

Problem

1.1 Prepare a set of curves like Fig. 1.5 showing the percent ionization
of cesium as a function of temperature.

1.2 APPLICATIONS OF PLASMA PHYSICS

Having discussed the properties that define a plasma and some of the
many naturally occurring and man-made plasmas, we now want to note
some of the potential and (in some cases) existing practical uses for
plasmas. A quick list, chosen to give some idea of the wide range of
such applications, includes nuclear-fusion devices, magnetohydrody¬
namic (mhd) generators, plasma propulsion systems, thermionic con¬
verters, plasma amplifiers, gaseous lasers, arc jets, and fluorescent tubes.

Nuclear fusion

Among the items on this list the nuclear-fusion devices could have the
greatest impact from a practical standpoint, but they have also proved
to be the most difficult to develop. Nuclear fusion is the process
whereby two light nuclei combine to form a heavier one. In these
fusion reactions the number of protons and neutrons is conserved; but
when the initial nuclei are light ones (atomic mass number less than

f For example, G. Francis, “Ionization Phenomena in Gases,” Butterworth

Scientific Publications, London, 1960.
 1.2 APPLICATIONS OF PLASMA PHYSICS 15

20), the total final mass is a little less than the initial mass. Hence,
from Einstein’s familiar law, £ = me2, there is a large energy release.
Important examples of this are the reactions

He3 + n1 + 3.27 Mev about

H2 + H2 equal
H3 4--H1 4- 4.03 Mev probability (1-2.1)

H2 + H3 —» He4 + n1 + 17.58 Mev

H2 + He3 —> He4 + H1 + 18.34 Mev

where n1 is a neutron, H1, H2, H3 are the three isotopes of hydrogen,

and He3 and He4 are the two isotopes of helium.
The amount of energy released in these fusion reactions is very
high. For comparison, the energy released in a typical fission reaction
is about 200 Mev, but as the atomic weight of the fuel (U235 or Pu239) is
about 240, the energy release per unit mass is actually lower than in the
reactions in Eq. (1.2.1). A second example, which also illustrates the
enormous amount of power that will be available if a practical fusion
device is developed, is given by Bishop,! who notes that as much
energy is obtained from the fusion of the f g of deuterium (H2) in
1 gal of ordinary water as is obtained from the combustion of 300 gal
of gasoline.

The coulomb barrier

The basic difficulty in achieving fusion is that the process requires the
interacting particles to approach within a distance of order ICE14 m.
(The nuclear “radius” is generally taken to be 1.5 X 1CE15.A1/3 m, with
A the atomic mass number.) However, the incident particles, being
positively charged, repel each other in accordance with the coulomb-
force law. In order to estimate the kinetic energy the particles must
have before they can overcome this coulomb barrier, note that the
potential energy due to the coulomb interaction between two nuclei
(with atomic numbers! Z\ and Z2) is

ZxZ2e2 (1.2.2)
47re0A

When the distance between the particles R is 1(E14 m, one finds that
Uc = 1-44 X 105ZiZ2ev. Since the average kinetic energy of the parti¬
cles is §T (with T in electron volts), it is clear that fusion interactions
f See A. S. Bishop, “Project Sherwood,” Addison-Wesley Publishing Com¬
pany, Inc., Reading, Mass., 1958.
J Note that, for a given nucleus, Z is the number of protons while A is the
total number of protons and neutrons.
16 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

will generally not occur in great numbers until the temperatures come
close to 105 ev or 109°K. (Actually, since at any temperature, some
of the particles have energies well above average, some fusion occurs at
temperatures as low as about 4 X 107°Iv.)
This is borne out in Fig. 1.6, where the cross section for each of
the fusion reactions shown in Eq. (1.2.1) is plotted as a function of
energy. As might be expected from Eq. (1.2.2), other fusion reactions
involving nuclei with larger Z values require still higher energies to
overcome the coulomb barrier (despite their slightly larger radii).
Thus, as a result of the coulomb barrier, fusion reactions occur only
when the interacting particles have energies that are at least in the
10-kev range, which means that the plasma must have temperatures on
the order of 100 million degrees Kelvin. (For this reason this is usually
called a thermonuclear process.)

Heating and confinement of the plasma

This requirement of a very high temperature has three important

consequences:
1. Methods must be developed to heat the plasma to these high tem¬
peratures (using techniques less violent than a fission bomb).
2. The plasma must be confined without the use of a conventional
solid container (which vaporizes at temperatures of a few thousand
degrees Kelvin).

10

£Z
o

Fig. 1.6 Fusion cross sections

for the H2-H2, H2-H3, and
H2-He3 reactions as a function
1 of energy. Note 1 barn =
10-28 m2.
 1.2 APPLICATIONS OF PLASMA PHYSICS 17

3. Some care must be taken lest the energy loss due to radiation by
the fast-moving charged particles exceed the energy gain by fusion.

Each of these areas is explored extensively in the literaturef and

will not be covered in great detail here. However, among the methods
used to heat the plasma are:

1. Cyclotron and other r-f heating schemes, in which the energy comes
from an electromagnetic wave, usually at some particular resonant
frequency of the plasma where (as we shall see) energy absorption
is very high.
2. Shock heating, in which the plasma is compressed and heated by
a shock wave.
3. Magnetic heating (also called adiabatic compression), in which the
energy of the particles is increased by slowly increasing a magnetic
field in the plasma. A special case is the pinch effect, where the
magnetic field is the self-induced field due to a current in the plasma.

In addition various types of plasma jets or “guns” have been developed

in which particles are ejected with energies of several electron volts;
hence these jets, when injected into a fusion device, provide a partially
preheated plasma.
With this variety of techniques available, heating a plasma to
thermonuclear temperatures is quite feasible. However, confining the
plasma long enough for large numbers of fusion reactions to take place
presents more serious problems. In brief, since we shall see that
charged particles do not move easily across the lines of a magnetic
field, the confinement schemes all use some type of magnetic field (either
self-induced or externally applied). While a great many confinement
schemes have been suggested, most of them fall into one of the following
categories (shown in Fig. 1.7):

1. Pinch devices, where, as shown in Sec. 2.2, a current produces a

force tending to compress the cross-sectional area of the plasma
carrying it,J so that the plasma is both heated and confined
2. Mirror machines, which are linear devices with an axial magnetic
field to keep particles from the wall and magnetic “mirrors” of the
type described in Secs. 1.5 and 1.6 to reduce the number of charged
particles that escape at each end

f See particularly, D. J. Rose and M. Clark, Jr., “Plasmas and Controlled

Fusion,” The MIT Press, Cambridge, Mass., and John Wiley & Sons, Inc., New
York, 1961.
f In this category there are also inverse pinches, where the plasma is driven
outward by a current flowing through a conductor in the center of the plasma.
 18 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

3. Stellarator-type machines, where again the confinement of the

plasma is by an externally applied, largely axial field but with a
tube closed on itself (typically in a figure eight) to eliminate losses
of particles at the ends

h 01 obvious reasons these devices are also often characterized as either

open-ended or closed systems, depending on whether the tube closes
upon itself or not.
The major problem in the fusion program is that all these confine¬
ment schemes are, in one way or another, unstable, and small fluctua¬
tions (which are always present) from the desired configuration can
quickly lead to a rapid escape of the particles from their “magnetic
bottle.” Interestingly enough, since thermonuclear plasmas are highly
uuefied, their total heat content is fairly low; hence when the hot plasma
escapes and comes into contact with the surrounding vessel, the walls
of the vessel are not damaged, but the plasma itself is quickly cooled to
temperatures below the threshold for fusion.

Fig. 1./ Typical examples of the three basic confinement schemes, (a) Linear
pinch; (6) magnetic-mirror system ; (c) stellarator.
 1.2 APPLICATIONS OF PLASMA PHYSICS 19

Fig. 1.8 Interaction volume cleared by a

nucleus during the confinement time t.

The instability problem is a fundamental one and seems likely to

arise in any conceivable confinement scheme. However, under some
conditions the growth of the instabilities can be slowed down, hopefully
to the point where the confinement time is long enough for a practical
fusion device to operate. Progress in this direction is evident from two
examples.f (1) In the initial pinch experiments instabilities set in
within a few microseconds, and the plasma filament was quickly
quenched at the walls. However, by applying a strong magnetic field
in the direction of the current flow, the pinch was stabilized foi about
10~2 sec, a considerable improvement, though still not so long as one
would like. (2) While the magnetic-mirror devices initially confined a
plasma only for times of about 1 CP4 or 1CH5 sec, this figure was lengthened
to perhaps 0.1 sec using the fields produced by a complex array of
current-carrying conductors known (after their inventor, M. S. Ioffe) as
Ioffe bars. Unfortunately these experiments were all at very low num¬
ber densities, so that the product nr was still much too low for practical
purposes.

The nr criterion
For a practical device the confinement time r must be long enough
so that a significant percent of the fuel can undergo fusion. To estimate
how long this takes, consider the cylinder with cross-sectional area a,
where cr is the fusion cross section, and length vt, where v is the average
speed, v = (v2)112, of a nucleus. We may think of this cylinder as the
volume swept out by an average particle during the confinement time,
such that any other particle in this volume will undergo a fusion reaction
with the test particle. If n is the number density of the potential
collision partners, then the average number of particles in this volume
is 7iavr. Hence, to have a high probability of fusion we want

- ^ i (1.2.3)
ncrVT > 1

As noted earlier, o- is not significant unless the temperature is above

10 kev; in addition v also increases with temperature, since, as in
Eq. (1.1.4),

/3UrV/2 (1.2.4)
\ m J
f See L. A. Arzimovich, “Elementary Plasma Physics,” pp. 168-174, Blaisdell
Publishing Company, New York, 1965.
20 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

with m the nuclear mass. Hence if we assume that T — 300 kev, which
is about as high a temperature as seems feasible, then for the H2-H-
reaction, from Fig. 1.6 and Eq. (1.2.4), <r = 10~29 m2, and v = 6.5 X 106
m/sec so that the condition for a high probability of fusion [Eq.
(1.2.3)] becomes

nr > 1.5 X 1022 (i-2-5)

with n in particles per cubic meter and r in seconds. Note that in

making this calculation we have computed the probability of fusion for
a single particle moving with the average speed v. If we modify the
argument to take account of the fact that there are some particles
moving faster than v (and hence have a higher fusion cross section) and
some with a slower speed (and a lower cross section), the net effect is to
reduce this condition, but not by very much.
Values for n that are expected in a fusion device range from 1020
to 1024 particles/m3. Hence, from Eq. (1.2.5) confinement times
between 10-2 and 102 sec are adequate for a good percent of the H2
to undergo fusion, provided the plasma is heated to a high enough
temperature. In addition it is clear that both a high density and a long
confinement time are of equal importance; hence the product nr is often
used as a figure of merit for fusion devices.

Radiation losses

As noted earlier, in addition to the heating and confinement problems

some attention must be given in a fusion device to the energy loss by
radiation from the fast-moving charged particles in the plasma. Any
nonrelativistic accelerating charged particle radiates energy at the ratef

_dU_ = q1a1 .(1.2.6)

dt 67t e0c3

where q = charge, coul

a = acceleration, m/sec2
c = velocity of light
Although there are many factors tending to accelerate the charged
particles in a plasma, the two which are of predominant importance in
producing radiation are:

1. The acceleration of electrons in electron-ion collisions, which gives

rise to electron-ion bremsstrahlung radiation

t See for example, W. K. H. Panofsky and M. Phillips, “Classical Electricity

and Magnetism,” 2d ed., p. 359, Addison-Wesley Publishing Company, Inc.,
Reading, Mass., 1962.
 1.2 APPLICATIONS OF PLASMA PHYSICS 21

2. The centrifugal acceleration of electrons as they spiral in an applied

magnetic field, which gives rise to electron cyclotron radiation

Realistic calculations of the power loss due to radiation from any

given plasma are extremely difficult to make. Therefore we shall
merely discuss some crude estimates of the radiated power and of the
total generated powder to see how these compare. To start, a good
estimate for the power density of bremsstrahlung radiationf is

PT =* 1.5 X 10-38Z27iifte7y/2 watts/m3 (1.2.7)

with Te in electron volts. The frequency of the bremsstrahlung radia¬

tion covers a wide band but is typically on the order of the frequency

(1.2.8)

where again Te is in electron volts. Hence, for Tc in the 10- to 100-kev

range this radiation is in the form of x-rays, which are absorbed in any
solid outside the plasma.
It is less easy to estimate the total electron cyclotron radiation
since this depends, among other things, on the strength of the magnetic
field and the detailed reflection and transmission properties of the
plasma and its surroundings. However, the total power loss due to
cyclotron radiation can be comparable to, and in most cases even larger
than, that due to bremsstrahlung. Hence Eq. (1.2.7) provides only a
lower bound for Pr.
If a device is to be self-sustaining, the power generated by the fusion
reactions must exceed the power lost through radiation. We note that
the average power density generated Pg is given by

P0 = RfEf (1.2.9)

where Rf is the average number of fusion reactions per cubic meter per
second and Ef is the average energy release per reaction. To estimate
Rf note that, as in the discussion leading to Eq. (1.2.3), the probability
per unit time for a particle to have a fusion reaction is n,crv; hence, since
the number of particles per cubic meter is rq and each fusion involves
the reaction of two particles, if we neglect the statistical spread in the
velocities of the particles, we have to a good approximation

Rf = \n?av (1.2.10)

with a given in Fig. 1.6 and v found in Eq. (1.2.4).

t See for example, Rose and Clark, op. cit., p. 232.

 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS
22
Power P , watts/m

Fig. 1.9 Power generated

by H2-H2 and II2-H3 reac¬
tions and power radiated
by bremsstrahlung as a
function of temperature.
The curves are for n =
1.4 X 1021 particles/m3.
(Curves adapted from A. S.
Bishop, “Project Sher¬
wood”, p. 9, Addison-Wes-
ley Publishing Company,
Inc., Reading, Mass., 1958.)

For the example considered earlier of a deuterium (H2) plasma at

300 kev, we find| [from Eqs. (1.2.9) and (1.2.10) and the data after
Eq. (1.2.4)] that if m = 1.4 X 1021 nr3,

Pg = 4 X 107 watts/m3 (1.2.11)

showing that a moderate-sized fusion reactor could produce as much

power (100 to 1000 Mw) as a typical hydroelectric power station. For
this same example, from Eq. (1.2.7) the power radiated is easily found
to be

Pr = 1.6 X 107 watts/m3 (1.2.12)

which is clearly a sizable fraction of the total power output. Hence,

although these calculations are very rough, they suggest that a signifi¬
cant part of the power comes out in the form of bremsstrahlung x-ray

f A more accurate calculation taking account of the distribution in velocities

tends, as shown in Fig. 1.9, to about double P„.
 1.2 APPLICATIONS OF PLASMA PHYSICS 23

radiation and that, in addition, the radiation losses are a serious factor
in maintaining a self-sustaining fusion device. This is shown graphi¬
cally in Fig. 1.9, where PT is compared with Pg as a function of tempera¬
ture for both the H2-II2 and H2-II3 reactions. The crossover points
(where Pg exceeds Pr) are seen to be at about 4.5 kev for the H2-H3
reaction and 40 kev for the H2-H2 reaction.
Comparing the expression for Pg [Eqs. (1.2.9) and (1.2.10)] with
the bremsstrahlung-radiation loss shown in Eq. (1.2.7), we note that P0
is proportional to n,2 while Pr is proportional to Z2riine. Hence, for
a gas with no impurities, where ne = Znh the ratio of Pr to Pg is inde¬
pendent of ion number density, but if all else is equal, the radiation
loss is much more serious for the high-Z plasma than for the low. In
addition it is evident that, even for a hydrogen plasma, any high-Z
impurities will increase the bremsstrahlung radiation greatly. Thus,
most realistic fusion devices require the use of a hydrogen plasma of
such a high purity that this gas purification alone poses a major techno¬
logical problem, almost comparable to the heating and confinement
problems.

Problem

1.2 Calculate the total energy released if 1 g of deuterium is completely

converted to He4, H1, and n1 by having every He3 and H3 nucleon
formed in H2-H2 reactions also undergo a H2-He3 or H2-H3 reaction.

Mhd generators

A second important potential application of plasma physics is in the

conversion of thermal energy into electric energy by means of a magneto¬
hydrodynamic generator. While a rigorous discussion of this device
becomes quite involved, f the basic principle is quite simple and involves
the use of a perpendicular magnetic field to slow down a plasma jet,
with the resulting decrease in kinetic energy showing up as electric
energy that can be applied to an external load.
To see how this occurs, suppose, as in Fig. 1.10, that we drive a
plasma with a velocity v m/sec (in the x direction) across a magnetic
held of B webers/m2 (in the y direction). Then, if there is no external
load attached, so that no current can flow, i.e., we have an open circuit,
we find from the generalized Ohm’s law,% which we shall derive in Sec.
2.2, that there is a uniform electric field produced in the plasma with

t A detailed discussion can be found in G. W. Sutton and A. Sherman, “Engi¬

neering Magnetohydrodynamics,” McGraw-IIill Book Company, New York, 1965.
J The actual form for the law is J = <x(E + v X B), with a the conductivity
of the plasma. The term v X Bis often called the induced electric field, and for
the case of a generator it points in the direction opposite to E.
24 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

magnitude vB and pointing in the negative z direction. Now, as we

shall see in the next section, E, the electric field, is related to the
electrical potential <t> by the condition

— = —E = vB (1-2-13)
dz

Hence, upon integrating both sides from 0 to d, we find that the open-
circuit potential difference in volts between electrodes at z = 0 and
z = d is

4> = $(d) — <t>(0) = vBd (1.2.14)

When a resistive load is attached to the generator, as shown in

Fig. 1.10, the current flow is given by the familiar circuit form of Ohm’s
law

<f>
I = (1.2.15)
Rl + Rp

where Rl = load resistance

Rp = resistance of plasma, ohms
Hence the power delivered to the load (and extracted from the genera¬
tor) is just I2Rl, or

&Rl
Pl (1.2.16)
(Rl + Rp)2

which has a maximum when Rl = RP (as can be verified by setting

dPL/dRL = 0).
To calculate Rp recall that the electric resistance of a conductor

Fig. 1.10 Schematic diagrams for an mhd generator. Plasma flows in x direc¬
tion across magnetic field in y direction producing voltage <I> between electrodes
at z = 0 and z — d.
 1.2 APPLICATIONS OF PLASMA PHYSICS 25

is directly proportional to its length and inversely proportional to its

cross-sectional area and conductivity a (in mhos per meter). Thus,
from the figure,

_d_
RP = (1.2.17)
Aa

Hence the maximum power that can be delivered to the load is easily
found, from Eqs. (1.2.14), (1.2.16), and (1.2.17), to be

PL,m ax = ^ = fr'B'aV (1.2.18)

with V = Ad the total volume between the electrodes.

We should immediately note that this figure may often be much too
large because we have assumed that v, a, and B are constant, whereas
v, at least, decreases as the plasma flows through the generator, and
some of the plasma’s kinetic energy is converted into electric energy.
In reality the maximum available power is obviously the kinetic energy
that enters the generator per second. To calculate this quantity note
that all the plasma which enters the generator during a time interval At
lies initially in the volume element (see Fig. 1.11) with cross section
S and length v At. Since |pr2 is the kinetic energy available per unit
volume, the total power (energy per second) available is

Pin = iPv3S (1.2.19)

Clearly, it is not possible for Pi,,max to exceed 50 percent of this, since

from the circuit diagram in Fig. 1.10 the total power drawn from the flow
goes one-half to the load and one-half to internal heating of the plasma.
Moreover, if the power drawn is even a large fraction of the input
power, our analysis should be modified to take account of the variation
in v, a, and B along the generator.
Despite these qualifications, Eqs. (1.2.14) and (1.2.18) can still
be used to estimate the potential difference and power output of an mhd
generator. For example, in a gaseous plasma, with v = 103 m/sec,

Fig. 1.11 Mass flow into the generator.

26 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

B = 1 weber/m2 (or 104 gauss), a = 100 mho/m, d = 0.1 m, and

V = 0.1 m3, we find that 4> = 100 volts and PL,max = 2.5 X 106 watts,
which is comparable to the power output of a fusion reactor of the same
size. [However, the power input for this example is probably not
quite high enough to achieve this output, since, from Eq. (1.2.19), if
P = HE1 kg/m3 and S = 0.05 m2, Pin = 2.5 X 106 watts.]
In principle the mhd generator has great potential as a device for
generating electric power. From Eqs. (1.2.18) and (1.2.19) the basic
requirements are a high flow velocity and conductivity to maximize the
power output and a high density to ensure an adequate power input.
It is possible to create fairly dense high-velocity beams using gas dyna¬
mic nozzles like those on a chemical rocket. However, as we shall see
in Sec. 2.2, to have a high conductivity the gas must have a significant
degree of ionization.f At ordinary chemical-fuel temperatures of per¬
haps 2000°K, the percent ionization for an ordinary dense gas is
extremely low, as can be seen from the equilibrium curves for hydrogen
in Fig. 1.5. However, much work is being done using gases “seeded”
with elements like cesium that have a low ionization potential, so that
the total value for ne (and a) is somewhat higher.
A second practical difficulty is that the high power output from an
mhd generator is obtained by having a very high current through a
very low resistance. For the example we were using, from Eq. (1.2.17),

d2
Rl = Rr = 1(E3 ohm (1.2.20)
p V (j

while, from Ohm’s law,

<t>
I = 5 X 104 amp (1.2.21)
2/0

Thus, in any realistic application there would be the need for some
conversion equipment to put this power into a more usable form.

Plasma propulsion systems

In the mhd generator we have a device that converts kinetic energy

into electric energy by having a plasma driven across the lines of an
applied magnetic field. The reverse process, where electrical energy
is converted into mechanical energy, can also be accomplished by
applying both an electric and a magnetic field (perpendicular to each
other) across a plasma, as shown in Fig. 1.12. This process can be

f The formula is shown in Eq. (2.2.2) to be a = nee2T/me with r the average

time a free electron moves between collisions with an ion or neutral.
 1.2 APPLICATIONS OF PLASMA PHYSICS 27

Fig. 1.12 Schematic diagram

for a crossed-field plasma pro¬
_L x (direction of vD
pulsion device.

examined from two points of view. First, if we neglect collisions, then,

as we shall see in Sec. 1.4, an electric held E (in the negative z direction)
and a magnetic held B (in the y direction) produce a drift velocity vD
of E/B m/sec (in the x direction) on any charged particle, regardless
of its mass, in addition to a circular cyclotron motion about the lines
of the B held. Hence, by making E/B sufficiently large one can
obtain a plasma beam with a high enough velocity to act as an ordinary
rocket-propulsion system, f
For example, if E = 104 volts/m and B = 0.1 webers/m2 (or
103 gauss), vD is 106 m/sec, which is two orders of magnitude higher
than the exit velocity of a typical chemical rocket.
To obtain this large drift velocity, however, the charged particles
must have no collisions; hence the gas must be rarehed, and the
momentum density mnv of the beam is very small. To treat somewhat
denser gases, one must use the magnetohydrodynamic equations
(developed in Sec. 3.7), which show that when there is a current
density J flowing in the negative 2 direction and a magnetic field B
in the y direction, there is a force F = JB which acts on the fluid in
the x direction. Thus, for a dense gas, too, a high velocity can be
developed if a large current is maintained perpendicular to the applied
magnetic field.
While it is clear that a plasma propulsion system is conceivable,
one must examine the dynamics of a rocket to see why plasma propul¬
sion systems are in fact very attractive as a device for long space
flights. To analyze a rocket once it is beyond the earth’s gravitational
field, we use conservation of momentum to equate the momentum of
the rocket exhaust gas expelled to the rear per unit time with the force
driving the rocket forward. If M(t) is the mass of the rocket, then
— M, the rate of decrease of M, is the mass expelled per unit time. In
addition, if v is the velocity of the exhaust gas relative to the rocket,

f This result suggests that vD can be made as large as desired by letting

—> 0. However, to avoid collisions with the walls the radius of gyration |a| of
the charged particles about the B field must be small. Since the radius of gyration
will be shown [in Eq. (1.4.21)] to be inversely proportional to B, large magnetic-
field strengths (of order 1 weber/m2) are required to keep [a| small.
28 MOTION OF A CFIARGED PARTICLE IN ELECTROMAGNETIC FIELDS

then the actual velocity of the exhaust gas is v — u, with u{t) the
velocity of the rocket. Hence the force acting on the rocket is
— M(v — u), and from Newton’s law we have

4Mu = —M(v — u) (1.2.22)

or

u = (1.2.23)

In general v is a constant ; hence the solution to Eq. (1.2.23) is easily

found to be

» In (1.2.24)

with Uo and M0 the initial values of the velocity and mass of the rocket
and uf and Mf the final values. This result shows that a plasma
propulsion system (with v perhaps 100 times as large as in a chemical
rocket) can attain a given final velocity using a much smaller percent
of the mass for fuel than would be possible with a chemical rocket.
A second factor in the design of a rocket-propulsion system is the
ratio a of the final mass to the power output, that is,

Mf 2M i
a or M = (1.2.25)
iiu2 av‘

To introduce this factor into the expression for the final velocity, note
that if a rocket burns for a time t0, then

M o = Mf (1.2.26)

Hence, upon replacing M0 with this expression in Eq. (1.2.24), we have

(assuming u0 = 0),

uf = v In ^1 + (1.2.27)

with V2 = 2to/a.
 1.2 APPLICATIONS OF PLASMA PHYSICS 29

Fig. 1.18 Plot of final rock¬

et velocity as a function of
exhaust velocity.

For a fixed value of V, uf has the form shown in Fig. 1.13, where
it can be seen that uf has a maximum of about 0.8F, which is obtained
when v = 0.5F. Note that under these optimal conditions, from
Eq. (1.2.26), Mo/Mf = 5. Hence 20 percent of the initial mass can
be used for the electric propulsion equipment and the payload.
An important practical factor involved in plasma rocket-propul¬
sion systems is that they achieve their high final velocity by means
of a small acceleration over a long time period. Hence, the force they
provide is much too modest to overcome the gravitational force close
to earth, and chemical rockets must still be used as the first stage of
any plasma propulsion system (see Prob. 1.3).

Other plasma devices

In addition to the fusion, mhd generator, and plasma propulsion sys¬

tems, there are a number of other plasma devices which should be
mentioned. The thermionic converterf uses a cesium plasma between
two electrodes, one heated to emit electrons and the other cooled.
Because of the cesium plasma very high currents are produced, and a
significant portion of the thermal energy applied to the cathode is
extracted as electric energy. Also under development are plasma
amplifiers (using both solid-state and gaseous plasmas) for amplifying
electromagnetic signals. For the most part these devices use the fact
that a plasma beam passing through a stationary plasma (or two con¬
trastreaming plasmas) has inherent instabilities at some frequencies.
Hence, as we shall see in Sec. 4.7, a small applied signal at an unstable
frequency can grow rapidly as it passes through the plasma, the energy
for the growth coming from a slowing down of the beam.
Still other plasma devices include the gas lasers,% used to produce

f See, for example, S. W. Angrist, “Direct Energy Conversion,” Allyn and

Bacon, Inc., Boston, 1965.
t See, for example, G. Birnbaum, “Optical Masers,” Academic Press Inc.,
New York, 1964.
30 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

high-intensity coherent-light signals; arc jets, which provide tempera¬

tures two or more times as high as the hottest gas flames and which
can be used to melt metals like tungsten, carbon, or molybdenum as
well as for cutting or welding; the ordinary fluorescent tubes used for
illumination and for signs; and a number of specialized tubesf like the
thyratron, a grid-controlled thermionic arc-type rectifier, and the
ignitron, which is used for switching.
To summarize, there are an enormous variety of natural and man¬
made plasmas whose understanding and application require some
knowledge of plasma physics. It is this fact which justifies the chal¬
lenging task of trying to formulate and solve a set of equations which
will adequately describe the behavior of a plasma.

Problem

1.3 Prove that the initial acceleration on a rocket is

v(Mo - Mf)
-u(O)
M oto

Thus, for a chemical rocket, with v = 103 m/sec, t0 = 1 min, and

Mf « M0, w(0) = 16 m/sec2, which exceeds g = 9.8 m/sec2, the
acceleration due to the earth’s gravity. In contrast, compute
ii(0) for an optimal plasma propulsion system with v = 105 m/sec
and tQ = 100 days.

1.3 REVIEW OF VECTOR ANALYSIS AND ELECTROMAGNETIC THEORY

In this section we shall introduce the basic mathematical tools and the
fundamental equations of electromagnetic theory needed ■ in our
analysis of plasmas.

Coordinate systems

We shall first review the notation employed throughout the text. We

shall deal with orthonormal coordinate systems, where a!, a2, and a3
represent unit vectors pointing in three mutually perpendicular direc¬
tions. The coordinates are denoted, in general, as uh u2, and u3, and
the linear differential elements in the three directions are hi dui ai,
h2 du2 a2, and /i3 du% a3. For the three standard coordinate systems,

f See, for example, Cruft Electronics Staff, “Electronic Circuits and Tubes,”
McGraw-Hill Book Company, New York, 1947.
 1.3 REVIEW OF VECTOR ANALYSIS AND ELECTROMAGNETIC THEORY 31

rectangular, cylindrical, and spherical, these differential elements are

given by

Coordinate system hi dui hi dui hz du3

Rectangular (x,y,z) dx dy dz
Cylindrical (r,<j>,z) dr r dct> dz
Spherical (R,8,4>) dR R d9 R sin 9 dcf>

The following relationships among the coordinates x, y, z, r, <j>, R, and

9 are easily inferred from Fig. 1.14:

/?2 = r2 _J_ z2 y.2 — x2 _|_ y2

z = R cos 6 r = R sin 9 (1.3.1)

x = r cos <p y = r sin <j>

Vector operations and the summation convention

The vector A may be written in terms of its three components

A = Aiai + A1&2 + dUa3 = (1.3.2)

On the right we use the summation convention, whereby any repeated

subscript that denotes a direction is automatically summed. Hence

3
A,a, = J A&i (1.3.3)
i= i

There are two types of products of vectors. The scalar, or dot, product

A • B = AiBi (1.3.4)

~ <i>
Fig. 1.14 Angles and lengths used in
cylindrical and spherical coordinates.
32 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

and the vector 'product

ai a2 a3
A X B = Ai A2 Az
Bz b2 Bz

If a is the angle between A and B, then one finds that

A • B = AB cos a |A x B| = AB sin a (1.3.6)

Hence, if A and B are parallel, their cross product is zero, while if A and
B are perpendicular, their dot product is zero.

The operator V and some vector theorems

The operator del V is defined by the following three operations. First,
for </> any scalar function, the gradient of 4> is defined as

ai d4> a2 d<j> a3 d<j> (1.3.7)

h\dui hi diii h3duz

Next, if A is any vector function, then the divergence of A is

V • A = iii" (z— hihzA]i + -r— h\h$A 2 + -— h\hiAzj (1.3.8)

h\hihz \du\ OUi OUz /

and the curl of A is

h iai h2 a2 hz&z
1 d d d
(1.3.9)
hih2h3 dui dlh duz
h\A i hiA 2 hzA

Two theorems (which we merely state without proof) follow from

these definitions:

Gauss’ theorem: f V • A dY = <£ A • dS (1.3.10)

J J closed surf

and

Stokes’ theorem: / v x A • dS = A • d1 (1.3.11)

 1.3 REVIEW OF VECTOR ANALYSIS AND ELECTROMAGNETIC THEORY 33

In Gauss’ theorem dV denotes a volume element while dS is a surface

element on the surface completely surrounding the volume in the
integral on the left. By convention the direction of dS is always
chosen to point out of the volume, as in Fig. 1.15. In Stokes’ theorem
the surface in the integral on the left is open, while d 1 is a line element
on the circumference of the surface. The directions for dS and dl are
chosen according to the right-hand convention illustrated in Fig. 1.15.
(Thus, if dS points toward an observer, the line integration proceeds
in a counterclockwise direction.)

Problems

1.4 Derive a general formula for the laplacian V24> = V • (V<£).

1.5 Prove that V x (V<£) = 0 and V • (V x A) = 0.
1.6 Show that V • R = 3 and v X R = 0.

Maxwell’s equations and charge conservation

The fundamental equations which apply to electromagnetic phenomena

are the Maxwell equations which, in inks units, have the form

VXE=—B V x H = D + J
(1.3.12)
V • D = Pc V • B = 0

where E = electric-field intensity, volts/m

H = magnetic-field intensity, amp-turns/m
D = displacement current, coul/m2
B = magnetic induction, webers/m2
J = current density, amp/m2
pc = charge density, coul/m3
Note that the dot above a quantity is used to indicate a partial deriva¬
tive with respect to time. For obvious reasons the first pair of equa¬
tions are often called the Maxwell curl equations while the remaining
two are called the divergence equations.

Fig. 1.15 Direction associated with clS in Gauss’ and Stokes’

theorems.
34 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

For a medium containing several species of charged particles, the

charge density pc and current density J are given by the following
intuitively obvious definitions:

Pc = Xna9a (1.3.13)
a

and

J = lnaqaua d-3.14)
a

where na = number density of particles of type a, particles/m3

qa = charge of particles of type a, cord
u« = mean velocity of particles of type a, m/sec
If we take the divergence of the Maxwell equation involving
V x II and replace V • D by pc, note that we have

0 = pc + V • J (1.3.15)

This result is known as the charge-conservation equation, since it can also

be derived by equating the current which flows out of a volume with
the decrease in the total charge contained therein. To find the total
current flowing out of the volume in unit time we integrate J • dS over
the entire surface area; thus, using Gauss’ theorem, we have

Current out in unit time = J J • dS = /V • J dV (1.3.16)

The total charge contained in the volume is JpcdV. Hence, for the
volume element

Charge decrease in unit time JpcdV (1.3.17)

Equating these two quantities, we have

J (pc + V • J) dV = 0 (1.3.18)

which can hold for all volumes only if Eq. (1.3.15) is true.

Vector and scalar potentials

From the second of the divergence equations in (1.3.12) it is clear that

one can let

B = V x A (1.3.19)
 1.3 REVIEW OF VECTOR ANALYSIS AND ELECTROMAGNETIC THEORY 35

since V • B is then identically zero. A is called the vector 'potential.

We also have [from the first curl equation in (1.3.12)]

V x E = —B = — V X A (1.3.20)

so that

V x (E + A) = 0 (1.3.21)

Hence, in similar fashion we may let

E + A - -V4> (1.3.22)

since V x (V4>) is also identically zero. $ is called the scalar potential.

The minus sign in Eq. (1.3.22) is a convention which leads, as we shall
see, to 4>’s being the potential energy per unit charge for a stationary
field.

The constitutive equations

For many media I) and B are related to E and H by the simple con¬
stitutive equations

D = eE B = mH (1.3.23)

where e = dielectric constant, farads/m

p = magnetic permeability, henrys/m
The values for e and p depend on the medium, and they may be treated
either as empirical constants or as quantities that are computed from
an analysis of the individual particles in the medium.
For anisotropic media, the relationships in Eq. (1.3.23) may break
down, since 1) and E (or B and II) may lie in different directions. If
there is nonetheless a linear relationship between D and E, then we
still have for any component Dl

Di = tnE\ T" tiiEi + tizEz — tijEj (1.3.24)

In matrix notation this equation has the form

D = e • E (1-3.25)

where e is a 3 X 3 matrix with elements e,-y and is called the dielectric

tensor.
For some media the relationship between I) and E (or, more com¬
monly, between B and II) is not even linear, so that e and p are func-
36 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

tions of E and H. However, in this text, while we shall often deal

with dielectric tensors, we shall not have to worry about nonlinear
media.

Poynting’s theorem
To conclude this brief review of Maxwell’s equations, we shall derive
the Poynting theorem, which is an energy-conservation law for electro¬
magnetic fields. From the two Maxwell curl equations note first
that for a simple isotropic medium with n and e independent of time

H V x E = -II B = — ^-AyH2 (1.3.26)

ot

EVxH-ED+JE = ^ ^E2 + J • E (1.3.27)

(Note that, for any vector A, A • A = A2.) There is a vector identity

[which follows easily from Eqs. (1.3.4), (1.3.5), and (1.3.9)] which
states that

E-VxH-H VxE = -V-(ExH) (1.3.28)

Hence, if we subtract Eq. (1.3.26) from Eq. (1.3.27) and integrate

over a fixed volume, we find [after using Gauss’ theorem (1.3.10)]

- j (E x H) • dS = ~ f (}eE2 + iM//2) dV + f J • E dV (1.3.29)

This result is known as Poynting’s theorem, and the three terms all
have important physical interpretations. The one on the left is the
power flow into the volume (since —dS points into the volume); the
first term on the right is the rate of increase of the electric- and mag¬
netic-field energy in the volume (\tE2 and ^nIP are, respectively, the
electric- and magnetic-field energy densities); and the last term on
the right is the dissipative power loss in the volume. The vector on
the left, E x II, which is involved with the power flow of the electro¬
magnetic field, is called the Poynting vector.

1.4 MOTION OF A CHARGE IN UNIFORM FIELDS

The Lorentz force on a point charge

In the remainder of this chapter we shall be concerned largely with

the behavior of a single charged particle in electric and magnetic fields.
 1.4 MOTION OF A CHARGE IN UNIFORM FIELDS 37

The forces acting on such a particle are intimately related to the very
definitions of E and B. E is defined in terms of the force Fe produced
on a charge distribution placed in the field

Fe = JpcE dV (1.4.1)

If the test charge is very small, so that E is essentially constant over

the volume where pc ^ 0, then clearly

Fe = 'EjpcdV = qE (1.4.2)

with q = Jpc dV the total charge.

In similar fashion B is defined in terms of the force Fm produced
on a current distribution placed in the field

Fm = JJ x B dV = Jpcii X B dV (1.4.3)

Again, if the test charge is small, so that u and B are constant over the
volume where pc ^ 0, then

Fm = u X B\pcdV = qu X B (1-4.4)

From Eqs. (1.4.2) and (1.4.4) we conclude that the force on a point
charge (with velocity v) placed in an electromagnetic field is given by

F = q(E + v x B) (1-4.5)

This is called the Lorentz force on a charged particle.

Energy conservation in a stationary field

From Newton’s law, if the Lorentz force is the only force acting on a
charged particle, we have

m ^ = g(E + v X B) (1.4.6)

Taking the dot product of v and each side of this equation, we have

d (1-4.7)
^mv2 qx • E
dt
38 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

since, for any vector a, a • (a x b) = 0. Now for a stationary field

Eq. (1.3.22) shows that E = — V4>. Hence we have

d_ /d<i> dx d$dy d4> dz\

-i mv2 (1.4.8)
dt ^ \d.r dt dy dt dz dt)

or

({m»2 + qd?) = 0 (1.4.9)

Hi

This shows that the sum of the kinetic and potential energies stays
constant (with 4> the potential energy per unit charge).

Formal solution for a particle in constant E and B fields

We now consider the actual solution to Eq. (1.4.6) when E and B are
constants. Although this problem is rather simple, we shall treat it
in great detail, since many of the more complicated situations are
treated as perturbations from this problem. Choose the z axis to lie
in the B direction, so that B = Baz and E = Exax + Eyay + Eza,.
(In practice one can arrange this situation in the laboratory by wrap¬
ping a coil of wire around a tube containing a pair of large planar elec¬
trodes, as shown in Fig. 1.16.) The Lorentz equation (1.4.6) for this
case reduces to

i)x = I <«. + **>

Vy = £ (Ey - Bvx) (1.4.10)
m

Vz =

Fig. 1.16 Experimental configuration to obtain crossed uniform E and

B fields.
 1.4 MOTION OF A CHARGE IN UNIFORM FIELDS 39

Solving for vz, we readily find

Vz — — E zt + vz(0) (1.4.11)
m

where vz(0) is vz at t = 0. Thus the magnitude of vz increases without

limit unless Ez = 0; moreover, if Ez does equal zero (as in Fig. 1.16),
the particle drifts in the z direction with its initial velocity. (When
collisions of our charged particle with other particles are taken into
account, however, we shall see that these results are altered.)
To solve for vx and vV} take a time derivative of the vx equation
and substitute in for vy, to obtain

(1.4.12)

where coc = qB/m. Note that |coc| is called the gyrofrequency or

cyclotron frequency of the charged particle. For an electron, where
q = — 1.602 X 10-19 coul and m = 9.109 X 10~31 kg, |wc| = 1.76 X 10UB,
with B in webers'per square meter. Thus, in the earth’s magnetic field
(B = 0.5 X 10-4), the electron cyclotron frequency is about 107 rad/
sec, while in a typical laboratory field (B = 1), |coc( is in the microwave
range of frequencies.
Equation (1.4.12) is a simple inhomogeneous harmonic-oscillator
equation. The homogeneous solution is of the form U sin (coct — 4>),
with U and <£ the integration constants, while a particular solution is
clearly vx = Ey/B. Thus the complete solution is

(1.4.13)

To find vy we substitute this result for vx directly into the first part of
Eq. (1.4.10), to have

Vy

1.1.11

Thus, the velocities in the plane perpendicular to B involve oscilla¬

tions at the cyclotron frequency plus a drift velocity, which we can
write as

E x B
Vd = (1.4.15)
B2
40 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

Note that this drift velocity is independent of q and m. Thus all the
charged particles will drift along together in this case, and the net
current flow due to the drift motion of a collection of charged particles
is [from Eq. (1.3.14y J

Jr, = ^naqavD = pc\D (1.4.16)

Thus, there is no drift current for the (usual) case when the net charge
density in the plasma is zero, i.e., when the total charge densities of
the positive and negatively charged particles balance out.

Determination of the path

If we have only the B field, the solutions for v reduce to

vx = U sin (cod — <f>) vy = U cos (cod — <j>) vz = v*(0) (1.4.17)

The actual solution for the path of the particle is obtained by integrat¬
ing each of these equations and applying the initial conditions on the
velocity and position of the particle. Let us choose our coordinates
so that initially we have

Wx(0) = 0 Vy( 0) = To i>*(0) = 7,

To (1.4.18)
X(0) = -a 7(0) = 0 Z( 0) = 0 a =

with X, Y, and Z the position coordinates of the particle. With these

initial conditions we easily find

vx = Fo sin coct vy — Vo cos cod vz = V\

(1.4.19)
X = — a cos coct Y = a sin cod Z = Vxt

If we switch to cylindrical coordinates, we have

r = (X2 + F2)1/2 = |a|

Y
</> = tan-1 — cod (1.4.20)
X
Z = V it

Thus the particle moves in a helical path. The radius |a| is called the
cyclotron radius or radius of gyration, and, as noted above,

F0 _ V0m
a (1.4.21)
w, qB
 1.4 MOTION OF A CHARGE IN UNIFORM FIELDS 41

so that a increases with the mass and velocity of the particle but
decreases with the charge and the held strength. In addition, from
the expression for 0, note that $ increases with time for a negative
charge (wc < 0) and decreases with time for a positive charge. Hence,
as shown in Fig. 1.17, for an observer looking into the lines of a B held,
the positive charge spirals clockwise and the negative charge counter¬
clockwise. (Alternatively, if one’s thumb points in the direction of B,
the positive charge spirals like the fingers of the left hand and the nega¬
tive charge like the fingers of the right hand.)
When a perpendicular E held is present, the path of the particle is
obtained by integrating Eqs. (1.4.13) and (1.4.14). The exact orbit
depends strongly on the initial velocity. However, it is clear that the
path consists of both the spiraling cyclotron motion and the steady
drift in the direction perpendicular to E and B, as shown, for a typical
example in Fig. 1.17.

Problems

1.7 Plot the orbit for a particle when E = Eay and initially the
particle is located at the origin with zero velocity. Note the
difference in path for a positive and negative charge.
1.8 Plot the radius of gyration for an electron as a function of held
strength when F0 is the “thermal velocity” («T/m)1/2 and T has
the values 10-2, 1, 102, and 104 ev.

Drift in a gravitational field

The effect of a gravitational field upon the motion of a particle is, in a

formal sense, entirely analogous to the effect of an electric held. The

Fig. 1.17 Typical orbits for charged particles in crossed electric and
magnetic fields. Curves on the left are for E = 0.
42 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

gravitational force can be written as F9 = mg, with g the acceleration

due to gravity: hence the equation of motion for a particle in magnetic
and gravitational fields is

= q{\ X B) + mg (1.4.22)

This is exactly like Eq. (1.4.6) for the motion in magnetic and electric
fields except for the change from the electric force qYL to the gravita¬
tional force mg. Thus, the gravitational force is entirely equivalent
to an electric field mg/q acting on the charge. In particular, where
the drift velocity for E and B fields was shown to be E X B/B2 we now
have for uniform g and B fields

_ m(g x B)
(1.4.23)
qB2

Note that for this case the drift is in opposite directions for electrons
and positive ions and is directly proportional to the mass. Thus the
current density produced by the g and B fields is

njiUjig X B) p(g X B)
Jo = ^ "Ab = ^ B2 B'2
(1.4.24)

with p the mass density of the charged particles.

SUMMARY

In these two sections we have seen that the motion of a charged

particle in a uniform magnetic and electric (or gravitational) field (neg¬
lecting interparticle effects of all kinds) consists of three components:

1. A rotation about the B direction at the gyrofrequency |oc| and

with a radius |F0/wc|.
2. A drift velocity, \D = (E x B/B2) + m(g X B)/qB2.
3. A constant acceleration in the B direction unless g and E are
perpendicular to B. (In this last case the particle drifts in the
B direction with its initial velocity.)

Because of the steady rotation at the cyclotron frequency, there is at

any instant a center of rotation about which the particle is spiraling.
This is often referred to as the particle’s guiding center, and the path
that a particle would have if we neglected its cyclotron gyrations is
the path of the guiding center.
 1.5 MOTION IN A NONUNIFORM B FIELD 43

Problems

1.9 Calculate the gravitational-drift current at 300 km where there

are 1.8 X 1012 0+ ions and electrons per cubic meter and 5 = 1
gauss. You may assume g perpendicular to B.
1.10 How large must g be before the gravitational drift on an electron
equals the electric-field drift due to a field of 10 volts/m?

1.5 MOTION IN A NONUNIFORM B FIELD

Description of the field

We want to examine the motion of a charged particle in a steady

inhomogeneous magnetic field B(r), when the region where the particle
moves has E = J = 0. Before beginning, we note that the magnetic
field itself must satisfy Maxwell’s equations, which for this caset have
the form

V • B = v X B = 0 (1.5.1)

From the second part of Eq. (1.5.1) we may let

B = V<hm (1.5.2)

with 4>m the magnetic scalar potential; while from the divergence condi¬
tion we have

V2<t>m = v • B = 0 (1.5.3)

Thus the magnetic field in which the particle moves is not completely
arbitrary but must, in fact, be the gradient of a solution to Laplace’s
equation.
As an example, suppose we let

Bz = 50(1 + ax) (1.5.4)

with \ax\ small compared to unity.J Thus Bz is nearly uniform, but

it does have a small gradient, which we choose to put in the x direction.
Note, however, that B z alone does not satisfy the condition V x B = 0;
hence, together with the small gradient in Bz we must add a small
curvature to the field lines. The simplest extra component we can
f Note that for this case B is produced solely by external sources, and the
effect of any fields produced by the particles is assumed to be negligible.
t It should be noted that in this section we shall analyze the motion of a
particle near the origin. Hence x, y, and z will all be close to zero.
44 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

introduce to satisfy the curl condition is to let Bx = B0az, so that the

total magnetic field strength is

B = B0[azax + (1 + ax) az\ (1.5.5)

where, to assure that B is still nearly uniform, we require that \az\ also
be small compared to unity. The x component of B we shall call the
curvature term', the ^-dependent part of Bz we call the gradient term.

Problem
1.11 Show that the following magnetic fields violate Maxwell’s equa¬
tions: B = B{z) az; B = B(x) a2.

A digression on the magnetic field lines

The reader interested solely in the motion of a particle in the magnetic

held given by Eq. (1.5.5) may skip directly to Eq. (1.5.14), where we
substitute this expression for B into the Lorentz equation and solve
for v. However, in order to interpret one of the terms in the resulting
motion, it is useful to digress briefly and determine the path of the
magnetic field lines themselves. To do this we note that the held lines
are defined as being perpendicular (or orthogonal) to the curves of con¬
stant potential.f Hence, our procedure is to first find the set of equa¬
tions defining the magnetic equipotential surfaces and then to deter¬
mine the set of orthogonal curves for the held lines.
To find the equipotentials, note that since B = V4>m, we have,
from Eq. (1.5.5),

<f>m = B0z( 1 + ax) + const (1.5.6)

Hence the equipotentials are dehned simply by letting B0z( 1 + ax)

equal some constant. To plot the held lines, we must find the set
of functions T that are orthogonal to <f>m. Fortunately, as this is a
two-dimensional problem, T is given directly by the Cauchy-Riemann
equations

cbk dT d4>
^ = ~aB0z (1.5.7)
dx aT = B"(1 + al) dz dx

From the first of these we have

d' = B0x( 1 + ±ax) + Ffiz) (1.5.8)

f See for example, J. A. Stratton, “Electromagnetic Theory,” pp. 160-162,
McGraw-Hill Book Company, New York, 1941.
 1.5 MOTION IN A NONUNIFORM B FIELD 45

while from the second we have

T = ~±aB0z2 + Fs(x) (1.5.9)

with Fi(z) and Fo(x) “arbitrary” functions. Comparing Eqs. (1.5.8)

and (1.5.9), however, it is clear that

'h = B0[x(l + ±ax) — Aaz2] + const (1.5.10)

Hence, the equation for each field line has the form

x(l + Yax) ~ -kaz2 = const (1.5.11)

For the field line passing through the origin (x = z = 0), the
constant in Eq. (1.5.11) must be zero, so that near the origin

z2 = — (1 + iax) (1.5.12)
a a

where in the last step we use the assumption that |ax| « 1. We can
show that, to a good approximation, near the origin this is the arc of a
circle with radius a-1 and centered at x = a-1, z = 0. The equation
for such a circle is

(x — a-1)2 + z2 = or1

or, restricting attention to that part of the arc near the origin,

z2 = —— (1 tax)**— (1-5.13)
a a

Hence, as long as \ax\ <5C 1, this circle has the same equation as the
field line passing through the origin.

Problem

1.12 Verify that, to a good approximation, near the origin the equipo-
tential lines are radii of circles centered at x = a-1, z = 0 as shown
in Fig. 1.18.

Velocity of the particle when B has gradient and curvature

The motion of a particle in this B field can be found by making a

perturbation expansion in a. The Lorentz equation for the particle is,
46 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

from Eqs. (1.4.5) and (1.5.5),

vx — wcvy(l + olt)

iiy = —<j>c[vx + a(xvx — zvj)} (1.5.14)

qB o
VZ = —GJcOLZVy
m

where, in order to keep their effects separate, we have underlined those

terms arising from the curvature term in B. Since all the terms pro¬
portional to a are small, the solutions for the path and velocity should
approximately equal their values when a is zero. Therefore we proceed
in the following steps:

1. Obtain the zero-order solution (the solution when a = 0).

2. Substitute the zero-order solution into those terms which con¬
tain a.
3. Repeat the problem, to obtain a first-order solution.
4. Check to make certain that the perturbations on the zero-order
result are indeed small.

Step 1 has already been done. We recall that for a = 0 we used

the initial conditions

t>*(0) = 0 *b(0) = Vo vz(0) = V i

X(0) = -a 7(0) = 0 Z( 0) = 0 a

a -l

Equipotentials

Field lines

Fig. 1.18 Equipotentials

and field lines near the
origin.
 1.5 MOTION IN A NONUNIFORM B FIELD 47

and found in Eqs. (1.4.18 and 1.4.19) that

vx° = V0 sin wct vy° = Vo cos wct vz° = Vi

(1.5.16)
A0 = —a cos u>ct Y° = a sin wct Z° = V it

where the superscript zero is used to emphasize that these are the
zero-order results. Substituting these results into the terms propor¬
tional to a, we have

vx = wc(vy + avy°X°) = coc(vy — aV0<x cos2 UCt) (1.5.17)

Vy = —Wc[vx + a( X°vx° — Z°vz0)]
= — uc[vx — a(±aVo sin 2uct + Vi2t)] (1.5.18)
vz = —cocaZ°Vy° — — o>caV0Vit cos wct (1.5.19)

To find vx and vy we again take a time derivative of Eq. (1.5.17) and

replace vy by Eq. (1.5.18) to have

vx + uc2vx = ao)c2(faT0 sin 2uct + V 2t) (1.5.20)

This is again a simple inhomogeneous harmonic-oscillator equation,

with the general solution U sin (coct + </>). Adding to this a particular
solution which matches the terms on the right, we have

vx = U sin (a)ct + </))— ^aaVo sin 2uct + aV\2t (1.5.21)

with U and 0 the integration constants to be determined by the initial

conditions. From Eq. (1.5.17) we can now calculate vy directly

vy = — -f- 4aaFo(l ~V cos 2oict)

uc

TT 2
= U cos (o:ct + 0) H-— + ^aa\ o(l — cos 2a>cf) (1.5.22)

Applying the initial conditions on vx and vy [Eq. (1.5.15)], we find

0 = 0 and U = Vo — aVi2/coc, so that

(T/ aVV\ . . (1.5.23)

vx = ( Vo-) sin cact ±aaV0 sill 2aoet + aV i2t
\
48 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

and

(
F0 — ) cos wct + ^—|- -^aaV0( 1 — cos 2wct) (1.5.24)

Finally, from Eq. (1.5.19), if we integrate and apply the initial condi¬
tion on vz, we obtain

vz = Vi + aaV \{\ — cos a)ct — Loct sin a>ct) (1.5.25)

As noted earlier, the underlined terms in these equations are those

arising from the curvature introduced to satisfy the Maxwell curl
equation, while the remaining terms proportional to a are due to the
gradient in Bz. Thus, had we assumed (in violation of the curl con¬
dition) that B = B0az(l + ax), we should have obtained these solu¬
tions minus the underlined terms. Conversely, had we kept only the
curvature term, letting B = B0(az + axaz), we should have obtained
only the underlined terms (plus the terms independent of a).

Interpretation of the results

Before considering the motion of the particle in detail we can make

several observations about the equations for v. First, it is notable
that the perturbation terms in v due to the gradient in Bz are all pro¬
portional to aVo = F02/wc, while those due to the curvature are pro¬
portional to either aVi or Fi2/coc. Other than this, however, for times
on the order of l/wc all these perturbation terms are of nearly equal
importance.
Considering each component of the velocity separately, we note
that of the three perturbation terms in Eq. (1.5.23) for vx, the first two
merely show small additional oscillations of the particle at the cyclotron
frequency and at twice the cyclotron frequency. The final term, how¬
ever, is interesting in that it provides precisely the motion needed to
make the particle follow the curvature in the magnetic field line.
Thus, from this one term, we have

vx = aV\H X = ±aV i2t2 = ^aZ02 (1.5.26)

which is just the (approximate) equation for the held line passing
through the origin, obtained in Eq. (1.5.12). Thus, one effect of the
curvature is to provide an acceleration which causes the particle to
stay rotating about the same “magnetic held line.”
Of the perturbation terms in vy, the hrst two again show small
 1.5 MOTION IN A NONUNIFORM B FIELD 49

additional oscillations of the particle at wc and 2wc rad/sec. In addi¬

tion there is a steady drift velocity

VD aya F|<zlo +
a
a.-G^Yd- IV
01 r.
(1.5.27)

which arises partly from the gradient and partly from the curvature.
Note that this drift is in opposite directions for oppositely charged
particles, because of oic. The factor aay can be expressed in terms of
either the gradient in Bz, VBZ = aB0ax, or in terms of the radius of
curvature, R = — oVa*, since with B0 = B0az we have

B0 X R X B0
aay (1.5.28)
B02 R2B0

With this notation,

To2 Bo X TBZ IV R X Bo
(1.5.29)
VD 2cc Bo2 ^ oic R2B0

which is a useful form, since it is easily applied to other coordinate

systems. The path of a particle with this drift velocity is shown in
Fig. 1.19.

Fig. 1.19 Typical orbits for charged particles with a small gradient or curvature
in B.
50 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

To examine the effects of the inhomogeneity upon the motion in

the z direction, we integrate Eq. (1.5.25) to obtain

2 sin u>ct
1 + OM (iff- (1.5.30)
Z = Z°
( COS U)ct
Wct

This result shows that the position of the particle along z can vary
from Z° = Vit by a factor of order oca, which, by hypothesis, is very
small. However, since these variations in Z fluctuate and depend
markedly on the exact orientation of the initial velocity component
perpendicular to B0, they are not classified as a true drift of the particle
in the sense used in Eq. (1.5.27).

Problem

1.13 Estimate the curvature drift velocity for an electron in the earth’s
magnetic field when R is the earth’s radius, B0 = 1 gauss, and I i
is the “thermal velocity” of an electron (kT/m)112, with T = 300°I\.

Motion of a particle in a converging field

Thus far we have seen what happens to the motion of a particle when
the magnetic-field intensity changes with position perpendicular to
the “basic” field direction (the z axis, in our problem), i.e., when Bz
is a slowly varying function of x or y. Now consider the case when
Bz is a slowly varying function of 2

Bz = B0{1 + az) (1.5.31)

Experimentally this condition is obtained by making the number of

turns per meter on the coil producing the field (see Fig. 1.16) a slowly
increasing function of z. In order to satisfy Maxwell’s divergence
equation, V • B = 0, B must have some additional component. If
the symmetry of the problem prohibits any <f> dependence (as is gen¬
erally the case, experimentally) we have, in cylindrical coordinates,

V-B = llrBr + dp = 0 (1.5.32)

r dr dz

or

-r- rBr = —rB0a (1.5.33)

dr
 1.5 MOTION IN A NONUNIFORM B FIELD 51

Solving, we have

aB0r F(z)
Br = (1.5.34)
2 + r

with F(z) an “arbitrary” function which we set equal to zero, so that

Br does not diverge at r = 0. Switching back to rectangular coordi¬
nates, we have

B = B'o[—±axax — %ayay + (1 + az) a,] (1.5.35)

Hence the equation of motion for vz becomes

i>z = }uca(Xvy — Yvx) (1.5.36)

The treatment of vx and vy is not crucial to our discussion; hence we

need not consider them now. To find vz, we again use our perturbation
approach and replace all the terms on the right by their zero-order
approximations, since the entire term is proportional to a. Using
Eq. (1.5.16), this'yields

vz = icoca(X°Vy° - Y°vx°)

— —±ucaaVo = — 02 (1.5.37)

Hence

Vz=Vl- ±aV0H (1.5.38)

showing that there is a deceleration of the particle which is proportional

to the square of the perpendicular velocity and to a.

Application to a magnetic mirror

Assuming that the particle has some initial transverse velocity, it is

evident that the particle will slow down as it moves into a region of
increasing field strength. Ultimately, if the field becomes strong
enough, vz drops to zero, and the particle reverses direction. There¬
after, since the particle retreats to a region of decreasing B, vz will
increase again in magnitude, but the particle will have changed direc¬
tion or “been reflected” by the magnetic mirror.
One should note here that, from the energy-conservation equation
(1.4.9), the total kinetic energy ±mv2 must stay fixed. Thus, at all
times

v±2(t) + v*2(0 = fV + V i2 (1.5.39)

52 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

with vx the velocity of the particle in the direction perpendicular to

B0. Hence, at the time of “reflection” the particle’s kinetic energy
is all contained in vx, while the maximum value for vz as a particle
drifts into a region of decreasing B occurs when vx drops to zero.

Problem
1.14 Investigate energy conservation for a particle in a converging field
by calculating first-order values for vx and vy and then calculating
±mv2 as a function of time.

SUMMARY
Although our discussion of the motion of a charged particle in non-
uniform B fields has been limited to two rather special problems, these
contain most of the physical aspects of any more general problem, pro¬
vided B does not change markedly over a distance of the order of the
cyclotron radius a. For when B is slowly varying, we can always
approximate B in the neighborhood of the particle by an expansion in
which B is represented by an equation like Eq. (1.5.5) or (1.5.35) (or
a sum of both). This approximation will be good, that is, ax, ay,
and az will all be small compared to unity, for times up to a few cyclo¬
tron rotations; thereafter one can start the problem anew and expand
the field about its value in the neighborhood of the particle’s new
position. Hence, if one is satisfied to employ a computational scheme
of this sort, our solutions can be applied to any nonuniform B field
problem.

1.6 ADIABATIC INVARIANTS; APPLICATIONS; TIME-VARYING FIELDS

The magnetic moment

In this section we want to show that when an externally applied

magnetic field changes slowly with space or time, the magnetic moment
pi for a charged particle stays nearly constant. We note first that,
for a loop carrying a current I around an areaf A, the magnetic moment
is, by definition,

M = /A (1.6.1)

Now, if B points toward the observer, we have seen that a positive

charge spirals clockwise and a negative charge counterclockwise with
a radius |a|. Hence, for either case the current flow (which by con-

t Note that this area A has no connection with the vector potential, A, intro¬
duced earlier. In fact, in the remainder of this text, we shall never again need to
use the vector potential, so that A will almost always denote an area.
 1.6 ADIABATIC INVARIANTS; APPLICATIONS; TIME-VARYING FIELDS 53

A*

A B

Fig. 1.20 Current and area for a charged

particle in an applied B field.

vention assumes a motion of positive charge) is clockwise, as shown

in Fig. 1.20. However, if we look upon the path traced by the current
flow as a perimeter about A, then according to the right-hand conven¬
tion (discussed in connection with Stokes’ theorem) A points in the
oppositef direction to B.
For a single charge rotating about B we thus have an equivalent
current loop with area

\ = — . (1.6.2)

and current

I = tv±
q (1.6.3)
‘lira

where v± is the velocity in the direction perpendicular to B and r is

the charge per unit length averaged over the loop. (In this we
assume that the path is a closed loop, so that motion in the B direction
and drifts due to inhomogeneities are neglected.) From Eqs. (1.6.1)
to (1.6.3) we have (using a = vjwc)

_ _ qav±B _ _ w±B (1.6.4)

M _ 2B B2

with w± = \mv_l2 the kinetic energy of the motion perpendicular to B.

Adiabatic invariance of \x

Suppose we consider now the time derivative of p = wJB when B is

a function only of time, so that

dtx . w± w± £ (1.6.5)
di = ^ = -B~B^B
t Note again that B is an externally applied field, not the field produced by the
spiraling charge (which is in the direction of A).
54 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

Since B(t) is assumed known, we need only calculate w±. do do this

exactly would, in general, be very difficult. However, we can get a
good approximation to w± by setting

Aw± (1.6.6)
ht

where Awx is the change in w± during one revolution of the particle and
At = 27t/coc is the time of one revolution. Then, assuming, still, that
the particle’s motion is a closed loop, we have

Awx = /F • d\ = fq(E + v X B) • dl (1-6.7)

Now d\ = v dt, so that (v x B) • dl = 0. Applying Stokes’ theorem

(1.3.11) to the remaining term, we have

Au>± = fq(V X E) • dS = — JgB • dS (1.6.8)

since V x E = — B. But, as noted earlier, to a good approximation

the surface is just a circle of radius |a| pointing in the direction oppo¬
site to B. Hence
qBirv j
Awx = qBira2 (1.6.9)
C0c

and

qBvx2 w±B
w, = (1.6.10)
c hr
Using this result in Eq. (1.6.5), we have m = 0.
The proof which we have presented applies to the case when B
is uniform in space but with a time-dependent amplitude. Hence the
motion in the direction of the field is unaffected, and the change in w±
arises from the work done by the E field, which always accompanies a
time-varying B field. In addition, throughout the analysis we neg¬
lected the uniform motion in the direction of B. However, when this
component of the velocity is nonzero, we need only change to a coordi¬
nate system moving with the constant velocity vz. In this moving
coordinate system the particle’s path is in a plane, as assumed in our
proof, and the electric and magnetic fields are unchanged. (This is be¬
cause the corrections involve v2 X B, which is zero.f) The major as¬
sumption in this analysis is that “not much changes” during a single
t See for example, W. K. H. Panofsky and M. Phillips, op. cit., pp. 160-165.
 1.6 ADIABATIC INVARIANTS; APPLICATIONS; TIME-VARYING FIELDS 55

rotation of the particle. This requires the change in B during one

cycle, B At, to be much less than B; hence, like any other process which
changes slowly with time, this is called an adiabatic process, and the
result, dn/dt = 0, is called the adiabatic invariance of ju.
The approximate invariance of n can also be shown for spatially
varying B fields. For this case we return to Eq. (1.5.37), where we
found that, for Bz = B0( 1 + az),

dBz/ dz
mvz 9^ — aw± wx (1.6.11)
~BT

Now, for a stationary B field, with E = 0, we have shown that the

total kinetic energy stays constant. Therefore

dw± clwz dvz w± dBz

— mvz —T- = TT—- (1.6.12)
dt dt dt B dz

But for Bz a function of z, the total derivative is

dBz _ 6BZ dz _ dBz

(1.6.13)
dt dz dt dz Vz

Hence we have, from Eqs. (1.6.12) and (1.6.13),

dw± ^ w± dBz
(1.6.14)
dt B0 dt

so that

d± = I _ w± dB ~ (16 15)
dt B dt B2 dt

The conclusion drawn from these examples is that ju (or w±/B)

stays nearly constant for B fields which vary slowly with space or time.
An important practical application of this residt is in the magnetic
heating of a plasma, discussed in Sec. 1.2, where the energy of the
charged particles is raised by gradually increasing the magnetic-field
strength in an applied field.

Problem

1.15 Let B = B0h(t)a, with hit) a slowly varying function.

(a) Show that Maxwell’s equations are satisfied if we let

E = ±B0h(Yax — A"a„)
56 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

(6) With these E and B fields, show that

= ±much(Yvx — Xvy)

(c) Next, show that for X, Y, vx, and vv given by the zero-order
solutions (valid when h = 0) wJB stays constant.
{d) Find the restrictions on h(t) required by V X H = ecE.

Conservation of flux through the orbit

The magnetic flux passing through a surface is defined as

Flux = /B • dS (1.6.16)

Thus, for a particle spiraling about the field lines

Tm2v±2 %rm w±
Flux = Bira2 (1.6.17)
~cflB q2 B

Thus, the conservation of wJB or p may also be thought of as a con¬

servation of the flux 'passing through the cyclotron orbit of the particle.
Note also, that, since Ba2 is a constant, an increase in B leads to a
decrease in the cyclotron radius \a\.

The reflecting field

As one application of the adiabatic invariance of p, we can easily cal-

cidate the magnetic-field strength required to reflect a particle by a
magnetic mirror. First, if we let w±(0) and 5(0) denote the initial
perpendicular kinetic energy and magnetic-field strength, then, at any
later time

w± _ w±(0)
(1.6.18)
~B ~ 5W

For reflection to occur, however, all the initial kinetic energy must be
transformed into wL \ that is, at reflection, w± = Wtot(0). Therefore, at
reflection we have

wtot(0)
5 = 5(0) (1.6.19)
w±( o)

For a plasma where the particle velocities are initially isotropic

the average value for WJtot(0)/w;j.(0) is -§; hence a reflecting field of 1.550
 1.6 ADIABATIC INVARIANTS; APPLICATIONS; TIME-VARYING FIELDS 57

will reflect one-half the particles, and one of several times B0 will
reflect almost all the particles. However, those particles which
initially are moving in or near the z direction [so that wtot » tt>x(0)]
will escape even if B is very large.
A term commonly used in this situation is the pitch angle a
between v and the 2 axis. If v is the total speed of the particle,
sin a = v±/v, so that, from our definition of m, we have

w± /LiB
sin2 a (1.6.20)
Wtot Wtot

Thus, for a time-invariant B field, where wtot is constant, the invariance

of p with position leads to the condition

sin2 a
= const (1.6.21)
~B~

which again shows that the pitch angle increases toward 90° (where
reflection occurs), as the particle moves into a region of increasing
field strength.

The longitudinal adiabatic invariant

There is one other adiabatic invariant which is useful to apply to the
problem of particles trapped between two moving magnetic mirrors.
This is the so-called longitudinal adiabatic invariant

Jvdl (1-6-22)

where the limits on the integral run over one period of oscillation back
and forth between the mirrors, and where it is assumed that the
separation between the mirrors changes slowly with time. To analyze
this problem we consider the idealized situation shown in Fig. 1.21.

B
-►
V
m
Uniform field

L
—-2

Moving Stationary
magnetic magnetic
mirror mirror

Fig. 1.21 Moving magnetic system used to derive longitudinal

adiabatic invariant.
58 MOTION OF A CHARGED PARTICLE IN ELECTROMAGNETIC FIELDS

Assume there is a large region of length L containing a uniform B field.

At one end assume that B increases to form a magnetic mirror, while
at the other end B increases with both space and time to form, in
essence, a magnetic mirror which moves toward the stationary mag¬
netic mirror. Since this latter mirror involves time-varying B fields
(and, hence, electric fields), it can lead to a change in the kinetic
energy of the particle.
If we let Al be the longitudinal adiabatic invariant defined in
Eq. (1.0.22), then neglecting the “end effects” at the two mirrors
and the small change in \vz\ after one reflection from the moving mirror,

Al — 21 i>s, | L (1.6.23)

with vz the drift speed of the particle in the long region where B is
uniform. The rate of change of AL is clearly

(1.6.24)

From the figure it is evident that dL/dt = — vm, while to calculate

d\vz\/dt we set

d\vz\ A|vz| n 5 95)

dt — At

with A11;*,| the change in \vz\ in one reflection from the moving mirror
and At = 2L/|v*| the period of oscillation between the mirrors. To
find A|v*|, consider the transformation to a coordinate system moving
with velocity vm. Denote this moving coordinate system with a
prime, so that

vz = v' + vm (1.6.26)

In the moving coordinate system the “mirror” is stationary; therefore,

after reflection, the speed of the particle is unchanged, although the
velocity is in the opposite direction; that is,

(1.6.27)

with the subscripts 1 and 2 denoting before and after reflection.

Transforming back to our stationary coordinate system, we have, from
Eqs. (1.6.26) and (1.6.27),

Vz2 Cm iPzl Vm)

 1.6 ADIABATIC INVARIANTS; APPLICATIONS; TIME-VARYING FIELDS 59

or

— —t-'zi + 2 vm (1.6.28)

Since Vzi is in the negative z direction, it is obvious that the change

in speed is

AH = Im - \i'zi\ = 2vm (1.6.29)

Applying this result to Eq. (1.6.25), we have

d\vz\ ^ vm\vz\
(1.6.30)
dt — L

so that, in Eq. (1.6.24),

= -2\v,\vm + 2\vz\vm = 0 (1.6.31)

and A L is constant.
The approximate invariance of AL permits one to compute very
simply the energy of a particle trapped between the two mirrors, since

ttiAl2
wz = ±mvz2 (1.6.32)
~8L~2~

Thus wz is inversely proportional to L2 and increases rapidly as L

decreases. This method of accelerating a charged particle is called the
Fermi mechanism, since it was suggested by Fermi as a possible
explanation for the origin of high-energy cosmic rays. An alternative
theory for the acceleration of cosmic rays is due to Alfven, who notes
that since

wl
M = = const (1.6.33)

the energy of charged particles can be increased if they simply move

in a region of increasing B fields.
chapter two

MAGNETOIONIC
THEORY
chapter two

2.1 THE LANGEVIN EQUATION

Equation for the average electron velocity

In the preceding chapter we considered in some detail the motion of a
single charged particle in various electromagnetic fields assuming no
interaction whatever between the particle and any other particles in
the medium. The behavior of the plasma as a whole was rarely con¬
sidered, and when it was, we generally assumed that every charged
particle was behaving in essentially the same way except for possible
differences in charge or mass. In reality, as we shall see in later
chapters, there are strong electrostatic forces which tend to maintain
a plasma in a state of near (if not total) neutrality, i.e., with nearly
zero total charge density. Hence the next simplest approach we can
use to study a plasma is to assume that we have a gas containing n0
free electrons, no singly charged positive ions (or no/2 positive ions,
each with charge ze), and n\ neutral molecules per cubic meter. To
find the average velocity u at any point in space for the electrons in this
plasma when there are electromagnetic forces acting we use the
Langevin equation

mix = —e(E + u x B) — mv\x (2-1.1)

where m is the electron mass, e is the magnitude of the electron’s

charge, v is called the electron collision frequency and is the number of
collisions per second which the average electron has with the heavy
particles in the plasma, and E and B are the electric and magnetic
fields which act upon the electrons and may be either externally applied
or self-induced by currents in the plasma.
The first two terms in this equation are obtained by averaging the
Lorentz force equation over all the electrons in a small volume element.
The last term, however, is new and is intended to represent the effect
upon u of collisions between electrons and heavy particles. Thus
if we use Eq. (2.1.1), when E and B are zero, we have

u = — vu (2.1.2)
 2.1 THE LANGEVIN EQUATION 63

which has the solution

u (t) — u(0)e_,,< (2.1.3)

Hence, if there are no external forces acting, the collision term causes
the average electron velocity to approach zero at a rate governed by
the collision frequency between the electrons and the heavy particles.
Note that this does not mean that the velocities of the individual elec¬
trons approach zero due to collisions with the heavy particles but only
that the velocities become completely random, so that the average
velocity is zero.
In addition to Eq. (2.1.1) one might expect to use an analogous
equation for the average ion velocity of the form

m.ui = e(E + u; x B) — damping term (2.1.4)

However, since m,-, the ion mass, is typically about 104 times larger
than the electron mass, the acceleration of the ions by electromagnetic
fields is generally, very small. Therefore, while there are some prob¬
lems in which the motion of the heavy ions is important, we shall
neglect these for now and assume that the average velocity of all the
heavy particles stays zero.

Linearization of the equations

Equation (2.1.1) provides an equation for the average electron velocity

u. Naturally, this flow of electrons corresponds to a current density

J = — n0eu (2.1.5)

so that the Maxwell curl equations become

V x E = —B V X H = D — n0eu (2.1.6)

We now assume that, aside from the effects due to this electron current,
the plasma medium acts like a vacuum with D = eoE and B = MoH,
so that in terms of just E and B we have

V X E = -B V X B = Mo n0eu (2.1.7)

with c = (moCo)_1/2 the velocity of light in a vacuum. Equations (2.1.1)

and (2.1.7) provide three vector equations involving only E, B, and u;
thus our problem is now well defined. However, the equations in
64 MAGNETOIONIC THEORY

their present form are not yet quite tractable because of the nonlinear
term u X B (involving the product of two variables), which enters
Eq. (2.1.1).
This difficulty can generally be eliminated by the following
technique. First, divide B into two parts

B(r,«) = Bo + b(r,0 (2.1.8)

with B0 a constant independent of space and time. Since B0 is con¬

stant, Eiq. (2.1.7) becomes

V x E = -b (2.1.9)

Now, for traveling plane waves we shall show in Sec. 2.4 that E, b,
and n are all proportional to exp (tk • r — iut). Hence, the opera¬
tions V x and d/dt become

cl;r ^i/ tlz

a* ay a.
AAA Lx liy kz fk X (2.1.10)
dx dy dz

and

d
(2.1.11)
= ~lW

Applying these operators to Eq. (2.1.9), we have

k X E
l> = - (2.1.12)
CO

Referring back to Eq. (2.1.1), we thus have in the troublesome Lorentz

force term

E + uXB = E + uxB0 + uX —— (2.1.13)

CO

where the last term on the right is the nonlinear one. Note, however,
that the magnitude of this nonlinear term is always less than \ukE/u\,
whereas the first term on the right has magnitude \E\. Hence, the
nonlinear term can be neglected, provided \uk/co| is much less than unity
 2.2 ELECTRON CONDUCTIVITY AND MOBILITY 65

or

CO
u\ « (2.1.14)
k

In general, co/k, the complex phase velocity, is of the order of c,

whereas u, the average electron velocity, is generally much smaller.
Hence, for most cases the nonlinear term can he neglected. However,
we should note that under certain resonant conditions that we shall
encounter, |u| becomes large while |cu//c| becomes small; hence, for
these resonant conditions the nonlinear term becomes very important,
and the residts of the linear analysis are not really valid. In addition,
there are many steady-state flow problems where the nonlinear term
must also be retained.
For most wave-propagation problems, however, and for many
steady-state problems as well, one can safely use the linearized Lorentz
force term.

2.2 ELECTRON CONDUCTIVITY AND MOBILITY

Generalized Ohm’s law

As our first application of the linearized Langevin equation, we con¬

sider the steady-state (or dc) current produced by a uniform electric
field. From Eqs. (2.1.1), (2.1.7), and (2.1.8), we have, with all time
derivatives set equal to zero,

0 = — — (E + u X B0) — m
m (2.2.1)
V X E = 0 Vxb = MoJ

Multiplying the first of these equations by n0e/v, we obtain

» /T~' , n \ n»e2 (2.2.2)

J — <70 E + U X Bo) (7o — -
mv

which is often referred to as a generalized Ohm's law, with <r0 the dc

conductivity of the plasma. In addition, when v —> 0 (so that <r0
becomes infinite) the current J can remain finite only if

E + u X Bo = 0 (2.2.3)

Hence, Eq. (2.2.3) is often called Ohm's law for a plasma with infinite
conductivity. Note that if we take the cross product of each term
66 MAGNETOIONIC THEORY

with B0, then, since (u X B0) X B0 = — u±£02, we have

E X Bo
>h (2.2.4)
B02

in agreement with our earlier results on particle drifts in an electro¬

magnetic field.

Perpendicular conductivity and Hall conductivity

The generalized Ohm’s law is not yet in a form where one can easily
calculate J for a given E field. Therefore, in Eq. (2.2.2) we choose our
coordinates so that Bc lies in the 2 direction, and in addition we replace
u by — J/en0, to have

eB 0
J —■ (JoE H-J X a3 LCr (2.2.5)
v m

Writing out each component of this equation more explicitly, we have

w,
fj x V to Ex

CO
~Jx + Jy T0B y (2.2.6)
V

J. Tq Ez

From the last of these equations it is evident that a0 governs the flow
of current in the direction of B0, while from the first two equations
we find

JX — (TLEX + OflEy Jy = (Tj_Ey ~ G RE X (2.2.7)

with

cror2 <T0C0CV
(Tk (2.2.8)
o>c2 + r2 C0C2 + V2

In these equations ax is called the 'perpendicular conductivity, since it

governs the current flow in the direction of E when E is perpendicular
to B0, while aH, called the Hall conductivity, governs the current flow
perpendicular to both E and B0. In Fig. 2.1 we show a± and oH as a
 2.2 ELECTRON CONDUCTIVITY AND MOBILITY 67

function of ctjv. It can be seen that for wc/v > 1 the perpendicular
conductivity falls off very rapidly, so that very little current can be
driven directly across the lines of a magnetic field. The Hall con¬
ductivity also falls off with increasing wc/v but at a much slower rate.

Electron mobility

The electron mobility /ie is just the ratio of the average electron velocity
to the electric-field strength in a steady-state problem with no B field.
Thus from the first equation in (2.2.1) we have

u e
He = (2.2.9)
E mv

Check on the linearization

In our analysis of this steady-state problem we have used the linearized

Langevin equation where we assume

|u X b| « \E\ (2.2.10)

However, from the final equation in (2.2.1) we can calculate b in any

problem and see whether |u x b| is indeed much smaller than E. As
an example (which incidentally also demonstrates the pinch effect
discussed in Sec. 1.2), consider the simple configuration shown in
Fig. 2.2, where B0 = 0 and E = Eaz. From the linearized analysis
in Eqs. (2.2.6) and (2.2.7) we have

J = a,Eaz (2.2.11)

Fig. 2.1 Hall and perpendicular conductivities plotted as a

function of the ratio of cyclotron to collision frequency.
68 MAGNET0I0NIC THEORY

Fig. 2.2 Self-induced B field and pinch effect in a linear

discharge.

Hence, from the last equation in (2.2.1), we havej

(V x b)z = - rb^, = hq(ToE (2.2.12)

r dr

If we restrict our attention to the interior of the plasma, where E

is essentially constant, we have, upon integrating Eq. (2.2.12),

K = (2.2.13)

with C an integration constant which we set equal to zero, so that b is

finite at r = 0. Thus, as expected, the current in the plasma produces
a magnetic field that circles through the plasma. From this result and
Eq. (2.2.11) we note that the nonlinear term in the Langevin equation
is

Jxb H()Oo2E2r
u X b = (2.2.14)
n0e 2 n0e

Applying this result to Eq. (2.2.10), we find that the linearization

procedure is justified provided

2 n0e
E « (2.2.15)
ju0cr02r

f Note that the term in (V X bh involving d/d<f> is zero because of the sym¬
metry in the 4> direction.
 2.3 CONDUCTIVITY TENSOR AND DIELECTRIC TENSOR 69

To put this in terms of numbers, recall that <x0 = n0e2/mv. Hence,

if we use for v just the electron-ion collision frequency, which we shall
derive in Sec. 5.5, we have (assuming In A = 8.8),

v = 6.1 X IQ"9
^3oo yi*
n0 (2.2.16)

and

(2.2.17)

with T in degrees Kelvin. | If we substitute this value for <r0 into

Eq. (2.2.15), we obtain finally the condition

300V n0 1
E « 1.2 X 106 volts/m (2.2.18)
T ) 1020 r

Although this linearization condition is satisfied in many laboratory

and natural plasmas, it does break down in most fusion problems,
where there are very high temperatures and relatively low electron
number densities.
Before concluding this discussion note that, from Eq. (2.2.14), the
force — e(u x b) due to the interaction of the current and its self-induced
field is directed radially inward and tends to “pinch” the plasma.
This inward force can be estimated from Eq. (2.2.14) when the lineari¬
zation condition is satisfied; however, for the cases of real interest, the
pinching force is actually quite large, and one must use a much more
detailed analysis.

2.3 CONDUCTIVITY TENSOR AND DIELECTRIC TENSOR

Dc conductivity tensor

In the last section we found that the current produced by a stationary

electric field in the presence of a dc magnetic field (in the z direction)
is given by [see Eqs. (2.2.6) to (2.2.8)]

Jx — a±EX + <JhEv Jy = —onEx + a±Ey ^ ^

Jz <T 0 Ez
f This is a maximum value for cr0 (and hence a conservative estimate for E)
since, more generally, v also includes a term for electron-neutral collisions.
70 MAGNETOIONIC THEORY

Hence, this may be written as

a • E (2.3.2)

with

O’x <rH 0
ct = <7 j_ 0 (2.3.3)
— <JH

0 0 (T o

the dc conductivity tensor.

Ac conductivity tensor
Equation (2.3.1) is the solution to the linearized Langevin equation
when u = 0. If we instead assume that there is a sinusoidal dis¬
turbance, so that u and E vary with time as e~wt, we have, from
Eq. (2.1.1),

u — icon = —- (E + u X Bo) yu
m

or

0 =-- (E + u X Bo) — (v — ico)u (2.3.4)

m

This is identical to the dc equation except for the change from v to

v — iw. Hence one again obtains solutions of the form J = cr • E,
with a still given by a matrix, as in Eq. (2.3.3), only now we must
replace v by v — ico in each matrix element. Thus we have

, _ 'ftpe2 _ npe2(v T ho) ^ 3 5)

<7° m(v — ico) m(i'2 + co2)

as the analog to the dc conductivity, n0e2/mv, and similar frequency-

dependent transverse and Hall conductivities analogous to a± and oh-

Problems

2.1 Consider the following equations for average electron and ion
velocities when B0 = 0:

mu = —eE — mv( u — U)
MV = eE — mv(V — u) M m
 2.3 CONDUCTIVITY TENSOR AND DIELECTRIC TENSOR 71

where M = mass of ion

U = average ion velocity
For u, U, E all proportional to e~iat, prove that

, _ n0e2E 1 + (m/M)
mv 1 + (m/M) — i( w/v)

Thus, for m <<C M, the effect of the ions is truly negligible here for
all co. (This is not the case if B0 ^ 0.)
2.2 Prove that if a solid has equal numbers of electrons and holes
and if mu = me and vh = ve, then the conductivity tensor has
the form

0 0
u = <Tj. 0

0 (To

Dielectric tensor

Thus far in our treatment of sinusoidally varying fields we have used

only the Langevin equation. If we next consider the second Maxwell
curl equation in (2.1.7), we have, using the conductivity tensor a,

V X b = — i'(coMoeo)E + MoJ

— idfJLO ^CqE -p — cr • E^ (2.3.6)

This equation can also be written in the form

V x b = — fco/io£ * E (2.3.7)

where, if we let I be the unit matrix (with diagonal elements equal to

unity and all other elements zerot), then

(2.3.8)

The tensor e which enters here is called the dielectric tensor for the
plasma.
The use of the dielectric tensor in a sense represents a different
approach to treating the plasma from that used till now. Up to this
point we have represented the plasma as a collection of particles moving
about in an otherwise evacuated medium. Thus, in Maxwell’s equa-

t Note that for any vector A, I • A = A.

72 MAGNETOIONIC THEORY

tions, we set

D = e„E B = moH (2.3.9)

as would be the case for a vacuum, and the effect of the plasma shows
up through the currents arising from the motion of the charged par¬
ticles. If, instead, we adopt the dielectric tensor given in Eq. (2.3.8),
we essentially set

D = e • E (2-3.10)

and thereby characterize the plasma as a dielectric medium whose

internal behavior (average velocity, particle currents, etc.) we need
know nothing about. Thus, through the use of the dielectric tensor
we switch from a microscopic picture involving the detailed motion
of the particles in the plasma to a macroscopic picture, concerned
solely with the gross properties of the plasma.

2.4 WAVE PROPAGATION WITH B0 = 0

Plane waves
We now want to use the Langevin equation for a plasma (plus Max¬
well’s equations) to study the propagation of plane waves in an
unbounded plasma. Before beginning the analysis, however, it is
useful to review briefly some of the fundamental features of waves.
A typical plane wave traveling in the z direction has the form

ip = A cos (kz — wt) (2.4.1)

where k = wave number

co = angular frequency of oscillation
From the plot of ^ as a function of space and time in Fig. 2.3 it is clear

Fig. 2.3 Plots of the amplitude of a plane wave as a function of space and time.
 2.4 WAVE PROPAGATION WITH Bo = 0 73

that the wavelength X and the period T of the oscillations are given by

2ir 1
X (2.4.2)
T CO
/
with / being the frequency in cycles per second. For any wave there
is a functional relationship between k and co which depends on the
medium and which is known as a dispersion relation. Hence, in Eq.
(2.4.1) we can, alternatively, think of k as being a function of some
applied frequency co or of co being a function of some specific wave
number k.
For a plane wave traveling in some arbitrary direction specified
by the unit vector a

yp = A cos (k • r — coi) (2.4.3)

where k = ka is called the propagation vector. (Note that for our

previous case, with a = az, k • r = kz as expected.) Because of the
argument of the oosine in Eq. (2.4.3), the planes of constant phase are
defined by the condition

k£ — u>t = const £ = r • a (2.4.4)

and the phase velocity vph, which is the velocity of propagation d£/dt
of these planes, is found [by differentiation of Eq. (2.4.4)] to be given
by the familiar relation

vph = l (2.4.5)

For convenience it is generally useful to use complex numbers and

have yp proportional to exp i(k • r — cot). However, when this is done,
it is always understood that the real part of yp is the true physical
quantity that one would measure in an experiment.

Wave packets and group velocity

In practice it is not possible to have a plane wave with one specific

value for co and k such as we have written in Eq. (2.4.1). Instead any
real disturbance consists of waves having some spread of wave num¬
bers, so that for a disturbance traveling in the z direction we have

\p = f A{k)eikz~iut dk (2.4.6)
74 MAGNETOIONIC THEORY

where A(k) is the amplitude (per unit wave number) of the waves in the
disturbance with wave number k. If the range of values for k is small
and centered about some specific value k0, we can let

k — ko k and CO = COo 4*

(2.4.7)

where co0 = co(/c0). If we substitute these expressions for k and co into

Eq. (2.4.6), \p separates into two factors

x[/ = (z,t) (2.4.8)

where

M(z,t) = I A (k) exp in z elk

(2.4.9)

Equation (2.4.8) has the typical form for a wave packet, such as
the one shown in Fig. 2.4, where co0 is the carrier frequency and M(z,t)
is the amplitude of the packet. If the total range of wave numbers is
small, so that for all k

1
2
kH «1 (2.4.10)

Fig. 2.4 A typical wave packet. The solid curve is ^{z), and the
dotted one is the amplitude factor M(z,t).
 2.4 WAVE PROPAGATION WITH B„ = 0 75

then the underlined term in the exponential (and all subsequent terms
which would involve still higher powers of k) are negligible.f Hence
the planes of constant packet amplitude are defined by

z t = const (2.4.11)

The velocity of these planes of constant M, dz/dt, is called the group

velocity, and is easily found (by differentiating with respect to time) to be

v0 =
(2.4.12)

It should be clear from this derivation that the concept of the group
velocity is not too precise, since for any dispersive medium (where w
is not a linear function of k), t eventually becomes large enough to
violate Ecp (2.4.10).
In addition, some modification of both the phase and group
velocity is required when there is damping. The most common type
of damping is simple exponential decay of a signal in space (for a
traveling wave) or in time (for a standing wave). In the former
case k has an imaginary part, and in the latter case w has one. Thus
for damped traveling waves k = 13 + ia, so that for a wave in the z
direction ^ = Ae~az+iPz~iat, and the phase velocity becomes w//3. The
concept of group velocity can also be extended; however, we shall not
go into that problem here since we never have to calculate the group
velocity of damped waves in this text.

Problem

2.3 (a) Show that for a nondispersive medium vph = v0.

(b) Find M(z,t) for a nondispersive medium when

A(k) = exp K = k — kn A = const

and make a rough plot of the disturbance \p when k0 » A.

Development of an equation for E

Turning now to the problem of waves in a plasma, we note, first, that

if we assume an time dependence for b, E, and u, then, from the

t This simplification also occurs if co is a linear (or nearly linear) function of k,

so that d2co/dfc2 and all higher derivatives are zero (or very small).
76 MAGNETOIONIC THEORY

two Maxwell curl equations and the linearized Langevin equations

(2.1.7) and (2.3.4), we have

V X E = icob (2.4.13)

V X b = i —; E — (2.4.14)
c2

II = (E 4 n X Bo) (2.4.15)
m( v — iw)

We can easily reduce this to a single vector equation involving just E

by first taking the curl of Eq. (2.4.13) and using Eq. (2.4.14) to obtain

V X (v X E) —— E — IW/XoTloCU (2.4.16)
cl

To eliminate u, we could solve Eq. (2.4.15) for u in terms of E; how¬

ever, it is simpler to recall that J = — n0eu, and we have already found,
in Eqs. (2.3.2) to (2.3.5), that J = a • E. Hence

E
ii —a • — (2.4.17)
n0e

Results with B0 = v = 0
Because of the complexity of Eq. (2.4.17), there is some advantage in
treating the problem in stages. First, if there is no dc magnetic field
and collisions are neglected (B0 = v = 0), we have, from either
Eq. (2.4.15) or (2.4.17),

le E
n =- (2.4.18)
mco

If we substitute this into Eq. (2.4.16), we have

(a>2 — we2) E
V X (V X E) (2.4.19)
c2

where coe = (n0e2/me0)1/2 is known as the plasma frequency. Note

that, if n0 is expressed in particles per cubic meter, we find ue = 56.5n01/2,
so that for electron number densities between 1012 and 1022 particles/m3,
fe = w,/2tt lies between 9 and 9 X 105 Mc/sec. If we recall from
Chap. 1 that the Debye length \D = 69.0(T/n0)112, we have the very
 2.4 WAVE PROPAGATION WITH Bn = 0 77

useful identity

1/2
coeXD = 3.90 X KEF1'2 = (2.4.20)

where k is again Boltzmann’s gas constant.

From Eq. (2.4.19) it is easy to find the dispersion relation between
k and w for the plasma. We simply assume that there are plane waves
traveling in the x direction, so that the space dependence of E is of
the form eikx, and the operator V x is given by

V x = ikax X (2.4.21)

In addition, let us set

E = Exax + E, (2.4.22)

with Ex and E, the longitudinal and transverse components of E. With

these substitutions-, we have (see Fig. 2.5), from Eq. (2.4.19),

(2.4.23)

or

(,k2c2 - oi2 + we2)E, = 0 (2.4.24)

and

(cd2 — Ue2)Ex = 0 (2.4.25)

x (direction of wave propagation)

f ar * ( ax x E,)

Fig. 2.5 Geometry for ax x(a* X E()-

78 MAGNETOIONIC THEORY

The first of these results shows that for a transverse wave (E, ^ 0) the
dispersion relation is

k2c2 = co2 - co,2 (2.4.26)

while, for a longitudinal wave (Ex ^ 0), we have, from Eq. (2.4.25),

co2 = coe2 (2-4.27)

Considering first the dispersion relation for transverse waves, note

that k2 is positive for co > co, and negative for co < co,. Hence for fre¬
quencies below the plasma frequency k is imaginary and E, is exponen¬
tially damped, since, for k = ia, we have E, °c e~ax~iat, so that the wave
dies out with increasing values of x. Such a wave is called evanescent
and can be shown to involve no average power flow (see Probs. 2.4
and 2.5). A plot of a as a function of co is shown in Fig. 2.6. In
addition, from Eq. (2.4.26) we can also calculate the phase velocity
(co divided by the real part of k) for the transverse waves

c
vph = - = [1 - (co,2AP)]1/2 " > (2.4.28)
^oo co < co.

A plot of phase velocity vs. frequency is also shown in Fig. 2.6, where
it can be seen that vvh is infinite for co < co, and then decreases rapidly
toward c as co > ». In the “propagation band” (co > co,) we also
note that if Eq. (2.4.26) is differentiated with respect to k, one finds

C2 = VphV0 (2.4.29)

Fig. 2.6 Plot of phase velocity, group velocity, and attenuation factor for trans¬
verse waves with B0 = v = 0.
 2.4 WAVE PROPAGATION WITH B0 = 0 79

so that the group velocity is given by the dotted curve in Fig. 2.6.
Note th^t although vPh is always greater than c, the velocity vg that a
signal would have is less than c, as required by the special law of
relativity.
For the longitudinal waves, Eq. (2.4.27) shows only that oscilla¬
tions (so-called plasma oscillations) can occur just at the plasma fre¬
quency. In Sec. 3.4 when we reexamine the wave-propagation prob¬
lem using a somewhat more complicated model for the plasma, we
shall see that this result is the limit one obtains when the electron
temperature goes to zero. Hence this is often called the cold-plasma
result. At this point, however, we can say nothing about the wave
number or phase velocity for the longitudinal plasma waves.

Problems

2.4 If A and B are two vectors each proportional to e prove that

(Re A • Re B) = JL
2 Re (A • B*)

and

(Re A x Re B) = \ Re (A x B*)

Note that Re denotes “the real part of” and for any function /(f)

Thus, the average value of the Poynting vector, for E, H °c e l0,t, is

<S> = 1 Re (E x H*)

2.5 Prove that (S) is zero for an evanescent wave.

The Fresnel interface problem

One of the most important properties of a rarefied plasma, i.e., where

collisional effects are slight, is that it totally reflects incident electro¬
magnetic waves with frequencies below the plasma frequency, whereas,
for co > coe, the reflection coefficient drops almost to zero. Ilius the
plasma acts as a high-pass filter. To see how this comes about, we need
only consider the reflection problem shown in Fig. 2.7. As indicated
in the figure, the incident medium is assumed to be a vacuum (or air),
so that k02c2 = co2; while the reflecting medium is a collisionless plasma,
so that ki2c2 = co2 — coe2. From Snell’s law (ko sin do = k\ sin di)
80 MAGNETOIONIC THEORY

we have

(2.4.30)

Hence, as indicated in the figure, 0i > 00-

In addition, the transmitted angle 0i must satisfy the condition
sin2 0! < 1, so that from Eq. (2.4.30) we must have

sin2 0O < 1 — (2.4.31)

CO

or

< co cos 0O condition for transmission (2.4.32)

Thus at normal incidence (0O = 0) the signal frequency must be at

least as high as the plasma frequency for the wave to get into the
plasma, while for other angles it must be still higher.

Application to the ionosphere

In the ionosphere the electron plasma frequency increases with altitude

until a maximum value wcr, the so-called critical frequency, is attained.
This maximum is reached, generally, at a height of about 300 km, and
thereafter the plasma frequency falls off slowly with increasing altitude.
Hence a radio wave transmitted vertically upward with frequency w
below wcr is reflected at an altitude when we = w. By measuring the
transit time for the wave to travel from a ground antenna to the reflec¬
tion point and back down to the ground as a function of frequency,
one can easily calculate the plasma frequency as a function of altitude

Fig. 2.7 Ray paths for plane waves incident upon a

vacuum-plasma interface.
 2.4 WAVE PROPAGATION WITH Bo = 0 81

up to the point where u„ is reached. Such measurements are called

ionosondes or ionospheric sounding measurements. Thereafter, since
w > u>cr, the wave will for the most part travel right through the
ionosphere provided it is transmitted close to the vertical so that
cos do exceeds cofr/co. Accurate measurements of the plasma frequency
above the critical height have come only recently via “topside” sound¬
ings made from a satellite traveling above the critical height and by
fairly delicate measurements of the scattering of electromagnetic
waves transmitted upward with « greater than wcr.

The reflection point

We have seen in this section that a wave incident upon a semi-infinite

plasma is totally reflected for <x> < coe. Hence o>e is often called a
reflection point. From Fig. 2.6 note that this is also the frequency
range where /3 (the real part of k) is zero and vph is infinite. More
generally, as shown in Prob. 2.6, one finds that the power transmitted
into a semi-infinite slab is always zero if /3 is zero|, so that any frequency
for which /3 = 0 is termed a reflection point.
Problems

2.6 Consider the full Fresnel problem at normal incidence. Let

E< = a ye***-™* Er = Erave-ik<>x-iut E* = E tayefk'x~iat

(a) Prove that

Hi = — azeikox~iat Hr =-— Eraze~ikl‘x~iut

k F
Ht = —

uyo

(b) From the boundary conditions that Ey and Hz are continu¬

ous at z = 0, prove that

p _ 2ko -rp _ k<i ki

1 ki + k0 r ko + ki

(c) Show that

Re (Et X Hf),-0 _ EtEfP

Re (Ef x Hf)*=0 k0

where /3 is the real part of kh so that T = 0 for /3 = 0.

f Interestingly enough, if the medium with 0 = 0 is finite, we show (in Prob.
2.7) that some energy can be transmitted through the slab. This is known as the
tunneling effect.
82 MAGNETOIONIC THEORY

2.7 (a) Consider electromagnetic waves incident normally on a

plasma slab of thickness l (with a vacuum for x < 0 and
x > 1) and prove that the ratio of the power transmitted out
of the plasma to the incident power is (for E,■ = ayeik°x~iat,
as above)

T = EtE*

with

Et = l\ 0—ikl
and
(1 — n)(l — n l)eikl -f- (1 + n)(l + n l)e
kc
n = —
w

(b) For w < we, where k = ia and n = irj, show that

EtE* ~
cosh2 al + ^(17 — t] 1)2 sinh2 al

so that some power is transmitted through the slab.

2.5 THE EFFECT OF COLLISIONS

It is fairly easy to include the effect of collisions in the dispersion

relation for transverse waves. For this case (with Ba still zero) we
have from Eq. (2.4.15)

ie E
(2.5.1)
mu>[ 1 + i(v/u>)]

so that the wave equation (2.4.16) becomes

We2 |
V X (V X E) E (2.5.2)
c2[l +i(v/ w)]j

This is formally equivalent to having a “complex plasma frequency,”

= <41 + i(v/u)]~112 in place of the usual plasma frequency
hence the dispersion relation for the transverse waves becomes

we 2
kV = CO2 - (2.5.3)
1 + i(4«)
 2.5 THE EFFECT OF COLLISIONS 83

while, for the longitudinal waves we find

co.
(2.5.4)
1 + i(v/a)

The principal effect of the collisions is to produce a damping of

the waves. Thus, for the longitudinal waves we have

off + ivu — coe2 = 0 (2.5.5)

or

iv + \/ 4coe2 — v2
CO = (2.5.6)

For all values of v, the frequency co has a negative imaginary part.

Hence, the waves, which are proportional to e~wt, are damped, since
for co = cor — ia we have

Ex = const (2.5.7)

Problem
2.8 Plot cor and a as a function of v/ue.

Determination of damping or growth

Before considering the transverse-wave dispersion relation (2.5.3) we
want to discuss some general results concerning dispersion relations
having the form

k2 = A + iB (2-5.8)

with A and B real. First, if we separate k into its real and imaginary
parts, we have

k = (3 + ia (2.5.9)

with and a both real. Hence

k2 - (/32 - a2) + 2i(3a (2.5.10)

or

A = P2 - a 2 B = 2(3a (2.5.11)
84 MAGNETOIONIC THEORY

From the second part of Eq. (2.5.11) we see that if B is positive, 13

and a have the same sign, while if B is negative, they have opposite
signs. The former case corresponds to damped oscillations, while the
latter case implies a growing wave, as can be seen by recalling that for
waves proportional to elkx~lwl we have

gikx—iut _ gi(3x—iiotg—ax (2.5.12)

Hence the sign of d determines the direction of wave propagation!

(d > 0 implies propagation in the + .r direction, and d < 0 implies
propagation in the —x direction), while the sign of a determines
whether the amplitude decreases or increases with increasing x.
Thus if d and a are both positive, the wave travels in the +x direction
and is damped; if they are both negative, the wave travels in the —x
direction and is also damped; but, if one is positive and the other
negative, the wave is exponentially growing. Thus the sign of the
imaginary part of k2 is crucial in determining whether one has growing
or decaying waves. From Eq. (2.5.11) one can also find d and a in
terms of A and B. The results are easily found to be

p = . + gJ-±A a2 = ^A- ^ -— (2.5.13)

These equations permit one to calculate k explicitly once k2 is known.

In similar fashion, one can easily verify that for a dispersion
equation having the form

CO 2 = A + iB (2.5.14)

where we let co = cor — ia, then for B negative the wave is damped,
while for B positive it grows.

Damping of the transverse waves

Returning now to the dispersion relation for transverse waves [Eq.

(2.5.3)], we have

C0e2 fcO<,2(,/co)
k2c2 = co2 (2.5.15)
1 + (,/co)2 + 1 + (,/co)2

hence B is positive, and for all frequencies we have damped traveling

waves. Detailed calculations of the phase velocity or damping factor

f Note that we assume here that the phase and group velocities are in the same
direction (as they are in all the dispersion equations we shall encounter here).
 2.5 THE EFFECT OF COLLISIONS 85

require the use of Eq. (2.5.13) and quickly lead to complicated alge¬
braic expressions, which we shall not bother to go through here. The
results, however, are rather surprising and well worth describing.
First, for w < coa (where «„2 = 3wc2/4), the damping factor a is a maxi¬
mum when v = 0 and decreases continuously with increasing v. (This
is, perhaps, not too surprising since is below the plasma frequency;
hence for o> < con, a is already rather large when v = 0.) Next, for
w > a plot of a as a function of v has the form shown in Fig. 2.8,
where it can be seen that the damping factor first increases with
increasing v, then reaches a maximum, and thereafter falls off toward
zero as v —* <x>. The maximum for a is found to be when

(2.5.16)

while, from Eq. (2.5.15), when v —> <», we have

W -> CO2 (2.5.17)

v—> co

so that the plasma behaves like a vacuum, in this limit. The reason
for this behavior is relatively obvious if one examines the ac con¬
ductivity for the plasma derived in (Eq. 2.3.5):

r nne2
a0 (2.5.18)
m{v — iu)

For large v, this approaches zero, so that no current flows in the plasma,
and the power absorbed (which is the product of the real parts of J
and E) goes to zero.

Fig. 2.8 Plot of a as a function of collision frequency. For

simplicity the calculation is made at a frequency satisfying
CO U>e-
86 MAGNETOIONIC THEORY

Problems
2.9 Verify that for w2 = A + iB the wave is damped when B is nega¬
tive and growing when B is positive.
2.10 Show that for w » coe the damping factor for transverse waves
is given by

~ a*Z 7 = -
“ — 2o>c(l + Z2) co

2.6 RESONANCES AND REFLECTION POINTS WITH B0 ^ 0

The Appleton-Hartree equation

Having examined the waves in a plasma when B0 = 0, we now con¬

sider the analogous development for the case when there is a uniform
magnetic field. To start we recall that from Maxwell’s equations
and the Langevin equation we have found [in Eqs. (2.4.15) and (2.4.16)]

Vx(VxE)-tE = — iwn0n0eu (2.6.1)

c2

and

u(co2 + icov) =-(E + u X B0) (2.6.2)

m

Although there are a number of ways to find the dispersion relations for
plane waves from these equations, probably the one which involves
the least algebra is to assume again, as in Sec. 2.4, that E and u are
both proportional to elkx~lwt, then use Eq. (2.6.1) to find E in terms ol u,
and finally use Eq. (2.6.2) to obtain an equation involving just u.
With this form for E, Eq. (2.6.1) becomes

CO2
— k2a* X (a, X E)-y E = — iwju0n0eu (2.6.3)
0

We next separate each vector into longitudinal and transverse parts, so

that E = Ex + E, and u = nx + u,. (Note that the subscripts x
and t after a vector indicate, respectively, a vector pointing in the x
direction and a vector perpendicular to ax.) With this substitution
we have a* x Ex = 0, so that, as shown in Fig. 2.5,

a* X (ax X E) = — E( (2.6.4)
 2.6 RESONANCES AND REFLECTION POINTS WITH Bo ^ 0 87

Hence, from Eq. (2.6.3), if we multiply through by c2/a>2 and let n be

the index of refraction

CO

we have the following two conditions:

in0eux
Ex (2.6.6)
C0€0

and

in0eut
(2.6.7)
coe0(l — n2)

Turning next to the Langevin equation (2.6.2), we note first that

u x B0 = (ux Hr X (Box + B0i)

= ux x Bot + u, x Box + u( X B0( (2.6.8)

where the first two terms on the right are perpendicular to a* and the
last is parallel to ax. Hence if we substitute Eqs. (2.6.6) to (2.6.8) into
Eq. (2.6.2) and remember that (n0e2/me0)1/2 is the plasma frequency
cog, we obtain from the x component of each term

lew , -r, ,
ux( CtT T" iwv) = - (u( X Bor) (2.6.9)
m

In similar fashion, we find from the transverse component of each term,

i te w
u, + IWV I =- (ux X Bor T u( X Box) (2.6.10)
m

Thus we now have three scalar equations involving just the three
components of u. To obtain a dispersion equation, we choose a set of
rectangular coordinates as indicated in Fig. 2.9, where ax is the previ¬
ously assumed direction of wave propagation and az is chosen per¬
pendicular to the plane formed by ax and Bn. If we let y be the angle
between a* and B0, we have

Box = Bo cos y a Bor = Bo sin y aw (2.6.11)

88 MAGNETOIONIC THEORY

Writing out Eqs. (2.6.9) and (2.6.10) explicitly and dividing through
by w2, we obtain an equation of the form

A • u = 0 (2.6.12)

where
1_

— iYsin y
b

0
1

A = 0 U — iY cos 7 (2.6.13)
iY sin 7 — iY cos 7 U — <3?

and

k2c2
U = 1 +i- X - n2 =
CO CO2 or
(2.6.14)
eB0 ^ = -
(J) x
Y =
TOO) 1 — n2

From Eq. (2.6.12) we have u = 0 (or no wave propagation) unless

det A = 0 (2.6.15)

Taking the determinant of A we have, by direct calculation,

(U — X)(U — <f>)2 — (U — <b) F2 sin2 y — (U — X) Y2 cos2 y = 0

(2.6.16)

or, after dividing by U — X

Y2 sin2 y \ , rr■> ( 9 , U sin2

Cf>2 2U J + U2 — 12 ( cos2 y + JJ _ Y ) = 0
“*( U - X
(2.6.17)

Fig. 2.9 Coordinate system used for plane waves with a dc

magnetic field.
 2.6 RESONANCES AND REFLECTION POINTS WITH B0 * 0 89

Note that this equation is quadratic, so that there will in general be

two solutions and therefore two types of waves that can propagate
at each frequency. Solving for <t>, we have

F2 sin2 1/2
7 F4 sin4 7
<£=£/ — 2 + F2 cos2 7 (2.6.18)
2 (17 - X) - 4(U - X)

Hence, from Eq. (2.6.14), the index of refraction n is given by

n2 = or k2c2 = co2 (2.6.19)

with <t> given by Eq. (2.6.18). This last result is the Appleton-Hartree
dispersion relation, which is used with considerable success to study
radio-wave propagation in the ionosphere, taking account of the earth’s
magnetic field. For the most part, the results predicted by the equa¬
tion remain unchanged even when more sophisticated models of a
plasma are used, provided the wave frequency is large compared to
the ion cyclotron frequency. For lower frequencies, the ion motion
becomes important, and, as shown in Chap. 3, there are quite different
types of waves which are not predicted by Eq. (2.6.19).

Resonances

Because of the complexity of Eqs. (2.6.18) and (2.6.19) it is not feasible

to calculate phase velocities and attenuation distances for arbitrary
angles y. Therefore we shall first find the resonances (where vPh = 0)
and the reflection points (where vPh = °°) for arbitrary angles, after
which we shall analyze the wave-propagation problem when k is either
parallel to or perpendicular to B0. Thereafter it will be fairly simple
to infer the phase-velocity plot for an arbitrary angle between k and
B0 without actually performing any further calculations.
As just indicated, the condition oj/k = 0 is termed a resonance.
For such a condition we have seen in Sec. 2.1 that the nonlinear term
in the Langevin equation is important, and our linear analysis breaks
down. The reason fern the term resonance, however, can best be seen
from Eq. (2.6.7), where we have

E = in°eUt _ (2.6.20)
‘ cueoU - n2)

Now, at a resonance, co/A; —> 0; hence

00 (2.6.21)
90 MAGNETOIONIC THEORY

Thus, for E( to be nonzero, u( must also approach infinity when there

is a resonance. In summary at a resonant frequency the phase velocity
drops toward zero while the drift velocity of the electrons approaches
infinity.
To find the resonant frequency for an arbitrary angle 7, note that,
from Eq. (2.6.14), 4> —•> 0 at a resonance. However, from Eq. (2.6.17)
it is evident that, for 4> = 0, the frequency must satisfy the condition

U sin2 y\
U 2 = Y2 ^cos2 7 + (2.6.22)
U - X)

This has a solution only when v = 0, since, for v 9^ 0, U is complex,

and one finds that there are no frequences where both the real and
imaginary parts of the equation are equal. For the collisionless case
we have

1 — X = T2(l — X cos2 7) (2.6.23)

or

w4 — aj2(we2 + coc2) + we2o)c2 COS2 7=0 (2.6.24)

Thus the condition for the resonance is

2\2 1/2
We2 + C0c2 We
+ we2wc2 sin2 7 (2.6.25)

A plot of these roots as a function of the angle 7 is shown in Fig. 2.10.

From the figure [or from Eq. (2.6.25)] it is evident that (1) the sum of
the squares of the two resonant frequencies always equals we2 + coc2;

Fig. 2.10 Frequencies which satisfy the necessary condition for a resonance as
a function of the angle 7 between B0 and the direction of wave propagation.
The curves on the left are for we > |oc|, the curves on the right for coe < jcoJ.
 2.6 RESONANCES AND REFLECTION POINTS WITH B„ ^0 91

(2) the high-frequency resonance increases with increasing y from the

larger of |coc| and ue to (ay2 + oy2)1/2, which is called the upper hybrid
resonance frequency; and (3) the low-frequency resonance decreases
correspondingly from the lesser of |coc| and ay to zero as y increases
from 0 to 90°

Problem

2.11 To get some picture of the motion of an electron at resonance

find the path of an electron when B = B0ax and

E = E0(ay sin coct + az cos uct)

Neglect collisions, and start the particle at the origin with zero
velocity. Note that this is a nonphysical problem since the fields
violate Maxwell’s equations.

Reflection points

To find two of the reflection points (where n is zero) we follow a pro¬

cedure entirely analogous to that used to find the resonances. First,
we note from Eq. (2.6.14) that, for n = 0, <i> = X. Hence, from
Eq. (2.6.16) the condition for n to equal zero is

(U - X)2 - F2 = 0 (2.6.26)

Equation (2.6.26) again has a solution only if v = 0. For this case

we have

(1 — X) = + Y

or

w2 + w|oy| — ay2 = 0 (2.6.27)

so that

+ |a>c| + (a>c2 -f 4we2)1/2

w = -^- reflection points (2.6.28)

(The minus sign in front of the square root is not considered, as this
leads to a negative frequency. Also, we use |wc| since a>c is negative.)
Thus, reflection points for any angle of propagation occur at two fre¬
quencies, wqi and co02, which lie one on each side of we and which are
separated by ay rad/sec.
92 MAGNETOIONIC THEORY

In addition at the plasma frequency (X = 1) when collisions are

neglected, we have directly from Eqs. (2.G.18) and (2.6.19) that, for
7 ^0,
<t>+ = 1 or n2 = 0
CO — coe (2.6.29)
= oo or n2 = 1

Hence, except for 7 = 0, there is a third reflection point exactly at the

plasma frequency.

Problems
2.12 Prove that co0i < coe < \/coc2 + coe2 < coo2, so that the reflection
points lie, respectively, below coe and above the hybrid resonance
frequency.
2.13 Derive an equation analogous to Eq. (2.6.18) for a system with
equal number densities of electrons and holes and with m* = m
and vh = v. What are the reflection points and resonances for
such a plasma?

2.7 WAVE PROPAGATION PARALLEL OR PERPENDICULAR TO B0

Propagation parallel to B0
To continue our discussion of waves in the plasma we now examine the
Appleton-Hartree equation for some special cases. First, for wave
propagation in the direction of B0, 7 = 0, so that from Eqs. (2.6.18)
and (2.6.19)

4> = U + Y U = 1 + i- Y = — (2.7.1)
w CO

2 .,
2 _ 2 _ _—_ (2.7.2)
1 + i(v/u>) + |o)c|/«

If we neglect collisions, we can easily find for the root with the plus sign
that there is no resonance but there is a reflection point (k = 0) when

-kl + k2 + 4av01/2 (2.7.3)

CO — CO01 — Q

Other than this, however, the dispersion relation for this wave is very
similar to the simple relation when Bo = 0, as can be seen in the plot of
phase velocity vs. frequency shown in the solid curve of Fig. 2.11.
 2.7 WAVE PROPAGATION PARALLEL OR PERPENDICULAR TO Bo 93

When we use the minus sign in Eqs. (2.7.1) and (2.7.2), there is a
resonance (/c = °o) for w = |coc|, as expected from Fig. 2.10, and a
reflection point (k = 0) at w02 = woi + |coc|. Furthermore, k2 is posi¬
tive both for co > co02 and for co < |coc|, so that the phase velocity is
finite for two ranges of frecpiency, as shown in the dotted curve of
Fig. 2.11. (In this discussion and in the figure we continue to neglect
collisions. Otherwise, as we have noted, there are no resonances or
reflection points, and some propagation can occur at all frequencies.)
Note that there is no resonance at coe for either of the waves,
although one is predicted there in Fig. 2.10. This is apparently due
to the fact that in deriving the resonance condition we multiplied by
1 — X, which is zero at w = we. At all other directions of propaga¬
tion, however, there is no difficulty with the resonance conditions.

Polarization of the waves

If we return for a moment to the basic equation for u [Eq. (2.6.12)],

note that for propagation parallel to B0, y = 0, so that we have for
the z component of that vector equation

-iYuy + (U - 4>)wz = 0 (2.7.4)

However, with given by Eq. (2.7.1) we have

—iuy + uz = 0 (2.7.5)

or

uy = ±iuz (2.7.6)

Fig. 2.11 Plot of phase velocity vs. frequency for waves traveling in the
same direction as B0, the applied magnetic field.
94 MAGNETOIONIC THEORY

Fig. 2.12 Plot of velocity vector as a

function of time for l.c.p. waves.

The plus-and-minus-sign solutions here [and in Eq. (2.7.2)] corre¬

spond, respectively, to left and right circularly polarized waves, as can
be seen from the following argument. As in Sec. 2.4, let k = P + ia,
so that

Ul = Uoei^-iat-ax (2.7.7)

Then, from Eq. (2.7.6) with the plus sign we have

uy = iuoe*13*-™1-011 (2.7.8)

However, we know that a measurement of uy or uz yields only the real

part of these expressions

Uz — Uoe ax COS (fix cot) (2 y

Uy = — u0e-ax sin (/3x — cot)

Now, as an example, consider these solutions at x = 0. We have

Uz — u0 cos cot uy = u0 sin ut (2.7.10)

which is a vector with magnitude Mo oriented as shown in Fig. 2.12.

Thus as t increases u( rotates according to a left-hand rule,t and the
wave is said to be left circularly polarized (l.c.p.). In similar fashion,
an analysis using the minus sign in Eq. (2.7.6) leads to n« rotating in
the opposite sense; hence this is called a right circularly polarized
wave (r.c.p.). To summarize, Eq. (2.7.6) shows that the two types

t Alternatively, for an observer looking into the lines of B0, u rotates clockwise
(like a -positive charge, as discussed in Sec. 1.4).
 2.7 WAVE PROPAGATION PARALLEL OR PERPENDICULAR TO Bo 95

of plane-wave solutions to the plasma equations when B0« = 0 are,

respectively, l.c.p. and r.c.p. waves, with the r.c.p. wave having the
resonance at |coc|, as might be expected from Sec. 1.4, where we saw
that a negative charge spirals counterclockwise in a uniform B field
(which is the same sense as the r.c.p. velocity vector).

Propagation perpendicular to B0
For wave propagation perpendicular to B0, 7 = 90°, so that, from
Eqs. (2.6.18) and (2.6.19), we have

*- u + mr^x)(-1 ± l) (2-7'1J)

and

k2c2 = “2 ~ 11 , ,•/ / n
+ l(jyco)
, r[COc 2/( — 1i^n/n/
+
2
± 1)/2(C02
1 ,•-iyj
+ lOJV — U>e )J
(2-7.i2)

As expected from Eq. (2.6.29), when we set v = 0, the dispersion rela¬

tion with the plus sign here has a cutoff at we. In fact, this equation is
identical to the dispersion relation (2.4.26) obtained for transverse
waves when B0 = 0. Hence a plot of phase velocity vs. frequency
(for v = 0) has the form shown earlier in Fig. 2.6 and repeated here
as the solid curve in Fig. 2.13. The reason for this becomes clear if
we return to the basic equation for 11 [Eq. (2.6.12)]. In the y compo¬
nent of this equation, for 7 = 90° we have

(U - 4>)Uy = 0 (2.7.13)

Hence, we can have a linear polarized wave (uy 9^ 0) for 4> = U, which
is precisely the dispersion relation given in Eq. (2.7.11) using the plus
sign. In other words, the root with the plus sign corresponds to a
wave involving electron velocities solely in the y direction. Since
this is also the direction of B0, the magnetic force term u x B0 is zero
for this wave, and the wave propagates as though J30 were zero.
If we neglect collisions, we easily find for the remaining root (with
the minus sign) in Eq. (2.7.12) that there is the expected resonance at
(coc2 + coe2)1/2 and the reflection points when co = w0i and w02. Hence
the plot of phase velocity vs. frequency has the form shown in the
dotted curve in Fig. 2.13, where it can be seen that there is propaga¬
tion for co > W02 and for u in a band between co0i and (coe2 + coc2)1/2.
For other frequencies the phase velocity is infinite, since k is imaginary.
96 MAGNETOIONIC THEORY

Propagation at other angles

The two special cases we have considered are both unique in that the
resonance at coe is absent at y = 0° and the low-frequency resonance is
absent at 7 = 90°. For all other angles of propagation there are both
high- and low-frequency resonances, so that for the usual case when cce
exceeds |coc| the curve of phase velocity vs. frequency resembles both
the low-frequency portion of Fig. 2.11 (with the resonance shifted to
some value below |«c|) and the high-frequency portion of Fig. 2.13
[with the resonance shifted to some point between aje and (coe2 + wc2)1/2].
This is shown in the top part of Fig. 2.14. Probably the most impor¬
tant points to note are (1) that no waves can propagate at frequencies
between the lower resonance frequency and the lower reflection point;
(2) that the reflection points are independent of 7; (3) that the lower
resonance frequency decreases with angle from |o>c| when k is parallel
to B0 down to zero when k is perpendicular to B0; and (4) that the
higher resonant frequency (which is missing only when k is exactly
parallel to B0) increases with angle but always lies between and
(cOe2 + CO,2)1/2.

For the case when is less than |wc|, which is unusual in a plasma,
we have noted earlier that the roles of we and |coc| are interchanged in
the resonances, so that the phase-velocity plot has the form shown in
the lower part of Fig. 2.14. The major difference in this case is that
since w0i is always less than we, there is no longer any frequency band
where waves cannot propagate.

Fig. 2.13 Plot of phase velocity vs. frequency for waves traveling in a
direction perpendicular to B0, the applied magnetic field.
 2.7 WAVE PROPAGATION PARALLEL OR PERPENDICULAR TO B0 97

The whistler mode

In this discussion we have seen that for the usual case when we is
much larger than |wc|, wave propagation at frequencies below |coc| is
restricted to a cone of angles about the direction of B0. The maximum
angle at which propagation can occur is given by the lower resonance
curve in Fig. 2.10. For uc2 « ue2 one easily finds [from Eq. (2.6.25)]
that

Tmax " COS * | | (2.7.14)

Thus, for a) = (\/3/2)|coc|, Ymax is 30°, and for « = |-|coc|, ymax is 60°.
When « is well below |w0|, the group velocity dw/dk is easily found to
increase with increasing frequency (see Prob. 2.14). Hence if a pulse

Fig. £.14 Plot of phase velocity vs. frequency for waves traveling in
an arbitrary direction relative to B0, the applied magnetic field.
98 MAGNETOIONIC THEORY

with a broad frequency band is transmitted along the direction of B0,

the high frequencies arrive at a receiver sooner than the low ones. If
these frequencies are in the 10- to 20-kc range and the receiver is
attached to a loudspeaker, one hears a descending whistle, lor this
reason, this extra mode of propagation at frequencies below |coc| is
called the whistler mode. Since the cyclotron frequency in the iono¬
sphere is about 1 Me, broadband electromagnetic signals in the 10-kc
range, which seem to be generated by lightning flashes, travel through
the ionosphere in this mode along the lines of the earth’s magnetic
field arid produce natural whistling noises when received with a simple
antenna and loudspeaker.

Problems

2.14 Prove that, for the whistler mode with v = 0, when co is much
less than either the plasma or cyclotron frequency, we have
Vph = C V«lwc| COS y/ue and v„ = 2c y/co|coc| cos y/ue, so that vPh

and va both increase as co1/2.

2.15 Prove that for the r.c.p. wave traveling in the direction of B0,
vg = 0 both at the resonance and at the reflection point. Hint:
Let CO = |coc| - e or CO = co02 + € and then take the limit as € -> 0.
2.16 From the results in Prob. 2.6, show that for waves incident
normally upon a vacuum-plasma interface, the power transmitted
into the plasma goes to zero both at a reflection point and a
resonance.

2.8 ALTERNATIVE DESCRIPTIONS OF DISPERSION RELATIONS

In the last few sections we have used the Appleton-Hartree dispersion

equation (with v = 0) to construct plots of phase velocity vs. frequency
for any angle of propagation relative to an applied dc magnetic field.
These phase-velocity plots are particularly easy to interpret and show
clearly the resonances and reflection points as well as the frequency
bands where waves cannot propagate. However, there are many
alternative figures that can be used either to supplement or to replace
these phase-velocity plots. In this section we shall discuss two of the
most common of these additional plots, the d versus co diagram and
the CMA diagram.

A typical d versus co diagram

Since d is merely the real part of k, it is fairly easy to calculate d as a

function of co for any of our dispersion relations. For example, for
 2.8 ALTERNATIVE DESCRIPTIONS OF DISPERSION RELATIONS 99

Fig. 2.15 Plot of 0 versus co for waves traveling in the same direc¬
tion as B0, the applied magnetic field.

waves traveling in the same direction as B0, the dc magnetic held, we

have noted [in Eq. (2.7.2)] that (for v = 0)

k2c2 = co2 — (2.8.1)

1 ± (|wc|/co)

From this equation one can easily obtain the plot of /3 versus co shown
in Fig. 2.15.
On the j3 versus co plot, the resonances are where f /? —> °° , and the
reflection points are where f3 = 0. From the plot one can also calcu¬
late both vph = co/13 and va = dco/d/3. In fact, by comparing the
curves with the line /3 = co/c, we note that:

1. If /3 is less than co/c, we have vph > c, while for (3 > co/c, we have
Vph ^ C.
2. If the slope d(3/dco is greater than the slope of the co/c curve (as is
the case in the figure), we have vg < c, as required by special
relativity.

Thus the /3 versus co curve, when properly interpreted, provides in a

convenient form a measure of both the phase and the group velocities.

f An exception to this rule is when co = 0, since in this limit vpt, can be zero
even when /3 is finite or zero.
100 MAGNETOIONIC THEORY

Fig. 2.16 The CM A diagram for the Appleton-Hartree equation. The solid
lines are the loci of the reflection -points, where vvh = «, and the dotted lines
are the loci of the resonances, where vph = 0.

The CM A diagram

The final diagram we shall describe is one introduced by Clemmow

and Mullalyf and then extended by Allis; J hence it is generally called
a CM A diagram. This diagram, which is shown in Fig. 2.16 in its
simplest form, displays all the reflection points and resonances as a
function of both X = we2/a>2 and F2 = coc2/o>2.
To see how this diagram is constructed, recall first that the con¬
ditions for a reflection -point were shown [in Eqs. (2.6.26) and (2.6.29)]
to be (for v = 0)

F2 = (1 — X)2 any angle y (2.8.2)

and

X = 1 any angle except 7 = 0 (2.8.3)

Hence the loci of the reflection points lie on the solid curves shown in
Fig. 2.16, which are defined by these two equations. In addition the

f P. C. Clemmow and R. F. Mullaly, in “Physics of the Ionosphere,” p. 340,

Physical Society, London, 1955.
t W. P. Allis, S. J. Buchsbaum, and A. Bers, “Waves in Anisotropic Plasmas,”
The M.I.T. Press, Cambridge, Mass., 1963.
 2.8 ALTERNATIVE DESCRIPTIONS OF DISPERSION RELATIONS 101

condition for a resonance was shown [in Eq. (2.6.23)] to be

1 - X X - 1
F2 = (2.8.4)
1 — X cos2 y A^ cos2 7 — 1

For 7 = 0 this is the straight line l"2 = 1, and for 7 = 90° it is the
straight line F2 = 1 — X. However, for 7 ^ 0 or 90° one can easily
verify that Y2 has two branches, for which both X and Y2 are positive.
(Negative values of Y2 or A" are, of course, not physically meaningful.)
I11 the first branch, which is the high-frequency resonance, Y2 decreases
monotonically from 1 when A" = 0 to 0 when X = 1. In the second
branch (the low-frequency resonance) F decreases from 00 when
A" = sec2 7 to sec2 7 when A —> °o. Hence, from Eq. (2.8.4) one can
easily construct the resonance curves for 7 = 0, 30, 60, and 90°. All
these (except the low-frequency resonances at 60°, where X and F2
are both always greater than 4) are shown as dotted curves in Fig. 2.16.
The two reflection-point curves and the resonance curves for 0 and
90° divide the XY2 plane into eight regions. With some additional
analysis one can make a polar plot of the phase velocity as a function
of angle 7 in each of these regions, so that plase-velocity information
is also included in the CMA diagram.
In summary, there are any number of ways of visualizing the
information contained in a dispersion relation like the Appleton-
Hartree equation. Each has its advantages, and it is probably usefid
to be familiar with a number of these different types of diagrams.
However, in the next chapter, when we again analyze wave propaga¬
tion in a plasma using a more detailed model of a plasma, we shall for
the sake of brevity use only the simple diagrams of phase velocity vs.
frequency.
chapter three

CONTINUUM
EQUATIONS
FOR A PLASMA
chapter three

3.1 DERIVATION OF CONTINUUM EQUATIONS

Macroscopic methods

In Chap. 1 we studied the motion of a charged particle in various

fields, and our plasma model, when we used one at all, consisted of a
collection of individual charged particles interacting with the field (but
not with each other). In Chap. 2 our model was a uniform gas con¬
taining n0 free electrons per cubic meter plus an equal charge density
of positive ions to maintain charge neutrality. The average electron
velocity at each point in space and time was determined by the
Langevin equation, while the average velocity of the heavy particles
(ions and neutral molecules) was assumed to be zero. Now we shall
use yet another model for the plasma, the so-called hydrodynamic or
continuum model, in which the plasma is treated as an interacting
mixture of electron, ion, and neutral-molecule fluids (or gases), each
with its separate set of mass-, momentum-, and energy-transport
equations.
Although a truly satisfactory derivation of continuum equations
can best be obtained from kinetic theory, as in Chap. 5, there are two
purely macroscopic methods which can be used to derive the transport
equations: (1) we may examine the change with time of some 'property
of a fixed mass of fluid in motion, or (2) we may examine the transport
of some property carried along by the fluid through a fixed volume
element.

Mass-conservation equation

To derive the mass-conservation equation for a fluid we use the second

of these methods and equate the mass which flows out of a volume
with the decrease in the total mass contained therein. To calculate
the total mass flowing out of the volume in unit time consider first
the flow across a single wall, as shown in Fig. 3.1. If the average
velocity for the fluid in the x direction is ux and the average density
in the element is p, then during unit time puxdy dz — pu • dS is the
mass flowing out through the surface indicated. Hence, for the entire
 3.1 DERIVATION OF CONTINUUM EQUATIONS 105

surface area, using Gauss’ theorem, we have

Mass out in unit time = J’pu • dS = J'V • pu dV (3.1.1)

The total mass contained in the volume is J'p dV. Hence, for the
volume element

Mass decrease in unit time (3.1.2)

Equating these two quantities, we have

J(p + V • pu) dV = 0 (3.1.3)

which can hold for all volumes only if

p + V • pu = 0 (3.1.4)

This result is called the mass-conservation equation. Note that, for

charged particles, this law (and its derivation) is entirely equivalent
to the charge-conservation law derived in Sec. 1.3.

Momentum-conservation law

In order to derive the momentum-conservation equation it is con¬

venient to use the alternative approach of considering a fixed mass of
fluid which moves through space. If the element is located at a
position R and has a velocity u at time t, then a short time later at
t + At the particle has clearly moved to a new position R + u At.
Therefore, if F is some property of the mass element, the rate of change
of F for the moving element, which we call the hydrodynamic derivative
DF/Dt, is plainly

DF AF F(R + u At, t + At) - F(R,t)

(3.1.5)
~Dt “ a™ At ~ At

Fig. 3.1 Mass flow across a

surface in the x direction during
time At.
106 CONTINUUM EQUATIONS FOR A PLASMA

Expanding F(R + u At, t + At) in a Taylor’s series, we find

DF dF , dF , dF . dF dF
+ u • VF (3.1.6)
TvT = ITT + Ux “I" a- uv “r XT Uz dt
Dt dt dx dy dz

To obtain a momentum equation let F be the x component of

momentum for the mass element Mux, with M the mass of the element.
Then, since M is constant,

DF _ j\,r Dux (3.1.7)

Dt Dt

is the rate of change of the x component of the element’s momentum.

From Newton’s law this must equal the forces acting in the x direction
on the moving element. These forces may be of many types. How¬
ever, initially we shall consider just two force terms:

1. An external force of the form Mfx(R,t), with f the force per

unit mass
2. Pressure forces due to the difference in pressure at different
points on the surface of the mass element

For a gas of identical particles with charge q, mass m, and average

velocity u in an electromagnetic field, the force per unit mass is just
the average Lorentz acceleration

f = — (E + u X B) (3.1.8)
m

To find the net force due to the pressure suppose the mass, at a time t,
is contained in the rectangular volume element shown in Fig. 3.2.
Then the net force acting in the x direction is clearly (recall pressure is
force per unit area)

V
(
(z -
5x
y* y>z)\ - p[■x
/ , Sx \
+ ~2> y>z) by bz (3.1.9)

But expanding both terms about p(R), we have

P(R) - - p(R)

dp M dp
bx by bz (3.1.10)
dx p dx
 3.1 DERIVATION OF CONTINUUM EQUATIONS 107

where in the last step we have noted that the mass M equals the
density p times the volume Sx by bz. From Eqs. (3.1.7), (3.1.8), and
(3.1.10) we have, for our simple gas of identical particles,

(3.1.11)

or, more generally,

~ = 2- (E + u X B) - - Tp (3.1.12)
Dt m p

Many other forces might be added on the right side of this equation:
the gravitational force on the mass element, the interaction force due
to collisions of these particles with particles of other species (which
we add in the next section), and various viscous forces due to stresses
acting on the mass element. However, in many problems the electro¬
magnetic force and the pressure-gradient force, which we have included
here, are of predominant importance, and one can safely neglect all
the other force terms.

Problem
3.1 Derive the mass-conservation equation by letting F — p and not¬
ing that M = p bV is constant. [You must actually calculate
D(bV)/Dt from a figure like Fig. 3.2.]

Fig. 8.2 Volume at time

x— jdx x x+ l&r
t of the mass element M.
108 CONTINUUM EQUATIONS FOR A PLASMA

The isothermal and adiabatic energy equations

Although one can derive a general energy equation using the same
techniques already used to derive the mass and momentum equat ions
we shall for now restrict attention to two special cases, the isothermal
and the adiabatic energy equations. As will be shown in the chapters
on kinetic theory, the scalar pressure p satisfies the condition

p = nuT (3.1.13)

where n = number density

T = temperature
k = Boltzmann’s gas constant = 1.38 X 10-23 joule/°K
Thus in an isothermal process (where T is held fixed)

dp = kT dn = Uis2dp Uis2 = - (3.1.14)

p

where (p/p)112, which enters here, has the dimensions of velocity and is
called the isothermal sound speed. More generally, when the tem¬
perature inside the plasma is not fixed, wTe shall merely assume (as
is often done in hydrodynamics) that any changes in pressure or volume
satisfy the familiar adiabatic lawf (commonly used in thermodynamics)

pp-r = const (3.1.15)

with 7 the ratio of specific heats at constant pressure and constant

volume. The ratio 7 is related to the number of degrees of freedom m
in a gas by the condition 7 = (2 + m)/m. Hence, for a monatomic
gas (where the only degrees of freedom are those of motion in any
direction), m = 3 and 7 = •§. This is the case that we shall deal with
in later chapters on kinetic theory, since it is difficult to treat the
inelastic interparticle collisions which occur in diatomic or polyatomic
gases with internal degrees of freedom.
If we differentiate Eq. (3.1.15), we have

p~r dp — 7pp~(T+1) dp = 0 (3.1.16)

or

dp = — clp = U2 dp (3.1.17)
P

f The conditions which are required to obtain this adiabatic law are investigated
in Sec. 4.2.
 3.2 TREATMENT OF PLASMA AS A MIXTURE 109

with l = (yp/p)1/2 the adiabatic sound speed for the fluid. From
Eqs. (3.1.14) and (3.1.17) we sec that the isothermal or adiabatic
assumptions mean that a change in pressure is accompanied by an
instantaneous, proportional change in the density. For many problems
this is a good approximation; however, as one might expect, in reality,
if the pressure change in a gas is very rapid, there is a time lag before
the density of the gas can respond. Hence for problems involving a
rapidly fluctuating pressure, one must replace these simple laws with
a more general energy equation.

3.2 TREATMENT OF PLASMA AS A MIXTURE

Addition of a collision term

The continuum equations we have developed thus far have the form,
for a gas of particles of type a,

Pa + V • (Pau„) = 0 (3.2.1)

+ <3-2-2>

and

= UJ VPa (3.2.3)

with Ua, depending on the problem at hand, either the adiabatic or the
isothermal sound speed. As we noted in our derivation of the momen¬
tum equation, there are other forces which may act on the mass element
of particles of type a. In particular, if we have a gas mixture, there
should be a force term acting on the mass element of particles of type a
due to collisions between particles of type a and particles of other
species. A simple way to take some account of this force is to assume
that the force per unit mass on the a-particle gas due to collisions with
particles of some other type (3 is proportional to the velocity difference
between the two gases, so that fap — — rap(ua — up), and we have

^
l~)t
^
Wla
(E + ua x B) - ^-
Pa
y Vap(ua - Up) (3.2.4)

The constant of proportionality vap in iap is called the collision

frequency for momentum transfer for particles of type a with those of
type /3. Since momentum must be conserved during interparticle
110 CONTINUUM EQUATIONS FOR A PLASMA

collisions, and since fa/3 is a force per unit mass, we have

Pafa3 + Ppfpa = 0 (3.^.5)

so that these collision frequencies satisfy the important condition

(3-2.6)
PaVa!3 — PWa V '

Equilibration of the velocities

To see the effect of this collision term consider a uniform gas mixture
(so that all spatial derivatives are zero) with no external forces. The
momentum equation for this case becomes

(3.2.7)

which suggests, at once, that at equilibrium (when u„ = 0) one must

have all the velocities equal. To prove this more rigorously, consider
the time rate of change of the fluid kinetic-energy density for the
entire gas mixture W

(3.2.8)

where we have used the fact that for a uniform gas, from Eq. (3.2.1),
pa = 0. Using Eq. (3.2.7) to eliminate ua, we have

(3.2.9)

Now, for any function which is summed over two indices, the result is
unchanged if we interchange the indices. Hence, for any function
fa& we have

(3.2.10)

or

(3.2.11)
 3.2 TREATMENT OF PLASMA AS A MIXTURE 111

Applying this identity to Eq. (3.2.9) and noting that pava0 =

we havef

W = ~ 2 ^ P«^(ua — up)2 (3.2.12)

Since each term on the right is positive, W is negative, and the kinetic
energy decreases until all the individual fluid velocities are equal.
To find the equilibrium value for all the individual fluid velocities,
consider the total fluid velocity

(3.2.13)

Clearly, when equilibrium is established and all the u0 are equal, we

have u = ua. However, we can also show that u is constant with time
(when there are no forces or gradients), since, from Eqs. (3.2.1),
(3.2.6), (3.2.7), and (3.2.11), we have

2 ^ PaVcpiUa — Up + Up — U„) = 0 (3.2.14)

or,/3

Thus, setting ua( °o) = u = u(0), we have for the equilibrium velocity,

u„(oo) = Y * u.(0) (3.2.15)

To summarize, we see that a collision term proportional to the

velocity difference between each pair of fluids provides a force tending
to drive each of the individual fluid velocities toward the total fluid
velocity u.

Problem

3.2 Solve Eq. (3.2.7) for ua(t) when there is a two-fluid mixture and
when there is a three-fluid mixture. (The three-fluid mixture
can best be solved by Laplace transforms.)
f As usual, the square of a vector is defined as the dot product of the vector
with itself.
112 CONTINUUM EQUATIONS FOR A PLASMA

Collision term for energy equation

In Chap. 5 we shall see that one can derive mass, momentum, and
energy equations for a gas mixture using the principles of kinetic
theory. It is shown there that, in an appropriate limit, one obtains a
collisional-force term of the form assumed here in Eq. (3.2.4). How¬
ever, at the same time one also obtains a collisional-transfer term in the
energy equation which tends to equalize the temperatures of the various
species in the mixture if, initially, they are not the same. Since the
use of anything but the simple adiabatic or isothermal energy laws
leads, in general, to severe complications, this more proper energy
equation is not generally used in plasma-dynamics problems, nor shall
we use it at this time. However, when dealing with problems involv¬
ing species with markedly different temperatures, the energy-transfer
terms can be very important and should be included, in some way,
in the analysis.

Problems
3.3 Show that if the energy equation for a homogeneous mixture with
no external forces is

2rnavaji
T„
-1 m„ + m,3
('Ta - Tf,) - ^ |u« - U/3|2

then the total thermal-energy density ^fnaKT„ increases with

a

time at exactly the-same rate as W decreases [see Eq. (3.2.12)].

3.4 Consider a homogeneous mixture of electrons and ions (let
mi = 104me) and make a rough plot of ue, m, Te, and Ti as func¬
tions of time when initially ue = uo, u% = 0, Te = 10 To, and
Ti = T0. Hint: Use Eq. (3.2.7) and the energy equation in
Prob. 3.3.

Comparison with magnetoionic theory

To see how the present set of equations compares with those used in
Chap. 2 in our discussion of magnetoionic theory, note that if we
assume that the flow velocities of the heavy particles are all zero, we
have, for the electron “gas,” from Eqs. (3.2.1) and (3.2.4),

p + V • (pn) = 0 (3.2.16)

and

u r n • Vu — — (E + u X B) — - Vp — ; u (3.2.17)
m p
 3.3 DIFFUSION 113

where all quantities refer to the electrons and v = 2pea is the total
electron collision frequency for momentum transfer with all types of
other particles. As noted earlier, Eq. (3.2.16) is entirely equivalent
to the charge-conservation law, since if we multiply by e/m and note
that p = mne, we have

Pc + V • J = 0 (3.2.18)

with pc = — ehe and J = —neeu. Equation (3.2.17) is similar to

the Langevin equation used in Chap. 2 except for the nonlinear term
u • Yu (which is called the inertial term in hydrodynamics) on the
left and the term

Ue2 VP
(3.2.19)
P

on the right. Since Ue2 is proportional to p/p = nT/m, the omission

of this last term in magnetoionic theory is often called the cold-plasma
approximation. The omission of the inertial term is not so easy to
appraise, but it is certainly justified when u and its derivatives are
small or when (as for transverse waves) u is perpendicular to Vu.

3.3 DIFFUSION

Simple electron diffusion

The pressure-gradient term which is added in the hydrodynamic model

provides a force which tends to smooth out any inhomogeneities in the
density of the fluid and produces either longitudinal waves or a diffu¬
sion of the particles. As an example of this consider the equations
for an electron gas [Eqs. (3.2.16) and (3.2.17)] when E and B are assumed
to be zero, when T is constant, and when the electron number density
ne is only slightly nonuniform, so that we have

ne(R,t) = no + n(R,<) |n| «n0 ^

p(R,0 = nT(n0 + n)

With these assumptions Eqs. (3.2.16) and (3.2.17) become

n + w0V • u = 0 (3.3.2)

kT
u + u • Vu Vn — p u (3.3.3)
mn0
114 CONTINUUM EQUATIONS FOR A PLASMA

As Si final assumption, let u (like n) be infinitesimally small,! so that

u • Vu is of second order and can be neglected compared to u and u.
The equations are now linear. Taking the divergence of Eq. (3.3.3)
and using Eq. (3.3.2), we have (after multiplication by n0)

kT „ . (3.3.4)
—n — V2n + vn
m

or

(3.3.5)
n = De V2n-n
V

where

(3.3.6)
De = —
mv

is called the electron free-diffusion coefficient.

Equation (3.3.5) is a damped-wave equation of a somewhat
simpler type than we shall actually examine in the next section.
However, before working with this full equation it is useful to compare
the magnitudes of the various terms in Eq. (3.3.5). For this purpose
let r and L represent the characteristic times and lengths over which n
changes, so that any spatial derivative is of order L~l and any time
derivative is of order r—1. This procedure is clearly very inexact, how¬
ever, it enables us to get a quick estimate of each of the terms in
Eq. (3.3.5), since we have

n n n
n ~ - De V2n — • (3.3.7)
r V t2v

Comparing the first and last of these terms, we note that if the
average number of collisions for each electron during the time r is
large, that is, vt 1, then the last term [in Eq. (3.3.5) or (3.3.7)] is
small compared to the first, and Eq. (3.3.5) reduces to the diffusion
equation

n = De V2n (3.3.8)

In other words, the number density is governed by a diffusion equation

when the changes in n are slow compared to the time between collisions.

f Note that, in the usual fashion, we also assume that all derivatives of n and
u are infinitesimal.
 3.3 DIFFUSION 115

The diffusion equation has solutions in which any initial non-

uniformities in the plasma are gradually smoothed out (as illustrated
in Prob. 3.5). The characteristic diffusion time td for this smoothing
to occur also follows from Eq. (3.3.7), since, for n — De V2n, we have

n Den
- - (3.3.9)
rd L2

or

L2
td (3.3.10)
De

Thus if, as in Prob. 3.5, there is some characteristic length L associated

with the number density, the diffusion time will be of order L2/De.

Problem

3.5 Consider the diffusion equation, D V2n = n.

(а) Let n = S(R)T(t) to find Tk(t) = const e~DkH and
(V2 + k2)S( R) = 0.
(б) Next, suppose n depends only on x and n(x,0) = n0(x) is
known. Show that S(x) — Ciceikx (k positive or negative)
hence

c{k)eikx~DkH dk and n0 = J c{k)eikx dk

(c) Using Fourier transform theory (outlined in Sec. A.l of the

Appendix), find c(k) and prove that

= 2(Dirty2 f-« n°(x')e-(x-x,)*liDt dx'

(d) Find n(x,t) when n0(x') = e-x'2/*°2 and note that td = x02/D
is the characteristic time for the diffusion to smooth out n.

Limits on free electron diffusion

In the analysis leading to the free-eleetron-diffusion equation there

are two crucial assumptions. First, in eliminating the n term in the
wave equation (3.3.5) we assumed that vtd » 1. If we use Eqs.
(3.3.6) and (3.3.10) to eliminate rD, we find that this condition reduces
to

kT
L2 » mv2 (3.3.11)
116 CONTINUUM EQUATIONS FOR A PLASMA

Now, as noted in Chap. 1, (?,KT/my<2 is the average random speed of

an electron in the plasma; hence (nT/m)in/v is approximately the
average distance an electron travels between collisions with heavy
particles. Thus the condition in Eq. (3.3.11) basically requires the
characteristic length L for the inhomogeneities to be much larger than
the mean free path for an electron-heavy particle collision.
A second crucial assumption used to derive the free electron-
diffusion equation (3.3.8) is that the electric field E is zero. However,
from Maxwell’s divergence equation

V • E = (3.3.12)
€o

it is clear that there is an E field generated whenever the electron

density differs from that of the ions. To obtain a crude estimate of
when the electric-field term is important we can again use dimensional
analysis to have,f from Eq. (3.3.12),

(3.3.13)
to

so that for the electric force per unit mass we have

eE e2nL
/e = -
(3.3.14)
m me o

Similarly, from Eq. (3.3.3), the diffusion force per unit mass is of order

kT kT n
fo = — Tn ~ (3.3.15)
mn0 mn0L

Hence, the neglect of the electric-field term is justified only iorfE «/o,
or

xTe o
L2 « = Ad2 (3.3.16)
n0e2

where, as we have noted earlier, \D is the Debye length, satisfying the

condition

(kT/m )1/2
Ad — (3.3.17)
ue

f Note that, as usual, we assume there is a uniform positive-charge back¬

ground n0e, so that pc = e(n0 — ne) = — ne.
 3.3 DIFFUSION 1X7

From the plot of Debye length in Fig. 1.1 we have seen that \D
is typically very small. Hence Eq. (3.3.16) is rarely satisfied, and for
most problems the electric-field term cannot be neglected.! Therefore
we shall next reexamine the diffusion problem, taking account of the
motion of both the ions and electrons and retaining the E field.

Ambipolar diffusion

To analyze this problem we assume first that the disturbances are

again small, so that, for a either e, for electrons, or i, for (singly
charged) ions, we have

na(R,0 = 7i0 + <(R,f) (3.3.18)

with n'a and the fluid velocities ua all assumed to be of very small
amplitude. With this assumption, we have for each species the mass
conservation equation

n'a + n0V • u„ = 0 (3.3.19)

and the momentum equation (assuming that the temperatures are

constant and Bo is zero)

(3.3.20)

where /3 takes on the values e, i, and n for electrons, ions, and neutral
molecules. Taking the divergence of this equation and using Eq.
(3.3.19), we obtain (after multiplying through by n0 and assuming
that the neutral velocity u, is zero),

t In fact, if L is small enough to satisfy Eq. (3.3.16), it almost surely violates

the first assumption [Eq. (3.3.11)]. This is easily seen if we replace kT/tu in Eq.
(3.3.11) by XnW, so that the two conditions become

\r>2coei
« L2 « \D2
V

When a)e is larger than v (as is the case for most plasmas), it is clear that the
assumptions used in the theory of free electron diffusion cannot both be met.
118 CONTINUUM EQUATIONS FOR A PLASMA

There are two equations of this form, one each for the electrons and
the ions. Therefore, if we replace V • E by the divergence condition

ein'i — w') (3.3.22)

V • E = —^-—

we obtain two coupled equations for the two variables ft' and n'e.
However, these coupled equations are generally too complicated to
treat in detail except by numerical means unless we make further
assumptions. To start we recall from Eq. (3.3.7) that the term na
can be neglected compared to h'avan when the characteristic time for
diffusion r„ satisfies the condition vanra » 1, so that the average electron
or ion has many collisions with neutral molecules during the time of
diffusion. With this assumption we have, from Eq. (3.3.21),

0 = -en0V • E — xTe V2n'e + mevei{n'e - n'i) + mevenn'e (3.3.23)

0 = en0V • E — uTi V2ft' + mt^e(ii' — ft') + mlvinn>i (3.3.24)

where we note, from the linearized version of Eq. (3.2.6), that

mevei = Trixie (3.3.25)

If V • E is again eliminated by using Eq. (3.3.22), the coupled

equations for n\ and n'e are still too complicated to solve easily. How¬
ever, one can obtain surprisingly good results by simply adding these
two equations (to eliminate the V • E and ft' — n'e terms) and setting
rr' = n'e in the remaining terms to obtain the diffusion equation

0= -k(Tc + Ti) VX + (m'Ven + miVin)h'e (3.3.26)

or

(3.3.27)

with

K(Te + Ti)
(3.3.28)
nie^en I nil V\n

the ambipolar diffusion coefficient. In practice, is shown in Sec.

5.5 to be of order
 3.3 DIFFUSION 119

so that

meven ^ /meTA1/2
(3.3.30)
TYliVm \7Yl%T{ J

which is usually very small, since we/mt- is of order 10-4. With this
approximation we have

(3.3.31)
Wli Vin 0 + r) - D< 0 + y)

where A, by direct analogy with Eq. (3.3.6), is the ion free-diffusion

coefficient. This calculation shows that the ambipolar diffusion rate,
for Te — Ti, is twice the rate at which the ions alone would diffuse
and a great deal smaller than the rate at which the electrons alone
would diffuse.
The physical interpretation of ambipolar diffusion is quite simple.
Although the electrons alone tend to diffuse much more rapidly than
the heavy ions, if ne differs significantly from n{, a large electric field
is established, which accelerates the diffusion of the ions and slows
down that of the electrons, so that, to a good approximation, both
species of particles diffuse together. This can be seen somewhat more
analytically if we again compare the force per unit mass due to the
electric field fe = qaE/ma and the “diffusion” force per unit mass
fd = — (^Ta/man0) Vn(,, which acts on one species. If we again let
L be the characteristic length associated with the gradients in the
plasma, then

E ^ Lejn'j - w')
V • E = — or (3.3.32)
eo Co

so that we easily find

U ^ LKK ~ K) (3.3.33)
Jd \D2(a)n'a

where A.o(a) = V^T „£0/n0e2 is the Debye length using the temperature
for species a. Since L2 is generally much larger than Ad2, it follows
that the electric-field force (which tends to equalize n' and n’e) becomes
very strong whenever there is any significant deviation from charge
neutrality.
120 CONTINUUM EQUATIONS FOR A PLASMA

Problem
3.6 Use Eq. (3.3.23) or (3.3.24) to demonstrate that when the ambi-
polar diffusion equation is used for n'e or n(, to a good approximation,

K - K XD2

K ~ u
so that, for \D2 « L2, the analysis is at least consistent.

The Einstein relation

From Eq. (3.2.17) note that the momentum equation for an electron
gas (when the flow velocities of the heavy particles are all zero and the
temperature is constant) has the form

r+- = Me(nE + rxB)-D( Vn (3.3.34)

where r = nu is the electron flux, and, as in Eq. (2.2.9), He = — e/mv

is the electron mobility. If the term involving the total derivative
of u is neglected, this equation shows that the electron flux is caused
by electromagnetic fields or by density gradients. The ratio of the
two coefficients, Ue/De, is known as the Einstein relation. From
Eq. (3.3.6) we have for this ratio

4“ =-%, Einstein relation (3.3.35)

De Kl

Effect of a magnetic field

The general problem of diffusion in a magnetic field is not well under¬

stood at present. However, it is relatively easy to extend our dis¬
cussion of free electron diffusion to include the effects of a uniform
magnetic field B0az. For this case, making the same free-diffusion
assumptions, namely, that E = 0, that the disturbances in number
density and velocity are infinitesimal, and that the diffusion is very
slow, so that vtd » 1, we have, from Eq. (3.3.34),

r = -DeVn + — T x a2 (3.3.36)
V

which is completely analogous to Eq. (2.2.5) for the current density J

due to an applied field E except that T replaces J, — De replaces <j0,
and Vn replaces E. Hence, just as we found J = a • E, now we have

I = - D • Vn (3.3.37)
 3.3 DIFFUSION 121

with [see Eqs. (2.2.6) to (2.2.8)]

Dl 0

D = -Dh Dx 0 (3.3.38)
0 0 De

and

Dy
D .L
y+ f2
(3.3.39)
DeVCOc
Dh
COc2 + F2

As a final point, to obtain the diffusion equation for this case, we

note that the mass-conservation equation (3.2.1) has the form

n -f v • r = 0 (3.3.40)

Hence, from Eqs.,(3.3.37) and (3.3.38) we have, by direct calculation,

n = V • (D • Vn)

(3.3.41)

Since Dx (like <rx shown in Fig. 2.1) falls off rapidly with increasing
values of |o>c|/f, this result indicates that the diffusion of particles
across the field lines is greatly reduced when |coc| is much larger than v.
Experimentally, it is found that while the diffusion is reduced by the
field, the reduction is not always so great as expected from this simple
theory (where D± is proportional to B0~2 for |wc| » f), and various
instabilities have been proposed to explain this anomalous diffusion
in the experiments.

Problems

3.7 In some discussions of ambipolar diffusion it is assumed that

= Ik. Show that if this were true, a dc magnetic field would
have no effect upon diffusion, and one would still find ne = Da V2we.
3.8 (a) Prove that plane-wave solutions to the diffusion equation
satisfy the dispersion relation k2D = ico, so that, for free
electron diffusion we have k2A2 — iuve, and for ambipolar
diffusion we have k2Av2 = iupin, where

Ae = (uTe/vie) 1/2 = (Pe/Pe)112

122 CONTINUUM EQUATIONS FOR A PLASMA

is the isothermal electron sound speed and

Ap = [k(T. + TJ/m,] ^ l(pe + Pi)/(Pe + pl)]1/2

is the isothermal plasma sound speed.

(b) Plot the phase velocity and damping factor as a function
of frequency for these waves.
(c) Are these waves longitudinal or transverse?

3.4 WAVE PROPAGATION IN A WARM PLASMA

Waves in an electron gas

In the last section we saw that the neglect of u« or n'a compared to
or vn'a in the continuum equations leads to a diffusion equation.
Now we want to retain this term in all our equations and examine the
plane-wave solutions to the equations for an infinite homogeneous
plasma. To start, let us consider just the equation for the electrons
when the heavy particles are all assumed to have zero velocity. For
this case, assuming small-amplitude adiabatic disturbances with

ne = no + neikx \n\ <$C no

11 ,= U <K ^ (3.4.1)
icot

B = Bo + beikx~iut

the mass- and momentum-conservation equations for the electron gas

[Eqs. (3.2.16) and (3.2.17)] become

— zcon + iknQux = 0 (3.4.2)

and

— iuvL = — — (E + u X B0) — — na, - m (3.4.3)

m n0

with Ue2 = yPeo/peo = ynTe/m the square of the adiabatic electron

sound speed and v the electron collision frequency. As in Sec. 2.6,
we next separate u, E, and Bo into longitudinal and transverse parts:
u = ux + u(; E = Ex + E(; B0 = Box + B0(. Then, since Maxwell’s
equations are unaffected by the change to the continuum model for
 3.4 WAVE PROPAGATION IN A WARM PLASMA 123

the plasma and u, is assumed zero, we still have [see Eqs. (2.6.6) and
(2.6.7)]

in0eux
Ex (3.4.4)
coe0

in0eut
E, (3.4.5)
coe0(l — n2)

Substituting these expressions for E and Eq. (3.4.2) for n into Eq.
(3.4.3) and multiplying through by ico, we obtain

ie<o
Uj(u2 — W2 + iwv — k2Ue2) = — (u, X B„,) (3.4.6)
m
2
ieu
U, W lWv\ = —
-I- icc (u* x B0, + u, x Bo*) (3.4.7)
1 ni m

These equations are identical to Eqs. (2.6.9) and (2.6.10), from

which we derived the Appleton-Hartree dispersion relations, except
for the single term k2Ue2 in the ux equation, f Hence with very little
effort we could derive a determinantal equation similar to the one used
in developing the Appleton-Hartree equations. Such an equation,
however, is now (because of the extra k2 term) a cubic equation in k2
with three roots whose exact solutions are too complicated to be of
much practical use. Therefore, as in Chap. 2 we shall restrict our
attention to two special cases:

1. Wave propagation in the direction of B0

2. Wave propagation in the direction perpendicular to B0

From these two cases and a knowledge of the reflection and resonance
points, it is again possible to infer the dispersion curves for waves at
an arbitrary orientation of B0.

Propagation parallel to Bo
For our first example we set B0 = Boax, so that B0, = 0, and Eqs.
(3.4.6) and (3.4.7) become

Wx(c02 — We2 + iu>V — k2Ue2) = 0

Ue e_Bo
+ iwv iwcwUz Uc = (3.4.8)
Uy
1 — n2 ) m

U)e2
uz — iwwcUy
1 - n2

f Since Uc2 goes to zero when T = 0, these equations are often called the warm-
electron-plasma equations to distinguish them from the cold-plasma equations in
magnetoionic theory.
124 CONTINUUM EQUATIONS FOR A PLASMA

From the first of these we see that, to have longitudinal waves (ux ^ 0),
the dispersion relation is

k2Ue2 = co2 - coe2+ iwv (3.4.9)

(Recall that using the Langevin equation, we had found this result
minus the crucial term on the left.) When collisions are neglected,
note that this longitudinal-dispersion relation is entirely analogous
to the simple transverse-wave dispersion equation (when B0 = 0)
except for the change from c2 to Ue2- Hence a phase-velocity plot
(for v = 0) has the familiar form (shown in Fig. 3.3) of a reflection
point at cce followed by a rapid decrease in the phase velocity toward
Ue as co increases beyond a>e.
The remaining two equations in (3.4.8) do not involve Ue and are
precisely the two transverse equations obtained in magnetoionic theory.
Hence, they lead, again, to the l.c.p. and r.c.p. waves described in Sec.
2.7 with phase velocities (for v = 0) as shown in Fig. 2.11 and repeated
here in Fig. 3.3. Thus the effect of the warm-plasma model for this
case is to add a new longitudinal (or acoustic-type) wave (in place
of the simple oscillations at the plasma frequency) to the two trans¬
verse waves we had already encountered.
The reason that the addition of the pressure-gradient term has no
effect on the propagation of the transverse waves is apparent if we recall
that we have used the adiabatic assumption, Vp = Ue2 Vp, so that Vp
is zero if the density is constant. Now, from mass conservation we
see in Eq. (3.4.2) that for small-amplitude waves traveling in the x

Fig. 8.8 Plot of phase velocity vs. frequency for plane waves traveling in the
same direction as B0, the applied magnetic field, when the warm-electron-gas
equations are used.
 3.4 WAVE PROPAGATION IN A WARM PLASMA 125

direction the fluctuations in the density are given by

kn0ux
n = - (3.4.10)
co

Hence, for a transverse wave (where ux = 0), the number density is

constant, and Vp = 0.

Problem

3.9 Make a rough plot of a (the imaginary part of k) as a function

of v for the longitudinal-wave dispersion relation. Contrast
the result with the damping of the transverse waves described
in Sec. 2.5.

Propagation perpendicular to B0

For our second example let B0 = B0a„, so that Eqs. (3.4.6) and
(3.4.7) become

Ux(c02 — CO e2 + ICOV k2Ue2) + ic0C0cUz = 0

Uy ^C02 iMt*+ *"")= 0 (3.4.11)

ue2 i • N n
— icococux + Uz • --, + icov ) - 0
1 — n2 )

The second of these equations shows that, as in the cold-plasma case,

for uy 9^ 0, we have a wave with the familiar transverse-wave dispersion
relation

k2c2 — co2 (3.4.12)

1 + i{v/u>)

This result is again unchanged since this is a purely transverse wave,

so that the pressure-gradient term goes to zero.
From the ux and uz equations above, we note again that ux and uz
must equal zero unless the determinant of their coefficients vanishes,
that is,

co2 — co2 + icov — k2U 2 lC0C0c

CO2 = 0 (3.4.13)
— icococ co2 + icov
1 — n‘
126 CONTINUUM EQUATIONS FOR A PLASMA

If we expand this determinant and neglect collisions by assuming that

v = 0, this dispersion equation becomes

k4C2Ue2 — fc2[c2(w2 — We2 — «c2) + U e2(u2 — ^e2)]

+ (co2 - coe2)2 - co2coc2 = 0 (3.4.14)

Since Ue = 4.9 X lO3^1'2 m/sec, it is clear that Ue2 is generally very

much less than c2, so that the underlined terms in Eq. (3.4.14) may
generally be neglected. Even with this simplification, the exact solu¬
tion is cumbersome to work with. Therefore we shall obtain approxi¬
mate roots to this equation by an approximation scheme (which we
shall also find useful in many later examples).
First, let us assume that there is a solution to Eq. (3.4.14) for
which \k2Ue2\ « w2. (This will be referred to as the high-phase-velocity
wave, since, for k2 positive, this condition implies that the phase
velocity is much greater than Ue.) With this assumption Eq. (3.4.14)
reduces to

fc2C2(c02 — I0e2 — COc2) ~ (a>2 — We2)2 — U2U, 2

= co2(w2 — C0e 0>( 2) — C0e2 (cO2 — Cde2) (3.4.15)

or

we
k2c2 = co2 (3.4.16)
1 — C0c2/ (co2 — We

Fig. 3.4 Plot of phase velocity vs. frequency for plane waves traveling per¬
pendicular to B0, the applied magnetic field, when the warm-electron-gas equa¬
tions are used. Note that in place of a resonance at (o)c2 + o><.2)1/2, there is a
transition from a basically transverse wave to a basically longitudinal one.
 3.4 WAVE PROPAGATION IN A WARM PLASMA 127

This is exactly the same dispersion relation (for v = 0) as we found for

a cold plasma [Eq. (2.7.12) with the minus sign]. Hence the phase-
velocity plot has the form shown in the dotted curve of Fig. 2.13 (and
repeated here as part of Fig. 3.4). Note, however, that a ■portion of
the curve, near the resonant frequency (we2 -f- coc2)1/2, is no longer a
valid solution, since' the phase velocity here is no longer large
compared to Ue-
We shall clarify this point shortly, but first, to obtain the second
approximate root to Eq. (3.4.14) let us assume that \k2c2\ »w2.
(This will be referred to as the low-phase-velocity wave, since, for k2
positive, this condition implies that the phase velocity is much less
than c.) With this assumption we have (for frequencies comparable
to, or greater than, ue) from Eq. (3.4.14),

k4c2Ue2 - k2c2(co2 - we2 - coc2) = 0 (3.4.17)

or

(3.4.18)

This result is identical to the longitudinal-wave dispersion relation

when B0 is zero [Eq. (3.4.9) with v = 0] except for the change in the
reflection point from coe to (a>e2 + toc2)1/2. Hence a plot of phase
velocity vs. frequency for this root has the form shown in the curve
with dots and dashes in Fig. 3.4. Again, however, the portion of the
curve in the frequency range near (a>e2 + wc2)1/2 is no longer a valid
solution, since it violates the low-phase-velocity assumption used to
derive the result.
To summarize, we see that both Eqs. (3.4.16) and (3.4.18) appear
to break down for co = (ooe2 + wc2)1/2. To obtain the proper solution
in this region we can solve Eq. (3.4.14) exactly when w = (oje2 + u)c2)1/2.
We have for this frequency, with no approximation,

k*c2Ue2 - k2U 2(a 2 - wcW = 0 (3.4.19)

so that

UM ± (t/AV + 4C/e2cWo>e2)1/2
k2 =
2c2Ue2

(3.4.20)
128 CONTINUUM EQUATIONS FOR A PLASMA

The solution with the minus sign is, of course, an evanescent wave,
while the phase velocity for the root with the plus sign is

1/2
CO (oJe2 + Uc2)cUe (3.4.21)
k

which generally lies between c and Ue, as indicated in big. 3.4.

This shows that for a warm plasma, when w = (o,e2 + wc2)1/2,
there is, in place of the resonance in magnetoionic theory, a transition
from a basically electromagnetic wave [for u < (u>e2 + wc2)1/2] to a
basically electroacoustic wave [for co > (we2 + wc2)1/2[.

Conclusions
Comparing Fig. 3.4 with the analogous cold-plasma result shown in
Fig. 2.13, it is clear that the principal difference lies in the absence of a
resonance at (we2 + wc2)1/2. Instead of dropping to zero, the phase
velocity drops toward the electron sound speed Ue. This same effect
is found for wave propagation at other angles where one of the reso¬
nances is eliminated, and, instead, there is a transition to the acoustic
mode.
The one remaining resonance is easily found (see Prob. 3.10) to
be at the frequency co = |wc| cos y, which is the value for the low-
frequency resonance found for the cold plasma in the limit when
we » |coc| and used in our discussion of the whistler mode. Hence a
typical plot of phase velocity vs. frequency for an arbitrary angle
between 0 and 90° has the form shown in Fig. 3.5.

Fig. 3.5 Plot of phase velocity vs. frequency for plane waves traveling
at some angle y relative to B0 when the warm-electron-gas equations
are used.
 3.5 LONGITUDINAL WAVES IN A FULLY IONIZED GAS 129

Problem

3.10 (a) From Eq. (3.4.8) form a determinantal equation analogous

to Eq. (2.6.12) in magnetoionic theory, with only the ele¬
ment An changed.
(6) From your result show that the “resonance condition”
(4> = 0, k —> oo) for the warm plasma is co = |coc| cos y.

3.5 LONGITUDINAL WAVES IN A FULLY IONIZED GAS

Equations for the fully ionized gas

In the last section we saw the effect of extending magnetoionic theory
by including the electron pressure-gradient term in the momentum
equation. Now we want to extend magnetoionic theory further by
considering the effect of the ions on the wave-propagation problem.
For a fully ionized gas the mass and momentum equations for the
electrons and for the ions have the form [see Eqs. (3.2.1) to (3.2.4)]

na + V • (naua) = 0 (3.5.1)

- (r (E + u. X B) - ~ vn. - - u„) (3.5.2)

E/ L ••('a ’voL

where a or /3 takes on the value e for electrons and i for ions (with
/3 5^ a), and where Ua is, again, the adiabatic sound speed for particles
of type a. Restricting ourselves again to small-amplitude waves, so
that

na = n0 + n'aeikx iat |n'| « n0

ua = uaeikx~iut ua « ^

(3.5.3)

B = Bo + be**1-1"'

we have, from Eq. (3.5.1),

, kTlQUax
na = -- (3.5.4)
CO

and, from Eq. (3.5.2),

UJ
— icoua = — (E + ua X B0) iknaax ^at)9(via *•»*) (3.5.5)
ma n0
130 CONTINUUM EQUATIONS FOR A PLASMA

For this case, the current term in Maxwell’s equations is

J = n0e(ui — ue) (3.5.6)

which is like the current in an electron gas except for the change from
ue to ue — Ui. Hence, in place of Eqs. (2.6.6) and (2.6.7) we obtain
from Maxwell’s equations

(3.5.7)

(3.5.8)

We can now obtain a set of equations involving only ue and u;

simply by replacing n'a and E in Eq. (3.5.5) by Eqs. (3.5.4), (3.5.7),
and (3.5.8). This leaves us with a set of six scalar equations from
which we can derive a dispersion equation (in the usual fashion) by
setting the determinant of the coefficients equal to zero. The resulting
equation is a fourth-order equation in k2, showing that there are now
four kinds of waves that can propagate in the plasma. Since the
treatment of the fourth-order equation is too complicated to discuss
in detail, we limit our analysis to the important case when B0( = 0,
that is, when wave propagation is in the direction of the applied dc
magnetic field.

Longitudinal waves
With B0 in the x direction, u„ X B0 is perpendicular to ax. Therefore
when we substitute Eqs. (3.5.4) and (3.5.7) into Eq. (3.5.5) we have,
from the x component of the electron and ion equations (after multi¬
plication by ioo),

Uex(u2 — k2Ue2 + IWVei ~ 03 e2) + Uix{03 e2 — 103Vei) =0

(3.5.9)
Uex(ui2 — i03Vie) + Uix(o32 ~ k2U\2 + io3Vie ~ 03(2) = 0

where w, = (e2n0/tn,:€0)1/2 is the ion plasma frequency. Note that,

from Eq. (3.2.6) and from the definitions of the plasma frequencies,

me
and (3.5.10)
Mi

In similar fashion, from Eq. (3.1.17),

 3.5 LONGITUDINAL WAVES IN A FULLY IONIZED GAS 131

Thus for a typical ionized gas the ion collision frequency, plasma
frequency, and sound speed are all much smaller than their electron
counterparts. This enables one to simplify much of the algebra
involved in obtaining the dispersion relations.
To have longitudinal waves (uex, UiX 9^ 0) the determinant of the
coefficients in Eq. (3.5.9) must be zero; that is,

(co2 - k2U2 - fie2)(CO2 - k2Ui2 - W) - SVfi;2 = 0 (3.5.12)

where fte2 = coe2 — ioovei and fl,2 = co;2 — iwvie, so that

777
^2 = a2— (3.5.13)
»!,■

High- and low-frequency limits

If we write out all the terms in Eq. (3.5.12), we have [using Eqs.
(3.5.11) and (3.5.13)]

k4Ue2Ui2 - k2
M)
U2(Ue2 + Ufi) - UfiU2 ( 1 + ^

+ CO4 - «2(SV + fii2) = 0 (3.5.14)

where the two underlined terms are very small and will henceforth be
neglected. Although it is not difficult to write down the two exact
solutions to this quadratic equation in k2, it is more convenient to
seek out separately some approximate solutions at high and low fre¬
quencies. For this purpose we again neglect collisions (so that
fl2,- = co2t-) and first assume that w2 » wt2[l + (Ti/Te)], so that Eq.
(3.5.14) reduces to

kiUe2Ui2 - k2U2c2 + w2(co2 - avO = 0 (3.5.15)

or

{k2U2 - a>2 + u2){k2Ui2 - U2) ^ 0 (3.5.16)

Note that in the last step we have used the approximation

£W » - ue2| (3.5.17)

which follows from Eq. (3.5.11) for w2 >>> coi2[l + (Ti/Te)]. From
Eq. (3.5.16) we easily see that for high frequencies the two dispersion
132 CONTINUUM EQUATIONS FOR A PLASMA

Fig. 3.6 Plot of phase velocity vs. frequency for the longitudinal waves
in a fully ionized plasma. This curve is for propagation in the direc¬
tion of B0, the applied magnetic field (and also holds when B0 = 0).

relations are

k2Ue2 = W2 - We2 WUi2 = CO2 (3.5.18)

The first of these is simply the longitudinal electron wave described in

the last section, while the second is a new wave, a longitudinal ion
wave, which travels at the ion sound speed. The phase velocity for
these waves is illustrated in the high-frequency portion of Fig. 3.6.
Next, for frequencies much less than co,[l + (Ti/Te)]1/2, Eq.
(3.5.14) reduces to

k'Ue'Ui2 + k2Ue W co2coe2 = 0 (3.5.19)

To find the approximate roots of this equation we divide by me/nii and

note that, to a good approximation, f we then have

(.k2Up2 - co2){k2U2 + co,2) = 0 (3.5.20)

where

7 k{T, + Tj) jk TeTi

U 2 = (3.5.21)
nii m { Te + Ti

f The approximation is only that Eq. (3.5.20) contains an extra term, — k2U^(t>2,
which, however, is small compared to fc2f7p2co,-2 [the second term in Eq. (3.5.19)
when we divide by mc/m,] for co2 « co,-2(l + Ti/T,).
 3.5 LONGITUDINAL WAVES IN A FULLY IONIZED GAS 133

Thus for low frequencies the two dispersion relations are

A:2 C7P2 = co2 /c2C/p2 = -cw2 (3.5.22)

of which the first is a wave traveling at the plasma sound, speed Up,
while the second is an evanescent wave (with infinite phase velocity).
The phase velocity for these solutions appears as the low-frequency
portion of Fig. 3.6. To complete the curve one need only solve the
dispersion equation at w2 = Wi2[l + (Ti/Te)] to have

k*Ue2US ^ cow or k2UeUi^±uue (3.5.23)

so that, for the plus sign, the phase velocity (after some simple algebra)
is found to be

£ = (CTC/p) 1/2 co = CO, ^1 + (3.5.24)

A more detailed analysis shows that the longitudinal waves with

phase velocities equal to Ue or C/; represent, respectively, acoustic-type
oscillations of just the electrons and just the ions, while the low-
frequency waves with phase velocity Uv represent an acoustic oscilla¬
tion of both the ions and the electrons. This comes as no surprise,
since for a gas of electrons and ions the combined sound velocity is
(yPtot/ptot)1/2; but Ptot = n0K(Te + 7\), while

Ptot = n0(me + = Worn,

Hence

yK(Te + Tj) 11/2

= C/p (3.5.25)
mi

so that C/p is the appropriate sound velocity for an acoustic wave in a

gas consisting of electrons and ions.

Problems

3.11 Verify Eqs. (3.5.16), (3.5.20), and (3.5.24).

3.12 Find the dispersion relation (neglecting collisions) for longitudi¬
nal waves in a plasma of electrons and holes with mn = me and
Th = Te.
134 CONTINUUM EQUATIONS FOR A PLASMA

3.6 TRANSVERSE WAVES IN THE DIRECTION OF B0

Development of the dispersion relation

If we return to the equation for u [Eq. (3.5.5)] and take now the
transverseportion, we have, after multiplying by i/w, neglecting
collisions, and setting B0 = B0ax,

ue( = <E(ue, — uu) — iYe(uet X a*) (3.6.1)

uit = — uet) + iYi(ui, X ax) (3.6.2)

where [as in Eq. (2.6.14)]

co<*2 eB o
<^>a — 9 19 9 ^ a
(3.6.3)
co“ — k2c2 ma co

with a equal to e for electrons and i for ions. From Eq. (3.5.10)
we have

4>i _ Yj _ me
(3.6.4)
Ye mi

so that, again, for an ionized gas some simplification is possible because

of the large mass difference. We next reduce Eqs. (3.6.1) and (3.6.2)
to a single vector equation involving just ue< by noting, from Eq.
(3.6.1), that

<Eu,( = ue((4>e — 1 — iYe x a*) (3.6.5)

Hence, if we multiply Eq. (3.6.2) by 4>e and use this result, we find

[ue((4>e — 1 — iYe X ax)](l — 4>, — iYi x ax) + = 0 (3.6.6)

As can be easily seen (from a drawing such as Fig. 3.7),

(uet X ax) X ax = — uet (3.6.7)

Hence, collecting all terms in Eq. (3.6.6), we have [after using the
identity 4\Fe = <EF;, which follows from Eq. (3.6.4)]

ue((<E + <E — 1 + YeYi) = i(Ye — Yi)uet x a. (3.6.8)

 3.6 TRANSVERSE WAVES IN THE DIRECTION OF Bo 135

Fig. 3.7 Geometry for (uet X a*) X a*.

Both of the underlined terms in this equation are very small, as

evident from Eq. (3.6.4), and could be neglected from here on; how¬
ever, we shall find it just as convenient to retain the F; term. Then,
writing out each component of Eq. (3.6.8), we have

Uey($e - 1 + YeYi) = l{Ye - F\)uez

(3.6.9
Uez($e ~ 1 + YeYi) = — i(Ye — Yi)uey

Comparing the ratio of uey to uez in each of these equations, we have

Uey _ u.
Or Uey2 = — Uez2 (3.6.10)
u,
3
N

Hence, as in the same problem in magnetoionic theory [see Eq. (2.7.6)],

we have

Uey i 1Uez (3.6.11)

where the plus- and minus-sign solutions here correspond again to

l.c.p. and r.c.p. waves, as shown earlier in Fig. 2.12.
To obtain the two dispersion relations we substitute Eq. (3.6.11)
into either part of Eq. (3.6.9) to have

Uez&e - 1 + YeYi) = ± (F. - Y%)u„ (3.6.12)

Hence, to have waves (w« ^ 0) we must havef

= 1 ± (Ye- Yt) - YeYi = (1 ± F«)(l + Ft) (3.6.13)

t Note that in the last expression on the right of Eq. (3.6.13) we use either
both upper signs (for the l.c.p. wave) or both lower signs (for the r.c.p. wave).
136 CONTINUUM EQUATIONS FOR A PLASMA

Hence, from Eq. (3.6.3)

w 2
k2c2 = w2 — -~—
<t>e

= co2 - n -L. i i /vi --TT (3-6.14)

(1 ± |wCe|/w)(l + Uci/U)

where |coce| = eB0/me and uci = eBo/rrii are, again, the electron and
ion cyclotron frequencies. This result is identical to the dispersion
relations obtained in magnetoionic theory [Eq. (2.7.2) with v = 0]
except for the underlined term in the denominator. Hence, for fre¬
quencies large compared to ccci (where the underlined term is negligible
compared to 1) the phase velocities obtained from Eq. (3.6.14) are
identical to those obtained in magnetoionic theory and shown in
Fig. 2.11 (as well as here, in the high-frequency part of Fig. 3.8).

Alfven waves
When we go to the low-frequency limit, i.e., when we set co <3C wci,
the dispersion relation (3.6.14) reduces for both the l.c.p. and r.c.p.
waves to

k2c‘ = w2 (l + (3.6.15)

Note that in the term on the right

coe2 _ n0rrii _ c2
(3.6.16)
| coce| C0ci e0B o2 -Va2

Fig. 3.8 Plot of phase velocity vs. frequency for the transverse waves in a
fully ionized plasma. This curve is for propagation in the direction of B0,
the applied magnetic field.
 3.6 TRANSVERSE WAVES IN THE DIRECTION OF B, 137

where Va is called the Alfven velocity and is given by

Bo
Va = 1/2 (3.6.17)
(Mo Pi)

For the usual case when Va is much less than c, Eq. (3.6.15) reduces to

k2VJ = co2 (3.6.18)

so that .both waves travel at the Alfven velocity Va, as shown in the
low-frequency portion of Fig. 3.8.

Resonances
Recall that resonances occur when k goes to infinity [or the denomina¬
tor in Eq. (3.6.14) goes to zero]. Clearly for the r.c.p. wave, where
we use the lower signs in the denominator, the resonance is still
exactly at the electron cyclotron frequency, while for the l.c.p. wave,
where we use the upper signs in Eq. (3.6.14;, there is a new resonance
exactly at the ion cyclotron frequency, as shown in Fig. 3.8. Again
this resonance is riot unexpected, since we have seen that the natural
motion of an ion in the field consists of left-handed spirals at the ion
cyclotron frequency, so that this resonance is entirely analogous to
the one at |uce| for the r.c.p. wave.
The appearance of this new resonance is one of the most important
features of this analysis. In fact the high absorption of waves that
occurs (when damping effects are included) at the ion cyclotron fre¬
quency is used as a means of heating a plasma in some fusion devices.

Problems
3.13 Prove that there is a small shift in the reflection points co0i and w02
when we take account of ion motions and that, from Eq. (3.6.14),
the reflection points occur at

OJ = -£{±(|wce| — wci) + [(1| + Wet)2 + 4cOe2]1/2 }

Note that this ensures that w0i exceeds coci and a>o2 exceeds |a>ce|.
3.14 For propagation at an arbitrary angle relative to B0, the resonances
in a fully ionized cold plasma are found to occur at

w2(w2 — |cocc|ojcj) -1
(w2 — wce2)(o>2 — Wd2) _

Use this formula to make a rough plot of the three resonant fre¬
quencies as a function of y with w« > |wc|. Your result will be
138 CONTINUUM EQUATIONS FOR A PLASMA

much like Fig. 2.10, but with two exceptions: (1) the curve that
starts at |u>c| falls only to wc(wi2 + ojCi2)1/2/(we2 + uce2)1/2 as 7 >90 ,
and (2) there is the third resonance, which decreases from coci
when 7 = 0 to zero when 7 = 90°.
3.15Starting with Eqs. (3.5.5) and (3.5.7), show that when collisions
are neglected there are only two resonances in a warm fully ionized
plasma and that these are at w = |wc«| cos 7 and w = uci cos 7.

3.7 LINEARIZED MAGNETOHYDRODYNAMIC EQUATIONS

Continuity equation
Thus far in this chapter we have been using separate equations for
each type of particle in the plasma. We have always used the set of
continuum equations for the electrons, and in some sections (when the
motion of the ions was not neglected) we also used the continuum
equations for the ions. Moreover, it is clear that we could have
included the possible motion of the neutrals by adding a set of con¬
tinuum equations for the neutral molecules. An alternative approach,
which we now consider, is to replace the set of individual-species equa¬
tions by a set of equations for the gas as a whole. These new equations
are known as the magnetohydrodynamic (mhd) equations, and in their
most general form they are entirely equivalent to the set of individual
fluid equations we have been using. However, under some conditions
(generally for steady-state or slowly varying problems) the mhd
equations are particularly convenient to use and lead to results which
are not easily obtained from the multifluid equations.
Since they are entirely equivalent, we shall derive the set of mhd
equations from the multifluid equations we have already used in this
chapter. In linearized formf these multifluid equations (3.2.1) to
(3.2.4) are

p'a T PaO^ ■ ua = 0 Pa ^ P<*0 (3.7.1)

u„ = — (E + u„ X Bo) - — - y vap(ua - U@) (3.7.2)

ma PaO A*

where p'a, u«, pa, and E are all assumed to be infinitesimal disturbances
and pao is constant. To obtain the mass-conservation (or continuity)

f To make this derivation from the nonlinear equations one must modify
slightly the pressures and temperatures in the multifluid equations to get a simple
set of mhd equations. This point is discussed more fully in Sec. 3.10 and in Chap. 5.
 3.7 LINEARIZED MAGNETO HYDRO DYNAMIC EQUATIONS 139

equation for the gas as a whole, we sum Eq. (3.7.1) over a to have

X Pa + X PaoV • ua = 0 (3.7.3)

Now, we let p and u be the density and velocity for the gas as a whole
such that

P — Po + p (3.7.4)

and

u
2 PaOUa
PO
(3.7.5)

[This last equation is, of course, the linearized form of Eq. (3.2.13).]
From Eqs. (3.7.4) and (3.7.5) it is evident that

P = X P« ar,d PoV • U = X PaoV • U« (3.7.6)

a a

Applying these to Eq. (3.7.3), we have

P d- p0V • u = 0 (3.7.7)

which is the continuity equation for the gas as a whole.

Momentum equation
In similar fashion to obtain a momentum equation for the gas as a whole
we multiply Eq. (3.7.2) by pa0 and sum over a to obtain

X PaOlTa X^'a0^a(^ "l- X Bo) X ^Pa XPa0^a^(^a ^^) (3.7.S)

a ct a a,p

Let us now consider each of these terms in their turn. From Eq.
(3.7.5) the term on the left equals p0u. Next, in the Lorentz force
term on the right

X n<*oqa = total equilibrium charge density (3.7.9)

140 CONTINUUM EQUATIONS FOR A PLASMA

and

(3.7.10)

For almost all problems, the undisturbed or equilibrium gas has no net
charge density; hence the sum in Eq. (3.7.9) vanishes. In the pres¬
sure-gradient term we let p be the total pressure (p = 2pa) so that
^ Vpa becomes just Vp. Finally, in the last term we recall from

Eq. (3.2.14) that this double sum vanishes. Thus, when we simplify
each term in Eq. (3.7.8), we reduce the momentum equation for the
gas as a whole to the simple form

pou = J x B„ - Vp (3.7.H)

Generalized Ohm’s law

The simple momentum equation for the gas as a whole is very useful,
but somewhat deceptive, since it provides only one equation involving
u and J, while the multifluid set of equations for a mixture with s
components has s individual momentum equations. Hence to ha\ e
an entirely equivalent system of equations for the gas as a whole we
must form s — 1 additional equations involving u or J (or othei com¬
binations of the u«). The simplest (and probably most important)
case to consider is a fully ionized gas, where, initially, our momentum
equations involve ue and ut. Hence, only one additional equation
is needed. In the mhd equations we transform to the new variables

_ pepue T p,pUt _ me\ie + mjxij (3.7.12)

me + ra»

and

J = en0(ui — ue) (3.7.13)

Inverting these equations, one easily finds (with no approximation)

that

mj (3.7.14)
U; = u +
en0(mi + me)

mi J (3.7.15)
u
en0(mi + me)
 3.7 LINEARIZED MAGNETOHYDRODYNAMIC EQUATIONS 141

Equation (3.7.11) clearly provides one equation involving u and

J. To obtain a second equation we merely substitute Eqs. (3.7.14)
and (3.7.15) into either the electron or ion momentum equation (3.7.2).
From the electron equation we find by direct substitution that

mj m,;J
u — E + u X B0
en0(m,i + me) m, en0(m,i + me)

- ^ — J (3.7.16)
Pe 0

If we use Eq. (3.7.11) to eliminate ii, we have (after multiplication

by me/e)

j — E u X Bo -- 1 [nii Vpe — me Vpt

€0(cde2 + W»2) ao ep o

- (w,- - me) J x Bo] (3.7.17)

where, as in Eqs. (2.2.2) and (2.4.19),

nQe2 n0e2
co w0 (3.7.18)
mvei mat0

Equation (3.7.17) reduces to the simple generalized Ohm’s law,

described in Sec. 2.2,

J = a0(E + u X Bo) (3.7.19)

whenever J on the left and the entire bracketed term on the right can
be neglected. Hence, Eq. (3.7.17) is also often referred to as a gen¬
eralized Ohm’s law. The bracketed term in Eq. (3.7.17) can be put
into an alternative form, if we eliminate J x B0, by using Eq. (3.7.11).
This leaves, for the bracketed term,

— [me Vpe — mi Vpi — {mi — me)p0u] (3.7.20)

cp o

The mhd approximaxions

Since we have made no approximations thus far, the momentum equa¬
tion and the generalized Ohm’s law [Eqs. (3.7.11) and (3.7.17)] are
entirely equivalent to the electron and ion momentum equations.
These equations, together with the two Maxwell curl equations, could
142 CONTINUUM EQUATIONS FOR A PLASMA

therefore be used to treat any problems involving motion of the ions

and of the electrons. In practice, however, these equations are rarely
used in their entirety. Instead, various approximations are generally
made on the basis of (often questionable) physical arguments which
permit one to eliminate many of the terms in this set of equations.
The most commonly used approximations involve a simplification
of the Maxwell curl B equation and the generalized Ohm s law. Recall
that the Maxwell equation (1.3.12) has the form

V x B = mo(€oE + J) (3.7.21)

Since we often have J = a • E, we can use dimensional analysis again

to set

((E~^ J~*E (3.7-22)

T

so that

(3.7.23)
J err

with cr and r some characteristic conductivity and time for the plasma.
Now €0 is a little under KR11 farad/m, while for most mhd problems
a is greater than 1. Hence, unless one is dealing with very short
characteristic times, Eq. (3.7.23) shows that one can neglect the dis¬
placement current e0E in Eq. (3.7.21).
The approximations made with regard to the generalized Ohm’s
law cannot be justified in such a straightforward fashion. Rather,
it is customary simply to assume that all time variations are exceed¬
ingly small and that one has a nearly cold plasma. 1 hen, after
neglecting time-varying and pressure-gradient terms in Eqs. (3.7.17)
and (3.7.20), the generalized Ohm’s law reduces to the simple formf
shown in Eq. (3.7.19). With these approximations one obtains the

t Various ones of the other terms in Eq. (3.7.17) are occasionally used in the
mhd literature. For example, the term involving J X B0) when retained, yields
the equation (assuming p0 = m<n0 and me <<C rrii)

J = cro[E + u X Bo - 0(J X Bo)] 0 = —

where the last term on the right leads to a Hall effect in mhd flow problems.
 3.7 LINEARIZED MAGNETOHYDRODYNAMIC EQUATIONS 143

following set of mhd equations:

P + PoV • u = 0 (3.7.24)
pou = J X B - Vp (3.7.25)
J = <r0(E + u x B) (3.7.26)
V x E = —B V x B = MoJ (3.7.27)

This set still contains one more variable than there are equations,
and for many problems one must add an additional relationship
between p and p. Since we expect the plasma to behave as a simple
gas for slowly varying problems, it is generally adequate to set

Vp = U2 Vp (3.7.28)

with U the isothermal or adiabatic sound speed for the fluid as a whole.
This equation can be simply derived from the individual-species energy
equations (Vpa — Ua2 Vpa) if we assume that (as for a perfect mixture
of gases) the percentage change in each species is nearly equal during a
compression or rarefaction. Then, for example, using the adiabatic
law for each species, we have

Vp = J Vpa = y 7
*-< PaO
Vpa (3.7.29)
a a

But now we set

Vp« Vp
- = - all a (3.7.30)
PaO p0

Therefore

Vp = J ypao ^ = t *1 Vp (3.7.31)
L-' P0 P0

in agreement with Eq. (3.7.28).

As we have emphasized, the most questionable equation in this
approximate set of mhd equations is the simple generalized Ohm’s
law [Eq. (3.7.26)], in which pressure-gradient and time-derivative terms
are omitted (despite their being retained in other equations). Hence
it is generally necessary to investigate the other terms in Eq. (3.7.17)
or (3.7.20) to see whether they are really small in a given mhd problem.
144 CONTINUUM EQUATIONS FOR A PLASMA

To conclude our discussion of the assumptions used in magneto¬

hydrodynamics we should note that these equations (like the single¬
fluid equations) assume that the plasma is large enough for the particles
to spiral about the lines of any dc magnetic held without constantly
hitting the walls. Thus in any mhd problem one should also compare
the cyclotron radii for both the ions and the electrons with the dimen¬
sions of the container for the plasma. From Sec. 1.4 recall that the
cyclotron radius a is equal to \vL/wc\, where v± is the particle s speed
in the direction perpendicular to the magnetic held B0. On the aver¬
age, if the distribution of velocities is isotropic (independent of direc¬
tion), then (v±2) = 2kT/vi, so that for singly charged particles

, N (2wcT)1'2 (3.7.32)
^= 7b«
If T is given in electron volts and B0 in gauss, we hnd for electrons
(a) = 3.4 X 10->T'i*/Bo m, while for protons (ap) = IAT1I2/B0 m.
Plots of these radii are shown in Fig. 3.9. For weak magnetic helds

Fig. 3.9 Plot of cyclotron radius vs. temperature for several values of
magnetic-field strength. The solid lines are for electrons, the dotted for
protons.
3.8 MAGNETIC PRESSURE, VISCOSITY, REYNOLDS NUMBER, AND DIFFUSIVITY 145

it can be seen that the proton cyclotron radius is fairly large, and for
heavier ions it would be still larger. Hence the effect of the walls in
inhibiting the cyclotron motion of the ions must often be taken into
account.

3.8 MAGNETIC PRESSURE, VISCOSITY,

REYNOLDS NUMBER, AND DIFFUSIVITY

Magnetic pressure

Although we shall not attempt here to cover many of the numerous

problems that have been studied using the mhd equations, we shall at
least introduce some of the terminology used in this subject. To
start, if we use Eq. (3.7.27) to replace J in the momentum equation
(3.7.25) and then use the vector identity

(V x B) x B = (B • V)B - (3.8.1)

we find that the momentum equation becomes

(B-V)B / , B2\
Pou ----V [p + ) (3.8.2)
Mo \ 2po/

The term 52/2/x0, which enters here, is called the magnetic pressure
since it plays a role entirely analogous to the hydrodynamic pressure p.
For a steady-state problem where the magnetic field lines are
straight (so that B is perpendicular to any gradient in B), Eq. (3.8.2)
states that p + B2/2uo is constant. This simple result provides a
rough basis [since a true steady state and an exact cancellation of
(B • V)B is not possible] for estimating the confining fields needed in
a fusion device. Ideally, outside the confinement region p is zero, so
that if B0 and B are the field strengths respectively outside and inside
the plasma, we have

B*_ Bl B2
(3.8.3)
V + or 0 = 1 Bo2
2mo 2po

where /3 = 2uop/B02 is the ratio of the plasma pressure to the magnetic

pressure of the confining field. Note that d is between 0 and 1, since
the field inside the plasma is less than Bo. (This follows from the
discussion in Sec. 1.6, where we saw that the magnetic moment u for
any charge points in the direction opposite to the applied field, so that
the field produced by each oscillating charge n • II points in the oppo¬
site direction to the applied field.)
146 CONTINUUM EQUATIONS FOR A PLASMA

Problem
3.16(a) Show that a plasma with a pressure of 100 atm requires a
field of at least 5 webers/m2 (5 X 104 gauss) for magnetic
confinement. (This is a pressure that is adequate for a
fusion device.)
(6) Show that if this field of 5 webers/m2 is produced by a current
flowing through a cylindrical column of plasma (as in the
pinch effect) with a radius of 10 cm, then the current needed
is 25 X 105 amp. (Of course these would be curved field
lines, so this calculation is just to show the order of magni¬
tudes involved.)

Magnetic viscosity and Reynolds number

In many mhd applications a problem of great importance is the behav¬
ior of the magnetic field itself. From the simple form of Ohm’s law
one can obtain a simple equation for B in the following manner. First,
take the curl of Eq. (3.7.26) to have

-V xE = Vx(uXB)--VxJ (3.8.4)
&0

If we replace V X E and J by the expressions in Maxwell’s equations

(3.7.27) and use the identity

V x (V X B) = — V2B (3.8.5)

(since V • B = 0), Eq. (3.8.4) reduces to

B = V X (u X B) + V2B I'm = — (3.8.6)

v O-QJUO

where, for reasons that we shall examine shortly, vm is called the

magnetic viscosity.
Of the two terms on the right side of this equation, the first is
called the flow term, while the second is called the diffusion term. In
most cases one or the other of these terms is of predominant impor¬
tance. To compare the terms we use dimensional analysis again, to
have

uB vmB
V X (u X B) I'm V2B (3.8.7)
17 L2
3.8 MAGNETIC PRESSURE, VISCOSITY, REYNOLDS NUMBER, AND DIFFUSIVITY 147

with L some characteristic length. Hence the ratio of the terms,

which is called the magnetic Reynolds number Rm, is given by

uL
Rm (3.8.8)
Vm

To compare vm and Rm with the usual hydrodynamic viscosity and

Reynolds number note that the Navier-Stokes equation of hydro¬
dynamics (which we derive in Sec. 5.8) is

^ = f - I Vp + ^[V2u + ^V(V • u)] (3.8.9)

with f the average external force per unit mass and vk the kinematic
viscosity (viscosity divided by density). When we compare this with
Eq. (3.8.6), it is clear that vm plays essentially the same role in the
decay of B that vk does in the decay of the fluid velocity u. Here the
Reynolds number is the ratio of the inertia term u • Vu to the viscosity
term vk V2u or, by dimensional analysis,

u2/L _ uL
(3.8.10)
vku/L2 vk

in complete analogy with Rm as given in Eq. (3.8.8).

Diffusion of the field lines

There are two “characteristic” types of behavior for the magnetic

field as governed by Eq. (3.8.6), depending on whether Rm is large or
small. First, for Rm « 1, the flow term is much smaller than the
diffusion term. Hence Eq. (3.8.6) reduces to

B = I'm V2B (3.8.11)

which is a diffusion equation (with vm the diffusion coefficient) entirely

analogous to the particle diffusion equations studied in Sec. 3.3. For
this case the initial magnetic held diffuses away with a characteristic
time TB, which we can again find from dimensional analysis by equating

B vmB
B - and vm V2B (3.8.12)
TD U
148 CONTINUUM EQUATIONS FOR A PLASMA

SO that

IS (3.8.13)
TD — -
Vm

Frozen field lines

When the magnetic Reynolds number is very large, the flow term
predominates on the right side of Eq. (3.8.6), and one finds a completely
different type of behavior, namely, that the magnetic field lines move
along exactly with the fluid rather than simply diffusing out in space.
To see how this comes about we integrate each side of Eq. (3.8.6) (after
neglecting the diffusion term) over an arbitrary open surface moving
with the local fluid velocity u. From Stokes’ theorem we have

JB • dS = Jv x (u X B) • dS = f(u X B) • d\ (3.8.14)

Now, using the vector identity,

(AXB)-C = -B- (AxC) (3.8.15)

we have

JB • dS + /B • (u x dl) = 0 (3.8.16)

Now we shall show in the next paragraph that for a surface moving
with velocity u

j ( f B • dS) = f B • dS + j) B • (u X d\) (3.8.17)

Hence, Eqs. (3.8.16) and (3.8.17) show that for the moving surface

JB • dS = const (3.8.18)

Thus the flux passing through the moving surface stays constant.
This is known as a frozen field condition, since the magnetic field lines
are, in essence, “tied” to the fluid and move along at the fluid veloc¬
ity u.
To complete this discussion we now want to prove the identity
given in Eq. (3.8.17). Suppose we consider a moving open surface S
3.8 MAGNETIC PRESSURE, VISCOSITY, REYNOLDS NUMBER, AND DIFFUSIVITY 149

such that, at time t, S(t) — Si, and at a time At later, S(t + At) = S2.
Then, as in Eq. (3.1.5),

Ji ( / B'dS) " JjS1, S Us. B(( + A,)'rfS - k B«>'rfS] <3-819>

If we expand B(£ + AO in a series about B(0, we have, in the limit

as At —> 0,

d
dt
(/s.ds)
(3.8.20)

To evaluate the second term here, we use the fact that V • B = 0.

Hence, for any closed surface at a time t, we have, from Gauss’ theorem,

(j)B • dS = f V • B dV = 0 (3.8.21)

We now apply this to the closed surface (shown in Fig. 3.10) consisting
of Si, S2, and the “sides” of length u At, to have

fs B(0 • dS - fs B(0 • dS + (j) B • (u At X d\) =0 (3.8.22)

(Note that the minus sign in the second term is due to the fact that
S2 points into the volume.) If we use this result in Eq. (3.8.20) and
note that

lim f B • dS = [, B'dS (3.8.23)

At—>0 J S* J Sl

u M->>

Fig. 3.10 Closed surface formed

by Si, S2, and the sides of length
u At. Note that we have arbi¬
trarily let Si and S2 point to the
left while u is to the right.
150 CONTINUUM EQUATIONS FOR A PLASMA

we have

^( f B • dS) = f B • dS + (£ B • (u X dl) (3.8.24)

which completes our proof.

Problems
3.17(a) Show for a typical mhd generator with u = 103 m/sec,
a = 100 mho/m, and L — 0.1 m that Rm « 1 and the
diffusion time is very short. (Hence inhomogeneities in the
field are quickly smoothed out.)
(6) In a fusion device (where inhomogeneous confining fields
should be maintained for a fairly long time) show that diffu¬
sion times on the order of 103 sec are possible when Te is of
the order of 104 ev and L ^ 1 m. Note that at these tem¬
peratures, as indicated in Eq. (2.2.17), ao = 4.6(T/300)3/2,
with T in degrees Kelvin. Show that, for such a device, the
field lines are essentially frozen.
3.18 Derive the equation pou = a0(E xB) + a0(u X B) X B - Vp,
and show that for E = 0 and p constant the velocity perpendicular
to B is given by ux = ux(0)e~tlT, with r = po/oqB2 = vm/\ a2.
Calculate r for a fusion device with n = 1022 protons/m3,
B = 5 webers/m2, amd T = 104 ev.

3.9 MHD WAVES

Development of an equation for u

Although the mhd equations have been applied to an extensive list of
problems, we shall consider here only one important example, magneto-
hydrodynamic waves. Using the approximate mhd equations, we shall
be able to obtain dispersion relations for waves traveling at an arbi¬
trary direction relative to B. In addition we shall see when the mhd
approximations break down by comparing the mhd results with the
more exact analysis (in Secs. 3.5 and 3.6) of wave propagation in the
direction of B0 when the full set of electron and ion fluid equations
is used.
Using the approximate mhd equations developed in the last two
sections and setting B = B„ + b(r,t), we have, from Eqs. (3.7.24) to
(3.7.28) and (3.8.6),

P d- Po^ • u — 0 (3.9.1)

Pou = J X B0 - f/2 Vp J = - V x b (3.9.2)

Po
 3.9 MHD WAVES 151

and

b = V X (u X B0) + pm V2b (3.9.3)

To develop a dispersion equation for small-amplitude waves we set

b, u, and p — p — p0 all proportional to exp (zk • r — iwt) For this
case, as shown in Eqs. (2.1.10) and (2.1.11), we can replace V by zk and
all time derivatives by —iw. Hence, from Eq. (3.9.1) we have

/ P0 /| \
p = — (k • u) (3.9.4)
CO

Similarly, from Eqs. (3.9.2) and (3.9.3) we have

— fwp0u = — (k X b) X B0 — zkt/2p' (3.9.5)

Mo

and

( — iw + rmk2)b = zk X (u X B0) (3.9.6)

If we now use Eqs. (3.9.4) and (3.9.6) to eliminate p and b in Eq.

(3.9.5), we have, after multiplication by zw/p0,

= *k Xlk. *_(u X X B- + U2(k • u)k (3.9.7)

POPo[l + l(Pm/u)kZ\

To simplify the analysis from this point we neglect collisions (so

that o-q = oo and vm = 0), and we introduce the vector Alfven velocity

Va = (MOPo)-1/2Bo (3.9.8)

With these simplifications, Eq. (3.9.7) reduces to

a)2u = {k X [k X (u X Va)]) X Va + f/2(k • u)k (3.9.9)

Although this result appears to be a very complicated vector equation

for u, we shall see that if we go through all the algebra involved in
writing out each term, the final result is remarkably simple.

Development of the dispersion relations

In order to proceed further we again introduce the coordinate system
shown in Fig. 3.11, where the wave travels in the x direction and z is
152 CONTINUUM EQUATIONS FOR A PLASMA

chosen perpendicular to the plane formed by k and Bo, so that

k = kax Va = V„(cos yax + sin yay) (3.9.10)

Hence, in the last term on the right side of Eq. (3.9.9) we have

U2(k • u)k = k2U2uxax (3.9.11)

Next, in the more complicated vector-product term on the right we

first let

A = u X V„ (3.9.12)

Now recall that for k = kax

k X (k x A) = — k2A, (3.9.13)

with A, the part of A that is in the plane perpendicular to ax. Using

these identities, we have

{k X [k X (u X Va)]} X va = —k2X, X Va (3.9.14)

By direct calculation, from Eqs. (3.9.10) and (3.9.12),

A, = Va[auUz cos 7 + az(ux sin y — uy cos 7)] (3.9.15)

Fig. 8.11 Coordinate system used for plane waves with

an applied magnetic field.
 3.9 MHD WAVES 153

SO that

— k2At X Va = k2Ya2[uz cos2 yaz

+ (ux sin 7 — uy cos 7) (sin 7a* — cos 7ay)] (3.9.16)

Applying Eqs. (3.9.11) and (3.9.16) to each component of Eq. (3.9.9),

we obtain finally

ux(w2 — k2Va2 sin2 7 — k2U2) + uyk2Va2 sin 7 cos 7 = 0

uxk2Va2 sin 7 cos 7 + uy(to2 — k2Va2 cos2 7) = 0 (3.9.17)

uz(co2 — k2Va2 cos2 7) = 0

From the last of these equations we see that there can be a wave
linearly polarized in the 2 direction (the direction perpendicular to
both k and B0) provided

k2Va2 cos2 7 = co2 (3.9.18)

This wave is generally referred to as the pure Alfven wave. Before

discussing this result, let us quickly obtain the remaining dispersion
relations. To have a solution where ux and uy are nonzero, the deter¬
minant of the coefficients of the first two parts of Eq. (3.9.17) must
vanish. Hence for these waves we have

(co2 — k2V 2 sin2 7 — k2U2)(co2 — k2V 2 cos2 7)

— /c4Va4 sin2 7 cos2 7 = 0 (3.9.19)

or

CO 4 - k2(U2 + Ha2)CO2 + k4U2Va2 cos2 7 = 0 (3.9.20)

Since we are primarily interested in calculating the phase velocity, we

divide through by fc4 and solve for co2//c2, to find

gj = |{(Ha2 + U2) ± [(Ha2 + U2)2 - 4V2U2 cos2 7]1/2} (3.9.21)

/c

The solutions with the plus-and-minus sign here are (for obvious
reasons) called fast and slow mhd waves.
All three of these mhd dispersion relations (3.9.18) and (3.9.21)
describe waves with constant phase velocity for all frequencies. Hence
154 CONTINUUM EQUATIONS FOR A PLASMA

in Fig. 3.12 we plot phase velocity as a function of the angle 7 between

k and B0 for each of the waves. From the figure note that there are
three kinds of mhd waves: (1) a pure Alfven wave involving oscilla¬
tions in the direction perpendicular to k and B0 with a phase velocity
that equals Va cos 7; (2) a fast mhd wave with phase velocity that
increases from Va (or U if U exceeds Va) when 7 = 0 to (1 a2 + L2)11
when 7 = 90°; (3) a slow mhd wave with phase velocity that decreases
from U (or Va if U exceeds Va) when 7 = 0 to zero when 7 = 90°.
When 7 = 0, note that two of these waves have the Alfven
velocity Va, while the third has the sound speed t in agreement with
the results found using the full two-fluid equations in Secs. 3.5 and 3.6
in the limit when o> is much less than both iOi and wCi and when Va is much
less than c. The same thing is also true at all other angles where our
mhd results agree exactly with the expressions obtained for waves
in a warm fully ionized plasma provided one lets w —► 0 and also
assumes Va <3C c. Obviously this is indicative of the breakdown of
some of the mhd approximations when w or Va is too large.

Problem
3.19 Note, from Eq. (3.9.7), that we can formally include the effect
of vm by replacing Va2 by

Va2
(V'a)2 1 + i(vm/ w)/c2

Use this fact to find the damping factor for each of the mhd
waves when vm is very small but nonzero.

Fig. 3.12 Plot of phase velocity as a function of the angle y between the
direction of wave propagation and B0 for the three types of mhd waves.
 3.10 THE NONLINEAR MHD EQUATIONS 155

3.10 THE NONLINEAR MHD EQUATIONS

Equations for mass and momentum

In the preceding sections we have limited our discussion to the linear¬
ized mhd equations. However, much of the work in magnetohydro¬
dynamics involves nonlinear flows or shock waves. Hence we shall
now show that one can derive simple mhd equations for mass and
momentum conservation from the multifluid equations even without
the linearization procedure used in Sec. 3.7. However, the generalized
Ohm’s law, which is already quite complicated for linear problems,
becomes even worse if nonlinear terms are retained.
The full multifluid equations for mass and momentum conserva¬
tion are given by Eqs. (3.2.1) and (3.2.4), which have the form

Pa + V • (paua) = 0 (3.10.1)
Duak
Pa naq«[E + (ua X B)]*
Dt

— ^ Pavap{uak — upk) (3.10.2)

Summing the first of these equations over all species, we have

p + V • pu = 0 (3.10.3)

where p and u are again the total density and total fluid velocity as
defined in Eq. (3.2.13). This simple result has precisely the form
one would obtain for a simple fluid with density p and fluid velocity u.
In similar fashion, if we sum the momentum equations for all the
species, we can obtain ultimately an equation of the form expected
for a simple charged fluid,

Duk dp0
PcEk + (J X B)fc (3.10.4)
p Dt dXk

In this equation pc = 2naqa is the total charge density, J = 2n„gaua

is the total current density, and D/Dt = d/dt + (u • V) is the hydro-
dynamic derivative for the total fluid; however, for reasons which are
clarified in Prob. 4.1, we must let the gradient of the total pressure p°
be given by
156 CONTINUUM EQUATIONS FOR A PLASMA

with wa the diffusion velocity,

(Note that this extra term involving the diffusion velocities is of second
order in wa; hence it drops out of the linearized mhd equations.)
To derive Eq. (3.10.4) we first sum the individual momentum
equations (3.10.2) over all a. Then, from the definitions of pe, J, and
dp°/dxk and the identity [shown in Eq. (3.2.14)] that

^ PaVaffUa ~ Up) - 0
(3.10.7)
a,0

we have

duak
Uak "T Uai —-
i )

To simplify this equation we must go through some algebra and use the
identity

(3.10.9)
£ PaVfa = X P“(U“ ~ U) = PU _ PU - 0

which follows from the definitions of w« and u. Working with the

term on the left side of Eq. (3.10.8), we add the term

(3.10.10)

[which is clearly zero from Eq. (3.10.1)] to obtain

(3.10.11)
dt

Next, we can replace each u« by wa + u and use the identity in Eq.

(3.10.9) to have
 3.10 THE NONLINEAR MHD EQUATIONS 157

The second term here is identically zero [from Eq. (3.10.3)], and the
third term exactly cancels the last term on the right side of Eq. (3.10.8),
so that we obtain the simple nonlinear momentum equation

P -^ = PcEk |(JxB)t-f (3.10.13)

Ut OX k

Nonlinear generalized Ohm’s law

To summarize we find that one can derive relatively simple mass- and
momentum-conservation equations for the fluid as a whole even if
we omit the linearization procedure used in Sec. 3.7. However, as
was emphasized in that section, if the gas mixture contains s compo¬
nents, Eq. (3.10.13) must be supplemented by s — 1 additional equa¬
tions involving u and J (or other combinations of the ua) to have a
completely defined system. To see what difficulties this leads to in
nonlinear problems we can briefly reexamine the case of a fully ionized
gas, where

PeTle —f- p,Ui

11 = -7 J = e(n;u, — neue) (3.10.14)
P

Inverting these equations, we find

(3.10.15)

(3.10.16)

where p is the reduced mass, memi/(me + ma). To obtain an equation

analogous to the linearized generalized Ohm’s law (3.7.17) we multiply
Eq. (3.10.2) by ma/qap and sum the resulting equation over both the
electrons and ions to obtain (after using the identity pevei = PiVu)

ep
ynlpi
Dui
In mePe T)f j
D uA
E + u X B — — (m*
ep
Vpi me V pe)

+ (We _|_ m,i)(ue — u,) (3.10.17)

If we use Eqs. (3.10.15) and (3.10.16) to eliminate ue — m in the last

term of this equation, we find (with no approximation)

.1 — eu (ne — nz) riie2

Last term a = - (3.10.18)
a Wle^ei
158 CONTINUUM EQUATIONS FOR A PLASMA

Hence, Eq. (3.10.17) reduces to

m (pi ^ - me (^pe ^ + Vpe^

+ eu^e-^ (3.10.19)
a

The terms on the left are the usual ones retained in magnetohydro¬
dynamics, while all the terms on the right are generally discarded.
Clearly, the omission of all these terms is not always justified, hence,
one should check the magnitude of these terms to see whether they
are truly negligible in a given mhd problem.
chapter four

THE BOLTZMANN-
VLASOV EQUATION
FOR A PLASMA
chapter four

4.1 DISTRIBUTION FUNCTION AND THE BOLTZMANN EQUATION

The distribution function

In previous chapters we have either considered the motion of individual
particles, assuming there were no interaction or collective effects,
or else we have treated the plasma as a mixture of continuous fluids.
In this chapter we shall begin to develop the statistical approach for
studying the plasma, where we recognize that the macroscopic prop¬
erties of the plasma are due to the average behavior of many individual
particles, but where each individual particle’s motion is governed by
the usual laws of mechanics. For a system containing Na particles
of type a, a complete description of the a particles would require us
to specify the positions and velocities of each particle; but to obtain a
statistical description of the plasma, we require a knowledge only of
the distribution function for the particles.
Before we define the distribution function, we must introduce the
notation we shall use here. Suppose we let x be a point in space with
coordinates Xi, x2, x3, and we let v be a velocity vector with rectangular
components Vi, v2, and v3. Then, as shown in Fig. 4.1, d3x is a cube
with sides dxi, dx2, and dx3, and d3v is a cube (in velocity space) with
sides dvh dv2, and dv3. Now, if dNa is the number of particles of type a
lying in the volume element d3x and whose velocities fall in the velocity
element d3v at time t, then the distribution function /«(x,v,0 is given by

dNa (4.1.1)
/a(x,v,i)
d3x d3v

Note that for a particle to be included in dNa its position coordinates

must lie between Xi and x» + dxit and its velocity components must lie
between Vi and + diu, where i takes on the values 1, 2, and 3.
From a knowledge of fa one can compute most of the macroscopic
variables for the particles of type a. For example, the number density
na(\,t) is the number of particles per unit volume and is given by
 4.1 DISTRIBUTION FUNCTION AND THE BOLTZMANN EQUATION 161

where 5Na is the total number of particles in the volume element 8V

at x, while 5V itself must be small by macroscopic standards but large
enough to contain many particles. [Otherwise na(x,t) is likely to be
a discontinuous function in space and time.] If we go to the limit
where 5F equals dzx, then 8Na is the integral of dNa [in Eq. (4.1.1)]
over all velocities; hence 8Na — d3x ) d3v and

I fa(x,v,t) d3v (4.1.3)

where the integration (like all subsequent integrations with no limits

indicated on the integrals) extends over all possible velocities.

Calculation of other average values

If Q is any function of the velocity of the particles of type a, then the
average value for Q at any point x and time t is given by

(4.1.4)

To see that this is consistent with our usual definition of an average

let us consider a particular example. If we were asked to find the
average velocity ua for the particles at x and t we would choose a small
volume including the point x, add together the velocities of all the
particles there, and divide by the total number of particles in the
volume. If we let dzx be the small volume at x, then, clearly, the total
number of particles is na(x,t) d3x, while the total sum of the velocities
equals the number of particles with a given velocity v times v itself
integrated over all possible velocities. Hence we have

(4.1.5)

Fig. 4.1 The volume element d3x and the velocity-space element dh.
162 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

as expected from Eq. (4.1.4). An obvious (but important) conse¬

quence of Eq. (4.1.4) is that, for Q independent of velocity,

(Q)a = Q if Q = Q(x,0 (4-L6-

so that, for example, (ua) = utt.

Definitions of other macroscopic variables

Many other macroscopic properties for the particles of type a can be
defined in terms of averages like Eq. (4.1.5). Before discussing these,
however, we must introduce the random velocity

which is the difference between the actual velocity of a particle and

the mean velocity for all the particles of type a. FromEqs. (4.1.5) and
(4.1.6), c„ satisfies the important identityt

(c„) = (\)a — u« = 0 (4.1.8)

As noted several times earlier in the text, the temperature is a measure

of the average random kinetic energy; by definition, therefore, we letj

| kTu = }ma(ca2) (4.1.9)

Other macroscopic quantities such as the pressure tensor or the

heat-flow vector involve the flux of some quantity Q across a surface
element moving with the mean velocity ua. 4he flux of Q is defined
as the net amount of Q carried across the surface element per unit
time and per unit area by the particles of type a. To calculate it,
consider the moving surface element AS shown in Fig. 4.2 with its
normal in the fcth direction. All the particles which have velocities
between v and v + d3v and which cross AS between t and t + At
must lie initially! in the volume element (shown in the figure) with
height cak At and base AS. From our definition of fa, the number of
a particles in the volume is facak A t AS d3v, so that if Q is some property
(like the mass, momentum, or energy) of each particle, then the total
amount of Q carried across AS during the time interval At is this num-

f Note that in this equation and elsewhere we omit the subscript next to an
average when it is redundant, that is, (ca) = (ca)«.
f With this definition Ta equals the absolute temperature in thermodynamics.
§ Note that, to make this calculation, we assume that dvk At is negligibly small
and that collisions during the time interval At are also of negligible importance.
 4.1 DISTRIBUTION FUNCTION AND THE BOLTZMANN EQUATION 163

ber times Q integrated over all possible velocities

JQfaCak At AS dh> (4.1.10)

Hence the flux T(Q), which is the rate of transport of Q per unit time
and area, is just

r(Q) = !QfaCakd3v = na(Qcak) (4.1.11)

The pressure tensor and heat-flow vector

The pressure tensor and heat flow vector can now be calculated quite
easily. The 'pressure-tensor element pajk is the flux of the jth component
of momentum across a surface pointing in the kth direction (so that
Q — mavf) and the heat-flow-vector component qak is the flux of random
or thermal energy across a surface pointing in the A:th direction (so that
Q = maca2). Hence, if we use the facts that vj = caj + uaj and that
(caf) = 0, we have, from Eq. (4.1.11),

'Paik Pa^Pj^ak) Pa( ”T klaj)Cak)

= Pa(cafak) (4.1.12)

Fig. 4-2 Volume containing, initially, those particles with velocity

v that pass through the surface AS during time At. Note that AS is
assumed to point in the k direction, so that = ca cos 6.
164 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

and

r i0 /c 2C \ (H. 1.1.5;
yak "2Pa\^a ca«7

The scalar pressure pa is one-third the trace of the pressure tensor.

Hence, from our definitions of pajk and Ta [Eqs. (4.1.9) and (4.1.12)]
we have (using the summation convention described in Sec. 1.1 where
any repeated direction index is summed)

Pa = %Paii = iPa(ca2) = UukT a (4.1.14)

Hence, pa automatically satisfies the ideal-gas condition of thermo¬

dynamics.
From these definitions it is evident that we can calculate the
macroscopic number density, mean velocity, pressure, temperature,
and so forth, provided we know the distribution function.

Problem
4.1 For a gas mixture one can define an alternative random velocity,
c° = v — u, where u = 2paii0/2pa is the velocity for the gas as a
whole. Similarly one can define an alternative temperature
Ta°, pressure pa°, and so forth simply by replacing ca in Eqs.
(4.1.9) and (4.1.12) to (4.1.14) by c°. Prove that

Vaik — P ~P PaLCaiU'ak

and pa° = pa + }pawa2, where wa = (c°)a = ua - u. Note that

when paa = pabik, we have

dP°aik dpa
dXi dXk
+ dx
Pa^ax^ak

which is the expression used in Sec. 3.10 to obtain the nonlinear

mhd equations.

The Boltzmann equation

To develop an equation for the distribution function /a, recall again
that dNa(x,v,t) is the number of particles whose positions and veloci¬
ties are essentially x and v at time t. Hence, if there were no collisions,
then at a short time At later each particle would move from x to
x + v At, and each particle’s velocity would change from v to v -f a At,
where a is the acceleration due to external forces on a particle at x
with velocity v. Therefore any difference between dNa(\,\,t) and
 4.1 DISTRIBUTION FUNCTION AND THE BOLTZMANN EQUATION 165

dNa(\ + v At, v + a At, t + At) is due to collisions, and we may set

[/«(x + v At, v + a At, t + At) — fa(\,\,t)] d3x d3v

Ol Jc d3x d3v At (4.1.15)

where (dfa/dt)c is the time rate of change of fa due to collisions. Now

expanding the first term on the left about fa(x,\,t), we have

/a(x 4vA(,v + aA«,i + At) = fa(x,x,t)

+ (iT2 vi + At + terms of order (At)2 (4.1.16)

\oxz dVi dt J

Thus in the limit as At goes to zero, Eq. (4.1.15) reduces to

df* dfa

dt
+ v■
dxi
+ di (4.1.17)

which is known as the Boltzmann equation for fa.

Of course Eq. (4.1.17) is incomplete because we have not calculated
the collision term on the right. This task, which will require us first to
review the dynamics involved in the collision of two particles, is
described in great detail in Chap. 5. In this chapter we shall examine
some consequences of the Boltzmann equation which do not depend
on the exact form of the collision term, and we shall treat some appli¬
cations of the equation when collisions are of negligible importance.
Hence it will not be necessary to use a detailed expression for (dfa/dt)c.

Problem
4.2 In deriving the Boltzmann equation we have assumed that the
elements d3x' d3v' and d3x d3v are equal. Actually, from the
theory of Jacobians, we have

d3x' dh/ = |J| d3x d3v

where J is called the Jacobian and is the determinant of the 6X6

matrix made up of all the partial derivatives of x' and v' with
respect to x and v. [For comparison with a similar calculation,
see Eqs. (5.1.23) and (5.1.24).] Show that, with x' = x + v At
and v' = v + a(x) At, \J\ — 1 + terms of order (At)2, so that our
results in Eqs. (4.1.15) to (4.1.17) are still correct in the limit
as At —> 0.
166 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

4.2 TRANSPORT EQUATIONS

General equation
In the preceding section we saw that macroscopic variables such as ft,
u, T, and p which enter the continuum equations can all be calculated
from the distribution function fa. The distribution function, in turn,
satisfies the Boltzmann equation (4.1.17) derived above. Hence, as
one might expect, it is possible to derive all the continuum equations
(like the mass, momentum, and energy equations used in Chap. 3 or
the viscous-stress and heat-flow equations we shall use in Sec. 5.7
and 5.8) from the single Boltzmann equation. In this section we shall
first derive the general transport (or continuum) equation for an
arbitrary function Q, and thereafter we shall use three specific Q’s
which lead to the mass, momentum, and energy equations.
To start, we have the Boltzmann equation

+ v. dU,ndfa = (d_h\ (4.2.1)

dt 1 dXi ‘ dVi \ dt )c

If we now multiply each term by some function Q(x,\,t) and integrate

over all velocities, we have for the first term

dzv ~ f fa ^ d*V
I = £/ <*• dt

= jt na{Q)a ~ na(Q)a (4.2.2)

where we have simply used the definition of an average given in Eq.

(4.1.4). (Note also that the limits of integration do not depend on
space or time; hence any derivatives can be moved in or out of the
integrals.) In the second term we have

/ Vi^ Qd3v
dx;
dQ\ (4.2.3)
na(Qvi)a na (Vi
dXi/a

where we have noted that x, v, and t are all independent variables, so

that dvi/dxi (or any other such partial derivative) is identically zero.
Next, in the third term we have (assuming a; independentf of vf)

dQ
d3v (4.2.4)

f Note that this restriction does not rule out the force due to a magnetic field,
a = (q/m)(v X B), since one can easily verify that a; is still independent of vx.
[For example, ax = {q/mfivyB, — vzBv), which is independent of i^.l
 4.2 TRANSPORT EQUATIONS 167

Now consider the integration over dvi in the first integral on the right.
We have (omitting the subscript i throughout) for one of the integrations

(4.2.5)

which we set equal to zero, since fa must be zero when v is infinite, as

there are no particles with infinite velocities. [In general we shall see
that Q is proportional to some power of v while fa is often proportional
to qxp( —const v2); hence Qfa approaches zero rapidly for large v.]
Thus, for the third term we have

[ Q d3v = —na (at (4.2.6)

JdVi \ dVi/a

so that the general transport equation [obtained by summing Eqs.

(4.2.2), (4.2.3), and (4.2.6)] is

^na(Q)« - na{Q)a + na(Qv%)a

- naU - na !di P) = 5Qa (4.2.7)

\ dXi/a \ dVi/a

with

•o--/0®.* <4'2'8)

the collision integral.

Mass-conservation equation
Equation (4.2.7) is a general result which holds for any arbitrary
function Q(x,v,t). To derive the mass-, momentum-, and energy-
conservation equations, we let Q — ma, macak, and \maca2, respectively.
Thus for the mass equation, with Q = ma, we have

(Q)a = rna (Qri)a rnauai

d = dQ = d_Q = Qv (4'2'9)
dXi dVi

In addition, if we neglect those collisions which lead to ionization or

recombination, then the mass density mana for each species is unchanged
168 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

by collisions, and

(4.2.10)

[This result also arises very simply when we use the Boltzmann expres¬
sion for (dfa/dt)c derived in Chap. 5.] Hence, for this case if we sub¬
stitute the expressions in Eq. (4.2.9) into the general transport equa¬
tion (4.2.7), we find

d . d (4.2.11)
= 0
ft p“ + a5

pa + V ■ (PaO = 0 (4.2.12)

which is just he usual mass-conservation equation used in Chap. 3.

Momentum equation
To derive the momentum equation, we let

Q = macak = ma[vk — uak (x,i)] (4.2.13)

and note that

. . _ . . dCotk dllak dCactc _ t.

<c-> = 0 = -ua "djq" ~~
(4.2.14)

where dik is the Kronecker delta, which equals one for i = k and zero
otherwise. Using these identities, we have for each of the terms in the
transport equation (4.2.7)

j-na{Q)a = 0

Ha(Q)0: — pa'dak

T Ha(Q?’j')a = T pa^CakiCai T" tta,')) - Paik (4.2.15)

OXi OXi oX i

_ n /,,. — q /(c . _i_u .) dUak\ = u . dUak

“\ 1 dxi/a ~ Pa V ai + ai) dxi / Pa ai dx

Ha \ T \ = Pa (di$ik)a Pa(P'fc)a
\ OVi/a
 4.2 TRANSPORT EQUATIONS 169

Adding these together, we obtain the momentum equation

(4.2.16)

where

(4.2.17)

and D/Dt = d/dt + uai(d/dx/) is, again, the hydrodynamic derivative.

Note that this reduces to the momentum equation (3.2.4) used
in Chap. 3 when three conditions are fulfilled:

1. (a)a is the Lorentz force per unit mass, (qa/ma)(E + na X B).

2. The pressure tensor has the scalar pressure for its diagonal ele¬
ments and zero for all off-diagonal elements, that is, paik = pa&nc-
3. The collisional momentum-transfer term reduces to

5Qak = ~pa X v«f>{uak - Uf,k) (4.2.18)

The first of these follows at once, since the average external force
per unit mass due to electromagnetic fields is

— (E + v x B)a = — (E + u„ x B) (4.2.19)
ma ma

Condition 2, however, follows for a gas only when viscous effects are
negligible. This will be clarified in Chap. 5, when we show that

Paik — Paik Pa^ik

(4.2.20)

is proportional to the viscosity of the a gas. The quantity Paik which

appears here is called the traceless pressure tensor, since, from Eqs.
(4.2.20) and (4.1.14), its trace I\a is zero. Condition 3, finally, is a
result which is also not valid in general, although, as we shall show in
Chap. 5, the result does arise in the limit when the flow-velocity
difference among the various gas species is small and each gas has a
so-called maxwellian velocity distribution.
170 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Energy equation
To derive the energy equation we again work directly with the general
transport equation. We let Q = -|maca2 and note that

na(Q)a = #P« Q — a • ua
(4.2.21)
dQ dUaj dQ_
= 'ffl'dCaj = macai
dXi dXi dVi

Using these identities plus those in Eqs. (4.2.14), we have for each of
the terms in Eq. (4.2.7)

d , d
jtMQ)a - dtsVa

PaX^-a * (^a) ^

d d d
— na(QVi)a = — |pa(ca2(ca, + Uai)) = (?«* + %PaUai)
dXi dXi OXi (4.2.22)

/ <3Q\ dtlaj // i \ \ dllaj

xVi te)a = Pa Itei ^Cai + Uai)Caj' “ p"y "to7

/ Ctt "T \ = Paifli^a-i)

uVi/ a

The last term here goes to zero, since for a velocity-independent force,
(a,iCai) = a,i(cal) = 0, while for the force due to a B field (the only
common velocity-dependent force)

<„ • c„> = <(v X B) • c„>

= <[(ca + ua) x B] • ca) = 0 (4.2.23)

where both terms vanish here since (ca X B) • ca = 0 and (c„) = 0.

When all the terms in Eq. (4.2.22) are added together, the energy
equation becomes

ft |p- + V * q« + v • fp«ua + Van = 5Qao (4.2.24)

where 8Qa0 is the collision term, fQa(dfa/dt)c d3v. We can rewrite this
in a form more reminiscent of the adiabatic energy equation used
earlier by noting that the last two terms on the left side of this equa-
 4.2 TRANSPORT EQUATIONS 171

tion can be written [using Eq. (4.2.20) for paij] as

S 11 a j duaj
V 2Pa^-a 1
~“ Paij 2P* Ua ~\~ 2'Ua * Vpa + Paij (4.2.25)
dXi dXi

With this substitution the energy equation becomes (after multiplica¬

tion by -§)

+ ipaT • ua = § (sQao - V • qa - Paij (4.2.26)

Next, from the mass-conservation equation (4.2.12), we have (after

expanding V • paua)

V • ua = - - ^ (4-2.27)
Pa Dt

so that, from the terms on the left side of Eq. (4.2.26), we have

DPa _ & Va DP<* _ 5/3 &v -5/3

(4.2.28)
Dt 3 Pa Dt pa DtVapa

Thus, an alternative energy equation is

^ PaPa~bl3 = %Pa~blZ ^5Qa0 - V • q« - Paij (4.2.29)

which reduces to the adiabatic energy equation (with 7 = •§) used in

Eq. (3.1.15) when all the terms on the right vanish. For an inviscid
gas we have noted that Paij is zero; similarly for a gas with no thermal
conductivity q„ is zero. Hence the energy equation reduces to the
simple adiabatic law only in the limit when viscous, thermal-conduc¬
tivity, and collisional-transfer effects are all negligible.
Since pa = naKTa, one can easily transform Eqs. (4.2.26) and
(4.2.27) into an equation for the time rate of change of Ta. The
resulting equation is

DTa dUaj
(4.2.30)
^nan “h N • CJa ""I- Paij 8QaO
Dt dXi

which reduces to a diffusion equation for Ta when ua and 8Qa0 are zero
and when qa = — X VTa, with X the thermal conductivity.
172 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Problem
4.3 Derive Eq. (4.2.30) for the temperature. When the equation
reduces to a diffusion equation, show that if X = 5i<pa/2mava,
then the characteristic diffusion time is given by

3 L2
T 5vth2

where, as we shall see in Chap. 5, va is the self-collision frequency

and vth = (K.Ta/ma)1/2-

SUMMARY
In this section we have seen that just as most macroscopic variables
can be computed from fa, so too the equations for the time rate of
change of these macroscopic quantities can be found from the Boltz¬
mann equation. An important point (which is evident in the three
transport equations we derive here) is that the set of transport equations
derived from the Boltzmann equation always includes more unknown
functions than equations. Thus the equation for the time rate of
change of pa includes ua; the equation for the time rate of change of
ua includes pa; and so forth. Any finite set of transport equations is
therefore always insufficient to provide a closed set of equations for a
problem, and it is necessary to introduce some type of approximation
(or truncation) scheme to eliminate some of the functions or to express
some of the functions in terms of others. A systematic approach for
working with the transport equations is discussed in Chap. 5.
In the remainder of this chapter we shall describe some problems
where we work directly with the Boltzmann equation itself and solve
for the distribution function. In principle this is always the most
satisfying technique, since one can calculate any macroscopic variable
once fa is known. However, as might be expected, the Boltzmann
equation is really too difficult to solve in general, and before we can
obtain a solution to any problems of interest, many approximations
must be made. It is for this reason that the transport equations
continue to play an important role even in kinetic theory.

4.3 THE BV EQUATION AND THE RELAXATION MODEL

The BV equation
We have seen in previous chapters that although a plasma is a mixture
of electrons, ions, and neutral molecules, one can explain many phe¬
nomena fairly well by considering just the motion of the electrons.
 4.3 THE BV EQUATION AND THE RELAXATION MODEL 173

Therefore, as our first attempt to work with the Boltzmann equation

we shall apply it to the so-called Lorentz gas, an electron gas in a sta¬
tionary background of heavy particles which has just enough uniformly
distributed positive charge density to make the plasma macroscopically
neutral when the plasma is undisturbed. If we let / be the velocity
distribution function for the electrons, then, from the Boltzmann
equation (4.1.17), we have

with a the acceleration due to external forces, which, for an electro¬

magnetic field, would be the Lorentz acceleration

a = - - (E + v X B) (4.3.2)
m

with E and B the slowly varying fields which enter Maxwell’s equations.
The first decision to be made at this point is how to treat the
collision term in Eq. (4.3.1). The simplest assumption is merely to
neglect collisions and set

+ + = ° {4-3'3)

This collisionless Boltzmann equation is called the Boltzmann-Vlasov,

or BV, equation. This type of approximation is not so restrictive as
one might suppose, since, as we shall show in Chap. 6, a significant
portion of the effects of the interparticle interactions on / is actually
accounted for in the Lorentz force term (which, it will be recalled, is
related by Maxwell’s equations to the current and charge density in
the plasma).

The Krook collision model

When it is desirable to take some limited account of collisional effects,
the next simplest technique is to use the simple relaxation model|

©.--'tf"" . . (4'3'4)

where f0 is some equilibrium velocity distribution function and r_1 is

the relaxation time for the electron velocity distribution. To see the
t Henceforth, we shall refer to this as the Krook model, since it has been
extensively developed and used in recent years by Krook and his coworkers.
174 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

significance of this model consider the solution to the Boltzmann

equation when there are no external forces or spatial gradients, and
when /o and r are independent of time. For this case we have

% = -v(J ~ fo) (4-3-5)

or

(4.3.6)
/ + vf = vfo

This is a simple inhomogeneous differential equation with homogeneous

solution ce~vt and particular solution /0. Hence the complete solu¬
tion is

/ = ce~vt + fo (4-3‘7)

Now if at t = 0 we have / = /(v,0), then we easily find

/(v,f) - fo = [/(v,0) - foie-'* (4.3.81

which shows that the difference between / and /0 decays to zero expo¬
nentially at a rate proportional to v.
To summarize, the Krook (relaxation) model provides a very
simple collision term which tends to drive / toward some equilibrium
distribution /<, at a rate governed by the relaxation time v~l. This
seems, intuitively, to agree with the role one might expect collisions
to play, and, moreover, as we shall see in Chap. 5, there are many
examples one can cite where the model leads to almost the identical
results obtained by using the full Boltzmann collision term.

Application to collision integrals

Although we shall use the Krook collision model primarily for problems
where we solve directly for the distribution function /, it is also of
interest to note that the model greatly simplifies the evaluation of all
the collision integrals that enter the moment equations. From Eq.
(4.2.8) recall that

•«-/«©* (4'3'9)

Hence, using the Krook model for (df/dt)c and the definition of an
average in Eq. (4.1.4), we have (for v independent of velocity)

5Q = -v(g - go) (4.3.10)

 4.3 THE BV EQUATION AND THE RELAXATION MODEL 175

where

9 = ffQ dh = n(Q)
(4.3.11)
go = SfoQ d3v = n(Q)o

If g is any quantity that is conserved by collisions (so that 5Q

must be zero), then from Eq. (4.3.10) we see that g must equal go,
and /o must satisfy the integral condition

J/oQ dh = n(Q) (4.3.12)

For a simple gas, i.e., a gas having only one species of particles, the
mass, momentum, and energy densities cannot be changed by collisions
(because of the conservation of these quantities in elastic collisions);
hence/0 must satisfy the three conditions

m Jfo dh = p
mffov dh = pu (4.3.13)
±mjf0v2 dh = ~2-p(t’2) = + *-pu2

where p (or n), u, and T are the true density, mean velocity, and
temperature in the gas.
A common choice for/0 is the local maxwellian velocity distributionf

ne' ~c2/a2 2 kT
fo = c = V — u, (4.3.14)
7r3/2a3 m

One can easily verify that this function satisfies all three conditions in
Eq. (4.3.13), so that 8Q is zero in the mass, momentum, and energy
equations. [The integrals involving f0 in Eq. (4.3.13) are easily
handled if one changes the variable of integration from v to c — u
with dh = d3c.\

Limitation on the Krook model

Further insight into the Krook model is obtained by examining the

collision integral when g is not conserved by collisions. Again it is
helpful to consider the sirqple example when there are no external
forces or spatial gradients, so that the Boltzmann equation has the
form shown in Eq. (4.3.5). If we multiply through in this equation
by some function Q(v) (note Q and/o are taken, for this example, to be
f Note that we shall show in Chap. 5 that the maxwellian velocity distribution
is the equilibrium distribution for any gas.
176 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

independent of time) and integrate over all velocities, we obtain (for

v independent of v) the simple transport equation

g=-v(g-go) g = n(Q) (4.3.15)

where go (like fo and Q) is independent of time. This equation is

identical to Eq. (4.3.5) itself, except for the change from/to g. Hence,
as in Eq. (4.3.8), if at t = 0 we have g = gf(0), then the solution to
Eq. (4.3.15) for g is just

g(t) - go- to(0) - go]e— (4.3.16)

This result, which could also have been obtained by multiplying

Eq. (4.3.8) by Q(v) and integrating, shows that with this simple relaxa¬
tion model everiy average value (Q(v)) approaches equilibrium at the same
rate. This is a direct consequence of the model’s simple form. When
a more careful analysis of collisional effects is made, one finds that the
relaxation time for various averages varies to some extent even for a
simple gas and that the relaxation process in a mixture is often much
different from that described by Eq. (4.3.16).
This point is easily illustrated by considering the equilibration
process for electrons in a weakly ionized gas (where electrons collide
primarily with neutral molecules). For this case 5Q is not generally
zero in the momentum or energy equation for the electrons, since
collisions between particles of two different species generally change
the momentum and energy density of one species at the expense of
the other. Thus, if we assume that when equilibrium is reached, the
electrons have no drift velocity and their mean energy is 0, then
from Eqs. (4.3.11) and (4.3.16), for Q = v and we find

u(t) = u(0)e vt
(4.3.17)
±m(v2)t — f«T0 = (|m(v2) 0 — ^Kl\)e~vt

with (v2)t the average value of v2 at time t. Thus, in accord with Eq.
(4.3.16), the Krook model causes both the drift velocity and the mean
energy to approach their equilibrium values in a time of order v~l
(where v for this case would be the electron-neutral collision frequency).
In fact, however, an analysis of the collision process using a more
detailed collision model reveals that while the equilibration time for
the mean velocity is approximately v~l, that of the mean energy is
much longer, approximately M(2mv)~1, with M the neutral-molecule
mass. This is easily understood by remembering that an elastic
collision between a fast electron and a slow neutral molecule is much
 4.4 APPLICATION TO LONGITUDINAL ELECTRON WAVES 177

like the collision when a ping-pong ball is thrown at a bowling ball;

i.e., the light object bounces off in a different direction, but with its
initial speed practically unchanged. In other words, if Av and Ac
represent, respectively, the change in the electron’s velocity and speed
due to an electron-neutral collision, then for most collisions (as we
shall see in Chap. 5 and in Sec. A.6 of the Appendix)

Av
v
« 1 (4.3.18)

Hence, after a time long enough for an average electron to have one
collision with a neutral molecule, we can expect the average velocity
of the electrons to change significantly but not their average speed
(or kinetic energy). For this reason, the Krook model in this simple
form is really applicable only to cases where self-collisions are of
predominant importance.

4.4 APPLICATION TO LONGITUDINAL ELECTRON WAVES

Development of the dispersion equation

As our first application of the BV equation we shall reexamine the

problem of the propagation of small-amplitude longitudinal waves in
an electron gas with no uniform applied magnetic field. We have
previously treated this problem in Sec. 3.4, using the warm-plasma
continuum equations. Therefore a major point ot this section (and
the next) will be to emphasize those effects which arise when the BV
equation is used and which were missing when we used the continuum
equations. Since we are concerned with small-amplitude longitudinal
waves, we let

/ = /o(v) + fi(\)eikx~iuit I/ll «/o (4A1)

E = Eaxeikx~iut

so that the wave travels in the x direction. Substituting this into the
BV equation (4.3.3) and linearizing, we have

— iufi ikvxf i —E — = 0 (4.4.2)

m dvx

so that

__ ieE(dfo/dvx) (4.4.3)
1 mk[(w/k) — Cx]
178 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

In addition, from the Maxwell equation involving V • E, we have

V • E = - = e<yUi ~ ^ (4.4.4)
Co fo

Now recall that

ne = tfd*v = J/0 dzv + jfieikx-^ dh (4-4.5)

The first integral here is just the unperturbed or equilibrium electron

number density which must equal ni, since we have assumed that the
plasma is macroscopically neutral. Hence, in Eq. (4.4.4) we have

ikE = - - J/i dh (4-4-6)

If we now replace/i in this equation by the relation in Eq. (4.4.3) and

divide by ikE, we find

« = tv* r (dfo/dvx) dh (4.4.7)

n0k2 J vx — (w/k)

where n0 is the equilibrium electron (or ion) number density and

ue = (n0e2/me0)1/2 is, again, the electron plasma frequency. This
result is our dispersion equation (despite its marked difference from the
dispersion equations encountered in our work with the continuum
equations), since, if /o(v) is given, one can complete the integration
and obtain a relationship between k and co.

High-phase-velocity limit

A useful alternative form of the dispersion equation can be obtained

by an integration by parts. We let

b
dV = UV
a }>dU (4.4.8)

where

U -
(4.4.9)
dV = ^ dvx V = f0
dvx
 4.4 APPLICATION TO LONGITUDINAL ELECTRON WAVES 179

Hence, for the vx integration in Eq. (4.4.7) we have

(d/o/dvx) dvx = f0 J (1 CLVX

J-*> vx — co//c vx — co/k +/: (v*. — Co//c V2
(4.4.10)

The first term on the right is clearly zero since f0 vanishes when
vx = +oo. Hence the full dispersion equation (4.4.7) reduces to

(4.4.11)

where the subscript 0 next to the average denotes an average calculated

using/0.
Before attempting to evaluate this average for some explicit
choices of /0, we first examine the limit when co//c —> oo, that is, when
the phase velocity is very high compared to the velocity of all (or
almost all) of the particles. From Eq. (4.4.11) we have

(4.4.12)

But, for any |e| < 1, recall that

n 1 = 1 + 2e + 3e2 + 4e3 + • ■ • (4.4.13)

(1 - e)2

so that the dispersion relation becomes (for |vx| < \u/k\)

oj2 — coe2 (1 +■ 2(vx)o~ + 3(t>x2)o—j +•'■') (4.4.14)

In this high-phase-velocity limit all terms proportional to some

power of fc/co are exceedingly small. Hence, to a high approximation
we have just

co2 = co 2 (4.4.15)

[It will be recalled that this is also the cold-plasma result shown in
Eq. (2.4.27).] To consider small corrections to this result consider the
next two terms in the dispersion equation. First, if the plasma is
180 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

stationary in equilibrium, then (vx)o = Uxo — 0, and the first term

vanishes. Next, if we assume that the equilibrium velocity distribu¬
tion is isotropic, i.e., if there is no preferred direction in equilibrium,
then, from the definition of temperature [Eq. (4.1.9)]

(4.4.16)
(vx2)o — (Cz2)o — -g-(C“)o m

where T0 is the temperature of the electron gas at equilibrium. Hence,

for this case the dispersion relation reduces to

= . • •) (4.4.17)
y m ad /

To compare this with previous results, note that the underlined term
here is very small (since w/k has been assumed to approach infinity),
hence, we can replace w in just this small term by we (the value for w
when the term is zero) and obtain

0,^(0/ + —°H2 (4.4.18)

m

which is known as the Bohm-Gross dispersion relation. Note that this

result is identical to the warm-plasma result shown in Eq. (3.4.9)
when collisions are neglected and when 7 is set equal to 3. As we
noted in Sec. 3.1, 7 is related to the number of degrees of freedom of a gas
m by the condition

2 + m
7 = (4.4.19)
m

For monatomic gases with no internal degrees of freedom, the particles

have three degrees of freedom (corresponding to motion in the three
directions), and 7 = f. The Bohm-Gross result (7 = 3) corresponds
to the case when the electrons have but one degree of freedom, so
that, in some sense, they can move only in the direction of wave
propagation.

Problem

4.4 Use the BV equation (for each species) and Poisson’s equation
to obtain the following dispersion equation for small-amplitude
longitudinal waves in a fully ionized gas with equal ambient
 4.5 LANDAU DAMPING 181

number density n0 of electrons and singly charged ions:

i = i r/ \ . / \
k2 _\(vx — co/k)2/eo \(yx — CO//C)V,0_

What is the high-phase-velocity limit for this case?

4.5 LANDAU DAMPING

Dispersion relation for maxwellian /0

In the last section we applied the BV equation to the problem of

longitudinal oscillations in an electron gas and obtained the dispersion
relation

Ue2 r (dfo/dvx) d3v

(4.5.1)
n0k2 J vx — w/k

Now we shall show how this equation can be evaluated for the impor¬
tant case when /0 is the maxwellian velocity distribution for a station¬
ary plasma (u =■ 0), so that, as in Eq. (4.3.14),

n0e~v2lai 2 kT
/o =
(4.5.2)
7r3/2a3 m

The major difficulty in handling the integration in Eq. (4.5.1) arises

because of the pole when vx = oj/k. Therefore it is actually more
convenient to work with the dispersion relation obtained using the
Krook collision model [Eq. (4.3.4)]. For this case the Boltzmann
equation for the electron gas with

/ = /o + and E =

reduces to

e Fdf0 (4.5.3)
— iu/i + ikvxfi -r— -vfi
m dvx

which is entirely equivalent to the result obtained using the BV equa¬

tion except for the change from — iufi to (—ia> + v)fi- Hence the
dispersion relation for this case is identical to Eq. (4.5.1) except for a
change from u> to co + iv, and we have

(dfo/dvx) d3v (4.5.4)

1 =
v. — (co T iv)/k
182 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

With/o given by Eq. (4.5.2), we have

dfo (4.5.5)
-2^2/o
dvx

so that the dispersion equation reduces to

— 2coe2 /■» vxe ’’*2/°2 dvx r«

1 = e-<’al dvy e-^2/a2 cfo. (4.5.6)
x3/2/c2a5 J- °° vx~aC J-»
with C = (co + iv)/ka. The second two integrals here each equal
r / a; hence, letting i;* = ^o, we have
7 1 2

— A’d2/
(4.5.7)
_fc2"

where ko is the reciprocal of the Debye length defined in Eq. (3.3.17),

so that

fcD2 (4.5.8)
kT/rn

and where I is the integral

CO + iv
(4.5.9)
I = C =
ka

Evaluation of the integral

To complete the calculation we must evaluate I. This integral cannot
be handled by the usual technique of integrating around a contour
closed by an infinite semicircle around the upper or lower half plane,
since the integrand diverges when q = ±i°°. Therefore, to evaluate
7, we first note that

—— 1 H-—r, — Id-—Ty2 d 2 -rd (4.5.10)

q — C q — C q — G q1 — C2

hence we have

(4.5.11)
 4.5 LANDAU DAMPING 183

Of the three terms on the right, the first is a well-known integral which
equals unity; the second integrand is odd, and its integral therefore
vanishes; and the third integral is one which we must still evaluate.
Hence, to this point, we have

1 = 1+ C2G( 1) (4.5.12)

with

G{s) - x-» /;, dq (4.5.13)

The parameter s has been introduced here in order to transform

the integral into a differential equation. Thus, from Eq. (4.5.13)

sq*
[" q2e
7T l/2 dq
/-« q2 -

Tr-i/2
( C2 \ ,
[ * ^ 1
S-112 .
- C2G(s) (4.5.14)

If we multiply this equation by exp (sC2), we have

^Ge‘ct = _s-i/VC2 (4.5.15)

as

so that, upon integrating on both sides from s = 0 to s = 1, we have

G(l)eC2 - (7(0) = - s~1,2es°2 ds (4.5.16)

To evaluate (7(0), note that, from Eq. (4.5.13),

G(0> ^ (4.5.17)

This integral (which no longer contains the troublesome factor e q) can

be integrated around the usual path, indicated in T ig. 4.3, consisting
of the real axis and a semicircle about the upper half plane. 1 he
poles (at q = iC) are located off the real axis, as shown in the liguie,
since C = (o> + iv)/ka. Moreover, the contribution from the semi¬
circular path of integration is of order radius-"1 and vanishes as this
184 THE BOLTZMANN VLASOV EQUATION FOR A PLASMA

radius approaches infinity. Hence from the theory of residues we have

(5(0) = 27i-i res (g = C)

C 7
(4.5.18)
2ir1/2f lim
*c q- C2 c
so that in Eq. (4.5.16) we have

• l/2

(5(1) = -e-c2 fQl r1'2^’ ds + e~C2 (4.5.19)

In the remaining integral we make one last transformation and let

s^C = 2 ds = (4.5.20)

so that

1/2
t7T
2 (4.5.21)
Gd) - -§ /%>'-<’* + c
D-C

Treatment of the dispersion relation

If we substitute this result into Eq. (4.5.12) for I and then apply that
result to the dispersion relation given in Eq. (4.5.7), we find

w + iv
(l - 2C foC ez2~C2 dz + i7r1'aCe-c‘) C =
ka
(4.5.22)

Fig. 4-3 Contour of integration used to evaluate (?(0).

 4.5 LANDAU DAMPING 185

The integral which remains here is known as the dispersion f unction and
has been tabulated, f while the last term, which is imaginary, is known
as the Landau damping term. To calculate k as a function of w from
this dispersion relation ohe may use the following procedure:

1. Choose an arbitrary C.
2. Find the (tabulated) value for the dispersion function.
8. The dispersion relation then gives k2.
4- Find co from the definition, C = (u> + iv)/ka. (Actually, in most
calculations v is neglected, so that C = co/ka.)

Although one can use this formal procedure to obtain detailed

calculations of k as a function of co (or vice versa), it is also of interest
to compare this result with the Bohm-Gross dispersion relation
obtained earlier. For that case we let |co/fc| —> =o ; hence, to make
the comparison we must examine Eq. (4.5.22) in the limit when
|C| —> oo. As our first step we expand the dispersion function in the
series

11 = 2 C fj ez2-C2 dz

= l-j—-—I- ——!-•••+ 0e~C2 (4.5.23)

' 2C2 ' 4C4 '

where Oe~°2 denotes terms of order exp( — C2). The derivation of this
expansion will be given shortly, but assuming that it is valid, we can
use it in Eq. (4.5.22) to find

k2 = kD2 + 4I1 + ’ ’ ‘ + 0e_C2 “ ^1/2C'e“C*) (4.5.24)

Now C2 = w2/k2a2 (in the limit when we set v — 0), and kD2 = 2o>e2/a2.
Hence this dispersion relation easily reduces to

co2 = coe2 (l + + • • • + 0C2e~cl - 2i^l2C*e-c^J (4.5.25)

Landau damping
Again, since C has been assumed to be very large, we have, to lowest
order, w2 = a>e2, and

C2 ^ ^ |C| -> * (4-5-26)

k2a2

f B. D. Fried and S. D. Conte, “The Plasma Dispersion Function,” Academic

Press Inc., New York, 1961.
186 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

so that, as expected, the high-phase-velocity limit occurs at frequencies

very close to the plasma frequency. In addition, if we return to Eq.
(4.5.25) and replace each C by we/ka, we have

ZkTW 2f7r1/2av5 utik2 a2

CO2 = COe2 + + (4.5.27)
m k3a3

The first two terms here are the Bohm-Gross result [Eq. (4.4.18)].
However, the imaginary term in the dispersion relation is new. Note
that for a standing-wave problem (where k is real) this result shows
that w2 has a negative imaginary part, so that, from the arguments in
Sec. 2.4, there is temporal decay. This is also evident here since, from
Eq. (4.5.27), to lowest nonvanishing order in real and imaginary parts,!

iirll2coe3 (4.5.28)
e we — ia
k3a3

The negative imaginary part in o> leads to temporal decay, since the
waves are proportional to

gikx—iut __ gikx—io)et—at (4.5.29)

This decay term a, which arises in the absence of collisions, is called

the Landau collisionless damping factor.
The physical mechanism responsible for Landau damping is
simply the interaction of the electrons with the electric field associated
with the wave, Eaz cos (kx — ut). Since there are assumed to be no
collisions, the velocity of each electron in the plasma satisfies the
equation

mvx = —eE cos (kx — wt) (4.5.30)

with x the position of the electron at time t. By considering solutions

to this equation one finds that, for some electrons, the average kinetic
energy increases with time due to this force term, while, for others, the
average kinetic energy decreases with time. The gain or loss of energy
depends on both the exact initial position and velocity of the electron,
but for a uniform mixture of electrons having an initial maxwellian
velocity distribution, one finds that the total average kinetic energy
of the electrons increases slowly with time at a rate which just balances

f Note that we use the result that, for |e| « 1, (1 + e)1/2 SS 1 +

 4.5 LANDAU DAMPING 187

the decrease in the electric-field energy due to the Landau damping

factor, t

Expansion of the integral

To complete our discussion of this problem we must derive the series
representation of the integral I\ shown in Eq. (4.5.23). We have

h = 2C Jo° e*2~C2 dz (4.5.31)

Hence, if we change to the new variable x = C2 — z2, we find

(4.5.32)

Since x is less than C2 over the entire path of integration, we can

expand [1 — (x/C2)]~112 in the series

(4.5.33)

with

1 X 3 X • • ' X (2w - 1) (4.5.34)

2"n!

We can now integrate each of the terms in this summation, since we

easily find from a series of integrations by parts that for any integer n

, n\ n n(n — 1)
2 + +
(
-C
e x dx £f2n e 1 C2
~C^
(4.5.35)

If we substitute Eqs. (4.5.33) to (4.5.35) into Eq. (4.5.32) we find

1 X 3 X_X- -1) _|_ ^erms 0f order e °2

h = 1 + 2
n= 1
(2 C2)n
(4.5.36)

which is the expression used in Eq. (4.5.23). Although this is an

asymptotic series expansion (and actually diverges when n-^ <*>) one

f See, for example, T. H. Stix, “The Theory of Plasma Waves,” McGraw-Hill

Book Company, New York, 1962.
188 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

can use the first few terms in the expansion to obtain a good estimate
of I\ provided C is large.

Problems
4.5 Prove that, for small values of C,

2C2 4C'4 8C6

h = 2C2[1
0 3 + 15 105 + -)

so that, as C —* 0, k2 = — ko2 or

k7
- = —03,
m

which is the low-frequency limit of the continuum result obtained

using the isothermal sound speed.
4.6 Use a perturbation approach (in which, to lowest order E = 0)
to show that for a particle with initial velocity v0 in the x direction
and initial position x0, the velocity perturbation due to an electric
field Eax cos (kx — ut) is given by

V\ = — [sin (kxo — at) — sin kxo] a — kvo — 03

ma

Hence the perturbation is large only when Vo approximately equals

the phase velocity 03/k.

4.6 LANDAU DAMPING OF TRANSVERSE WAVES

The techniques described in the last two sections can be extended

easily to treat more than just longitudinal waves in a plasma. There¬
fore in this section we shall analyze first the Landau damping of
transverse waves in fhe relatively simple case when there is no dc
magnetic held, and then we shall indicate two techniques by which
one can extend the analysis to treat the much more complicated prob¬
lem of wave propagation in the presence of a uniform B held.

Analysis with B0 = 0
To find the general dispersion equation for small-amplitude waves in
the electron gas we set

/ = fo(v) + fi(v)eiks-iat l/il «/o (4.6.1)

 4.6 LANDAU DAMPING OF TRANSVERSE WAVES 189

Note that for simplicity we assume heref that /0 is a function just of

the magnitude of v; hence we have the very useful identity

dfo = dv_dfo =
(4.6.2)
dvi dvi dv v 0

In addition we assume initially that there are no steady E or B fields,

so that the electric and magnetic fields have the form

|Tpikx—uat (4.6.3)

If we now substitute Eqs. (4.6.1) and (4.6.3) into the BY equation,

dl + df_ - -(E + X B); = (4.6.4)

dt dXi m dVi

we obtain, to first- order in the small quantities fh E, and b,

-tw/i + ikvxfi - - Ei P = 0 (4.6.5)

m di'i

Note that the magnetic-field term does not enter, since, from Eq.
(4.6.2),

(v x B)ip- = (v x B) • = 0 (4.6.6)
dVi V

From Eq. (4.6.5) we havej

= i(e/m)Ej dfo (4.6.7)

1 co kvx dVi

To complete the specification of the problem we use the two

Maxwell curl equations, which, for the fields given in Eq. (4.6.3),
reduce to

kax X E = wb (4.6.8)

kax Xb = — ^ E f/ioJ (4.6.9)

f For some problems this assumption cannot be used, and the resulting analysis
is somewhat more complicated.
t Note that, again, if we use the Krook collision term, the only change is from
to to a; + iv in this equation. For this reason, where needed, we shall evaluate all
poles as if we had used the Krook model.
190 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

with J (which equals —nee(\)) given byf

J = —ejfi\d3v (4.6.10)

If we use Eqs. (4.6.8) and (4.6.10) to eliminate b and J, Eq. (4.6.9)

reduces (after multiplication by c2/ co) to

E - n2E, = — [ /iv cl3v (4.6.11)

coe0 J

where E, = E — Exax is the transverse part of E and n = kc/ co is

again the index of refraction. For fi given by Eq. (4.6.7), the x
component of this equation is

E = _ [ d/o v*d*v (4.6.12)

x Uqu J dVi co — kvx

while the transverse part of the equation is

E =_
n0co(l
^E-—~ n2) Jf a/°
_
dVi co
Vi dh -
— kvx
(4.6.13)

Two of the terms vanish in each sum on the right side of Eqs. (4.6.12)
and (4.6.13) since they lead to integrals of the form

r VjVj dfp d*v (4.6.14)

J v dv co — kvx

which have odd integrands with respect to either Vi or Vj (since, for

i *j, both cannot be vx). Hence Eq. (4.6.12) and the y component
9

of Eq. (4.6.13) reduce at once to

CO.2 E f Vx(dfo/dvx) d*v

Ex (4.6.15)
noco 1 J kvx — co

cce2Ev r Vy(dfo/dVy) d3v

Ey (4.6.16)
noco(l — n2) J kvx — co

[The 2 component of Eq. (4.6.13) is just like the second of these

equations.]

t Recall that since /0 is an even function, J/0v d3v is zero.

 4.6 LANDAU DAMPING OF TRANSVERSE WAVES 191

In Eq. (4.6.15) note that

r vx(dfo/dvx)
J kvx — co
d,v

l /1; (tor=n +!)dHl

co f
(dfo/dvx) d3v
(4.6.17)
k J kvx — co

since j df0/dvxd3v is identically zero. Substituting this result into

Eq. (4.6.15), we see that for longitudinal waves (Ex ^ 0), we once again
have the dispersion relation [see Eq. (4.4.7)]

C0e2 r (dfo/dvx) d3v

(4.6.18)
n0k J kvx — co

This equation has been analyzed thoroughly in the last two sections
and need not be discussed further here. However, from Eq. (4.6.16)
we see (after an integration by parts) that for transverse waves (Et 9^ 0)
we have the new dispersion relation

coe2 ■ r fo d3v
(4.6.19)
1
n0co(l — n2) J kvx — co

If we again let /0 be the maxwellian velocity distribution

n0e-”2/a2 (4.6.20)
7r3/2a3 m

we find, after integrating over vy and vz,

l = ~W*2C_ (4.6.21)
7t1/2co2(1 — n2) Q~C

where once again C = u>/ka and q = vx/a. Multiplying numerator

and denominator by q + C and noting that qe~qt/(q2 + C1) is odd (so
that its integral vanishes), we find

CO e2C2 (4.6.22)
1 = G(l)
co2(l - n2)

where G(s) is the integral defined in Eq. (4.5.13). In the limit when
C is very large we have seen [Eqs. (4.5.21) and (4.5.23)] that

1/2
ITT
2
i( 1+J_ + ± + + c -C
(4.6.23)
(7(1) c! V ^ 2C* 4C* T )
192 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Hence, in this limit, Eq. (4.6.22) reduces to

k2c2 = co2 — coe2 + t7r1/2coc2Ce_C2 (4.6.24)

which is identical to the dispersion relation obtained in magnetoionic

theory (with no collisions) except for the Landau damping term,
iir1,2we2Ce~C2.
Since C = w/ka is of order c/a for frequencies above the plasma
frequency, this damping factor is exceedingly small for nonrelativistic
plasmas. In fact, one can argue that for this case the Landau damping
term is really zero, since the integration over vx should really extend
only from —doc, while the phase velocity is greater than c. Hence
the pole at vx = vph (or q = C) is really outside the path in the contour
integration. In any event, whether one uses Eq. (4.6.24) or not, the
important conclusion is that Landau damping is much less important
a factor for these high-phase-velocity transverse waves than for the
lower-phase-velocity longitudinal waves. However, by the same
argument one woidd expect the Landau damping again to be impor¬
tant for the low-phase-velocity waves that can propagate when there
is an applied magnetic held. For this reason we shall briefly consider
that problem next.

Results for B0 small

When there is a uniform magnetic held present, the wave-propagation
problem becomes much more complicated. Hence rather than treat
the problem in great detail, we shall merely indicate two approaches
by which the dispersion relation can be obtained. We note hrst that
when /i and E are still proportional to eikx~io,t but the magnetic held
is of the form

Bn + beikx~iat (4.6.25)

we have, from Eqs. (4.6.4) and (4.6.6), to hrst order in fh E, and b,

— fco/i + ikvxf i — F df°

Bi — +I (v
(v v
X R df1
'I —
r»o)i (4.6.26)
m dVi dVi

Thus, in place of Eq. (4.6.7) we now have

_ i(e/m)Ej d/0 i{e/m)(\ X B0)t- dft

(4.6.27)
^ w — kvx di\ co — kvx dVi

Although Eq. (4.6.27) is a complicated partial differential equa¬

tion, we shall show shortly that one can find an exact solution to it.
 4.6 LANDAU DAMPING OF TRANSVERSE WAVES 193

For many applications, however, this exact solution is not really

necessary, and it is adequate to have a solution in the limit when B0
is small. For this case we set

fi = /i(0) + £o/i(1) + B o2/d2> + • • • (4.6.28)

Substituting this expansion into Eq. (4.6.27) and matching powers of

Bo, we find that/i(0) is the field-free result [Eq. (4.6.7) or the first term
on the right side of Eq. (4.6.27)], while

s (i) _ _ (e/m)2Ei(y X B0)y d_ vty(v) (4.6.29)

^1 (w — kvx)B o dvj co — kvx

where, from Eq. (4.6.2), g(v) = f'Jv. Now by direct calculation

d v,g(y) _ 1 |" s / s , ViVjg'ji') , Sjxkvjg{v) (4.6.30)

di>j w — kvx w — kvx v w kvx

If we use this expression in Eq. (4.6.29) and note that (v X B0) • v is

zero, we have •

(■e/m)2g(v)Ei
/i(1) kvx){\ X B0)i + kvi(\ X B0)J (4.6.31)
B„(« - kvxy u

Since Maxwell’s equations are unchanged by the B0 term, we still

have, as in Eq. (4.6.11),

E - n2E, = — [ fiv d3v (4.6.32)

coeo J

Hence, to obtain our dispersion relations (to first order in B0) we need
only substitute our approximation /i = /ico) + B0f i(1) into this equa¬
tion and carry out the integrations over v. If we let

Bo = B 0(a, cos 7 + av sin 7)

with 7 the angle between B0 and the direction of wave propagation,

then most of the integrals are zero (because the integrands are odd
with respect to vy or vz). In addition, if we assume that /o is the
maxwellian velocity distribution, we find that
194 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

so that after carrying out the integrations over vy and vz, we can reduce
Eq. (4.6.32) to the form

A•E = 0 (4.6.34)

where

An = (1 + XTi) A13 = —iXY sin y(Tz + T4 + Tb)

A22 = Azz — 1 n2 XT2

(4.6.35)
A2z = —Az2 = iXY cos yTz
An = —iXY sin yT4 zli2 = An = 0

In these coefficients, as in Eq. (2.6.14), X — we2/co2 and I — eBo/mco.

The factors T1 to T6, however, are all the final integrals (written in
terms of C — w/ka and q = vx/a)

2C r q2e~gl r _ C r er* dg
Ti
J
7T1/2 ' g-cd? 12 ~ ir112 J q-C

C2 r e~g2 dq T _ f e~q2 dQ (4.6.36)

Tz
7T1'2 J1 (q - cy 4 tt1'2 J (q - cy
2C2 r q2e~q2
T,
7T1/2 .1 (q - C)2

The integrals T1 and T2 have been evaluated previously t in terms

of (7(1) in our discussions of Landau damping when B0 = 0. The
other integrals are all simply related to T1 and T2, since

T3 = C2 AA
dC c
_ ifiZ At A (4.6.37)
t4
2 dC2 C

t5 — C2 A. ^1
= dC~C

fNote that an equivalent form for T1 is

_ 2C2 r” qe~q'1
dg
7J — » q — C

= 2C2[1 + C2G(1)]

while, as in Eq. (4.6.22),

r2 = <7*G(i)
 4.6 LANDAU DAMPING OF TRANSVERSE WAVES 195

While it is not feasible here to analyze the resulting dispersion relations

in general, if we take the limit when C °° and neglect the Landau
damping terms (due to the poles in these integrals), we easily find that

rj10 r'^/ | T 3 T5 1 (4.6.38)

so that A reduces to

1 - X 0 — iXY sin y
A = 0 1 - n2 - X iXY cos 7 (4.6.39)
iXY sin y —iXY cos y 1 - n2 - X

Setting the determinant of A equal to zero, we find (after dividing by

1 - X)

(1 - n2 - X)2 - B{1 - n - X) - C = 0 2 (4.6.40)

with

sin2 y
B = X2F2 C = X Y cos2 2 2 7 (4.6.41)
1 - X

Hence we have

n2 = 1 — X - ± (}B + C)1/2
2 (4.6.42)

This result differs somewhat from the Appleton-Hartree dispersion

relation [Eqs. (2.6.18) and (2.6.19)] found using the Langevin equa¬
tion. However, this difference is solely due to our having made an
expansion to first order in B (or Y). Thus if we expand Eq. (4.6.42)
0

in a Taylor’s series about F = 0 and retain only the term linear in F,

we find

n 2 = 1 - X ± XF cos 7 (4.6.43)

while, to the same level of approximation, in the Appleton-Hartree

equation we have

n = i-—-^ 1 — X ± XF cos
2 7 (4.6.44)
1 ± F cos 7

An expansion of Eq. (4.6.42) and of the Appleton-Hartree equations

to terms of order F2 would lead to different results for each case.
196 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

However, this is not surprising in view of our having kept only terms
of order B0 (or Y) in the expression for /j.

Exact evaluation of/i

To complete our discussion of the application of the BY equation to

the wave-propagation problem, we shall examine briefly the pioblem
when there is a large applied magnetic field, so that the expansion of
/i as a power series in B0 is not justifiable. Rather than work directly
with Eq. (4.6.27) for fh it is convenient now to use a new coordinate
system where Bo is in the z direction and the direction of propagation
is in the yz plane (see Fig. 4.4). In this new coordinate system we can
again let y be the angle between B0 and the direction of wave propaga¬
tion, so that B = Roaz and k = k{az cos y + ay sin y).
Note that in this coordinate system we also have

vx = v sin 9 cos cf> vv = v sin 9 sin vz = v cos 9 (4.6.45)

Hence

dfi^dVxdfi.dVydfi.dtudfi
d<f> d(f> dvx d<t> dvv d4> dvz

(4.6.46)

y
Fig. 4-4 Coordinate sys¬
tem with az in the direc¬
tion of B0 and ax perpen¬
dicular to k and B0.
 4.6 LANDAU DAMPING OF TRANSVERSE WAVES 197

Through the use of this identity, the BV equation now reduces to the
form [see Eq. (4.6.27)]

/ | \ r ^ X? 0 I * d/1
(co - k • vi = — Ei --h iuc — (4.6.47)
m di’i d4>

with wc = —eB0/?n. This differential equation in <f> has the formal

solution

eEi r<t> d/0

h = ~ 7 exp [(w — kzvz)(4> — <t>')
ma>r J/■_ 00 di COc

+ ky(vx — v(.)] i d<t>' (4.6.48)

as can be verified by taking a derivativef with respect to </>. Note

that, the lower limit is chosen to be — °°, since one may assume (from
the Krook model, for example) that w has a small positive imaginary
part, while from its definition wc is negative. Hence the integrand in
Eq. (4.6.48) is zero at <p' = — °°.
To actually work with Eq. (4.6.48) note that v' is the vector with
components v, d, and <t>'; hence

dfo Ejv'j df0

J% dv[ v dv

= dh [sin e(Ex cos + Eu sin 0') + cos 6EZ) (4.6.49)

dv

By substituting this expression into the integral for/i it is possible to

evaluate]; fi in terms of E and v. Because of the term

kyv'x = kyv sin 9 cos </>'

in the exponential, the result involves Bessel functions, so that the

final calculation of J/iv d3v in Eq. (4.6.32) is difficult to handle except
in some special cases (see, for example, Prob. 4.7).

f To take the derivative of fi recall that the formula for the derivative of an
integral is given by

s jC Fix’’x> dx' -FlMxM t - FiMx)’x] £ + //!' £dx'

which follows directly from the basic definition,

dG ,. G(x + *) - G(x)
—— = lim -
dx e_^o e
J See for example, I. B. Bernstein, Phys. Rev., 109:10 (1958) or E. G. Harris,
J. Nucl. Energy, pt. C, 2:138 (1961).
198 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Problem
4.7 Perform the integration indicated in Eqs. (4.6.48) and (4.6.49)
for waves traveling in the direction of Bo (so that ky = 0) to find

ie 1
h = A2 - 1
mwc

aVr + i§fo\ E*dfo\

dvy dvxJ A dv2j

with A = (w — kvz)/uc. Show, from Eq. (4.6.32), that this

choice for /i leads to three wave solutions: the usual (Landau
damped) longitudinal waves and l.c.p. and r.c.p. transverse waves.

4.7 THE TWO-STREAM INSTABILITY

Choice of /0

In our discussion of Landau damping, we noted that the damping of

the wave is due to the fact that for the particular case of a uniform
electron gas with a maxwellian velocity distribution, the interaction
of the particles with a small-amplitude oscillating longitudinal electric
field leads to a gain in the average kinetic energy of the electrons and,
consequently, to a damping of the electric-field energy of the wave.
For many other cases, the opposite result occurs, and the average
kinetic energy of the particles actually decreases, while the electric
field in the wave grows. Such a situation obviously leads to a growing
wave. Physically, however, we recognize that small statistical fluctua¬
tions are always present in the plasma. Hence any disturbance which
grows (at the expense of the ambient kinetic energy in the plasma)
will almost certainly develop spontaneously and cause the initial
velocity and/or spatial distribution of the electrons to change. For
this reason, such an initial state is called unstable.
Although there are numerous examples available in the literature
of plasma instabilities, we shall here describe just one important
example, the two-stream instability. For this problem we consider a
zero-order distribution that consists of two contrastreaming beams of
electrons. Although the instability arises under a wide range of beam
conditions, we consider first the very simple case when each beam is
uniform, each has the same number density, and each travels in the x
direction with the same speed V but in opposite directions. If we
further assume that each particle in each beam has in zero order
exactly the beam velocity, then the distribution function can be simply
 4.7 THE TWO-STREAM INSTABILITY 199

written as

fo = in08(vy)8(vz)[8(vx — V) + 8(vx + F)] (4.7.1)

where the 8 functions shown here are discussed in Sec. A.l of the
Appendix.

The dispersion equation

As given earlier, the dispersion equation for small-amplitude longitudi¬

nal waves in any electron gas described by the BV equation with the
Krook collision modelf is [from Eqs. (4.4.10) and (4.5.4)]

fo dh (4.7.2)
0 = w + iv
n0 J (.kvx — ft)2

Substituting Eq. (4.7.1) into this dispersion equation and integrating

over each of the 8 functions, we obtain

1 1
_(kV - ft)2
+ (.kV + ft)2
(4.7.3)

or

{k2V2 - ft2)2 = We2(/c2F2 + ft2) (4.7.4)

We can solve this equation either for standing waves (where k is real)
or traveling waves (where a> is real). Since it is somewhat easier with
the Krook model to treat traveling waves, we rewrite Eq. (4.7.4) in
the form

fc4F4 - k2V2(2Q.2 + coc2) + ft2(ft2 - we2) = 0 (4.7.5)

which has for its solutions

lc2V2 = ft2 + |we2 ± (2ft2we2 + iwe4)1/2 (4-7.6)

In order to investigate these solutions we can examine first the

high-frequency limit when |ft| ^S> w«, so that Eq. (4.7.6) reduces to

k2V2 ^ ft2 (4.7.7)

j The use of the simple relaxation model is not really justified here, since fo
is not a maxwellian distribution and collisions do not tend to damp out deviations
between / and/o. However, as in the last sections, we shall always make our cal¬
culations in the limit as v —* 0.
200 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

or, for the wave traveling in the positive x direction,

kV ^ 12 + 2~1/2«e = co + 2_1/2oje + iv (4.7.8)

In this limit the real and imaginary parts of k are both positive for
each root; hence both waves are slightly attenuated. The phase
velocity for each wave approaches the beam speed V as w approaches
infinity, but for finite frequencies the phase velocity is less than V for
the root with the plus sign and greater than I for the root with the
minus sign, as shown in Fig. 4.5.
The growing wave appears when we examine the low-frequency
limit, where |122| « we2/8, so that Eq. (4.7.6) reduces to

2122
k2V2 ^ iue2 1 + — ± (4.7.9)
U>e

For the root with the plus sign this becomes

k2V2 ^ o>e2 + 3fl2 (4.7.10)

which is a damped wave (since the imaginary part of k2 is positive)

with a very low phase velocity; however, with the minus sign, the
dispersion relation is

k2 V2 == —fl2 = v2 — co2 — 2 icci; (4.7.11)

Since the imaginary part of k2 is negative for this root, from the dis¬
cussion in Sec. 2.5 it follows that this is a growing wave. It is worth
emphasizing the importance of using the Krook relaxation term.
Had we used just the collisionless Boltzmann equation, the dispersion

Fig. 1^.5 Plot of phase velocity and growth rate vs. frequency for waves
traveling in two contrastreaming plasmas.
 4.7 THE TWO-STREAM INSTABILITY 201

relation here would have been k2V2 = — co2, and we should not have
been able to demonstrate clearly that the wave is growing rather
than evanescent, f
In Figure 4.5 we show a plot of phase velocity and growth factor
(the imaginary part of k) as a function of frequency for these two waves
in the contrastreaming plasma beams in the limit when v = 0. As
shown in the figure, the root with the minus sign grows for all fre¬
quencies below the plasma frequency. This can also be seen from
Eq. (4.7.6), where, for v = 0, the root with the minus sign is of the
form

k2V2 = A - B (4.7.12)

where

A2 = co4 + coW + icoe4 (4.7.13)

and

B2 = 2coW+iwc4 (4.7.14)

so that, for co2 <C co,.2, B exceeds A and k2 is negative, dhe maximum
growth rate occurs when co2 = f coe2, as can be verified by examining
the derivative of k2 with respect to co. 1 his amplification effect has
been verified experimentally and serves as the basis for the two-
stream amplifier.

Problems
4.8 Derive Eq. (4.7.3) from Poisson’s equation and .the macroscopic
cold-plasma mass and momentum equations for two beams
of electrons where 711,2 = + n\:2elkx^l0,t, v\x = I + vxelkx lat,
V2x = —V A- v'2eikx~iat, Ex = E'eikx~io,t, and V » v'. Note that
one should neglect collisions; hence the result will have co rather
than H.
4.9 Verify that the maximum in the growth rate occurs when
w — (3)i/2w<, What is the number density, mean velocity, and

f We should note that this conclusion, though it follows directly from the use
of the Krook model, does not agree with an analysis (using the BV equation) by
P. A. Sturrock [Phys. Rev., 112:1488 (1958)], who concludes that for contrastream¬
ing plasmas there is only temporal growth and not spatial growth. However, this
difference is probably only a question of the choice of a moving or stationary coor¬
dinate system, as discussed by R. J. Briggs, “Electron Stream Interaction with
Plasmas,” The M.I.T. Press, Cambridge, Mass., 1964.
202 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

temperature for each beam and for the mixture of the two beams:
[Find these by working directly with fo as given in Eq. (4.7.1).]

The beam-plasma amplifier

Another simple but important example of the instability of two plasma
beams is the beam-plasma amplifier. For this case one of the beams
(with number density rib again has velocity 1 , while the other beam
is actually a stationary plasma (with number density np). Hence the
zero-order distribution function is assumed to be

fo — 5(vy)d(vz)[nb5(vx — V) + npd(vz)} (4.7.1.))

Substituting this distribution function into the dispersion equation for

small-amplitude longitudinal waves [Eq. (4.7.2)] and integrating over
each of the 5 functions, we find

Wb2 , C0p2
(4.7.16)
(kV - ft)2 + TF

where co6 = (nbe*/me0)1/2 and cop = {npe2/mt0)I/2 are the plasma fre¬
quencies for the beam and for the stationary plasma, respectively.
From Eq. (4.7.16) we have

- e>! - ^ (47-I7)

Hence, if we take the square root of each side and go to the limit as
v —> 0, we have

COb
kV = co 1 + (4.7.18)
(co2 - CO,2)1'2

When co exceeds cop, k is real, so that the waves are neither damped nor
growing. However, for co below cop, we have

IWb
kV = CO 1 + 1/2
(4.7.19)
(cop2 — CO2)

so that the phase velocity co/Re k is V for both roots, but one root is
damped, and the other grows at the rate

COCOb
Im k = (4.7.20)
F(cop2 - CO2)1'2
 4.8 THE DEBYE POTENTIAL PROBLEM 203

A plot of phase velocity and damping (or growth) factor is shown

in Fig. 4.6. It can be seen that the growth rate for the unstable wave
is greatest for frequencies just below cop. In addition, the phase
velocity for the wave with the plus sign in Eq. (4.7.18) is a relatively
simple function that is zero at cop and rises iponotonically toward V as
co—-» co. The phase velocity for the other root, however, is more
complicated. It is also zero at cop but decreases to — °o as

CO > C0e = (cOj,2 + C062) 1/2

Then at co = coe the phase velocity jumps to + 00 before falling off

toward V as co —> °o.

4.8 THE DEBYE POTENTIAL PROBLEM

The BV and Poisson equations

The next problem to which we apply the BV equation is known as the

Debye potential problem or the potential about a charge in a plasma.
We suppose that a point charge Q is placed, for convenience, at the
origin and that the medium in which this “test particle” is embedded
is a plasma with equal numbers of electrons and ions. We know that
this charge will cause a rearrangement of the particles in the plasma
since it attracts particles of the opposite sign and repels like charges.
Therefore, the questions we ask are (1) what are the resulting steady-
state velocity distributions for the electrons and the ions? and (2)
what is the electric potential 4> that is established near the charge Q?

CO

Fig. 4.6 Plot of phase velocity and growth rate vs. frequency for wave travel¬
ing in a beam-plasma amplifier.
204 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Since this is a steady-state problem, as shown in Sec. 1.3, we can

set

E = -V4> (4.8.1)

In addition we can neglect the time-derivative terms in the BY equa¬

tions for the electron and ion velocity distributions, so that we have
(assuming just an E held in the Lorentz force term)

Sfe d4> dfe 0 (4.8.2)

Va — + —
J dxj me dxj dvj

and

e d4> dfi
1) ■ dfi- (4.8.3)
1 dxj mi dxj dVj

To these two equations which involve fe, fi, and 4> we again add
the Maxwell equation involving V • E (Poisson’s equation) to have

Pc
V • E = (4.8.4)
«0

In this equation V • E = -V24>, and the charge density is just that

due to the electrons and ions plus the charge density of the point
charge, Q8(x)8(y)8(z). Hence, for this case we have

V24> = —
£0 J
f
(fe — fi) dzv — — 8(x)8(y)8(z)
£0
(4.8.5)

The equation for <J>

Equations (4.8.2), (4.8.3), and (4.8.5) provide three equations for fe,
fi, and 4>. As our first step in solving the set of coupled partial differ¬
ential equations, we note that one formal solution for/*, and f, in terms
of 4> is given by|

\mav2 + qa$\
fa = A a exp (4.8.6)
«T )
where a takes on the values e for electrons and i for ions and where Aa
is a constant. We can easily verify that this satisfies Eqs. (4.8.2)

f Note that this distribution goes into a maxwellian velocity distribution

(with zero drift velocity) when <i> is zero.
 4.8 THE DEBYE POTENTIAL PROBLEM 205

and (4.8.3), since for this solution by direct calculation

v dfa _ dfa
- 1) = 0 (4.8.7)
3 dxj ma Ox, dvj K 1 OXj

In addition we can evaluate the constants Ae and Ai by noting that if

the total number of electrons and of singly charged ions each equals N,
then

N = jj fad3vd3x

= Aa J e~v‘‘,aa' d3v J e~qailKT d3x aJ =

2 kT
Via

= AaTr3l2aa3 J d3x (4.8.8)

Hence, the two constants, Ae and Ai, are given by

N • . N
A, — Ai =
7T 3l2ae3f ey d3x IT 3/2a,i3fe y d3x
(4.8.9)
e<t>
T rn
Kl

We next use Eq. (4.8.6) for the distribution functions to reduce

Eq. (4.8.5) to an equation involving only <t>. Note that, from Eqs.
(4.8.6) to (4.8.9),

ffa d3v = Aae-qa*lKTfe~vt,aat d3v

(>—Qa$ I *T
(4.8.10)
fe—qa<t>UT

If we substitute this result into Eq. (4.8.5), we obtain, with no

approximation,

V24> = — F(4>) - - 5(x)5(y)5(z) (4.8.11)

£o eo

where

ey e4>
7 = (4.8.12)
W) = fey d3x fe~y d3x kT
206 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

If we want to proceed further by analytic (rather than numerical)

techniques, we must now introduce two approximations.
First, we assume that |y| is much less than 1. (1 his means that
we restrict our analysis to regions where the potential energy ±e$ is
small compared to the average thermal energy of a particle.)
For |y| « 1, we can set e±y £= 1 ± y to find

F(4>) =
1+7
7(1 + 5)
1 7 =*
7(1 - 5) ~ 7
% (7 - 5) (4.8.13)

where 7 = J d3x is the total volume of the system, and

5 = y y 7 d3x |5| « 1 (4.8.14)

Even with this simplification, the equation for <t> is still too complicated
to treat analytically unless we also have

|5| « |y| (4.8.15)

We shall examine the significance of this condition very shortly, but

assuming it to be true, we finally obtain

F($) ^ 1^, (4.8.16)

so that the partial differential equation for [Eq. (4.8.11)] reduces to

V24> - kD>2<i> = — - 8(x)8(y)8(z) (4.8.17)

to

where kD> is 2| times the reciprocal of the Debye length, so that

2co„2 = 2(N/V)e2
(4.8.18)
kT/wi K.TtQ

The Debye potential

To complete our problem we must now solve Eq. (4.8.17) for the
potential 4>. Since the source of the potential (the point charge Q) is
spherically symmetric and located at the origin, <t> should have no 6
or </> dependence. Moreover, in spherical coordinates, when 4> depends
 4.8 THE DEBYE POTENTIAL PROBLEM 207

only on the radial distance R,

V2$ = 1 d2(-R^l (4.8.19)

R dR2

Hence, for R ^ 0, Eq. (4.8.17) reduces to

l R$ - kD'^ = 0 (4.8.20)
R dR2

or

— R® ~ kD-2m) = 0j (4.8.21)
aKz

which is a simple differential equation for hM? with the solution

E4> = A exp (kD'R) + B exp ( — kD'R) (4.8.22)

If we are dealing with an unbounded (or very large) system, then

the potential due to the charge Q at the origin must go to zero for
large R. Therefore the constant A must be zero, and we have

_ B exp (-kD’R) (4.8.23)

R

The final constant B is found by integrating the full differential

equation (4.8.17) over a sphere of radius e centered at the origin and
then going to the limit as t —> 0. We have

V24> dV - [ kD'2$ dV = - - f 8(x)8(y)8(z) dV (4.8.24)

Jr <e )r<€ e0 JR<*

To evaluate the first term on the left we first apply Gauss’ theorem
to obtain

V24> dV = [ V4> • dS (4.8.25)

Jr <e JR <e

On the surface of the sphere with radius e

(4.8.26)
dS = t2 sin d dd d<\> ar
208 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

while

/ d<t>\ Bexp( — kD'e)n 1v

(4.8.27)
= a'(as), = ~a'-4-^ + !)

Applying these expressions to Eq. (4.8.25) and taking the limit as

e —> 0, we have for the first term

f
J R <(
V24> dV = -
J0
r B [exp ( — kD>e)](kD>e + 1) sin 6 dd d<t>
JO
-> -47rB (4.8.28)

The remaining two terms in Eq. (4.8.24) can be treated much

more easily. For the second term on the left we have (recall
dV = R2 sin 0 dR dd d<t>)

— kD'2B J* R sin 6 exp ( — ko’R) dR dd dcf)

= —4irkD'2B Jj R exp ( — kn'R) dR—^0 (4.8.29)

while, for the term on the right, the integration over the 5 functions
yields unity, so that the entire equation reduces to

— 4tt £ = — ^ or B = -P- (4.8.30)

e0 47T€o

Substituting this expression for B into Eq. (4.8.23), we obtain, finally,

for the potential due to the charge Q located at the origin

Q exp( — kD'R)
(4.8.31)
47veoR

a result which is called the Debye potential.

Interpretation of the Debye potential

Note that if the point charge Q were placed in a vacuum, coe (and
therefore kD') would be zero, so that the potential would reduce to
the familiar coulomb potential, 4>c = Q/^-k^R. Therefore, we can
 4.8 THE DEBYE POTENTIAL PROBLEM 209

express the Debye potential as

<t> = <t>c exp ( — ko'R) (4.8.32)

which shows that <t> is much less than the ordinary coulomb potential once
R exceeds the Debye length. Hence, in a crude way, one can say that
a charged particle in a plasma interacts (in steady state, at least) only
with particles less than one Debye length away. These particles are
said to lie in the Debye sphere surrounding the charged particle.
To complete our discussion of this important problem, we now
can reexamine the two approximations used to obtain the Debye
potential: (1) that |y| = \e$/kT\ <3C 1 and (2) that

|S| = |F_1/y d3x\ « |t|

To check the first approximation, note that with <f> given by Eq.
(4.8.31) and Q = e,

e3 exp ( Icd'R') Xd exp ( /c/yE) g 33)

^ 47T€oE/<:T 3 RN d

where ND = |ttXDzne = 1.37 X 106T3l2/ne112, with T in degrees Kelvin

and ne in particles per cubic meter, is the number of electrons in a
Debye sphere, mentioned in Sec. 1.1. If Nd is very large, which is
the case for virtually all plasmas, it is evident that y is much less
than 1 (except when R is less than XD/ND). However, when ND is
not large, one cannot expect the potential near a charge to have the
Debye form.
To check the second approximation, one can show by direct cal¬
culation (Prob. 4.10) that, for a spherical volume of radius R0, if one
assumes 4> to be given by Eq. (4.8.31), then one finds

- = 1exp (kD>R)[ 1 - (1 + kD'Ro) exp (-kD'R0)] (4.8.34)

7 E0

Hence 5 can be neglected compared to 7 for distances comparable to,

or less than, Xd provided the volume of the plasma is much larger than a
Debye sphere, that is, R03 » XD3. For larger distances the neglect of
8 is not justified, but at these large distances 4> is already negligibly
small; hence the error is not of great significance. To summarize, we
find that the Debye potential is consistent with the approximations
used to derive it provided we restrict attention to distances between
Xd/N and Xd and provided the volume of the plasma is large com¬
pared to a Debye sphere.
210 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Problems
4.10 Verify Eq. (4.8.34) by evaluating 6 = V-'jy d3x over a sphere
of radius E0 and letting <t> be the Debye potential [Eq. (4.8.31)].
4.11 Using the collisionless hydrodynamic momentum equations for
the electrons and ions

Pa
I)U a __
^Pa

and Poisson’s equation, derive the approximate steady-state

partial differential equation for <f> [Eq. (4.8.17)] which results
when a stationary uniform plasma is perturbed by a point charge
Q placed at the origin. You may assume that, in steady state,
the fluids are stationary, with uniform temperature T0, and that
the number density for each species has the form no + n'a(x)
with K| « n0. State clearly all the approximations required
to obtain the result.
4.12 Repeat the derivation of the Debye potential when the elec¬
trons and ions have different temperatures and show that for this
case, with suitable approximations, we again have

4> = 4>c exp ( — kD’R)

but with

so that, for Te » Ti} the Debye length is determined by the ion

temperature.
4.13 Prove that if the Debye potential is assumed to hold for all R,
the total charge enclosed in a sphere of radius Ri about Q is
Q(1 + kD'R\) exp ( — ko’Rx). Hint: It is not necessary to use
fe and fi in this calculation.

4.9 PLASMA SHEATHS

Physical basis for the sheath

In the preceding section we analyzed the potential due to a point
charge located in a plasma, and we have seen that, in equilibrium, the
point charge attracts enough oppositely charged particles (and repels
enough particles with the same charge) so that the potential due to the
charge is reduced by the factor exp ( — kD'R) from the value it would
 4.9 PLASMA SHEATHS 211

have in a vacuum. In this section we shall discuss a closely related

phenomenon, the plasma sheath that arises in a plasma near a wall.
Although a truly satisfactory mathematical treatment of this
effect cannot be presented here, the underlying factors responsible
for a plasma sheath are fairly simple to understand. First, charged
particles hitting a surface are for the most part lost to the plasma.
(Electrons either recombine there, or if the surface is a metal, they
can enter into the conduction band. Ions, on the other hand, gen¬
erally recombine at the surface and return to the plasma as neutrals.)
The actual number of particles hitting a wall per unit time is easy to
calculate (provided we know the velocity distribution function near the
wall). For example, if we consider a plasma bounded on the right
by a surface parallel to the yz plane, as shown in Fig. 4.7, then, to
start, we note that the wall is hit only by particles moving to the right
(vx > 0). For these particles with vx > 0, we note further that the
ones that hit the surface element dS during a short time interval At
with some specific velocity v are all initially located in the volume
element with height vx At and base dS (and with sides parallel to v)
(note Fig. 4.7). From our definition of/ the number of particles with
velocity v in that volume element is just/^* At dS d3v. Hence, Ta, the
total number of particles of species a that hit the surface per unit time
and per unit area, is given by

ra = ffa(y)vx dh vx > 0 (4.9.1)

where the limits of integration are from — oo to + °° for vy and vz

and from 0 to <» for vx.
To obtain a typical value for Ta, note that for a maxwellian
velocity distribution, where

(4.9.2)

Fig. 4-7 Volume element containing,

initially, those particles with velocity v
that strike the surface element dS during
time At.
212 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

we find

na rx r°° r°
ra = vxe~v*laa2 dvx dvy dvz
7r3/2a«3 J-« J -» yo

na r« (4.9.3)
vxe~Vi,,aa2 =
7r1/2aa jo

From this calculation it is evident that if initially the plasma near the
wall has equal number densities for the electrons and ions, then I\>
far exceeds I\ as a result of the large difference between (Te/me)1/2 and
(Ti/m,i)1/2 in most plasmas. For this reason, a wall in contact with a
plasma rapidly accumulates a negative charge.
The potential associated with this negative charge provides an
attractive force for the ions and a repulsive force for the electrons in
the plasma, so that I\ increases and 1% decreases, until eventually an
equilibrium is established where I\ and re are equal and the negative
potential at the wall has a fixed value.

Formulation of the problem

To summarize, we have seen in a qualitative way that a surface in

contact with a plasma tends to develop a negative potential large
enough in magnitude to equalize the rate at which electrons and ions
strike the surface. This process depends strongly on the particular
geometry in question; hence one cannot write down a general result
valid for all cases, nor, for that matter, can we analyze any single case
completely. However, we can at least set up the problem mathe¬
matically and obtain some approximate results. To start, we shall
assume that the wall bounding the plasma is a plane surface infinite
in the y and z directions and passing through the point x = xq. In
addition we shall assume that because of this symmetry all the vari¬
ables in the problem are independent of y and z. Finally, we shall
restrict our attention to the steady-state problem when the sheath
has already been established.
As in the Debye potential problem, the set of equations to be
solved are the BY equations for the electron and ion velocity dis¬
tribution functions and Poisson’s equation. Because of our assump¬
tion that there is no dependence on y, z, or t, the BV equations reduce
(with E = — V4>) to

„ V± . A df? =o (4.9.4)
1 dx me dx dvx
 4.9 PLASMA SHEATHS 213

and

V*J-
ox - —TTY1
rrii ox ovx = 0 (4-9.5)

In addition, since the charge density in the plasma is just that due to
the electrons and ions, Poisson’s equation has the form

V • E = - (4.9.6)

or

d24>
— (ne — rii) ~ [ (fe ~ fi) d3V (4.9.7)
dx2 to to J

Equations (4.9.4), (4.9.5), and (4.9.7) provide three equations

for the three functions fe, fi, and <!>. Hence, in principle at least, one
can solve the problem in terms of some set of boundary conditions.
One boundary condition is that in equilibrium there is no charge
buildup at the wall (x = x0), so that

re(x0) = Ti(x0) (4.9.8)

A second boundary condition might be obtained by letting the point

x = — oo lie well in the interior of the plasma, where the electron and
ion number densities equal some common value n0 and the potential
<t> is a constant (which we shall set equal to zero). Unfortunately, a
straightforward approach such as this is too difficult to carry out by
simple analytic techniques. Hence, we shall restrict our discussion
to two very crude approaches to the problem—one (which is very
similar to the analysis used in the Debye potential problem) to obtain
an estimate of the potential that develops at the wall, and the second
(which is based on the moment equations) to obtain a good estimate
of the potential and the charged-particle number densities inside the
sheath.

Estimation of 4* at the wall

We showed in Eqs. (4.8.6) and (4.8.7) that one solution to Eqs. (4.9.4)
and (4.9.5) is given by

n0 2kT
aa2 (4.9.9)
U =
Tr3l2aa3 ,T)
=
ma
214 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

where the constant in front has been set equal to no so that the electron
and ion number densities both equal n0 in the interior of the plasma,
where <t> = 0, and where we have also assumed, for simplicity, that
Te = Ti. We should note at once that several objections to using
Eq. (4.9.9) can be raised. First, for this distribution function the
probability of particles’ moving toward or away from the absorbing
wall is equal; that is, fa(vx) = fa{-vx). However, for the ions in
particular it is evident that near the wall almost all particles must
be moving to the right (since no charged particles are emitted by the
wall). Similarly, the mean velocity, as calculated from Eq. (4.9.9),
is zero. Yet, again we expect that there should be a net drift of both
electrons and ions toward the wall in order to replenish the charged
particles that are lost there.
Because of these inadequacies we shall not attempt to use Eq.
(4.9.9) for fe and /, in Eq. (4.9.7) and thereby obtain a differential
equation for 4> alone.f (This is left to Prob. 4.14.) However, we
shall use this distribution to obtain an approximate value for the
potential on the wall. First, from Eqs. (4.9.1) and (4.9.3) we have
[using Eq. (4.9.9) for/e and fi]

If we apply this result to the boundary condition at a;0 [Eq. (4.9.8)],

we have [after canceling the common factor no(nT/2ir) 1/2 and letting
4>(x0) = $„]

(4.9.11)

or

(4.9.12)

f The error involved in such an analysis is often not so bad as one might fear,
since Eq. (4.9.7) involves the integral of fe and/,- over all v, which is not too sensitive
to the exact functional form of the distribution functions.
 4.9 PLASMA SHEATHS 215

Taking the logarithm of each side and solving for <l>w, we find

mi
$W (4.9.13)
me

This result is in qualitative agreement (for the case when Te = Ti)

with the wall potential calculated by other means, f despite the obvious
limitations to the assumed distribution functions.
From Eq. (4.9.13) we note that the ratio of the magnitude of the
potential energy |c4>| to the thermal energy kT for a particle near the
wall is given by

kM = ! ln (4.9.14)
kT t me

which is about 2 for protons and 3 for heavier ions. Thus, the wall
potential is on the order of the thermal energy (divided by e) of the
particles in the plasma.

Description of sheath structure using moment equations

Since we have seen that the solution to the BV equations given in

Eq. (4.9.9) is not really justified for the sheath problem, we shall
attempt now to analyze the problem using the mass and momentum
equations for the plasma [Eqs. (4.2.11) and (4.2.16)]. If we assume
again that there is no dependence on y, z, or t, that viscous and colli-
sional effects are negligible, that the temperature in the plasma is
uniform, and that u lies in the x direction, these equations reduce to

TlaXla 9 (4.9.15)
ax

f A common argument, for example, is to note that for 4> negative any positive
ion that passes through the edge of the sheath will continue and hit the wall. If
the velocity distribution at the sheath were a maxwellian with no drift, the flux
of ions would be [from Eq. (4.9.3)]

On the other hand, the electrons with insufficient velocity are repelled by the
potential and do not hit the wall; thus the number that do hit is still assumed to be
given by the flux estimate in Eq. (4.9.10), Te = n0(i<T/2meir)'12 exp (e4>u,/K/).
Such a calculation (because of the reduction in r») leads to a value for <t>„, which is
twice the value in Eq. (4.9.13). However, this is probably a lower bound for I\,
since the plasma at the edge of the sheath must have a net drift velocity toward
the wall, and for a drifting maxwellian I\ is easily found to be larger than I
Hence any modification to include this drift velocity would move the value for
|<l>u,| closer toward our original estimate (see, for example, Prob. 4.15).
216 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

and

dua dp a
Paf\^x) a
paUa~dx dx
d$> „ dna (4.9.16)
— naqa — — k1
dx dx

If we write out Eq. (4.9.16) explicitly for the ions and electrons,
we have

dui kT drii
— rriiUi
6 dx dx 7ii dx

due kT dne
= meue
dx
+ ne dx
(4.9.17)

Although it is possible to solve Eq. (4.9.15) and (4.9.17) for n* and ne

in terms of 4> with no approximation, f the customary procedure is
to simplify the analysis by neglecting the two underlined terms in
Eq. (4.9.17). This approximation will be investigated more carefully
later. However we may note here that, from Eq. (4.9.15),

dna
ua (4.9.18)
dx

Hence the ratio of the magnitudes of the two terms on the right side
of either part of Eq. (4.9.17) is

kT dna
na dx kT
(4.9.19)
dua maua2
'W^'OL^JOL j
dx

Thus the neglect of the two underlined terms in Eq. (4.9.17) requires
that the ratio of the thermal energy to the kinetic energy due to the
flow of plasma be large for the electrons and small for the ions, so that

meue2 « kT « MiUi2 (4.9.20)

We shall see later that these conditions appear to be satisfied in the

sheath.

t The analytic solution, however, gives <t> in terms of na, and one must use
numerical techniques to invert the equation and find na(<t>).
 4.9 PLASMA SHEATHS 217

With the omission of the underlined terms in Eq. (4.9.17), one

can easily obtain simple expressions for ne and rii in terms of 4>. If we
integrate the second part of that equation, we have

e$(x) = kT In ne(x) -f const (4.9.21)

Hence, using the condition that ne = no when 4> = 0, we find

ne = n0 exp (4.9.22)

Note that this result is also what we should have obtained using Eq.
(4.9.9) for fe. (This is not surprising since the drift velocity is neg¬
lected in both cases.)
To find n, we integrate both the first part of Eq. (4.9.17) and
Eq. (4.9.15) to find

e4>(x) + ±m,iUi2(x) = Ci rii(x)ui(x) = C 2 (4.9.23)

Applying the boundary condition that at x = — <x>, 4> = 0, Ui = u0,

and rii = n0, we find

Ci = m-iUo2 C2 = n0Uo (4.9.24)

so that, after eliminating ui} we have

(4.9.25)

This result, because of the importance of the drift velocity, is markedly

different from the result, n, = no exp ( — e<&/nT), that we should have
obtained from Eq. (4.9.9), since, for 4> < 0 (as it is in the sheath), we
now find n; decreases slowly as we approach the wall, rather than
increasing (see Fig. 4.8). I his is due to the fact that the negative
potential causes Ui to increase as the ions approach the wall, while
the flux riiUi stays constant.

The equation for 4> and the Bohm criterion

To complete the problem we now substitute our expressions for ne and
m [Eqs. (4.9.22) and (4.9.25)] into Poisson’s equation (4.9.7) to have

(4.9.26)
218 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

Wall

Fig. 4.8 Plot of ion and electron number densities and of the potential in the
sheath region between a plasma and a wall (assuming = — 2kT/e).

Recall now that |e<t>| ranges from zero in the plasma to a value of order
kT at the wall and that [from Eq. (4.9.20)] we have already assumed
that miUo2 is larger than kT. Hence, if we restrict attention to the
region near the plasma edge of the sheath, we may assume that |e$| is
small compared to both kT and FmiUo2, and we can expand the terms
on the right side of Eq. (4.9.26) to find

d24> ^ 4> (4.9.27)

dx2 X2

where Ad = (e0KT/n0e2)1/2 is again the Debye length.

Since we have already assumed that kT <5C mpio2 [see Eq. (4.9.20)],
it follows that X is real and approximately equal to Ad- If we solve
Eq. (4.9.27) and apply the boundary condition that $(—°°) = 0,
we find

$ = Aexlx (4.9.28)

so that |<t>| grows approximately as exp (x/Ad) as we move from the

plasma into the sheath.
Note that if kT were greater than miU02, X would be imaginary,
and 4> would be an oscillating function. Hence, the condition, kT <
miUo2, is known as the Bohm criterion for the formation of a plasma
sheath, t

f See I). Bohm, in A. Guthrie and R. K. Wakerling (eds.), “The Characteristics

of Electrical Discharge in Magnetic Fields,” chap. 3, McGraw-Hill Book Company,
New York, 1949.
 4.9 PLASMA SHEATHS 219

From the moment method there is no obvious way to determine <t>

at the wall. A number of approximate schemes have been suggested,
but all give results which (for Te = Tt) are in essential agreement with
the simple estimate given in Eq. (4.9.13). In addition, there is no
consistent way to calculate u0, the drift velocity at x = — °o toward
the wall. We can, however, estimate u0 by noting that, from Eq.
(4.9.15), the particle flux naua must be constant. Hence, we can
equate UqUo (the flux at x = — oo) and T (the flux at the wall). From
our moment equations alone we have no estimate of T, however.
Therefore, as an approximate value we use the expression calculated
inEq. (4.9.10), to find

(4.9.29)

[Recall that from Eq. (4.9.8) the expression on the right has the same
value for electrons and for ions.]
With this expression for u0 we can easily verify that the Bohm
criterion is satisfied, since we have (for e$w ~ —2kT)

(4.9.30)

Hence our assumption that |4>| grows exponentially rather than oscil¬
lates has some justification.

Summary of the sheath conditions

In this section we have attempted to analyze the conditions inside a
plasma sheath using Poisson’s equation and the mass and momentum
equations for the ions and the electrons. In addition, we have used
some crude kinetic-theory approximations to estimate 4v, and u0 (since
these cannot be obtained using the moment equations). Before
attempting to assess the validity of all the approximations used in the
calculation, let us summarize the results shown in Fig. 4.8. I'irst., we
have found that the potential decreases from 0 at x = - °° to
<t>w = —2xT/e (for a plasma of electrons and protons) at x0. The
decrease in 4? has been shown to be proportional to exp (x/X) (with
X slightly larger than the Debye length) well away from the wall.
However, as shown in Prob. 4.16, the decay in is not too different
from an exponential even close to the wall.
The ion and electron number densities, which are given by Eqs.
(4.9.22) and (4.9.25), both fall off as 4> decreases in the sheath, but
with ne decreasing at a much more rapid rate than rq, so that at the
220 THE BOLTZMANN-VLASOV EQUATION FOR A PLASMA

wall for a gas of electrons and protons

rii(xo) = 0.8n0 ne(xo) == 0.14n0 (4.9.31)

It is this difference between n; and ne which is the most prominent

distinction between the sheath and the plasma itself, since in the
plasma the charge separation is always very small.
Finally, because of the absorption of charged particles at the wall,
there is a net drift of the plasma toward the wall at a speed u0, which
(for electrons and protons) from Eq. (4.9.30) is about three times the
ion thermal velocity (KT/nq)1/2.

Assessment of the approximations

In working with the conservation equations two crucial approxima¬
tions were used in order to obtain simple expressions for ne and nl
in terms of <f>. As shown in Eq. (4.9.20), these approximations require
that kT be large compared to meue2 and small compared to nqu;2. In
order to check these approximations we note first that, from the mass-
conservation equation, the product na(x)ua(x) must equal, for all x,
the value n0u0. Hence, since the minimum value for ne is n0 exp
(eQw/nT) while the maximum value for rq is no, we have

ue =
UqUo
< u0 exp
/ (4.9.32)
ne V -*t)

and

n0u0 ^
Ui = - > Uq (4.9.33)
Ui

If we use these inequalities together with Eq. (4.9.29) for u0,

we find

kT kT 2e$w
> 2ir < 2tt exp 0.11 (4.9.34)
meue2 m%Ui2 kT

so that our neglect of the underlined terms in the momentum-conserva¬

tion equations (4.9.17) is well justified for virtually all values of x.
The most unsatisfactory part of the calculation, therefore, lies in
the crude methods used to compute the potential at the wall and Uq.
As we have indicated, the value for for the case when Te = 7\,
seems to be very insensitive to the methods used to make the calcula¬
tion. However, this is not the case for the common situation in the
 4.9 PLASMA SHEATHS 221

laboratory, when Te and Ti are markedly different. Hence, the quan¬

titative aspects of this discussion of plasma sheaths should not be
taken too seriously although the qualitative picture of the sheath is
certainly correct.

Problems

4.14Show that if we use Eq. (4.9.9) for/e and/,, the differential equa¬
tion for <t> reduces (in the region when e4> <3C kT) to

d24> _ <f>

From this equation, and the conditions that <t> = 0 at a; = — °°

and <h(x0) = — («T/4e) In (rru/me), construct a rough plot of 4>,
m, and ne as a function of x.
4.15 Prove that if the ions at the edge of the sheath have a maxwellian
distribution with drift velocity u0ax [note Eq. (4.3.14)], then

'Uo
Ti n0 [exp ( — r2) + 7r1/2r(l + erf r)] r =
di

Also prove that this function increases monotonically with

increasing u0.
4.16 Show that, to a good approximation, near the wall

d2<S>l n0e n0ex2

or 4>i T- CiX -f- C2
dx2 eo 2t0

Use this solution together with the prior result, well away from the
wall,

4> = 4>2 = Aex/Xc

to obtain a good estimate of 4> for all x. To simplify the analysis

let x0 be at x = \D; let <t> = 4>2 for x < 0; and let <t> = <t>i for
x > 0; and determine A, C\, and C2 by the conditions 4>i(Ad)
= 4>,t, - -2xT/e, <U(0) = <1>2(0), (d$i/dx)o = (d<S>2/dx)o to find

3 kT
3 + ^+^ 4*2 —
4e
+ Ad + Ad2

(Note that 4>2 agrees closely with 4>i, even in the region from 0
to Ad.)
chapter five

THE BOLTZMANN
EQUATION FOR
A PLASMA
chapter five

5.1 TWO-BODY COLLISIONS AND THE BOLTZMANN COLLISION TERM

Center of mass and relative velocities

In the preceding chapter we saw that many of the physically important
parameters of a plasma can be calculated if one knows the velocity
distribution function for each species of particles in the plasma.
We then derived the Boltzmann equation

(5.1.1)
dt + »< ir
% dXi + “.ir
dVi =

for the distribution function but we did not attempt to derive the
collision term (dfa/dt)c- Instead, we either neglected collisions and set
(dfa/dt)c = 0, or we replaced (dfa/dt)c by the simple Krook collision

model

= - *(/« - fao) (5.1.2)

In this chapter we shall take a much more accurate account of colli-

sional effects by using the collision term originally proposed by Boltz¬
mann, which attempts to take full account of the individual two-body
elastic collisions that occur in the gas. Hence, as our first step, we
must review the dynamics of a two-body elastic collision.
Let m and mi be the masses of the colliding particles, and let
v, \i and v', v( be the velocities of m and mi before and after the
collision. The center-of-mass velocity (before the collision) is defined as

mv + miVi
Co mo = m + mi (5.1.3)
m0

while the relative velocity g (before the collision) is

g = Vi — v (5.1.4)
 5.1 TWO-BODY COLLISIONS AND THE BOLTZMANN COLLISION TERM 225

Because of the momentum- and energy-conservation laws, we can show

that after an elastic collision the center of mass velocity is unchanged,
while the relative velocity changes in direction but not in magnitude.
To prove these statements we note first that, from conservation
of momentum, we have

m0 c0 = mv + mivi = mV + miv( = m0 c'0 (5.1.5)

so that Cq, the center-of-mass velocity after the collision, clearly equals
c0. Next, if we let g be the reduced mass

mm i (5.1.6)
m0

then by direct calculation using Eqs. (5.1.3) and (5.1.4)

-i(m0c01 2 + ng2) = %(mv2 + mivx2) (5.1.7)

Hence, by conservation of kinetic energy the expression on the right is

conserved in a collision so that

i(m0c02 + ng2) = i(m0c'02 + ng'2) (5.1.8)

Since we have already shown that Cq = c0, this reduces to

g2 = g> 2 (5.1.9)

so that g', the magnitude of the relative velocity after the collision,
exactly equals g.
In view of the conservation of Co and g it is useful to describe
a collision in terms of a coordinate system where g lies in the 3 direction,
as shown in Fig. 5.1. (We shall refer to this as the gz coordinate sys¬
tem.) In this figure the angle x between g and g; is called the scattering
angle, while e defines the collision plane (the plane formed by g and g ).
The scattering angle x (or, equivalently, the vector k) is the only
quantity that depends on the details of the collision. I he technique
for evaluating x is detailed in Sec. A.2 of the Appendix. Here it is
sufficient to note that for force laws involving only the distance
between the particles x depends on three factors:

1. The interparticle force law between the colliding particles

2. The magnitude of the relative velocity
226 THE BOLTZMANN EQUATION FOR A PLASMA

3. The value of 6, the impact parameter (the minimum distance

between the centers of the colliding molecules that one would
have if there were no collision)

From Fig. 5.1 it is evident that in the gz coordinate system

gx = gv = 0 gt = g
(5.1.10)
g'x = g sin x cos e g'y = g sin x sin e g'z = g cos x

Relationships between initial and final velocities

We have seen how a collision affects the center-of-mass and relative
velocities. Now we want to derive the relationship between the final
velocities v' and v( after the collision and the initial velocities v and Vi.
First, by inverting Eqs. (5.1.3) and (5.1.4), which give Co and g in
terms of v and vi, we find

(5.1.11)

so that, for the final velocities we have, analogously,

(5.1.12)

Now we have shown that c'0 = co, and from Fig. 5.1 we can set

g' = g - ak (5.1.13)

Fig. 5.1 Relationship between g and g' in the gz coordinate system.

 5.1 TWO-BODY COLLISIONS AND THE BOLTZMANN COLLISION TERM 227

where k is a unit vector pointing from g' to g and

= g sin ix = Q cos £(*■ - x) = g • k (5.1.14)

Therefore, in terms of c0 and g (which depend only on v and we

have

v' = c0 — — (g — ak) vi = c0 + —- (g — ak) (5.1.15)

This result thus allows us to compute the final velocities if we know

the initial velocities and the scattering angle %•
For some cases we may know the final velocities and the scattering
angle but not the initial velocities. To find v and Vi we must first
obtain an expression for g in terms of g'. To do this we let
g' • k = \a! [in direct analogy with Eq. (5.1.14)]. Now, if we take
the dot product of k with each term in Eq. (5.1.13) and also use Eq.
(5.1.14), we obtain the useful identity

g' . k = g • k — a = — -ia or a' = —a (5.1.16)

Hence, from Eq. (5.1.13) we find that g is given by

g = g' + ak = g' - a'k (5.1.17)

If we use this expression for g in Eq. (5.1.11) and recall that Co = c0,
we find that

This result, which is entirely equivalent to merely mterchanging

primed and unprimed quantities in Eq. (5.1.15), enables us to com¬
pute v and vj if we know the final velocities and the scattering angle.

Inverse collisions

The collision process we have just been analyzing is often called a

direct collision and may be represented by the diagram

(5.1.19)
m(v) + rai(vi) —> m(v') +
b
228 THE BOLTZMANN EQUATION FOR A PLASMA

which shows that, for the given impact parameter b, the velocities of
the particles change from v and vi to and v'j. We now want to
prove that the inverse of Eq. (5.1.19) also holds, so that if the initial
velocities in the collision are v' and v( and the impact parameter is
still b, then the final velocities are v and vi, provided that the initial
positions of the particles are selected in such a way as to leave k
unchanged.
For this inverse collision the scattering angle is again x, since the
impact parameter, interparticle force law, and relative speed are all
the same as for a direct collision. Hence, depending on how we choose
the initial positions of the two particles, the vector g', which is now the
initial relative velocity, can swing either back to g or down to an angle
of 2x relative to g. The former case, which arises if we arrange the
particles initially so that k is unchanged, is called an inverse collision.
This is illustrated in Fig. 5.2, where we plot the path of particle mi
relative to particle m (located at the origin 0 in each plot) for a direct
and for an inverse collision.
If we denote the velocities after the inverse collision by a double
prime, we have

g" = g (5.1.20)

Fig. 5.2 Illustration of the path of particle rat relative to par¬

ticle m (assumed to be at the origin) for a direct and inverse
collision. Note that at any time the relative velocity is
tangent to this path, so that the scattering angle x is the angle
between the asymptotes of the path.
 5.1 TWO-BODY COLLISIONS AND THE BOLTZMANN COLLISION TERM 229

Since the center-of-mass velocity is always conserved, we also have

c'o = c0. Hence, from Eq. (5.1.11), we have directly that

v" = v and v'/ = Vi (5.1.21)

so that the inverse collision does indeed have the form m(v') +
mi(v[) —> m(v) + mi(vi).
b
To complete this discussion of two-body collisions we shall prove
the useful identity

d3v' d3v[ — d3v d3Vi (5.1.22)

From the theory of Jacobians, recall that

d3v' d3v[ = \J\ d3v d3Vi (5.1.23)

where J is the determinant

dv'x dv'x dv'x

dvx 6Vy dvu
K K . ,
(5.1.24)
J = dvx dVy

dv'u dv’u
dvx dvu

If one uses Eq. (5.1.15) to calculate all the partial derivatives, one
finds, after a tedious but straightforward calculation, that J = — 1,
so that the identity is proved. However, the result \J\ = 1 can also
be inferred from Eqs. (5.1.15) and (5.1.18) without any calculation.
To see this recall that we can also set

d3v d3vi = \J'\ d3v' d3v[ = \J\ \J'\ d3v d3v 1 (5.1.25)

where J' is identical to ./ but with all the primed and unprimed quanti¬
ties interchanged. Now, from the symmetry of Eqs. (5.1.15) and
(5.1.18) one can easily prove that a partial derivative of any component
of v or vi with respect to any component of \' or Vj equals the same
partial derivative with primes and unprimes interchanged. [I'or
example, dvx/dv'ly = dv'Jdvly = 2(mx/m,)kykx.] Thus each element in
230 THE BOLTZMANN EQUATION FOR A PLASMA

the matrix J' equals the corresponding element in J, and therefore

J' — J. Applying this result to Eq. (5.1.25), we have

|J|2 =1 or \J\ = 1 (5.1.26)

which is the desired result.

The Boltzmann collision term

The expression for (dfa/dt)c used by Boltzmann is

(W), = 2 J Wn - db * <*"». (5.1.27)

where/a = fa(x,\,t) is the velocity distribution function for particles

of type a and fa = fax,\i,t) is the velocity distribution for the par¬
ticles with which the a particles collide. (Hence 0 is summed over
all types of particles in the plasma.) The distribution functions
fa = /«(x,v',f) and fa = /s(x,v(,£) are just fa and fp evaluated at the
velocities after a collision.
The first part of this term represents the rate of increase of
/„(x,v,£) due to all the inverse collisions of the form

m«(v') + mp{\i) —> ma(v) + mp(x0 (5.1.28)

b

which create an a particle with velocity v, while the second part of

the collision term is the rate of decrease of fa(x,\,t) due to all the direct
collisions involving a particles with velocity v. A detailed derivation
of this term is given in Sec. A.3 of the Appendix, along with a discussion
of the assumptions used in the derivation. Here we shall note only
that the derivation is much more justifiable for a dilute gas of small
neutral molecules than for charged particles in a plasma or for dense
gases or liquids. Hence, we must use it with caution for problems
involving a plasma and not expect to obtain results with a very high
degree of accuracy.

5.2 THE H THEOREM

One of the most important features of the Boltzmann collision

term is that it tends to drive the distribution function irreversibly
 5.2 THE H THEOREM 231

toward an equilibrium state, where = fafpi- To demonstrate

this we can show that the functiont

H — // In / dh (5.2.1)

which is related to the negative of the entropy of the gas, decreases

monotonically with time until / reaches equilibrium. For simplicity
we shall prove this for a one-component (or “simple”) gas with no
spatial gradients and no external forces. (The extension to a mixture
is indicated in Prob. 5.1.) For this case we have

H = //(In / + 1) dh (5-2.2)

where, from Eqs. (5.1.1) and (5.1.27),

f= /(/'/( - ffi)gb db de dhy (5-2.3)

Using this expression for / in Eq. (5.2.2), we have

H = /(/7( - //0 (In / + l)gh db de dh dhy (5.2.4)

The remainder of the proof involves the use of two identities.

First, we note that when there is a double integration, one can always
interchange the dummy variables of integration without changing the
result (just as earlier we noted that one can interchange the dummy
summation indices in a double summation). Hence for any function
F(v,vi) we have

/F(v,vO dh dhi = i/[F(v,Vl) + F(vi,v)] dh dhl (5.2.5)

Next, we also note that if we make a transformation in which v and Vi

are replaced by v' and vj, then from our analysis of the inverse collision
problem v' and vj will be transformed back into v and vx. Hence,
since we have also shown [in Eq. (5.1.22)] that dh' dh[ = dh dhu
we have for any function F(v,vi;v',v[)

JF(v,vi;v',vj) dh dhi = iJ[F(v,V!;v',vj)

+ F(v/,vj;v,vx)] dh dhi (5.2.6)

f Note that for problems with spatial gradients, Htot = Sf In / d3v dhr, so
that the H used here is Htot Per un't volume.
232 THE BOLTZMANN EQUATION FOR A PLASMA

If we apply these identities in sequence to Eq. (5.2.4), we find

H = l/Cf'f[ - ffi) (In ffy + 2)gb db de d3v dh,

= iKf'fi - ffi) ln Sr, db de dh (5.2.7)

J Ji

The integrand here is negative semidefinite, as can be seen by noting

that g, b, and all the distribution functions are positive, while

(ffi ~ ffi) and (ln//i - ln/'/O (5.2.8)

always have opposite signs (unless both are zero). Hence their product
is negative semidefinite, and we have

H < 0 (5.2.9)

where the equal sign arises only when

fh = f'fi (5-2.10)

This result shows that when one uses the Boltzmann collision term, H
always decreases (or, conversely, the entropy always increases) with
time until ffi =; /'/(. When this occurs, there is no further change
in the system, since, from Eq. (5.2.3), / = 0.

Problems

5.1 Prove that for a mixture (where H = V/«ln/a), H always

a

decreases until f'af'pi = fafpi for all a and /3-

5.2 Calculate the difference /'/£i — fafpi when both fa and f$ are
maxwellian velocity distributions of the form

n7e_cW<V Cy v u')'
= T3'2ayz , 2 KTy
my

and y = a or j3. Show that this difference goes to zero if and only
if ua = up and Ta — T0.
5.3 (a) Show that for a simple gas, if / = C(x,t)e~0<-I’t)v2, then /
satisfies the Boltzmann equation if and only if /3 is constant
and C(x,t) = Cie~2U(z)fflm, where U(x) is the potential of any
 5.2 THE H THEOREM 233

external force (ma. = —VU), and Ci is a constant. Hint:

You need only substitute this solution into the Boltzmann
equation and show that one obtains four conditions:
C = & = dfi/dxi = 0 and In C + 2 Ufi/m = const.
(b) Express C and /3 in terms of n and T and show that T is
constant and n(x) is proportional to exp (—U/kT), as
assumed in the Debye potential problem.

The reversibility paradox

The H theorem predicts that the entropy always increases with time.
While this agrees with the second law of thermodynamics, the result is
nevertheless not compatible with the fundamental laws of mechanics,
which are reversible. Thus in a system of N interacting particles, if
we reverse the velocity of each particle, the laws of mechanics require
each particle in the system to retrace its former path. To see how the
H theorem leads to a paradox, consider the behavior of the following
two systems of N particles numbered from i = 1 to N. Let system I
be a nonequilibrium gas where the initial conditions for each particle
are given by x^O), v/(0) and where the initial value for H is H (0).
[Assume also that the Velocity distribution at t = 0 satisfies the con¬
dition /r(v) = /r( — v).] For system II, let the initial conditions be
defined by

In other words, each particle in the second system starts off initially
with the position and speed of the corresponding particle in system I
at time r but with the velocity reversed. Hence, from the laws of
mechanics,

I!
H H

System I System II

T T

Fig. 5.3 Plot of H versus time for two systems.

234 THE BOLTZMANN EQUATION FOR A PLASMA

Now according to the H theorem, we must have

H\t) < H\0) (5.2.13)

since system I started from a nonequilibrium state; but clearly this

means that Hu(t) > Hn(0), so that, as in Fig. 5.3, Hn increases with
time in violation of the H theorem.
There are truly two points involved in this paradox. First, since
the fundamental equations of mechanics are reversible, one may ask
how the Boltzmann equation ever becomes irreversible. The answer
to this is not too difficult. As noted in Sec. A.3 of the Appendix, a
number of assumptions are made in deriving the Boltzmann collision
term, and one of these, which involves the neglect of correlation effects
in calculating the probability that two particles will collide, leads to
the irreversibility in the collision term. Thus the BBGKA equations,
which we shall use in Chap. 6 are still reversible and do not satisfy
the H theorem if no approximations are made.
The second part of the paradox is to ask why the second law of
thermodynamics (or the H theorem) agrees with nearly all macroscopic
experiments, since we have seen that one can think up systems which
would violate these laws. The resolution of this part of the paradox
comes from the theory of ensembles in statistical mechanics. While
we cannot cover this topic in detail here, we may recall briefly that
there are three levels of description for a gas of N particles:

1. The microscopic description in terms of the position and velocity

of each particle
2. The statistical description in terms of/(x,v)
3. The macroscopic description in terms of a small number of
macroscopic parameters

In general there are many possible distribution functions which might

describe a single macroscopic state, just as there are many microscopic
states which lead to the same distribution function. Therefore, in
statistical mechanics, for a given macroscopic system, each individual
microscopic state (that has the proper set of macroscopic parameters)
is declared to be equiprobable. Hence the probability for having a
particular distribution function /(x,v) is proportional to the number
of microstates which lead to that particular /, subject, of course, to
the constraint that we consider only distribution functions that lead
to the proper set of given macroscopic variables. In statistical
mechanics the entropy is shown to be proportional to this probability
for having a particular distribution; hence, the state having maximum
entropy (or minimum H) is also the most probable one (subject to the
 5.3 MAXIMUM ENTROPY AND THE MAXWELLIAN DISTRIBUTION FUNCTION 235

constraints imposed). Thus the H theorem states that a system moves

with time toward its most probable state. The theorem obviously
cannot apply to every single experiment, since we have seen, from the
laws of mechanics, that we can always conceive a particular system
which violates the theorem for at least a short time. However, if one
repeats, an experiment many times, each time starting with the same
initial macroscopic state, then, for the whole set of data, the H theorem
will almost certainly hold (unless there are some special circumstances
such as very small numbers of particles).
This aspect of the paradox can perhaps be clarified by a simple
example. Consider a gas of N particles enclosed in a volume V, and
suppose that initially all the particles are contained in one corner by a
membrane. Suppose we then break the membrane and then measure
the number density of particles at some time £o later. If the time to
is reasonably long, our intuition suggests that the particles will have
distributed themselves uniformly over the entire volume. (This is
shown in Prob. 5.4 to be the state of minimum H for this gas.) How¬
ever, in principle at least, the particles might have been arranged
initially in such a way that at time t0 they all have moved to some small
subvolume even' smaller than their initial container. For this case
one would find that H had increased. However, unless the total num¬
ber of particles is exceedingly small, the chance of this last experi¬
ment’s actually occurring (which could be calculated) is very slim
so slim indeed that one may never observe such an event, even in a
long series of experiments.

5.3 MAXIMUM ENTROPY AND THE MAXWELLIAN DISTRIBUTION FUNCTION

Derivation of the equilibrium distribution function

We have seen that the Boltzmann collision term tends to drive the
distribution function toward the equilibrium value where H is a mini¬
mum and the entropy is a maximum. In addition, in Prob. 5.2 it is
shown that if each gas in the mixture has a maxwellian velocity dis¬
tribution with the same temperature and drift velocity, then the gas
is in equilibrium. One can further show by a number of methods that
the maxwellian velocity distribution is the only distribution for which
(dfa/dt)c — 0. For example, we have seen that at equilibrium II is a
minimum; hence, for a one-component gas

hH = 5// In / d3v = 0 at equilibrium (5.3.1)

where the symbol 5 before any quantity denotes a variation in the

quantity due to a small change in the distribution function /. In
236 THE BOLTZMANN EQUATION FOR A PLASMA

varying /, however, one cannot violate mass, momentum, or energy

conservation for the gas as a whole. Hence for a uniform gas we must
keep the total mass, momentum, and energy densities fixed, so that

5(number density) = J bf d3v = 0 (5.3.2)

^(momentum density) = m/v bf d3v = 0 (5.3.3)

and

5(energy density) = \mfv2 bf d3v = 0 (5.3.4)

Equation (5.3.1) is a standard variational integral which can be

solved for / subject to the constraints shown in Eqs. (5.3.2) to (5.3.4)
by the method of Lagrange multipliers. However, to perform the
calculation most simply we must make some simple transformations.
First, we change the variable of integration to c = v — u in each
equation and formally carry out the variation indicated in Eq. (5.3.1),
to have

bH = f bf (In f + 1) d3c = 0 (5.3.5)

Next, we multiply Eq. (5.3.2) by the Lagrange multiplier X to have

Xf bf d3c = 0 (5.3.6)

From this result it is evident that Uif bf d3c also equals zero; hence,
if we multiply each component of Eq. (5.3.3) by a Lagrange multiplier
yi/m, we have

7iS(Ci + Ui) bf d3c = 0 (5.3.7)

or

7ifci bf d3c = 0 (5.3.8)

Finally, from the equation for the variation of the energy density
[Eq. (5.3.4)], we have, after multiplication by the Lagrange multiplier
2/3/m,

dj(c2 + 2aui + u2) bf d3c = 0 (5.3.9)

 5.3 MAXIMUM ENTROPY AND THE MAXWELLIAN DISTRIBUTION FUNCTION 237

However, since we have already shown in Eqs. (5.3.6) and (5.3.8) that
the second and third terms here integrate identically to zero, we have

/3/c2 5/ dzc = 0 (5.3.10)

To find / we now sum Eqs. (5.3.5), (5.3.6), (5.3.8), and (5.3.10)

to obtain

J 5/(ln / + 1 + X + tid + 13c2) d3c = 0 (5.3.11)

Since we have taken account of all the constraints, the variation in /,

5f, is now completely arbitrary. Hence this integral can equal zero if
and only if

Inf = -(1 + X + lid + dc2) (5.3.12)

or

j — (Je-y .ci-ffc* (5.3.13)

where C = exp (—1 — X). To evaluate the five constants in this

solution for /, we can use first the condition that (ck) = 0, for each
choice of k; hence

n(ck) = fckCe-y*'-^ d3c = 0 (5.3.14)

This condition implies that yk must be zero, as can be seen by noting

that, with one integration by parts (where in the formida

fUdV = UV| - fVdU

with no summation intended, we let U = e~ykCk and d 1 = cke &Ck dck)

we have

n(ck) = ~^C f e-y^ d*c = 0 (5.3.15)

Since the integrand here is positive definite, this means that yk must
be zero. (Alternatively, one might set C = 0 or d = 00, but either
of these assumptions makes the distribution function equal zero.)
The final two constants are obtained from the definitions of the
number density and temperature

3/2
d3c = C (5.3.16)
n
238 THE BOLTZMANN EQUATION FOR A PLASMA

and

:iv^-T = f fc2 dzc = C f c2e_(3c2 dzc

_dn _ 3C /tr\3/2
(5.3.17)
d(3 _ 2/3 \p)

Solving these two equations for & and C, one easily finds

m
/3 = or2 C = (5.3.18)
27r tr3/2a3

so that the equilibrium distribution reduces to the maxwellian

distribution

Tie c2 la2
(5.3.19)
7r3/2a3

Problem

5.4 (a) Calculate Htot = // In / dzv dzx for a gas with a maxwellian
distribution. Show that if we let S = — kHtot, then S satisfies
the thermodynamic relations

(S)„ = f and =I
with iV the total number of particles, V the volume, and
E — §NkT the total energy.
(b) Prove that if / = [w(x)/7r3/2a3] exp ( —i>2/a2) and a is assumed
constant, then f/tot is a minimum when w(x) is constant.
(Assume gas has N molecules in a fixed volume V.) Hint:
Use a Lagrange multiplier technique.

The maximum-entropy approach

From our discussion at the end of the last section it is evident that the
maxwellian velocity distribution, besides being the equilibrium solu¬
tion to the Boltzmann equation, is also the most probable distribution
one would have if the macroscopic state were specified by n, u, and T.
For nonequilibrium problems one can still perform a variational cal¬
culation of the type just shown, to find the most probable distribution
(subject to the constraints provided by a set of macroscopic param-
 5.3 MAXIMUM ENTROPY AND THE MAXWELLIAN DISTRIBUTION FUNCTION 239

eters describing the nonequilibrium state), even though the resulting

distribution function does not satisfy the equilibrium condition
f’afp'i = faffti for all a and /3.
Such a situation can be very simply illustrated by calculating the
most probable distribution function for a nonequilibrium gas mixture
when one knows the number density, mean velocity, and temperature
for each species. To solve this problem most easily, note that

H = lH Ha I fa In fa dh (5.3.20)
<a

Hence, if we can independently minimize each Ha, we shall also

minimize H.
When Ha is at its minimum, 5Ha = 0, so that, as in Eq. (5.3.5),

8Ha = J 5/a(In fa + 1) d3ca = 0 (5.3.21)

Next, since n«, u„, and Ta must all remain fixed as fa is varied, we can
set <5n«, 5(c„), and 5(3naxT«/m«) all equal to zero, to find

XaJ 5fa d3Ca = 0

Tat/ 5faCaid3Ca = 0 (5.3.22)
da/ 5/aCa2 dZCa = 0

Equations (5.3.21) and (5.3.22) are identical to the equations we solved

to obtain the maxwellian velocity distribution for a one-component
gas [Eqs. (5.3.5), (5.3.6), (5.3.8), and (5.3.10)]. Hence, they lead, in
identical fashion, to the solution

nae~Cal /o“2 2 uTa (5.3.23)

/. = 7r3'2a«3 ma

where each gas has a maxwellian velocity distribution but with its
own temperature and mean velocity. Note again that this is not an
equilibrium distribution (unless the temperatures and mean velocities
of all species are equal), since, as shown in Prob. 5.2, the collision teim
for such a mixture does not vanish. However, it is the most probable
distribution with these constraints.

Problem
5.5 Find the most probable value for / for a one-component non¬
equilibrium gas when the known macroscopic parameters aie the
240 THE BOLTZMANN EQUATION FOR A PLASMA

number density n, the mean velocity u, and the three kinetic

“temperatures” of the random motion in each direction
Ti = (mcfi/n).

5.4 LOWEST-ORDER TRANSPORT EQUATIONS FOR A GAS MIXTURE

The moment method of solving the Boltzmann equation

When we use the Boltzmann collision term, the full Boltzmann equa¬
tion has the form

(5.4.1)

Although it is difficult to work with this equation because of the com¬

plexity of the collision term, there are a number of perturbation methods
that can be used. The first technique we shall use is a simple form of
the moment method. Suppose we formally expand f(x,\,t) in a series
of the form

(5.4.2)
p

where/0(x,v,<) is some “appropriate” zero-order velocity distribution,

Mp is a complete set of functions in v, and p is a vector to indicate
that the summation is generally over more than one index.
While such an expansion is possible with any complete set of
functions Mv, in practice one aims to choose a set such that the op
become very small for all but the first few coefficients. If it is possi¬
ble to do this, then one can truncate the series (by setting all but the
first few ap = 0) to have the summation in Eq. (5.4.2) extend over a
finite number of terms, so that the approximate form for / involves
only a finite number of functions of x and t (the nonzero ap and any
functions of x and t in /0). Thereafter one can use the same number
of independent moment or transport equations of the kind developed
in Sec. 4.2 to obtain a closed set of equations to determine all the
functions introduced in the approximation.!
While this method is straightforward in principle, its application
in practice is fairly complicated. Therefore we shall begin with a

t In fact, since the truncated solution for / is generally not an exact solution
to the Boltzmann equation, the results obtained are not independent of the choice
of the transport equations to be used. While this problem has not been thoroughly
investigated, it appears that the best results are obtained when the least compli¬
cated Q’s are used to obtain transport equations.
 5.4 LOWEST-ORDER TRANSPORT EQUATIONS FOR A GAS MIXTURE 241

relatively simple example of the method, where we use just the first
term in Eq. (5.4.2) for each species of particles in the plasma. Although
this is an exceedingly crude level of approximation for/a, we shall see
that it leads to a full set of mass, momentum, and energy equations
(including collisional interaction terms) for each species—much like
the equations used in Chap. 3 but with the collision terms for momen¬
tum and energy transfer given explicitly (instead of in terms of some
empirical collision frequencies).
For this example we set

nae~Cai,aa'1 2 kTc
u = ca = v — u aj (5.4.3)
7r3/2aa3 ma

where na, u«, and Ta may all be functions of position and time. (Recall
that this would be our most probable form for fa, if we “know” values
for na, ua, and Ta.) For this case we have five functions of position
and time for each species (since ua is a vector); hence we can use the
five transport equations for the mass, momentum, and energy of each
species that are given in Eqs. (4.2.12), (4.2.16), and (4.2.29) but which
we rewrite here for convenience.

Pa + V • (paUa) - 0 (5.4.4)

Duak (5.4.5)
Pa(&k)a ^ Paik &Qak
pa~i)r
D
PaPa
—5/3
V • qa (5.4.6)
Dt

where

(5.4.7)

(5.4.8)

Simplification of the equations

These equations were initially derived with no assumption about the
form of/„. Hence they involve many parameters other than na, ua,
and Ta (or pa = naKTa). However, if we now use the maxwellian
form for/a assumed in Eq. (5.4.3), we can reduce all the other terms
to expressions involving just the densities, mean velocities, and tem¬
peratures (or pressures) in the mixture. Proceeding term by term,
242 THE BOLTZMANN EQUATION FOR A PLASMA

we first assume that the external forces are purely electromagnetic, so

that, as in Eq. (4.2.19),

(a)a = — (E + u„ X B) (5.4.9)
m

Next, recall that paik = pa(CaiCak), so that, using Eq. (5.4.3) for /«,

Paik — 'Ola J j a^ai^ak d Ca

= Pa . f e-Callaa'!cmcak dzca (5.4.10)

7r3/2a«3 J

This integral is zero when i ^ k, since the integrations over cai- and
Cak are of the form

J dx (5.4.11)

which is identically zero because the integrand is an odd function.

However, for i = k the integration over each direction other than cai
is of the form

e x2laa2 dx — irH2aa (5.4.12)

while the integration over cai is of the form

x,,a 0 o , aj dll aa3TT112

p jCL(X~op 2 pi'll* — _ _ — -
(5.4.13)
I- 2 daa 2

Hence, by direct calculation, we have, for this form of fa,

Pa^a ~ ^
Paik = ^ Pa^ik (5.4.14)

From this result, it immediately follows that

I aij = Paij Pa&ij 0 (o.4.1o)

so that the traceless pressure tensor, for this case, is zero. Similarly,
for the heat-flow vector, we have, by definition, qak = -ipa(ca2cak), so
 5.4 LOWEST-ORDER TRANSPORT EQUATIONS FOR A GAS MIXTURE 243

that with fa given by Eq. (5.4.3)

Qak — j Ca^Cakfa d3Ca

= 2^k? f d’e° 0
=
(5.4.16)

since the integral over cak again has an odd integrand. In summary,
we see that when fa is a maxwellian velocity distribution, the traceless
pressure tensor and heat-flow vectors are identically zero, so that the
transport equations given in Eqs. (5.4.4) to (5.4.6) reduce to

Pa ^ ‘ (paUa) — 0
(5.4.17)

DUak dpo (5.4.18)

Dt = naqa[Ek + (ua X B)fc] ~ jfk +

Z) (5.4.19)
PaPa~bli = fPa 5/3 &QaO
Dt

However, to complete the problem of reducing these equations to

terms involving only na (or pa), ua, and pa, there remains the non¬
trivial problem of evaluating the collision integrals bQak and bQao.

Evaluation of the integrals over collision coordinates

When we replace (dfa/dt)c by the Boltzmann collision term (5.1.27),
the general collision term bQ — fQ(dfa/dt)cd3v becomes

(5.4.20)
8Q = J / (J'J'pi - fMQgbdb dt dh dh'
e

Now we note again that in an integral over both v and Vi we can

interchange primed and unprimed quantities in the integrand without
altering the result. Hence,

ff'MQgb db de dh d3tq = ffafpiQ'gb db de d3v d3vi (5.4.21)

so that an alternate form for bQ (which we shall actually use) is

bQ = ^ / fafmiQ’ - Q)db db de dh d*Vl (5.4.22)

p

As our first step in evaluating bQak and bQao we integrate Q' — Q

over e. To perform this integration we recall that in the gz coordinate
244 THE BOLTZMANN EQUATION FOR A PLASMA

system [Eq. (5.1.10)] we have (see Fig. 5.4)

g = ga.3 g' = g[sin x(ai cos e + a2 sin e) + a3 cos x] (5.4.23)

where ai, a2, and a3 are unit vectors in the x, y, and z directions in the
gz system. Hence, to find the fcth component of g and g' in some
external coordinate system we take the dot product with a* to have

Qk = <7«8*
(5.4.24)
9k = {/[sin x{oL\k cos e + a2k sin e) + a^k cos xl

where aik = at • ak for i = 1, 2, or 3. With these expressions we

easily find that

- g'k) de = 2irga3k(l - cos x) = 2ir0*(l - COS x) (5.4.25)

This result enables us now to carry out the e integrations in both

8Qak and 8Qa0. For 8Qak recall that Q = macak• To rewrite this in
terms of c0 and g, recall [from Eq. (5.1.11)] that

Q = ma{vk — uak) nia (5.4.26)

Of these quantities, only gk is changed in a collision. Therefore we

have

M (9k — g'k) de = 2717x^(1 — cos x) (5.4.27)

Fig. 5.4 The vectors g, g', and a* in the

X gz coordinate system.
 5.4 LOWEST-ORDER TRANSPORT EQUATIONS FOR A GAS MIXTURE 245

with n = mamp/m0, again, the reduced mass. Similarly, for 5Qa0, we

have Q = \mac2- Hence, in terms of c0 and g we find

Q = iwa[(c0 - u«) • (c0 - U„) +^g2 - ~ (Co - U.) • g] (5.4.28)

Again, of these quantities only g changes in a collision, so that

(Q' - Q) de = g(c0i - Uai) f** (flf* - Qi) de

= 27Tg(c0,' — uai)gi(l — cos x) (5.4.29)

The deflection angle x, which relates the direction of g' to g,

depends, as noted earlier, on b, g, and the interparticle force law.
However, rather than going through the details of integrating 1 — cos x
over all impact parameters, we shall merely define the cross section

Sapm(g) = 2ttf (1 - cos<« x)b db (5.4.30)

Detailed calculations of these cross sections are presented in Sec. A.4

of the Appendix, where we show that for an inverse-power interpar¬
ticle force law of the form force = K/rp, with K a constant and r the
distance between the particles,

Safim = 2tt (j) nnAl(v)gn n = - ^5-4-31)

The constant At(p) is a dimensionless number which is typically of

order unity, and the power n, which is plotted in Fig. 5.5, varies from
-4 (for coulomb forces) to 0 (for hard spheres).

Fig. 5.5 Plot of —n versus p for inverse-power interparticle force

laws.
246 THE BOLTZMANN EQUATION FOR A PLASMA

To summarize, we find after integrating the collision integrals

over b and e that, from Ecjs. (5.4.22), (5.4.27), (5.4.29), and (5.4.30),

sQak = y m f fafpiSap^ggic cPv d3vi (5.4.32)

0

and

SQao = ^ H f faf?iSa^v(coi - um)gig cPv (5.4.33)

8

In the next two sections we shall discuss three techniques for

handling the remaining integrations in these collision integrals:

1. The simple calculation that arises for Maxwell molecules, when

the interparticle force law is of the form / = K/rb
2. An approximate calculation for all inverse-power laws in the
limits when |u„ — ug| is much less than a0 = (aj + ap2)112, which
shows that 5Qak and bQa0 approach the simple forms

(5.4.34)
0

and

8Qao -> y ~P-aV-^ (T0 - Ta) (5.4.35)

V mo
with vap, the collision frequency for momentum transfer, given by

8pga0 r«
Vap = e~x2/Sap(1)xb dx (5.4.36)
3m07r1/2 Jo

3. An exact calculation for coulomb interactions which shows how

a new phenomenon, the runaway effect, comes into play when
|u0 — ug| is much larger than a0 for long-range interparticle
force laws

5.5 CALCULATION OF COLLISION FREQUENCIES

Maxwell molecule results

A particularly simple calculation of 8Qak and 8Qa0 arises when the
interparticle force law is of the form K/rh. (Particles with this force
 5.5 CALCULATION OF COLLISION FREQUENCIES 247

law are called Maxwell molecules.) For this case, from Eq. (5.4.31),
the cross section is proportional to g~l, so that

5Qak = y Cap f fafpigk d3V dh

J
1 (5.5.1)
B

8Qa 0 == y Cap j fafpiicoi — Uai)gi dzv dhi (5.5.2)

6

with

Cap = 2tt(Km)1/2^1(5) (5.5.3)

Now, from the definitions of na, rip, ua, and up, we have for any fa
and fpi,

Jfafp\gk dh> dh>! = ffafpi(vn - vk) dh dhx

= ffa(npupk — upvk) dh

= nanp(upk — uak) (5.5.4)

Hence, even without the assumption of a maxwellian velocity distribu¬

tion for/„ and/,3, 8Qak can be written for this case in the form (assumed
in Sec. 3.2 on physical grounds)

hQak = X PaVaf)(Upk — Uak) (5.5.5)

B

but with the collision frequency given explicitly now by the expression

vap = 2tt(Xm)1/2 —^i(5) (5-5-6)

VfloL

The exact evaluation of 8Qao can also be achieved without making

any assumption about/„ and fpi. The calculation, however, involves
a little more algebra. We first rewrite the integrand in Eq. (5.5.2)
in terms of ca and c,si by noting, from Eq. (5.1.3), that

_ _ maca + mpcpi + mpjup — ua) (5.5.7)

while

(5.5.8)
g = C0! — C„ + (up — U«)
248 THE BOLTZMANN EQUATION FOR A PLASMA

This leaves an integral of the form

5Qa0 = y — [ fafpAmpCfn2 - maca2 + mp(u? — ua)2

Aw Mq J
0
+ A] d3v dhi (5.5.9)

where A is the sum of all terms proportional to cai or Cpu and in our
usual notation for the square of a vector

(U/S — Ua)2 = (llfl — Ua) • (Ujj — na)

From the definition of c,

ffaCad3v = jfpiCpi dzVi = 0 (5.5.10)

hence all the terms in A integrate to zero.

For the remaining terms, we have from the definitions of na, n$,
Ta, and TP}

SQao = ^ f — maca2 + mfiup — ua)2] dh>

= X 3k(T^ - Ta) + m^(u(3 — ua)2] (5.5.11)

7 m°

Hence, if we replace Cafj by Eq. (5.5.3) and use ra/3 as defined in Eq.
(5.5.6), we have

8Qao = Y [3k(T0 — Ta) + m.f)(- iia)2] (5.5.12)

o 0

The calculation we have just completed shows that for Maxwell

molecules we obtain relatively simple expressions for 8Qak and 8Qa0
in terms of just the number densities, mean velocities, and tempera¬
tures of the various species in the mixture without having to make
any assumption about the distribution functions. The expression
for 8Qak was previously introduced in Sec. 3.2 on the basis of some
crude physical arguments. In that section we showed that 8Qak pro¬
vides a force tending to drive all the individual mean velocities toward
the total fluid velocity u. The term in 8Qa0 proportional to TB — Ta
plays an analogous role for the temperatures and tends to drive all
the individual-species temperatures to a common value. The second
 5.5 CALCULATION OF COLLISION FREQUENCIES 249

term in 5Qa0, however, is positive definite, and (as shown earlier in

Prob. 3.3) it increases the total thermal energy of the mixture at a
rate which just balances the decrease in the kinetic-energy density of
the flow (±2paua2) which occurs during the equilibration of the flow
velocities.

The low diffusion Mach number limit

For other force laws it is not possible to obtain simple expressions for
5Qak and 5Qa0 without specifying fa and fpi. Hence, in Eq. (5.4.32)
for 5Qak ,we replace fa and fp1 by their approximate maxwellian forms,
as given in Eq. (5.4.3), to obtain

fiQak (5.5.13)
L ir3da3dp3
e

To facilitate the integration we use the following new variables:

g dp2ca + aa2c?i
x = y =
do CloClaClp
(5.5.14)
11/3 “ Ua
£ = - do2 = da2 + dp2
a0

From these definitions one easily finds that the factor in the exponential

= (x - e)2 + ya (5-5-15)
da2 dp2

and (by calculating the Jacobian)

d3v d3vi (5.5.16)

d3x d3y
da3 dp3

When we substitute these results into Eq. (5.5.13) and perform the
resulting integration over d3y (which is now straightforward), we find

(5.5.17)
SQak ^ HQaPIc
(3

where

iQm = -/ exp[-(x tY^aP^XXk d3X (5.5.18)

250 THE BOLTZMANN EQUATION FOR A PLASMA

The final integration depends on the particular cross section

$«/3(1)(x) which is used for the collision between a and/3 particles. Exact
calculations can be made (in terms of error functions) for many
inverse-power interparticle force laws. However, since the exact
expressions are complicated, we shall examine here the simple (and
very useful) results that one obtains by expanding 8Qapk in a simple
power series of the form

SQapk = (SQaPk)e = 0 + €7 ^Q + ' ‘ * (5.5.19)

This expansion in small e (which is called the diffusion Mach number)

can be applied to a wide class of problems involving small-amplitude
waves or subsonic flows; however it cannot be used, for example, in a
problem involving high-velocity beams of particles.
Considering the first terms in this expansion, let Ik be the integral
involving e. Then, we have

/,(0) = f e-^Sa^xxk dzx (5.5.20)

This integral is identically zero, since the integrand is of the form

F{x)xk, which is an odd function of xk. [Note that F{x) is a function
of the magnitude of x, (xd + xff- + X 2)1/2; hence it is an even function
3

of each component of x.] For the next term in the expansion we have

rv T

= 2 J (xi — ei) exp [ —(x — z)'2]Sa^1)xXk d3x (5.5.21)

Hence, if we set e = 0 and apply the identity, which again holds for
any F(x) that depends on the magnitude of xf

J F(x)xiXk dzx = ^8ik J F(x)x2 dzx = ^ bik Jo F(x)xA dx (5.5.22)

we have

87T
e “’Sa/s^x6 dx (5.5.23)
3"

Applying this result in Eqs. (5.5.17) to (5.5.19) for 8Qak, we find that
8Qak can again be written (to this lowest nonvanishing order in e) in
f Note that in the last step of this identity we merely switch to spherical
coordinates, where d3x = x2 sin 6 dx dd d4>, and then integrate over 9 and <j>.
 5.5 CALCULATION OF COLLISION FREQUENCIES 251

the form

&Qak ^ Pa^ afiJ'U'ak Ufik) (5.5.24)

0

where, for an arbitrary force law, the collision frequency is given by

Sng/xQp (5.5.25)
Vaf! = e-^Sa^x5 dx
3ma7r1/2 Jo

Temperature dependence of vap

We have noted earlier [Eq. (5.4.31)] that the cross section for inverse-
power interparticle force laws is given by

Sa^ = 2 it ^yn/A1(p)(a0x)" (5.5.26)

where p is the inverse power in the force law and n = — 4/(p — 1).
For A*(p) to be finite, p must be greaterf than 2; hence, as shown in
Fig. 5.5, n lies between -4, for coulomb forces, and 0, for hard spheres
(where p —* °o). Using this expression for Sa(3(1), and noting that

fj e-*2xn+5 dx = |r(3 + |n) (5.5.27)

the expression for collision frequency [Eq. (5.5.25)] reduces to

— (l/2)n
87r1/2P3d.i(p)aon+ir(3 + vn)
Va& =
3m0 © (5.5.28)

The most important point to note in this expression is that va$ is

proportional to a0n+1 or to [(Ta/ma) + (Tp/mp)],''l2)(n+l)■ Hence, vap
has a temperature dependence which ranges between T~312 for coulomb
collisions (where n = —4) and T1^2 for hard-sphere collisions (where
n = 0). In addition, for collisions between electrons and heavy ions
or neutral molecules, it is evident that because of the large mass
difference, a0 = ae (unless Te is very much less than the temperature

f For coulomb forces, with p = 2, the divergence in A tip) is avoided (as shown
in Sec. A.4 of the Appendix) by considering only impact parameters less than the
Debye length \D. The justification for this is based upon our discussion of the
Debye potential in Sec. 4.8, where we noted that the coulomb potential 4>f between
two particles in a plasma is effectively reduced by the factor exp ( —r/Xn). Since
the full Debye potential is hard to use in calculating Ah one still uses 4>c but just
for b < Xz>.
252 THE BOLTZMANN EQUATION FOR A PLASMA

of the heavy particles). Hence it is the electron temperature which

generally determines the electron collision frequency.

Typical numerical values

To obtain actual numerical values for the electron-ion and elec¬
tron-neutral collision frequency, we can use the coulomb and hard-
sphere cross sections obtained in Sec. A.4 of the Appendix and sum¬
marized in Table 5.1, where

TZI2
A = 9.0A"d = 1.23 X 107 --/2 (5.5.29)

and <x is the sum of the radii of the colliding particles (for ven this is
essentially the classical radius of the neutral molecule). If we sub¬
stitute these values for A j and n into Eq. (5.5.28), we find

= 3.62 X lO-6 n;Tr3/2 In A (5.5.30)

and

= 2.60 X 10Vn„7V/2 (5.5.31)

Typical values for In A are about 10, while a is of the order of

10-10 m. Hence at 300°Iv we find

vei = 7 X 10~9rp ven = 4.5 X 10'15n„ (5.5.32)

so that at low temperatures electron-ion collisions are often of pre¬

dominant importance even when a gas is weakly ionized. At higher
electron temperatures, however, vei falls off while ven increases; hence
at 3 X 106°K, we have

Vei ^ 7 X 10-157ii ven ^ 4.5 X 10-13n„ (5.5.33)

TABLE 5.1

Cross section $(1) n Ai (p)

Coulomb 47r(e2/47T€ong2)2 In A -4 2 In A
Hard spherej 7rcr2 0 cr2/2

f The value of Ai for hard spheres is chosen artificially to

satisfy Eq. (5.5.26).
 5.5 CALCULATION OF COLLISION FREQUENCIES 253

Thus, at high temperatures, a small number of neutrals can be of great

importance even in a highly ionized gas.
If we also use the hard-sphere cross section to calculate vin, there
are three changes in Eq. (5.5.28) (compared to the calculation of ven):
(1) the sum of the radii of the colliding particles is now about 2a (since
the ion and neutral molecule radii are about equal, while the electron
radius is much smaller), so that Hi increases by a factor of 4; (2) the
value for m0 = ma + mp increases by a factor of 2, since mn = rrii;
(3) the value for a0 changes from

1/2

to

(2k)1/2

Hence, if Ti = Tn, we find that

meTA112 (5.5.34)
— ^ 2.8
V en mlTe)

so that, as noted earlier in the text, vin is generally much smaller

than ven.

Limiting value of 8Qao

To complete the analysis we consider the collision integral for energy
transfer 5Q«o in the limit as e —> 0. In Eq. (5.4.33) we replace/a and
/(3i by the approximate forms given in Eq. (5.4.3), and we again
rewrite the integral in terms of x, y, and t [as defined in Eq. (5.5.14)].
Most of the factors in the integrand are identical to those in 5Qak.
Hence using the identity

(5.5.35)
C Qi 'llai
a0
Vi + Gi(x)

where

2k (5.5.36)
Gi(x) {Tf> - Ta){xi - et) +^a0€,-
254 THE BOLTZMANN EQUATION FOR A PLASMA

we find

aaa0
8Qa0 = £ Mny°- / exp [-(x - e)2 - t/]Sap (i) Vi
a0
B

+ Gi(x) xxi dzx dzy (5.5.37)

The term involving yi is odd and therefore integrates to zero, leaving

8Qa0 = 2 SQapo (5.5.38)

with

8Qapo = / exP [-(* - «)2]^(1,(?t(x)®x,-d*x (5.5.39)

Comparing this result with Eq. (5.5.18) for 8Qapk, we see that

n(Tt3 ~ Ta) d8Qapi i mP sn

~\- - Oo€i Ok/api (5.5.40)
m0a0 dei m0

Since we have already evaluated 8Qapk, it is now quite simple to cal¬

culate 8Qapo- From Eq. (5.5.24) we easily find (to lowest nonvanishing
order in e)

8Qao = y K(Tp ~ Ta) + terms of order e2 (5.5.41)

P 0

Problems

5.6 Verify the identity shown in Eq. (5.5.35).

5.7 Expand Eq. (5.5.39) in a Taylor’s series about e — 0 and calculate
8Qao to order t2. Show that your result agrees with that used in
Prob. 3.3.

5.6 THE ELECTRON-RUNAWAY EFFECT

In the last section we analyzed 8Qa in the limit when e —> 0, and found
that for all inverse-power interparticle force laws

8QaPk e_>o'> Pa^ap(Upk Uak) (5.6.1)

 5.6 THE ELECTRON-RUNAWAY EFFECT 255

with vap given, for arbitrary force laws, by Eq. (5.5.25) or (5.5.28).
Hence, for small e it is clear that 5Qapk increases linearly with the
difference in the flow velocity. To illustrate the significance of this
fact consider a fully ionized gas in the presence of a constant electric
field. Assuming no spatial gradients and that B is negligibly small,
we have from Eq. (5.4.IS)

n, = E Uj) (5.6.2)
me

u. rie(ni Up) (5.6.3)

In general we should also use Maxwell’s equations to take account

of the electromagnetic fields set up by the currents in the plasma and
energy equations to take account of the increase in the electron and
ion temperatures to be expected when ue does not equal ut. How¬
ever, for this example we assume that E and the temperatures are
maintained constant by some external mechanism. If we let A be
the difference in' the fluid velocities, u,; — ue, then we find (by sub¬
tracting the first equation from the second)

A + vA = — v = Vei + Vie (5.6.4)

M

If A(0) is the initial difference in the flow velocities, then

A(0 = A(0)e-"( + — E(1 - e~vt) (5.6.5)

jlV

so that, after a time large compared to v~\ the difference in the flow
velocities becomes constant and equals eR/^v. At equilibiium the
current density, as might be expected from our analysis of the ( (in¬
ductivity in Sec. 2.2, is given by

J = n0eA = <r0E (5-6-6)

with ao the conductivity,

n0e2 n0e2 (5.6.7)

Co —-
nv Wle ^ ei
256 THE BOLTZMANN EQUATION FOR A PLASMA

Exact evaluation of 5Qapk

When e is not small and 5Qa0k for election-ion collisions is not given
by Eq. (5.6.1), this simple problem of the response of a plasma to a
uniform E field is greatly changed, and a new phenomenon arises, the
so-called electron-runaway effect, in which for a large but finite E
field J —► oo. Before proceeding with the mathematical analysis of
this problem, we should note that in a rough way the effect can be
anticipated from the fact that the coulomb cross section from Eq.
(5.4.31) is proportional to ^r4, with g the relative velocity of the collid¬
ing particles. Now in a strong E field, the electrons and ions are
accelerated in opposite directions, and for most of the colliding par¬
ticles the relative velocity g becomes very large. Hence, the collision
cross section becomes very small, and the “conductivity” (which is
inversely proportional to v) increases beyond the value calculated
using Eq. (5.5.30) for v.
To investigate this effect analytically, we must first evaluate
5Qapk for arbitrary t. From Eq. (5.5.18) and Table 5.1 we have for
coulomb interactions,

(5.6.8)

with

nanffe4 In A
(5.6.9)
47r5/2eoVa02

To evaluate this integral we choose a set of coordinates (shown in

Fig. 5.6) where the a3 axis lies in the direction of t. In this coordinate

Fig. 5.6 Coordinate system used to

a evaluate 8Qagk.
 5.6 THE ELECTRON-RUNAWAY EFFECT 257

system

t = ea3 X = x[sin 9(ax cos </> + a2 sin </>) + a3 cos 9} (5.6.10)

and

d3x = x2 sin 6 dx d6 d4>

Hence the k component of these vectors (in some other coordinate

system) is

tk = dike Xk = X’[sin 0(alk cos <f> + aok sin </>)

+ azk cos 9] (5.6.11)

where the aik are direction cosines of the fth axis in this collision-integral
coordinate system relative to the /cth direction in the external coordi¬
nate system (atk = a, • a*). Using Eq. (5.6.11) to handle the 4>
integration, we find

C III 5* g-*-*+ixt cos e sin Q dx do ^

2-irCeke e' (5.6.12)

dx e x2+2xs

with s = e cos 9. The integration over x is performed by letting

2 = x — s, to have

IT il2Ctke- (5.6.13)
foQaflk es2s(l + erf s) ds

where erf s is the error function

erf s = 2tt~112 f l e~xl dx = 2-k~"2 f°_s e~*2 dx (5.6.14)

The first term in this final integral is zero, since the integrand is odd;
while the second term (which has an even integrand) is easily evaluated
by integrating by parts. In the formula fU dV = UV| - J1 dU
we letj

U — erf s dV = se”2 ds (5.6.15)

dU = 27r-1/2e-s2 ds V = 4e*2

j To take the derivative of the error function recall the formula for the
derivative of an integral given in Sec. 4.6:

Ts C F(s’s'} ds' = F[s’Ms)] it ~ F[s,/l(s)11 + IfM sids'

258 THE BOLTZMANN EQUATION FOR A PLASMA

Fig. 5.7 Plot of \p versus e for coulomb forces. For comparison, the result
is compared with the straight line = e, which holds for Maxwell molecules
and is also the small e approximation for all force laws.

to find

SQafik = 7T^C ^ (5.6.16)

From Eqs. (5.5.28) and (5.6.9) and Table 5.1 we can rewrite 8Qap,
where 8Qag = [^ (SQ^fr)2]1/2, in the formj (useful for computation)
k

SQafi = aoPava^(e) T(«) =

3tt1/2 /
(erf e —
9f
~in e“f2 j
\
(5.6.17)

A plot of T(e) is shown in Fig. 5.7. The most significant aspect of

this figure is that T(e) reaches a maximum when e = 1 rather than
increasing indefinitely [as it would if the small-e result, T(e) = e, were
always valid]. It is this fact which leads, as we shall see, to the
electron-runaway effect.

Electron-runaway effect

Using the exact expression for 5Qei and 8Qie, suppose we reexamine the
interaction of the plasma with a steady electric field. The momentum
equations for the electrons and ions [Eq. (5.4.18)] with all spatial

f Recall that for the coulomb potential K = e2/4ire0.

 5.6 THE ELECTRON-RUNAWAY EFFECT 259

derivatives set equal to zero reduce to

jp C
&Qeik i (5.6.18)
Uek —--Uk ~I-
me Pe

6 77T . &Qiek
(5.6.19)
Uik = — -C/k H-
mi pi

Hence, by subtracting the first of these from the second and using
Eqs. (5.6.16) and (5.6.17) we find

eE a0d/'(€)rA
(5.6.20)
A - u il
A

with V = Vei + Vie. This equation has two classes of solution, depend¬
ing on the magnitude of E:

1. For eE/pa0v < 0.57 [the maximum value of T(e)] A rapidly

approaches the equilibrium value, where T(e) = eE/pa0v, so that
the two terms on the right side of Eq. (5.6.20) cancel, and A is
zero. (This is, of course, just a generalization of the simple
result obtained earlier for small e.)
2. For eE/paov > 0.57, however, the forcing term in Eq. (5.6.20),
eE/p, is always greater than the restoring term, a0vV, so that A
increases continuously, and J —»■ °° •

The minimum field strength, Ec — 0.57meaevei/e, which must be

maintained by an external force for runaway to occur is called the
critical field■ Using Eq. (5.5.30) for vei, we can write the expression
for the critical field as

e (5.6.21)
Ec = 0.193 In A
47re0Az32

Since In A is of order 10, this shows that Ec is on the order of the field
strength (in a vacuum) one Debye length away from a charge. In
addition, since A and XD2 are both proportional to temperature, the
temperature dependence of Ec is of the form

Ec * 71-1 In T (5.6.22)

Hence, runaway can occur for arbitrarily weak fields if the temperature
becomes sufficiently high.
260 THE BOLTZMANN EQUATION FOR A PLASMA

Problem
5.8 Prove that for an arbitrary inverse-power interparticle force law

Hence, for large e, <5Qa/3 —> 0 for all force laws with n < —2 or
p < 3; while for the other (shorter-range) force laws 8Qap—> °°
for large e, so that the runaway effect cannot occur.

5.7 THE THIRTEEN-MOMENT EQUATIONS FOR AN ELECTRON GAS

The thirteen-moment approximation

The analysis in the last three sections is all based upon the assump¬
tion that each species of particles in the plasma has a maxwellian
velocity distribution, so that/a is specified by just the five “moments,”
na, ua, and Ta. With this assumption we were able to derive transport
equations similar to those developed in Chap. 3 on macroscopic grounds.
However, the phenomenological collision frequencies used in Chap. 3
could be computed from a knowledge of the interparticle force laws,
and there were some effects (most notably, the electron-runaway effect)
that one would not suspect from the macroscopic theory. All these
results, however, still omit such phenomena as heat conduction and
viscous-damping effects in the plasma.
To analyze these factors we must extend the analysis in some way.
However, to reduce the algebraic detail involved in such an extension,
we shall, in this section, restrict our attention once again to an electron
gas in a stationary background of heavy particles which has just
enough uniformly distributed positive charges to make the plasma
macroscopically neutral in equilibrium. In addition, rather than
using the full Boltzmann collision term, we again use the Krook colli¬
sion model, (df/dt)c = — v(J — /0), whose use greatly simplifies the
collision integral.!

f For this case we must assume that v is the electron self-collision frequency,
since, as noted in Sec. 4.3, any attempt to interpret v as an electron-heavy particle
collision frequency leads to large errors in at least some of the collision terms.
Nevertheless, the results we shall obtain for the electron pressure tensor (or vis¬
cosity) and heat-flow vector (or thermal conductivity) in this section and the next
can be put into remarkable agreement with those obtained by exact calculations
for a weakly ionized gas if we do interpret v as the electron—heavy particle collision
frequency.
 5.7 THE THIRTEEN-MOMENT EQUATIONS FOR AN ELECTRON GAS 261

Two approaches can be used here, the original Chapman-Enskog

procedure for solving the Boltzmann equation and the more recent
13-moment approximation of Grad. Although the former method is
actually somewhat easier to apply when there is a weak (or no) dc
magnetic field and when the Krook collision model is assumed, it is
much more difficult to work with when there is a strong dc magnetic
field or when the full Boltzmann collision term is used. For this
reason we shall use the 13-moment method here, while outlining the
Chapman-Enskog procedure in Sec. A.5 of the Appendix.
The basic expansion procedure followed in the 13-moment
approximation is to set

/ = /0(1 + AiCi -T BijCiCj T CiC2Ci) (5.7.1)

where f0 (for the reasons indicated in Sec. 5.3) is the maxwellian

velocity distribution with the local number density, mean velocity,
and temperature of the electrons,

-c^/a2
ne (5.7.2)
/o
7r3/2a3

and where Aif C<, and B{j ( = £*), like n, u, and T, are functions of
position and time but not of v. (Hence we shall refer to them as
constants.) The reasons for expanding in a power series in c are two¬
fold: (1) the integrals involving / are all manageable, and (2) the coeffi¬
cients Ai, Bij, and C; all have a real physical meaning.
To show this we need only compute n, (ck), pkh, and qk in terms of
the constants. To simplify the calculation we note, first, the following:

TABLE OF INTEGRALS

J/o d3c = n (5.7.3)

mjf0CiCjd3c = p8ij = %pa28ij (5.7.4)
mffoCiCjCkCi d3c = \pa2(8ij8ki + 8ik8ji + SuSjk) (5.7.5)
mSfaC2CiCj d3c = %pa25ij (5.7.6)
m//oC4 d3c = i^-pa2 (,r> -7.7)
mjfoC2CiCjCkCi d3c = ApaA{8i}8ki + 8ikbji + 8u8jk) (5.7.8)
mjf0cicicjd3c = ^pa48n (5.7.9)
262 THE BOLTZMANN EQUATION FOR A PLASMA

The integrals in this table can all be verified very quickly by using
the following scheme. First, note that for /o given by Fq. (5.7.2)

dfo (5.7.10)
dCi

Hence, for any integral involving the product foCi we can first use this
identity and then integrate by parts to find [assuming/0F(c) vanishes
at Ci = ± °° ]

I f0CiF(c) d3c

(5.7.11)

This formula makes the calculation of the integrals almost trivial.

For example, to verify Eq. (5.7.4), where F(c) = mCj and dF/dCi mbij,
=

we have at once, from this formula and Eq. (5.7.3),

TOSfoCiCj d3c = ^a2mSijjf0 d3c = ^pa25ti = rphl] (5.7.12)

Similarly, for Eq. (5.7.5), where F(c) = mcjCkCi and

dF/dCi = m(8ijCkCi T 5ikCjCi T S,iCjCk)

we find, from the formula and the preceding result, that

mSfoCiCjCkCi dzc = ia2TOffo(8ijCk.Ci + 8ikCjCi + 5uCjCk) dzc

— + 8ikdji + Siidjk) (5.7.13)

From these results one can easily, verify all the remaining integrals
with very little effort.

Interpretation of the coefficients

Through the use of this set of integrals we can now calculate n, (ck),
Pkh, and qk directly from their definitions in Sec. 4.1. We have, for
/ given by Eq. (5.7.1),

n — //d3c = J/0(l + AiCi + BijCiCj + CiC2Ci) d3c = n( 1 + 4a2F„) (5.7.14)

 5.7 THE THIRTEEN-MOMENT EQUATIONS FOR AN ELECTRON GAS 263

This result shows at once that Bn, the trace of B, is zero. Next, we
calculate p(ck), to find

p(cfc) = mjfck dzc

= p(Ak + |o2Ct) (5.7.15)

Since (ck) must be zero, this result shows that

C„ = - ^ (5-7.16)
5 a1

Thus far we have not yet related the coefficients to any physical
quantities, but this fault is remedied if we calculate pki and qk. From
the definition of the pressure tensor, we have from Eqs. (5.7.1), (5.7.3),
and (5.7.5)

Vki = mifckci d3c

= p[5ki + i a3Bij(5ijdki + 8ik8ji -f- dudjk)]
— phi + pa2Bki (5.7.17)

(Note that in the last step we have used the fact that Bydij = Bu = 0.)
Comparing this result with Eq. (4.2.20) for the traceless pressure
tensor, it is evident that

P kl (5.7.18)
Bki
pa2

Similarly, from the definition of the heat-flow vector, we have from

Eqs. (5.7.1), (5.7.6), and (5.7.9)

qk = im\fc2ck d3c
5 pa2 (5.7.19)
IT (Ak + ^o-2Ck)
Eliminating Ck through the use of Eq. (5.7.16), we find

A_2 qk (5.7.20)
Ak pa2

To summarize, this analysis shows that the constants Ak and Ck

are proportional to the heat-flow vector, while Bkt is proportional to
the traceless pressure tensor. Hence/, as given by Eq. (5.7.1), is a
function of n, u, T (or p), P, and q. Although P is a 3 X 3 matrix,
264 THE BOLTZMANN EQUATION FOR A PLASMA

it contains only five independent components, since, from its definition,

Pij = Pji and Pa = 0, so that

Pn P13
P = P12 P 23 (5.7.21)
P13 — P11 — P 22

Thus / in this approximation depends on 13 independent moments.

Problems

5.9 Use the maximum-entropy approach to find the most probable

distribution for a simple gas when n, u, p, and q are all specified.
Under what conditions does this result approximately equal
Eq. (5.7.1)?
5.10 Verify the integrals given in Eqs. (5.7.6) to (5.7.9).

Viscous-stress equation

To determine the 13 moments in this approximation for / we need 13

independent moment (or transport) equations. Five of these are pro¬
vided by the mass-, momentum-, and energy-conservation equations
derived in Sec. 4.2,

5? + • u = ° (5.7.22)

DUk C rp i / w i 1 dpik
n, = [Pk + (u X B)fc (5.7.23)
Dt m p dXi

^^(Pp-»)+l(v.q + P,g) = 0 (5.7.24)

Note that in the momentum and energy equations we have set 8Q = 0,

in accord with the discussion of a simple gas in Sec. 4.3. (Note, how¬
ever, that this also follows by a direct calculation of

SQ = -vfQ(f - /o) dh (5.7.25)

for Q = mck or Q = ^me2.)

The additional transport equations are obtained most conven¬
iently by letting Q = mCkCj and Q = ±mc2Ck in the general transport
equation (derived in Sec. 4.2)

l m - MQ) + ± »<CM - n (v< g) - n (a, g) = SQ (5.7.26)

 5.7 THE THIRTEEN-MOMENT EQUATIONS FOR AN ELECTRON GAS 265

First, for Q = mckCj, we find for each term on the left side of this
equation

dpkj
Jtn{Q) dt

~n(Q) 0

(Pikj “t" UiPki) Pikj ~ p(CiCkCj) (5.7.27)

^n{ViQ)
diij . dUk
_b/3
n w P(‘ aS + p“ a7

— n (a,
dQ\ ■ p(djCk + cqe,) — — Mjk
dVi/

The collision integral, using the Ivrook model, is given by

5Q -- — vmj(J - fo)ckCj d3c

= —v(pkj ~ Phi) = -vPkj (5.7.28)

Hence this transport equation (which is known as the viscous-st) ess

equation) reduces to

dpkj d . s , dUj dUk

+ ^ (.Pikj + uiVki) + Pik ^ + Pa ^7 M kj - vPk] (5.7.29)
nr

This result has been derived for an arbitrary distribution function

/. However, for / given by Eq. (5.7.1), we have from the table of
integrals and Eq. (5.7.20)

Pikj = mSfclcjck d3c

(5.7.30)
— f (qkhj + qjhk + qSjk)

In addition, for a = — (e/m)(E + v X B), we find

Mjk = — ne((c X B)jCfc + (c X B)kCj)

so that in a rectangular coordinate system, where B lies along a2, we

have (with ojc = —eB/m)

M ii — 2ojcPi2 M 22 = — 2o)CP 12 M 33 = 0

M13 — WCP 23 M 23 — — 03CP13

Mn = WC(E 22 P ll)
(5.7.32)
266 THE BOLTZMANN EQUATION FOR A PLASMA

When Eqs. (5.7.30) and (5.7.32) are used, the viscous-stress equation
(5.7.29) involves just the 13 moments used in the approximation for/.

The heat-flow equation

To complete the required set of 13-moment equations, we let

Q = ±mc2ck. For this case the terms in the transport equation
(5.7.26) are found to be [for a = — (e/m)(E + v x B)]

dqk
hn{Q) dt
~n(Q) Uj(Pjk T ^pSjk)

(^Pikjj “f" 'U'iQk) Pikjj

k n{ViQ)
/ dQ\ dUj duk . , , q t x dUj
(5.7.33)
Pi'‘teI + 5ite- + fe‘ + lpJrt)“iaT.

— n (a.
dQ\ p(Uj(CjCfc T ^C25yfc))
7 dvJ

— [(E + u x B)j(pjk + fp8jk) + (q X B)fc]

8Q — vqk

When adding these terms together we note that the three terms pro¬
portional to pjk + |p8jk all reduce, because of the momentum equation
(5.7.23), to

Pjk ~l~ jl p 8jk dpij

(5.7.34)
p dXi

Hence, the resulting equation, which is called the heat-flow equation, is

dqk Pjk + ip8jk dp dll,

dt si + s; + ««*> + s
i dllk | C ,
(5.7.35)
+ 9itei + m(q XB)‘= -"I 1

This result, again, involves no assumption about /. However, when

Eq (5.7.1) is used for /, pijk can be written in terms of q, as shown in
Eq. (5.7.30), and we also find

Pika = m Jfc2CiCk (Pc

= ha2(5p8ik + 7Pik) (5.7.36)
 5.8 CALCULATION OF TRANSPORT COEFFICIENTS 267

With these substitutions, the heat-flow equation also involves just

the 13 moments used in the approximation for/.
To summarize, in this section we have introduced a distribution
function [Eq. (5.7.1)] that depends on the 13 moments, n, u, T (or p),
P, and q. To specify these 13 moments we can use the newly derived
viscous-stress and heat-flow equations [Eqs. (5.7.29) and (5.7.35) with
Pijk and pikjj given by Eqs. (5.7.30) and (5.7.36)], plus the mass, momen¬
tum, and energy equations (5.7.22) to (5.7.24).

Problems
5.11 Verify that one obtains the energy equation by letting/ = k (and
summing over k) in the viscous-stress equation (5.7.29).
5.12 Calculate the collision integral for the heat-flow equation when
the Boltzmann collision term is used and the Maxwell interparticle
force law is assumed, and show that, as here, the term is propor¬
tional to qk-

5.8 CALCULATION OF TRANSPORT COEFFICIENTS

Results for small dc magnetic field

The 13-moment equations are used in two distinct ways. First, and
most obviously, one can use the full set of equations to solve for all
13 moments in a flow, wave-propagation, or heat-transfer problem.
Although this approach is often very difficult, it has been carried
through for a number of problems, 4 he results of such analyses aie
generally much better than those obtained in the five-moment approxi¬
mation of Sec. 5.4, where and qk are identically zero and just the
mass, momentum, and energy equations are used. In addition, how¬
ever, the 13-moment equations can also be used as the basis for a
simple perturbation scheme in which the stress and heat-flow equa¬
tions are used simply to calculate Pkj and qk in terms of n, u, and T
(or p). When this is done, the remaining momentum and energy
equations reduce to the Navier-Stokes equations of hydrodynamics,
but with the viscosity and thermal conductivity given explicitly in
terms of v. It is this application of the equations which we consider
in this section.
The expansion procedure we use is to assume that v is very large,
so that we can replace Pkj and qk by a series of the fmm

Pkj c. + £ + 3 +
V VL
(5.8.1)

qk Da + ^+^ +
V V
268 THE BOLTZMANN EQUATION FOR A PLASMA

The other moments, however, are left independent of v. Hence, in

the viscous-stress and heat-flow equations (5.7.29) and (5.7.35), from
the terms proportional to v we have (assuming for the moment that
|wc| « v)

0 = —vCo and 0 = —vD0 (5.8.2)

so that Co and D0 are both zero.

To next order in the two equations (terms independent of v) we
have

(5.8.3)

(5.8.4)

To this order (terms independent of v) we have from the mass and

energy equations (5.7.22) and (5.7.24)

Dp _ 5 p Dp
-fpV • u (5.8.5)
Tri ~ 3~pDt

Hence, from Eqs. (5.8.1) and (5.8.3) we have, to this order,

dUj
Pkj = ^ih3v u (5.8.6)
dxk

and from Eq. (5.8.4)

5p d p _ dT 5 Kp
(5.8.7)
2v dxk p °dxk 2m v

The coefficients r/o and X0 which enter here are the viscosity and the
thermal conductivity of this simple electron gas. The ratio of X0 to 170
is easily seen to be 5k/2to. Had we performed this calculation using the
Boltzmann collision term (just for self-collisions), this ratio would
instead have been 15/c/4m, so that the error involved in using the sim¬
ple Ivrook model (where, as noted in Sec. 4.3, the relaxation time for
each quantity is exactly r_1) is only a factor of 1.5.
If we substitute the expressions in Eqs. (5.8.6) and (5.8.7) for
Pkj and qk into the momentum and energy equations (5.7.23) and
 5.8 CALCULATION OF TRANSPORT COEFFICIENTS 269

(5.7.24), we have (assuming rj0 is constant) in (5.7.23)

Duk e ,„ , „ -\ 1 dp 170
-7T7 =-[Bk + (u X B)fc---1- V2Ujfc + ~(V'u)
Dt m p dxk p 3 ax*
(5.8.8)

which is the Navier-Stokes equation of hydrodynamics. To the same

level of approximation the energy equation (5.7.24) is

,5/3 IL
Dt
VP
-5/3 _
•fXo
v 1 + ^ dxi
(5.8.9)

with Pa given by Eq. (5.8.6). When the electron gas is stationary

(u = 0), Dp/Dt goes to zero (because of mass conservation). Hence
the energy equation reduces to the familiar heat-conduction equation

UK? = fXo V2T (5.8.10)

Effect of the magnetic field on <lk

To simplify the analysis we have thus far assumed that wc <5C v, so that
the magnetic field has, to this order, no effect on Pkj and qk- Now
let us repeat the analysis for «c comparable to (or greater than) v.
To lowest order (terms proportional to v or wc) we still find

Co = D0 = 0 (5.8.11)

However, to next order the results are quite different. First, in the
heat-flow equation [Eq. (5.7.35)] for B = Bat, we now have, in place
of Eq. (5.8.4),

(5.8.12)
|p — - - wc(q X at)k = -D1
2 dxk p

or, to this level of approximation, letting D1 = qtv,

q--(qX af) = -X0 VT

(5.8.13)
V

where once again = —-eB/rn and X0 — 5/cp/2mr.

This is completely analogous to the equation for current flow with
a uniform magnetic field [Eq. (2.2.5)] but with q replacing J, Xo
270 THE BOLTZMANN EQUATION FOR A PLASMA

replacing (To, and — V71 replacing E. Hence, as in Sec. 2.2, we have

q = -A-VT (5.8.14)

with

Ax Ah 0
A = — A// Ax 0 (5.8.15)
0 0 Ao

and

\ _ A0r2 A0cocr
= (5.8.16)
Ax ~~ W„2 + C
This residt shows that for a fixed temperature gradient the heat
flow in the direction of B is unchanged. However, the heat flow is
reduced in the directions perpendicular to B and, in addition, there is a
Hall heat flow in the direction perpendicular to both B and VT.
Calculations of X± and \H as a function of magnetic-field strength are
shown in Fig. 5.8, where it can be seen that both fall off fairly rapidly
with increasing values of o>c.

Effect of the magnetic field on Pk]-

The effect of the magnetic field on the traceless pressure tensor is
slightly more complicated since we must deal with a tensor rather
than just a vector. In the viscous-stress equation (5.7.29) in place
of Eqs. (5.8.3) and (5.8.5) we find

p(-^-“+S+S)_Mh= (5-8'17)

Fig. 5.8 Hall and perpendicular thermal conductivities plotted

as a function of the ratio of cyclotron to collision frequency.
 5.8 CALCULATION OF TRANSPORT COEFFICIENTS 271

with Mkj given in Eq. (5.7.32). If we divide each term by v and recall
that 770 = p/v and that, to this level of approximation, P kj = Ci/v,
we have

> (5.8.18)
II
1

with fkj given by

, „s _
fkj - ?hjV • u
dUj
^ ■ d-ff± (5.8.19)
dxj

Writing out Eq. (5.8.18) explicitly in the coordinate system where

B = Baz, we have two groups of terms

P13 - - ^23 = VoflZ

V

P 23 H-- P13 — 110/23

(5.8.20)
V

P 33 = 11of33

and
II

1—1

Pn -2 -CP12
V

S* Pn + P„ - ^ Pn = Wi* (5-8.21)
V V

2 — P12 + P 22 = 770/22
V

The set of equations involving P 1.3, P23, and P33 is again exactly
like the current-flow equations with Pk3 replacing Jk, fk3 replacing Ek,
and 770 replacing cr0. Hence, as in Eq. (2.2.7), we have

P13 = V±/l3 + 7?/7/23 P23 = — vh/u + 771/23 (5.8.22)

with

77qf2 _ 770 03cv

(5.8.23)
Vx = T? // 9 1 2
03c2 + r2 03c + V1
272 THE BOLTZMANN EQUATION FOR A PLASMA

The second set of equations (involving Pn, P12, and P22) is easily
solved to find (using the identity fn + f22 = —/33)

p _ 7?o[/u ~h 2(tOc/y)/i2 ~~ 2(faJc/r)2/33]

11 1 + 4(wc/ f')2
D 7/o[/22 + 2(wc/t)/i2 — 2(wc/^)2/33]
(5.8.24)
^22_ l+4(ccA)2
P _ 7?o[/i2 + (Uc/v)(f22 — fu)]
12 ~ 1 + 4(co„A)2

In the limit when coc/v approaches zero these results reduce again
to the form

Pi} Vofij (5.8.25)

which leads, as we have seen, to the Navier-Stokes equation of hydro¬

dynamics. However, for very strong magnetic fields, when \ue/v\
approaches infinity, we have

-I 0 0
P — ^0/33 0 -i 0 (5.8.26)
0 0 1

From these results it is clear that viscous effects with a strong mag¬
netic field are generally quite different from those found using the
usual Navier-Stokes equation.

Problems

5.13 (a) Use the Navier-Stokes equation to calculate the steady-

state flow velocity ux(z) in a gas between two infinite plates
(parallel to the xy plane) when the plate at z = 0 is station¬
ary [mx(0) = 0] and the one at z — d moves in the x direction
with velocity ux(d) = u0. For simplicity, let p and T be
constant and let E = B = 0.
(6) Repeat the analysis, but let this be an electron gas with a
weak uniform magnetic field. Suppose first that B = B0az,
then that B = Boa^.
(c) Next examine the problem when B0 —> °o and P is given by
Eq. (5.8.26).
5.14 Find the dispersion relation for small-amplitude longitudinal
waves in an electron gas with B = 0 when the Navier-Stokes
equation is used for ue and Eq. (5.8.9) is used for the energy
 5.9 DISTRIBUTION FUNCTION FOR A STRONG E FIELD 273

equation. To simplify the analysis use a perturbation approach

where, to lowest order, rj0 and X0 are zero. Show that the damp¬
ing due to X0 and to rj0 is of comparable importance.

Limitations of the theory

In this section we have used the stress and heat-flow equations to
calculate the traceless pressure tensor and the heat-flow vector for an
electron gas. The results obtained, while certainly indicative of the
behavior expected in a real plasma, cannot be taken too literally,
because of two limitations in this analysis: (1) we have used the Krook
collision model rather than the full Boltzmann collision term; and (2)
we have neglected the motion of the heavy particles, which means
that the results are probably not very accurate for low-frequency
waves or flow problems (where the mhd equations would be a better
approximation).
Both of these restrictions can be eliminated by using the full
collision integrals-

5Q = X / faf^Q' ~Q)gh dh de dh d*vi (5'8'27)

0

for the stress and heat-flow equations. Although all the integrations
can be carried out for many inverse-power interparticle force laws,
the results are quite complicated and have not yet been reported in a
form which is easy to interpret physically. Here we note only that
in the stress equation, for example, where the Krook model leads to

-vPak] (5.8.28)

the full collision integral involves terms proportional to Pakj, Pm,

Tp - Ta, and (upk - uak)(upj - uaj). Thus, while one can again use
an expansion procedure to find the traceless pressure tensors in terms
of the number densities, mean velocities, and temperatures in the
mixture, the analysis is more complicated, and one must use the full
set of stress equations for all the species simultaneously.

5.9 DISTRIBUTION FUNCTION FOR A STRONG E FIELD

Expansion procedure for /

In this section we reexamine the problem of the behavior of a plasma
in a uniform steady electric field. We have already looked into this
problem twice, first using the Langevin equation in Sec. 2.2 to obtain
274 THE BOLTZMANN EQUATION FOR A PLASMA

a simple expression for the dc conductivity and, again, using the five-
moment approximation for the distribution function in Sec. 5.6 to
study the electron-runaway effect in a fully ionized plasma. Now we
shall apply the Boltzmann equation to a very weakly ionized gas, so
that the collision term for the electrons is dominated by the electron-
neutral collisions. (This is generally called a Lorentz gas.) The
questions we pose are the following: (1) what is the velocity distribu¬
tion of the electrons when there is an applied uniform electric field,
and (2) what is the current flow (or conductivity) in the plasma?
For simplicity we assume that there is no magnetic field present,
that the plasma is completely uniform (so that there are no spatial
gradients), and that a steady state has been attained (so that there
are also no time-derivative terms). With these assumptions the
Boltzmann equation for /(v), the velocity distribution function for
the electrons, reduces to

(5.9.1)

where for a weakly ionized gas we retain only the electron-neutral

collision term, so that

(5.9.2)

In this collision term F(V) is the velocity distribution function for the
neutral molecules, and, as usual, f and F' are the distribution functions
evaluated after the collision. It is assumed that F(\) is known, and
in practice we take it, later, to be a function of just the magnitude of V.
Although Eq. (5.9.1) appears to be relatively simple, it is still
too complicated to handle exactly. Hence we shall introduce another
expansion procedure for/. We let

f(y) ~ fo(v) + ~ fiiv) + dL± fety) -p... (5.9.3)

where /0, fi, fa, etc., are all functions of just the magnitude of v, so
that, for any one of them, say /p,

d/p = dy djf = Vk dfp (5.9.4)

dvk dvic dv v civ

This expansion procedure is equivalent to expanding / in terms of a

set of spherical harmonics in velocity space; hence the technique is
 5.9 DISTRIBUTION FUNCTION FOR A STRONG £ FIELD 275

generally referred to as the spherical-harmonic solution to the Boltz¬

mann equation.
Although it is possible, in principle, to retain as many terms as one
likes in this expansion for f, in practice one generally uses just the
first two terms. For this case Eq. (5.9.1) becomes

e_ df0 d fi
m v
Vk -J-
dv
+ ViVk ~J - + ft
dv v
(5.9.5)

where it is understood that the approximate form for / is used in

(,df/dt)c. ’ This equation can be reduced to a set of four coupled
integrodifferential equations by the following simple technique
(commonly used in any expansion procedure employing an orthogonal
set of functions). First we multiply each term in Eq. (5.9.5) by
sin 0 dd d(t> (where 0 and <j> are the usual angles in the spherical coordi¬
nate system shown in Fig. 5.9) and integrate over 0 (from 0 to tt) and
(from 0 to 2x). By direct calculation one can easily verify that

Jj sin 0 dd d(f> = 4ir

J J Vi sin 6 dd d<f> = 0
(5.9.6)
JJ ViVk sin 6 dd d 4> = A

Jj ViVjVk sin d d6 dcfi = 0

When these identities are applied to the integration over 0 and <t>,
Eq. (5.9.5) reduces to

eEk fvf_ d_ fk (5.9.7)

-4tr — ( - ^ t + /*) sin 6 dO d<t>
mv \3 dv v

Fig. 5.9 Spherical coordinate system used

x for v.
276 THE BOLTZMANN EQUATION FOR A PLASMA

The term on the left can be simplified by noting that

1 rf 2/ _ * A (5.9.8)
3v dv 1 k dv dv

Hence, Eq. (5.9.7) can be written in the form

4ireEk d_ 2,
sin 6 d9 d<fi (5.9.9)
3mv2 dv 1 k //(
The collision term on the right side of this equation (like most
collision integrals) is difficult to evaluate. Therefore, before attempt¬
ing to calculate it we shall first obtain a set of three additional differen¬
tial equations, so that we have four equations to determine /0 and the
three/i functions. To obtain these other equations we simply multi¬
ply Eq. (5.9.5) by Vj sin 9 d9 cl(j> and integrate over 9 and <f>. From
the identities in Eq. (5.9.6) we find

(5.9.10)

where j can take on the values 1,2, and 3 (or x, y, and z). We now
have enough equations to solve for /0 and the /,- functions. However,
before proceeding further we must evaluate the collision terms on the
right side of each equation.

Lowest-order results for (df/dt)c

When we use the approximate expression / = /0 + (vi/v)fi} the com¬
plete expression for the electron-neutral collision term reduces to

F gb db de d3V (5.9.11)

To evaluate this term we note first that the mass of the neutral molecule
M greatly exceeds the electron mass m and that

(5.9.12)

so that, on the average, in a collision, v » V. For this reason (df/dt)e

is calculated using an expansion procedure (outlined in detail in Sec.
A.6 of the Appendix) in which, to lowest order, the neutral molecules
are assumed infinitely massive and stationary. Thus, to lowest
order the electron in a collision is merely scattered through some
 5.9 DISTRIBUTION FUNCTION FOR A STRONG E FIELD 277

angle x, but its speed is unchanged (to conserve energy). Hence

we have

g = v = v' F(Vr) = F(V) f'o = /o /' = fi (5-9.13)

so that, in Eq. (5.9.11),

(^j = I (y' - Vi)fiFb db de d*V (5.9.14)

The integrations in (df/dt)e are now easily performed. Recall

from Eqs. (5.4.25) and (5.4.30) that

f(v' - Vi)b db dt = -vlSen<-l)(v) (5.9.15)

where Sen(l)(v) is the cross section for momentum transfer calculated

for a number of force laws in Sec. A.4 of the Appendix. In addition,
by definition,

fF(V) dzV = nn . (5.9.16)

with nn the neutral-molecule number density. Substituting these

results into Eq. (5.9.14), we find

(3/\ = _ffi (5.9.17)

\dt )c X

where by definition

(5.9.18)
\(V) = (UnSen^)-1

is the mean free path for the electrons. If we apply this result to the
collision terms on the right-hand side of Eqs. (5.9.9) and (5.9.10), we
find [after multiplying by -3y/47r and using the identities in Eq.
(5.9.6)] that

(5-9-19)

and

dfo (5.9.20)
dv s//"J («).““9=
278 THE BOLTZMANN EQUATION FOR A PLASMA

From the second equation here we see that each of the f}(v) functions is
simply related to f0(v) by the condition

_e\ E.dfo (5.9.21)

mv 1 dv

However, because of the null result in the first collision term, we shall
have to evaluate (df/dt)c to one higher order (where some account is
taken of the finite mass and velocity of the neutral molecules) in order
to complete the problem.

General expression for conductivity

Before proceeding with that calculation, we can use Eq. (5.9.21) to
obtain a simple expression for the dc conductivity of the plasma. We
recall that the current flow in a given direction is given by

Jk = —effvkdhi d3v — v2 sin 6 dO d<j> (5.9.22)

Hence, with/ = /0 + (vi/v)ft, we find [using the identities in Eq. (5.9.6)]

Jk = - ~ JQ vzfk dv (5.9.23)

If we replace fk in this equation by the expression given in Eq. (5.9.21),

we find, after an integration by parts, thatf

Jk — <to Ek (5.9.24)

with ao, the dc conductivity, given by

(5.9.25)

To complete the calculation of cr0 we must specify the interparticle

force law (so that X can be found), and for most cases we must also
know /o. However, if the interparticle force law is the inverse fifth-
power Maxwell molecule force law, the result, as usual, is independent
of/0, and cr0 reduces to the simple expression obtained in Sec. 2.2 from
the Langevin equation. To show this, we note that from Eqs. (5.4.31)

t As usual, the term Xy2/o is assumed to vanish at v = 0 or v = *, as, indeed,

it generally does, since At>2 is zero as v —> 0 and /0 usually goes to zero exponentially
as v —> oc.
 5.9 DISTRIBUTION FUNCTION FOR A STRONG £ FIELD 279

and (5.5.6) we have, for Maxwell molecules,

1 V
X (5.9.26)
V

with v the Maxwell molecule velocity-independent collision frequency

for momentum transfer in Eq. (5.5.6). Using this expression in Eq.
(5.9.25), we have

(5.9.27)

However, from the definition of the number density and the identities
given in Eq. (5.9.6), we find that for / = /0 + (Vi/v)fi

ne = // dh' = 47t Jo f0v2 dv (5.9.28)

Hence, for this case the conductivity reduces to the familiar result

<r0 = —2 (5.9.29)
mv

obtained using the macroscopic Langevin equation.

Solution for /0
For other force laws it is not possible to calculate the conductivity in
Eq. (5.9.25) unless one can find /0. However, in order to find /0 we
need to use a more accurate expression for (df/dt)c, so that the collision
term in Eq. (5.9.19) does not entirely vanish. This next-order calcula¬
tion of the collision term is carried out by a rather straightforward
expansion in powers of m/M [assuming also, from Eq. (5.9.12), that V2
is of order (m/M)v2\ in Sec. A.6 of the Appendix. Here we shall merely
state the result that to lowest nonvanishing order in the collision term
we find, in place of Eq. (5.9.19),

e_ „ d_ 2f _ _o ™ v% _ 3A2 /V c(fo\ (5.9.30)

m kdvV h M dv A 2 dv \\ dv )

with A2 = 2\iTn/M.
The determination of /0 is now straightforward. We first use Eq.
(5.9.21) to eliminate/* in the term on the left and let y = eE/m, to have

3m d^idfo _ 3A2 !L(V1 (5.9.31)

yl±(\V ‘MA
dv\ dv J M dv X 2 dv \\ dv J
280 THE BOLTZMANN EQUATION FOR A PLASMA

Integrating once on both sides with respect to v, we next find

d[o 3m iPfo _ 3 Ah3 df0 „ (5.9.32)

y2Xv
dv MX 2 A dv

The constant of integration C in this result can be set equal to zero by

noting that [for f0(v) a well-behaved function] each of the other terms
here goes to zero at v = 0. Hence, from Eq. (5.9.32) we easily find

dfo mv dv (5.9.33)
77 ~~ ~ (MX2y2/3v2) + kTh

so thatf

v' dv'
(5.9.34)
fo - C exp
MX2y2/3v'2 + kTu )
This expression for /0 is essentially our final result. The integration
in the exponential can be performed but only after the interparticle
force law (or A) is specified. Thereafter, the integration constant C can
be expressed in terms of the number density by using Eq. (5.9.28).
Finally, from Eqs. (5.9.25) and (5.9.21) we can calculate the conduc¬
tivity ao and the various /*■ distribution functions.

Results for Maxwell molecules and hard spheres

As an application of Eq. (5.9.34), let us evaluate /0 for two inter¬
particle force laws, Maxwell molecules and hard spheres. First, for
Maxwell molecules we have noted that A = v/v, with

v = 2Tr(A/m)1/2n„Ai(5) (5.9.35)

Using this expression for A in Eq. (5.9.34) and evaluating C by the use
of Eq. (5.9.28), we easily find that/0 is a maxwellian velocity distribu¬
tion of the form

nce-!,2/a*2
(5.9.36)
7r3/2a*3

but with

My2\
a *2 + 3r2 )
(5.9.37)

t Note that this result differs by a factor of 2 in the term with y2 from the dc
limit obtained for /0 when there is an applied oscillating E field. This is because
for an oscillating field (E2) = ^E2, whereas here we require no such average over
time [see, for example, H. Margenau, Phys. Rev., 69:508 (1946)].
 5.9 DISTRIBUTION FUNCTION FOR A STRONG £ FIELD 281

Thus, for this case, /0 is always maxwellian, but at a “temperature”

y* My2
Tn + (5.9.38)
3v2k

which increases rapidly with the applied field strength.

For hard spheres we have shown in Sec. A.4 of the Appendix that
= tra2. Hence the mean free path, X = (nrlSe„(1))_1, is independ¬
ent of velocity. The integration in Eq. (5.9.34) can be performed
exactly to find

U = Ce—(l + ^)"a (5.9.39)

where A = 2X272/3a2A2. However, this exact result is somewhat awk¬

ward to use; hence for strong fields one can expand Eq. (5.9.34) about
Tn = 0 to have

fo = fo(Tn = 0) + _q Tn + • • • (5.9.40)

The first term in this expansion,

Wr» = 0) = Cexp(-1^ti) (5.9.41)

is known as the Druyvesteyn velocity distribution. Conversely, for weak

fields one can expand Eq. (5.9.34) about y = 0 and easily obtain small
corrections to the maxwellian distribution found when 7 = 0.

Limits on the expansion procedure

In cutting off the spherical-harmonic expansion of / after just the first
two terms, we tacitly assume that this expansion for/ converges lapidly
on the true distribution function. As a rough test of this assumption
we can compare the magnitudes of fz and fo (when E is taken to be in
the z direction), since for a rapidly converging series this ratio would be
very small. From Eqs. (5.9.21) and (5.9.34) we easily find that

U_mXy (5.9.42)
fo xTn + (M\2y2/3v2)

This result suggests, first, that the expansion procedure converges more
rapidly for the slow electrons than for the fast ones. In addition, the
282 THE BOLTZMANN EQUATION FOR A PLASMA

convergence is apparently very rapid for all electron velocities when¬

ever m\y, the energy gain for an electron in one mean free path due co
the electric-field force, is small compared to nTn (which is two-thirds of
the thermal energy at the neutral-molecule temperature).

Problems
5.15 In the limit when Tn is negligibly small calculate a0, u (the average
velocity in the direction of E) and (v2) for an arbitrary inverse-
power interparticle force law. Show that, for a strong E field,
(v2) » w2, so that, to a good approximation,

m(c2) _ m(v2)
3k — 3k

5.16 Make a rough plot of Te as a function of y for Maxwell molecules

and for hard spheres.

5.10 THE FOKKER-PLANCK EQUATION

As we have noted in Sec. 5.1, the full Boltzmann collision term for
the rate of change of fa due to collisions between particles of type a
and of type /3 is given by

(5.10.1)

Because of the complexity of this full collision term, it is often useful

to work with some simpler collision model. In Chap. 4 and Secs. 5.7
and 5.8 we have used the Krook collision model, which is the simplest
model available. Now we shall discuss a second model, the Fokker-
Planck collision term, which we can actually derive from Boltzmann’s
expression for (dfa/dt)c when grazing collisions (where v7 = v) are of
predominant importance.
In this section we shall first present a short derivation of the Fokker-
Planck expression for (dfa/dt)c in terms of two coefficients (known as
the coefficients of dynamical friction and diffusion in velocity). There¬
after we shall evaluate these coefficients in terms of the cross sections for
momentum and energy transfer, Saf1) and Saf2), which are derived for a
number of interparticle force laws in Sec. A.4 of the Appendix. Finally,
to complete the section, we shall apply the general results to the special
case of electron-ion collisions.

Reduction of the collision term

Although we intend actually to calculate (dfa/dt)c itself, it is convenient
first to multiply both sides of Eq. (5.10.1) by an arbitrary function
 5.10 THE FOKKER-PLANCK EQUATION 283

Q(v) and integrate over all velocities to have (after an interchange of

primed and unprimed coordinates in the first term on the right)

/ Q (If), d*V = J(Q/ “ db de dh d*v' (5-10.2)

With this transformation only Q' on the right is a function of the

velocity after the collision. In this term we now expand v' about the
initial velocity v to have

Q' = Q(v) + AVi + \ Avi Avj + • • • (5.10.3)

dvi 2 dVi dvj

where Av = v' — v is the change in v due to the collision. If we substi¬

tute this expression into Eq. (5.10.2) and eliminate the terms involving
dQ/dVi and d'2Q/dV{ dvj by integrations by parts [with dV = (3Q/dvi) dv^
etc.], we find

fQ(f),d,” = ~ IQ*•(&*>-)
+ i J Q d3v (^dv9dl, Ja(AVi — • • • (5.10.4)

where

(AVi)av = ffp i A Vi gb db de d3v i (5.10.5)

and

(AVi Avj)av = Jf0i Aiu AVj gb db de d3vi (5.10.6)

Since Eq. (5.10.4) must hold for any function Q(v), we conclude
that

(5-ia7)

This is the Fokker-Planck collision term, and the quantities (Avt)av and
(AVi AVj)av are known respectively as the Fokker-Planck coefficients of
dynamical friction and of diffusion (or diffusion in velocity).
While the expansion procedure used to obtain the Fokker-Planck
collision term can, in principle, be extended to any number of terms, in
284 THE BOLTZMANN EQUATION FOR A PLASMA

practice only the two terms shown in Eq. (5.10.7) are ever used. For
this reason it is a reasonable approximation to (dfa/dt)c only when Av,
the change in velocity due to a collision, is small for most collisions.
While it is hard to state this condition very precisely, it is generally
assumed that the Fokker-Planck term is a good approximation for long-
range forces, such as the coulomb interaction between charged particles,
and a poor approximation for short-range forces, such as the interaction
of hard spheres.

Evaluation of Fokker-Planck coefficients

To continue our discussion of the Fokker-Planck equation we now
examine the dynamical-friction and diffusion coefficients defined in Eqs.
(5.10.5) and (5.10.6). We recall first, from Eqs. (5.1.11) and (5.1.12),
that

Av = x' — \ = — (g — g') mo = ma + mp (5.10.8)

m0

In addition, from Eqs. (5.4.25) and (5.4.30) we recall that

ff (gi ~ g[)b db de = glSa^1)(g) (5.10.9)

where Sapa)(g) is the collision cross section calculated for a number of

force laws in Sec. A.4 of the Appendix. Applying these identities to
Eq. (5.10.5) for (AV;)av, we find

(5.10.10)

The integration of the diffusion coefficient over b and e is also

relatively simple. However, the calculation involves a new collision
integral, which we must first evaluate, i.e., the integral

(5.10.11)

To perform this integration we note first that, from Eq. (5.4.24),

gi — g’i = Qi( 1 - cos x) - g sin x(au cos e + a2; sin e) (5.10.12)

where an and an are the dot products, a* • a, and a„ • a,-. (Note the
gz coordinate system repeated here in Fig. 5.10.) Hence, for the pro-
 5.10 THE FOKKER-PLANCK EQUATION 285

duct of (gr,- - g't)(g} - g') we find

{gi - Qi) (ft - g') = - cos x)2

+ g2 sin2 x(«ii«iy cos2 e + a2ia2j- sin2 e) + A (5.10.13)

where A includes all terms proportional to cos e, sin e, or cos e sin e.

When we substitute this expression into the integral shown in Eq.
(5.10.11), all the terms in A integrate to zero, while the terms cos2 e or
sin2 e integrate to x. Therefore if we use the identity akiakj = dtj (which
follows at once by considering the dot product of two unit vectors
a, and aj in an orthogonal coordinate system) and also recall that
azi = gi/g, we have

g2(auaij + a2ia2j) = g2{^a ~ azia3j) = g25{j — g$j (5.10.14)

Hence, for the e integration in Eq. (5.10.11) we have

(gi - g'i)(jgs ~ ft') di = Mg^Ai - cos x)2 + isin2 x(.g2*n - ftft)]

= 2tt[2^(1 - cos x) + i(g2Sij
- 3(/;0y)(l - cos2 x)] (5.10.15)

where, in the last step, we have noted that

(1 — cos x)2 = — (1 — cos2 x) + 2(1 — cos x)

Our result is now in a form that is easy to integrate over all impact
parameters. Recall that we have previously [in Eq. (5.4.30)] intro-

Fig. 5.10 The vectors g, g', and a, in

X the gz coordinate system.
286 THE BOLTZMANN EQUATION FOR A PLASMA

duced the collision cross sections

= 2x f“ (1 — cos' x)b db (5.10.16)

Hence, if we use this definition and the result obtained in Eq. (5.10.15),
we find, for our collision integral,

fj b db j*T (gfi - £')(&■ - g'j) de = 2gigjSa0{1)

+ - 3gigj)Sa^ (5.10.17)

Returning now to the evaluation of the diffusion coefficient, we

note that from Eqs. (5.10.6), (5.10.8), and (5.10.17) we have

(Avl Avy)av = (^J f fffiigi - g'i) {g, - g'j)gb db de d3vx

= f f{n[2g>g}S«f){1) + l-(g2&a - ^gxVi)Sapw]g d3vi

(5.10.18)

To summarize, we have shown to this point that if one expands (dfa/dt)c

in a series involving powers of the velocity difference, v' — v, one
obtains a Fokker-Planck equation of the form

(5.10.19)

with the coefficients (Ar,)av and (Ai>,- AVj)av given, respectively, by Eqs.
(5.10.10) and (5.10.18). These general expressions are so formidable
in appearance that one might well question whether the Fokker-Planck
collision model is any simpler to use than the full Boltzmann equation.
However in many specific cases, particularly those involving charged-
particle interactions, the final collision term can be reduced to a rela¬
tively simple result.

Application to electron-ion collisions

As a simple (but important) example, suppose we examine the

Fokker-Planck collision term due to electron-ion collisions. For this
case we have shown (in Sec. A.4 of the Appendix) that

Sapm = Cig (5.10.20)

 5.10 THE FOKKER-PLANCK EQUATION 287

where Ci = 47r(e2/47re0M)2 hi A and C2 = 47r(e2/47r€0M)2(21n A — 1). On

the average, the ion velocities are much less than the electron velocitiesf
(since (vS) = ZnTi/mi while (ve2) = 3KTe/me)\ hence we shall assume
for simplicity that the ions are all stationary and set

/s 1 = n0 S(vxi) &(vyi) 5(vzi) (5.10.21)

In addition, because of the large mass difference we also have

m0 = m, y me a02 = ae2 (5.10.22)

Using all these expressions in the equations for the Fokker-Planck

coefficients [Eqs. (5.10.10) and (5.10.18)], we obtain at once

n0C\Vj
(A Vi)a
v3
(5.10.23)

(AVi AVj)a = n0 2Cl “T +

with

bij oViVj (5 10.24)

COij — , v
V V6

Now that we have evaluated the coefficients, we return to the

actual collision term. If we write out each differentiation in Eq.
(5.10.19) explicitly and collect terms involving/, df/dvi} and d2//dvt dvh
we have

(df\ = /I d2(AVl A?7y)av

\dt )c ^ \2 dVi dVj

■ d2f
+ if (t~ (Al>i A^a
dVi \OVj
1
2 dVi dVj
(AAt^')av (5.10.25)

Each term on the right side of this equation reduces to a simple expres¬
sion when (Avj)av and (Aw,- Awy).v are given by Eq. (5.10.23). However
some algebra is required before we can obtain our final result. We
note first the following useful identities:

A = _ A hi = ill (5.10.26)
dViV3 dVj v3 dvj v v3

f As usual this assumption would break down in those rare cases when Ti/rry
is comparable to TJme.
288 THE BOLTZMANN EQUATION FOR A PLASMA

From these identities and Eqs. (5.10.23) and (5.10.24) we find that

( AVj)av (AV{ AVj)a.v 0 (5.10.27)

dVi dVi dv

while

2 n0Vi
— (AVi Avj)a (Ci - C2) (5.10.28)
dv.

Our calculation is now essentially complete. If we substitute these

expressions for the derivatives of the Fokker-Planck coefficients into
the collision term given in Eq. (5.10.25), we find

CzVj df d2/
n0
v3 dVi
+ dVi dvj
(5.10.29)

where

C3 = 3Ci - 2C2 (In A - 2) (5.10.30)

This residt is now in a form which can be used easily for many problems.
For other types of coulomb interactions, the Fokker-Planck col¬
lision term also reduces to a relatively simple form even if we cannot
assume that the j8 particles are heavy and stationary. To see this we
note first that since g = vx — v, if we have any function F(g), then

dF _ _ dF
(5.10.31)
dVi dgi

Applying this identity, together with the identities in Eq. (5.10.26)

(but with g replacing v), to the expression for the Fokker-Planck coeffi¬
cients in Eqs. (5.10.10) and (5.10.18), we easily see that for coulomb
forces, when Sa,s(0 = Cig~i,

_d_ d2
( A?^i)av (AVi AVj)av = 0 (5.10.32)
dVi di>i dVj

and

~ (Avi AvXv = -2 (^)2 f fl)1(C1 - Ci) p dhx (5.10.33)

 5.10 THE FOKKER-PLANCK EQUATION 289

If we substitute these expressions for the derivatives of the Fokker-

Planck coefficients into the full collision term (5.10.25), we have

mpC 1
(Ci — C 2) +
m0

1 d2/a / nipY2 (5.10.34)

' 2 dVi dVj \m0/

which generally reduces to a fairly simple result when fa is specified

and the integrals over vi are evaluated.

Problems
5.17 Investigate the Fokker-Planck coefficients for Maxwell molecules
(where the results are independent of fp 1) and compare the results
with Eq. (5.10.23).
5.18 Evaluate the full expression for (dfa/dt)c for electron-electron
collisions when/01 is assumed to be maxwellian. Hint: You may
refer back to Eq. (5.6.8) to handle the integrals.
chapter six

THE BBGKY
EQUATIONS
FOR A PLASMA
chapter six

6.1 THE LIOUVILLE EQUATION

In the last two chapters we have been using a statistical method for
describing a system of particles in terms ol the particle distribution
functions fa(x,\,t), where fa(x,v,t) dzx d3v is the number of particles
of type a in the volume element d3x with velocities in the range d3v at
time t. Although a knowledge of fa enables one to calculate macro¬
scopic variables such as the pressure, temperature, mean velocity, and
so forth, we have seen that some difficulties remain in this approach.
Mainly these difficulties arise in the attempt to include particle inter¬
actions or collisions in the partial differential equation for fa. The
various approaches that we have seen for handling the collision term
range from the collisionless BV equation and the relatively simple
Krook collision model to the full Boltzmann collision term and the
Fokker-Planck approximation to it. In all these, however, funda¬
mental assumptions are used (such as those delineated in our deriva¬
tion of the Boltzmann collision term in Sec. A.3 of the Appendix)
whose limitations are not easy to assess.
For this reason, in this chapter we shall introduce a different
approach (developed independently by Bogoliubov, Born and Green,
Kirkwood, and Yvonf), in which, initially, the particle interactions
are handled exactly. Needless to say, this exact treatment of the
interactions leads, as we shall see, to difficulties of other kinds that
make it virtually impossible to apply the approach to most of the
practical problems we have solved using the Boltzmann equation.
However, the method has been applied to some problems and can also
be used as a means of clarifying the approximations involved in using
the BV equation or any of the more detailed collision models. Before
embarking on our analysis of the BBGIvY equations we should note
that almost any analysis using the BBGIvY equations becomes
extremely involved. Hence, in this chapter we shall cover only the
derivation of the equations themselves and some of the general pro-

f It is for this reason that the resulting equations are called the BBGKY
equations.
 6.1 THE LIOUVILLE EQUATION 293

cedures that are used in working with them. The attempt to apply
the equations to actual problems represents an area of current research
activity which we shall not attempt to cover here.j

Derivation of the Liouville equation

To start our analysis consider a collection of N identical particles.

(The extension to a mixture is very simple, and will be handled later.)
We first introduce the N-particle probability or distribution function
f(N\ which is defined by letting

/(JV)(xi>vi;x2,v2; . . . xN,\N)t) d3Xi d3vi dzxi dhi • • ■ dzxN d3vN (6.1.1)

be the probability that at time t particle 1 ist at xx with velocity vi and

particle 2 is at x2 with velocity v2, etc. To shorten the notation we
shall often rewrite Eq. (6.1.1) in the form

N
/<"(!,2, . . . N,t) n d3xad3va (6.1.2)
a=l

Let us now consider the partial differential equation for /(iV).

If we let a be the acceleration on a particle due to both external and
interparticle forces, then, in the limit as At —■» 0, a particle at x with
velocity v moves during time At to x + v At, and its velocity changes
to v + a At. Hence the probability that the particle is at x + v At
with velocity v -(- a At at time t At exactly equals the probability
that it was at x with velocity v at time t. For this reason we can set§

/W(Xl + vi At, vi + ai At', . . . ; xa? + vw At, \n + a.v At) t + At)

= /w(l,2, . . . N,t) = /w (6.1.3)

f For a more complete discussion, the reader may consult D. C. Montgomery

and I). A. Tidman, “Plasma Kinetic Theory,” McGraw-Hill Book Company,
New York, 1964.
X Note that, as in the last two chapters, the phrase at x with velocity v is
intended to mean with position coordinates between Xj and xj + dxj and velocity
components between v,- and Vj + dvj.
N

§ To be more rigorous we should include the factor n

a = 1
d3xa dha on the right

side of this equation and the same factor (evaluated at t + A<) on the left. How¬
ever, as in our derivation of the Boltzmann equation in Chap. 4, these differential
elements differ only by terms of order (A<)2 and drop out in the limit as At —> 0.
294 THE BBGKY EQUATIONS FOR A PLASMA

If we expand the term on the left about its value when At = 0, we have

/w(xi + V! At, . . . ; t + At) = Ai

+ —— aai AM + At + terms of order (At)2 (6.1.4)

dvai ) at

Note that the summation here is over all particles (a = 1 to N) and

over each component of position and velocity (i = 1 to 3). Substitut¬
ing this result into Eq. (6.1.3) and neglecting terms of order (At)2 or
higher, we have

dfW | V dfW) = 0 (6.1.5)

dt + A dxai
a= 1 x

This result is known as the Liouville equation.

Multiparticle probability functions

Thus far we have used only f(N), the joint probability function for all
the particles in the system. If we want to find the probability that
some smaller group of particles has a particular set of positions and
velocities, all we have to do is integrate /(A0 over all possible positions
and velocities for all the other particles. Thus if we let

/(n)( 1, . . . n,t) = J /(A0

a = n +1
d3xad3va (6.1.6)

then plainly/(n) is the joint probability function for the first n particles
irrespective of the positions and velocities of all the remaining particles.
In particular, the probability function for just the first particle is

/(1)(1,0 = f f(N) n d*xad3va (6.1.7)

a— 2

This function, /(1), is directly related to the distribution function

f(x,\,t). To see this recall that f(x,v,t) d3x d3v is the number of par¬
ticles at x with velocity v. However, if all the particles are identical
and we have no way of distinguishing particle number 1 from any of
the others, then the average number of particles at x with velocity v
is just Nf(1)(x,x,t) d3x d3v. Hence Nj{l) is the average value of /, and
 6.2 THE EQUATION FOR f<>> 295

the equation for /(1) should be directly related to the Boltzmann

equation for /.

Problem

6.1 Write out the Jacobian matrix for the transformation from xa(t),
xa(t) to xa(t + At), xa(t + At), where a runs from 1 to N. You may
assume that aa (the acceleration on particle a) depends on the
positions of all particles. Show that the determinant has no
terms proportional to Ah

6.2 THE EQUATION FOR/(1)

Integration of the Liouville equation

To obtain a partial differential equation for/(I) we need only integrate

each term in the Liouville equation over the positions and velocities
of all the other particles. Referring back to Eq. (6.1.5), we have
for the first term

N
dfW df(1)
Ti = [] d3xa d3va (6.2.1)
/ dt
a=2
dt

since, as the limits are independent of time, d/dt can be taken outside
the integral. Next, consider the integration of the second term in the
Liouville equation. For simplicity we separate out the first term in
the summation over a to have

To =
J
f vu
V d-P n
dX" 8 = 2
d'x, dh» + ff
a=2 J
va% IP n
dXal 0.2
d% (6.2.2)

In each term of the sum there is an integration of the form

dfw +»
dxai = fw no summation over i (6.2.3)
/- “ dXai %r--

which is zero, since J/W) dxai must be bounded. However, as we do

not integrate over xu, the first term in Eq. (6.2.2) does not vanish,
and we find for this term

T dfw 11
n d+pd+fj
js df(1) (6.2.4)
1 2 VI i+r- = Vii
I dx u
Xlt p = 2
296 THE BBGKY EQUATIONS FOR A PLASMA

To complete the equation we now examine the integral of the

last term in Eq. (6.1.5). First, let us divide the acceleration into the
part due to external forces and the part due to the interaction with
each of the other particles; i.e., we set

aa = A„ + ^ aa7 (6.2.5)
7=1

with Aa the acceleration on the ath particle due to external forces and
aa7 the acceleration on the ath particle due to its interaction with the
7th particle. The prime on the summation over 7 is to note that the
term 7 = a is excluded from the sum, since the particle does not
interact with itself.
For the term involving Aa, the external force, we again separate
out the first term in the summation to have

N N N
T 3,ext = f dl,^1

11
n
/9 = 2
d3Xgd3V0 +
0
2f = 2
df™
dVai
[]
0 =2
d3xg d3v0

(6.2.6)

If we assume that A^- is independentf of vpi, then each term in the sum
again involves an integration of the form

+"
dvai = fw no summation over i (6.2.7)
dl’ai voti = — 00

which is again zero if //(A0 dvai is bounded. Since we do not integrate

over vu, the first term in Eq. (6.2.6) does not vanish; hence taking
the derivative out of the integral, we find

d/(1)
T 3,ext — Au (6.2.8)
dvu

Finally, for the term involving the interaction of particle a with

all the other particles we have

T3,int = ^ ^ / aayi~~ [] d3xpd3vp (6.2.9)

a=i 7=i J m 0=2

t Note that the only common velocity-dependent force is that due to a

magnetic field. However, for this case, as noted earlier in the text, the force is
proportional to v X B, so that, for example, Ax is proportional to vyBz — vzBv.
Hence, even for the magnetic force At is independent of vt.
 6.2 THE EQUATION FOR f<» 297

The acceleration on particle a due to its interaction with particle 7

generally depends only on the distance between the two particles.
However, with no additional difficulty we can also permit a„7 to depend
on the components of va and v7 which are perpendicular to aa7. How¬
ever, we will assume that aayi is independent of vtti and vyi (the velocity
components in the direction of the acceleration). With this assump¬
tion, if we separate out the term in Eq. (6.2.9) where a = 1, we find

N
dfw
3,int CLiyi
dVu
P[ d3xp d3Vp (6.2.10)
0 =2

since for any other choice of a we obtain another integration like that
indicated in Eq. (6.2.7), which is zero. In Eq. (6.2.10) we can again
take the derivative out of the integral and integrate over all coordi¬
nates except those of particle 1 and particle 7 to find

Tt,int = £ / aiyif(2)(l,y,t) dzxy d3vy (6.2.11)

7 = 2 11

with /(2) the two-particle joint probability function.

For those cases when both the interaction force and the two-
particle probability function have the same form for all 7, that is,
when all the particles are identical, each of the N — 1 terms in the
summation over 7 is equal, and we have

Tz,int = (2V — 1) \ W(2>(1,2,0 d3x2d3v2 (6.2.12)

oVu J

In addition, in many treatments of the BBGKY equations it is assumed

that both N and the volume V approach 00 but with the concentration
N/V held fixed. Hence the 1 in Eq. (6.2.12) is neglected compared
to N.
In summary we see that upon integrating the Liouville equation
over all coordinates except those of the first particle, we obtain, from
the sum of Eqs. (6.2.1), (6.2.4), (6.2.8), and (6.2.11),

a/(1) + „ufH + Aufl+ j*f = 0 (6.2.13)

dt dxii ovu
v=2
oVu J
298 THE BBGKY EQUATIONS FOR A PLASMA

In addition, for a system of N identical particles the last term here (in
the limit as N and V —> °°) reduces to

(6.2.14)

For this case, as there is no longer any need to specify any particular
particle, we can also replace xi, vi and x2, v2 by x, v and x', v', so that
the equation for /(1)(x,v,<) becomes

4- [ ajw d*x'd (6.2.15)

dVi J

with a,; the acceleration on a particle at x with velocity v due to another

particle at x' with velocity v'.

Discussion of the equation

This result, which is known as the BBGKY equation for/(1), bears a

marked resemblance to the Boltzmann equation. The first three
terms, in fact, are identical in form to the left-hand side of the Boltz¬
mann equation. However, the collision term in the Boltzmann equa¬
tion (which is where all the approximations and assumptions were
required) has been replaced now by the interaction term involving a,.
This term, like the Liouville equation itself, involves none of the
assumptions or approximations used in deriving the Boltzmann equa¬
tion. However, the penalty we pay for being so rigorous is a severe
one, since this equation for /(1) involves both /(1) and /(2). In similar
fashion, if we attempt to obtain a differential equation for /(2) (by
integrating the Liouville equation over the coordinates of all the
particles except the first two) the result involves /(3), and so forth.
Thus unlike the Boltzmann equation, which, while complicated, is at
least solvable in principle, this set of equations obtained from the
Liouville equation always contains one more function than there are
equations unless we simultaneously solve for all the /(n)’s from n = 1
to n = N.
This difficulty is reminiscent of the problem encountered in dealing
with the moment equations in Chap. 5; and, just as there are expansion
procedures which we have used for dealing with the moment equations,
there are similar methods available for obtaining solutions to the
BBGKY equations. In the following sections we shall look into some
of these approximation procedures and show, in particular, that they
provide some justification for both the BV equation and the Fokker-
Planck equation for a plasma.
 6.3 THE BV EQUATION WITH SELF-CONSISTENT FIELD 299

Problem

6.2 Prove (by integrating the Liouville equation over all variables
for the particles from n + 1 to N) that the BBGKY equation for
an arbitrary /(n) is given by

n N

where aay is the acceleration on particle a due to its interaction

with any particle 7(7 9^ a) and /(n+1) is a function of the coordi¬
nates of the first n particles plus particle 7.

6.3 THE BV EQUATION WITH SELF-CONSISTENT FIELD

We have already noted that the fundamental difficulty involved in

the BBGKY hierarchy of equations is that (except for when we use all
N joint probability functions) there is always one more probability
function than there are equations. Hence some scheme must be used
to truncate, or cut off, this inevitable dependence of /(n) on /(n+1). A
formal expansion procedure for eliminating this difficulty is outlined
in Sec. 6.0. However, to start, we shall first use some less formal (but
more physical) schemes for achieving this truncation of the set of
BBGKY equations. In the process we shall see that the resulting
equations provide, for the case of an infinite plasma, some justification
for both the BV equation, used in Chap. 4, and the Fokker-Planck
equation, which we discussed in Sec. 5.10.

Truncation of the fi equation

Probably the simplest method one can use for working with the
BBGKY equation is to assume a functional relationship between /(21
and /(1) in the interaction term in the /(1) equation. When this
approach is used, the assumption generally made is that the probability
functions for the two particles are completely independent. For this
case, as we have noted in our discussion of the Boltzmann collision
term in Sec. A.3 of the Appendix, the joint probability that particle 1
is at xi with velocity Vi and particle 2 is at X2 with velocity V2 is just
the product of the two individual probabilities so that

/2(1,2 ,t) =/<»(!, 0/(1)(2,0 (6.3.1)

300 THE BBGKY EQUATIONS FOR A PLASMA

(Note that this is like flipping two coins and asking for the probability
that both turn up heads. If the behavior of each coin is independent
of the other, then this probability is just the product of the two sepa¬
rate probabilities, | X i = }■) If we use this assumption, we imme¬
diately reduce Eq. (6.3.1) to an equation involving just f(1), so that
the truncation of the set of equations is achieved.
One interesting result which follows from this method of trunca¬
tion is that one can reduce the /(1) equation for a plasma to the BV
equation, used in Chap. 4. To see this, let us consider the simple case
of a plasma of N identical electrons in a uniform background of sta¬
tionary ions with number density N/V (to maintain overall charge
neutrality). If we consider just the electrons, then as N and V
approach infinity, we have, from Eqs. (6.2.15) and (6.3.1),

^ iP + ^ [Ai + NfafHx'y) d*x' dV] = 0 (6.3.2)

dt dXi &Vi

Because of the infinite volume we shall neglect any real external forces.
However, because of the ion distribution, we set

A = — — Eion (6.3.3)
m

where Eion is the electric field produced by the ions, so that (from
Poisson’s equation)

eN
V • Eion = (6.3.4)
V £q

Evaluation of the interaction term

To evaluate the interaction term in Eq. (6.3.2) we note first that for
electrons moving slowly (compared to the speed of light), to a very
good approximation,

e2R
(6.3.5)
Aire^mR3

with R = x — x' the vector pointing from x' to x as shown in Fig. 6.1.
If we now substitute this expression for a into Eq. (6.3.2) and multiply
through by N, we have|

e_df_
dl + v .v Ei I f«y) Ri d3x' d3v' = 0 (6.3.6)
dt ^ 1 dXi m dVi 4 TTti R 3

f Recall that A/(1) is the average value of the usual distribution function /.
However, to reduce the notation we shall here let / d3x d3v be the average number
of particles at x with velocity v.
 6.3 THE BV EQUATION WITH SELF-CONSISTENT FIELD 301

Comparing the last term here with the analogous term —{e/m)Ei
df/dVi in the BY equation, we see that they are equal if

e Ri
Et = E*°” - ~ //(<>') d3x' d3v' (6.3.7)
47T€o R3

Since E must satisfy Poisson’s equation, V • E = pc/eo, our proof of

Eq. (6.3.7) will be to show that

pc
Eic d3x' d3v' (6.3.8)
47re( / R*

To start, note that if R = |x — x'|, then one easily finds that

1_ R
V (6.3.9)
R R3

Hence, in the term on the left side of Eq. (6.3.8) we have

V • Eion + v— f
47T6o J
v2 4
H
d3x' dZv' (6.3.10)

Now we have shown (in Sec. A.l of the Appendix) that for an infinite
volume

V2 — 5(x — x') (6.3.11)

4ltvR

Fig. 6.1 Position vectors for

particles at x and x'.
302 THE BBGKY EQUATIONS FOR A PLASMA

where 5(x — x') is the three-dimensional delta function satisfying the

three conditions

n ( x x'
1. 5(x “ x } = (
0
00 = x'
(6.3.12)

f 1 x in the domain of
(6.3.13)
2. /S(x — x') dzx' = <
u /(x)
integration
otherwise
x in the domain of
3. J/(x')5(x - x') d*xr = ■ integration (6.3.14)
_ 0 otherwise

If we use Eqs. (6.3.4) and (6.3.11) in Eq. (6.3.10), we have, after

integrating over x',

V • Eion + ~r~— f /(x',v') V2 4 dzx' dV

47re0 J n

= e^~ ~ - [ /(x,v0 dV (6.3.15)

V to Co j

We easily recognize that the expression on the right is the average

charge density at x divided by e0. Hence Eqs. (6.3.7) and (6.3.8) are
verified, and we may rewrite Eq. (6.3.6) as the BY equation

= 0 (6.3.16)
dt ' dXi m di>i

with E the solution to Poisson’s equation

T-E-S *-*(7-//<**) (6.3.17)

In summary, we see here that the BV equation for an electron gas can
be derived from the BBGKY equation for /(1) when correlation effects
are neglected. Thus, although it includes 110 collision term explicitly,
the BV equation actually includes (in the E-field term) some of the
interparticle effects.

Problem

6.3 Derive the equation for /(1)(1) f°r an arbitrary collection of N

distinguishable charged particles and show that if one sets
/(2)(l,y) = /(1)(1)/(1)(t), one obtains a BV equation but with E
the field due to all particles except particle 1.
 6.4 CORRECTIONS TO THE BV EQUATION 303

6.4 CORRECTIONS TO THE BV EQUATION

Truncation of the/2 equation

Thus far we have described only the simplest possible method of

truncating the set of BBGKY equations, that of assuming that /(2) is
just a product of two one-particle probability functions. If we want
to take some account of correlation effects, one way to do this is to use
both the/(1) and/(2) equations but to truncate the set of equations at
that point by assuming a functional relationship for/(3) (which appears
in.the interaction term of the /(2) equation) in terms of /(1) and /(2).
When the truncation is achieved in this manner, the usual assumption
is to setf

/<3>(a,/3,7) =/(2)(a,d)/(1)(7) +/(2)(a,7)/(1)(|8) + /(2)(d,7)/(1)(«)

- 2/<»(a)/<1W1)(7) (6.4.1)

This expression for /(3) can be derived by means of a formal

expansion procedure for solving the set of BBGKY equations, as
shown in the next section. However, the result is also quite plausible,
since we certainly expect that the probability for three particles to
have some prescribed set of coordinates should be related to the joint
probability for any two of the particles to have their prescribed coordi¬
nates times the probability that the third particle also has its pre¬
scribed coordinates. (That is, /<3) should be related to the product
/(2)jd).) The least biased form thus involves a sum of all the per¬
mutations of the three particles among the probability functions /(1)
and /(2). This sum, by itself, is plainly too large, since/(3) must satisfy
the additional requirement that in the limit when the particles are
uncorrelated, /<»(a,/3,7) -> /(1)(a)/(1)(0)/(1)(7)- It is for this reason
that we subtract off the term 2fa) (a)f<-1) (P)f(1) (y).
When this expression for /(3) is used in the /(2) equation, the
resulting set of /(1) and /(25 equations is extremely difficult to work
with. However, it is perhaps worth noting the form that these equa¬
tions take in comparison with the simple BV equation obtained when
correlations are neglected.
To start we note that, with no approximation, one can let

/(2)(«,/3) = /(1WX)(/3) + P(a,P) (6A2)

where P is known as the pair correlation function. If we use this
expression for in the interaction term of the first BBGIvV equation,
f We shall see in the next section that this is equivalent to neglecting three-
particle correlation effects.
304 THE BBGKY EQUATIONS FOR A PLASMA

then for the simple electron gas we considered in the last section we
find [from Eqs. (6.2.15) and (6.3.7)] that

% + t.*L-±El*L d3x' dV (6.4.3)

dt, ' dxi m dVi

In this equation P is the correlation function for two electrons at x

and x' with velocities v and v', respectively, and / = Nf(i) is again
the average value of the distribution function.
The left side of Eq. (6.4.3) is just the BV equation, the result
obtained in the last section when we set P = 0, while the term on
the right is the correction due to correlation effects. Note that it is
this term which replaces the usual collision term in the Boltzmann
equation for a plasma.

Equation for the correlation function

Since we have expressed the correction to the BV equation in terms of

the correlation function, we next investigate the equation governing P.
To obtain this equation, we shall for simplicity neglect external forces;
however, other than this we shall initially use the full equations for
/(1) and /(2) (rather than the limiting forms when we assume all the
particles are identical and let N and V approach °o). From Prob. 6.2
recall that the equation for/(1) is

/(1)(«) + vai + T [ aaytfW(a,y) d3xy d°vy = 0 (6.4.4)

OXai ^ uL'ai J
7=1

where, again, the prime denotes that the term y = a is omitted in the
sum; while, from the same problem, the equation for/(2)(l,2) is

/(2) + ^t \_Vai dXai + aafK foai + X dVa. f Ctcyi/^il^y) d Xyd3Vy 3 = 0

=1L 7=3
0 7*cr

(6.4.5)

(Note that in the third term here, when a = 1, /3 = 2, and vice versa.)
In the first two terms of Eq. (6.4.5) we now replace/(2) by/(1)(1)/(1)(2)
+ P(l,2). In the first term, we have

/(2) = £ [/(1)(«)/(1)(fl] + P(1.2) (6.4.6)

«=l
/9
 6.4 CORRECTIONS TO THE BV EQUATION 305

Similarly, for the second term we have

V df&
dP‘ V
> vai -— = > vai /(D(/3) dfa)^ + — (6.4.7)
Lt
a=l
& 9^a
dXai
/S
Lt
(=1 ^ ' dxat ^ dxai

Now note that if we add together just the two underlined terms in
Eqs. (6.4.6) and (6.4.7), which are both proportional to /(1)(/3), we
have, from Eq. (6.4.4),

f} 7^ a
2 N

= - ^ 2 / a«u/(2)(a,7) d3x7 d3^7 (6.4.8)

a= 1 7=1
/3 p^or

To obtain the full equation for P, we replace the first two terms in
Eq. (6.4.5) by the expressions given in Eqs. (6.4.6) to (6.4.8). In
addition, in the summation on the right side of Eq. (6.4.8) we separate
out the term where y = P, so that the remaining terms run from
y = 3 to 7 = N. With these substitutions Eq. (6.4.5) becomes

a= 1 ' - —
0 ?*«

+ 2 dT-J
7=3
«^l/H>(«,S,T)-/<I>((3)/<!'(a,7)]<i%4xJ -0 (6.4.9)

If we now use Eq. (6.4.1) to eliminate/(3) and recall the definition

of P in Eq. (6.4.2), we have, in the final term of Eq. (6.4.9),

/<»>(«,/3>7) - /(1W2)(«,7) = /(1)(a)P(iS,t) + /(1)(t)P(«,/3) (6.4.10)

so that Eq. (6.4.9) now involves only/(1) and P (or /(2)). Even with
this truncation, however, the equation is still too complicated to use.
Hence two additional simplifications are generally made. First, in
the term (with a single underline) involving df(2)/dvai in Eq. (6.4.9) it
is assumed that correlation effects are small, so that

df« d/(1)(a) (6.4.11)

daffi aW(1)(«
dVai dVai
306 THE BBGKY EQUATIONS FOR A PLASMA

The argument here is based on a calculation of P and /(1) in equilibrium

(which we shall verify in Prob. 6.4) where one finds that /a,(Q!)/(1,(^)/
P(a,0) is of order ND, the number of particles in a Debye sphere, f
Second, we note that the term with a double underline in Eq. (6.4.9) is
identical to each of the N — 2 terms in the very last part of the last
term except that in the integration /3 is for particle 1 or 2 while y runs
from 3 to N. Hence, in the usual limit when N —> °° and all or
half if the ions are also considered—the particles are identical electrons,
this term is of order 1 /N compared to the entire sum and drops out.
Using these two assumptions and Eq. (6.4.10), we obtain finally

P + ^ | Vcti dP + W'W df"(a)

dx dVc
a= 1 '
0 r^a.
N
+ £ faayi[f(1)(.«)P(P,y) +/(1)(y)P(a,d)]d3^ dzvy} = 0

7=3
(6.4.12)

where for a system of identical particles the final summation is simply

replaced by the factor N — 2 = N times a single integral.

Problem

6.4 Show that for an electron gas in equilibrium (/ = P = 0) a solu¬

tion to Eqs. (6.4.4) and (6.4.12) is

g—J)a2/a2
/(1)(a) 7r3/2Ea3

and

ele~
Peq(a,P) = -/(1>(a)/(1>(d) R = Ix/s — xa
4-ireoKTR

A De~-RIXD
-/(1)(«)/(1) 08)
"3

Note that as usual,

a2 = 2 kT/wi Ad2 = KTto/nee2 Nd = %neTr\D3

with ne = N/V.

f The ratio is actually a function of the distance between the two particles and
becomes much smaller at small separation distances, which is where one would
expect correlation effects to be large. Hence, as we shall note shortly, it is some¬
times important to retain the full expression for /(2) in this term.
 6.4 CORRECTIONS TO THE BV EQUATION 307

Application to an infinite homogeneous plasma

Equations (6.4.4) for/(1) and (6.4.9) or (6.4.12) for P are used as the
basis for many current papers in the plasma physics literature. How¬
ever, as they cannot be solved exactly for most problems, they are
usually simplified by further approximations. For example, in many
papers the equations are applied to an infinite homogeneous medium, so
that/(15 is independent of position.
With this assumption it is apparent that dfw/dxai is zero. How¬
ever the assumption also leads to another, even more important result,
namely, that the average interparticle force on each particle is zero.
On physical grounds, this arises because with /(15 independent of posi¬
tion, the electrons are distributed smoothly over the entire volume (as
are the ions), so that the net charge density is zero everywhere, and
there is no electric field acting on the charges. To prove the result
formally, we note that since the integral of /(1) over all velocities and
positions must equal 1, it follows that for a homogeneous medium with
volume V (where V is allowed to approach °o)

1 = J/<«(7) d*xydzvy = F//(1) (7) dhy (6.4.13)

or

J/<«(7) dhy = F-1 (6.4.14)

Hence, for a plasma where aayi — — qaqyXayi/^irtomaX ayZ, we have (with

a change in variable from \a to xay = x7 — x„)

ff(1)(y)ac'yid*Xy d3vy = V-lSaayid*xay = 0 (6.4.15)

which vanishes when the limits approach 00, since the integrand is an
odd function of Xayi•
Because of these identities, the equations for /(1) and P [Eqs.
(6.4.4) and (6.4.12)] reduce to just
N

/(D(l) = - ^ / aiyip(.l,y) d3Xyd3vy (6.4.16)

and
2 ,
dP , d/««(a)
2
+ a= 1 '
& 9^a
Vai
dXa
— +
dv0
[f(l)(P)aa0i

+ 2V / a«yiP{P,y) dHydhyy, = 0 (6.4.17)

7=3
308 THE BBGKY EQUATIONS FOR A PLASMA

These equations, while still quite complicated, can be used for some
plasma problems. Probably the most notable work has been the reduc¬
tion of the equations to a Boltzmann-like equation for /(1) which is
called the Lenard-Balescu equation.f
In another interesting paper, TchenJ has used the full expression
for/(2) in the single-underlined term in Eq. (6.4.9), so that the third
term in Eq. (6.4.17) is, in his work, aapi dfw/dvai. Thereafter, by
solving Eq. (6.4.17) for P in terms of/(1) and substituting the result into
Eq. (6.4.16), he obtains (after a number of approximations) the Fokker-
Planck equation discussed previously in Sec. 5.10.
In this derivation, which is too involved to repeat here, each of the
terms in Eq. (6.4.17) plays a distinctive role. For example, in the
third term, where

a/(1)(«) dP(a,P)
/C1)(d) (6.4.18)
dVai dvai

the terms involving /(1)(a)/(1)(d) produce the main portion of the

Fokker-Planck collision term, while the term involving P(a,d) (which,
as we have noted, is omitted in most analyses) prevents certain diverg¬
ence difficulties at short interparticle differences. In addition, the last
term in Eq. (6.4.17) dominates at large interparticle separation distances
and produces a Debye shielding effect in a way which is more rigorous
than simply cutting off the interaction between particles at the Debye
length.

SUMMARY
To summarize we have seen here that it is possible to extend the simple
BV equation obtained by assuming/(2) (a,/3) = /(1)(«)/(1)(/3)- Toobtain
the extension one must take some account of correlation effects. For¬
mally this can be done quite generally, using Eqs. (6.4.4) and (6.4.9)
for/(1) and P but replacing/(3) by Eq. (6.4.10). As a practical matter,
however, we have seen that such an analysis is extremely complicated
algebraically even for the relatively simple case of an infinite homo¬
geneous medium. It is possible, however, that future developments in
this field of research will provide techniques by which analytical results

f The standard references for this work are R. Balescu, Phys. Fluids, 3:52
(1960) and A. Lenard, Ann. Phys., 10:390 (1960). The derivation can also be
found in two recent texts: Montgomery and Tidman, op. cit., and T. Y. Wu,
“Kinetic Equations of Gases and Plasmas,” Addison-Wesley Publishing Company,
Inc., Reading, Mass., 1966.
\ C. M. Tchen, Phys. Rev., 114:394 (1959).
 6.5 FORMAL PERTURBATION TECHNIQUE FOR WEAK CORRELATIONS 309

can be obtained from the BBGKY equations without these algebraic

difficulties.

6.5 FORMAL PERTURBATION TECHNIQUE FOR WEAK CORRELATIONS

The Ursell-Mayer expansion

In Sec. 6.2 we noted that the BBGKY equation for the rate of change
of the n-particle joint probability function has the form

with /(n+1) a function of the coordinates of the first n particles plus

those of particle 7. Since the equation for /(n) involves /(n+1), the
system of equations always involves one more unknown function than
there are equations (except when we work with all N of the equations).
In the last two sections we have described two ad hoc procedures for
truncating the set of equations. Now, to complete this very brief
survey of the BBGKY equations, we shall describe a more formal
method for truncating the set of equations but one which leads to essen¬
tially the same results as we have obtained by our ad hoc procedures.
The formal technique may be described as a weak-con elation expan¬
sion procedure, since it is based upon the assumption that the correla¬
tion effects between any two particles are small. To formulate this
idea in a concrete way, we introduce a full set of correlation functions
gi2, g 123, . . • , which are defined by letting

/(2)(1,2) = /(1)(1)/(1)(2)(1 +012)

/(3) (1,2,3) = /(1) (1)/(1) (2)/(1) (3) (1 + 012 + 0i;i + 023 + 0123)

=1 f)=a+l 7 = 0+1
In this system of equations, which is called an Ursell-Mayer cluster
expansion, there are three points to note: (1) we have made no approxi-
310 THE BBGKY EQUATIONS FOR A PLASMA

mations or assumptions, as these equations merely serve to define the

various correlation functions; (2) from Eq. (6.4.2) we see that ga$ is
related to P, the pair correlation function, by the condition

_ P(a,P) (6.5.3)
9a* /(1)(«)/(1)(d)

Hence, the equilibrium calculation discussed in Prob. 6.4 suggests that

gal3 is typically of order ND~' and therefore very small in most plasmas;
and (3) we should also note that the introduction of the correlation
functions in this manner is done so as to automatically satisfy the
condition that any two particles are uncorrelated if the distance between
them approaches infinity. If we demand that any correlation function
is zero when the distance between one particular particle (say, particle p)
and all the rest approaches infinity, we easily find that, for this case,
from Ecp (6.5.2),

/(n)(l,2, . . . ,n) =/(n_1)/(1)(p) x<*p 00 for all a (6.5.4)

where /(n-1) is a function of the variables of all the particles except

particle p. Hence, this type of expansion automatically satisfies the
requirement (imposed on physical grounds) that/(n) is just a product of
the separate probability functions/(n_1> and /(1)(p) when particle p is
far away from all the other particles.
These cluster-expansion formulas also help to clarify the truncation
procedure used in Secs. 6.3 and 6.4. From the first part of Eq. (6.5.2)
it is evident that the truncation of the /(1) equation in Sec. 6.3 was
achieved by neglecting the two-particle correlation function <712 in the
full expression for/(2). Similarly, in Sec. 6.4 the truncation of the/(2)
equation was achieved by neglecting just the three-particle correlation
function <7123 in the expression for/(3).

The formal expansion procedure

Starting from the cluster expansion shown in Eq. (6.5.2) there are a
number of formal approximation schemes that one can use to truncate
the hierarchy of equations. One which requires a minimal number of
assumptions is to assume that gal3 is of order Nd-1 and that ga$y is of an
even smaller order. The first of these is at least supported (except at
short ranges) by the equilibrium calculation in Prob. 6.4, but the
assumption about gm is one which we shall use without any real justi¬
fication beyond the feeling that three-particle correlations should be
less important than two.
 6.5 FORMAL PERTURBATION TECHNIQUE FOR WEAK CORRELATIONS 311

In any event, with these two assumptions we note that if we

neglect terms of order Nd~\ the first equation in Eq. (6.5.2) reduces to

/(2)(«,£) = /(1)(«)/(1)(|8) (6-5.5)

which is precisely the uncorrelated expression which, as we have seen in

Sec. 6.3, reduces the /(1) equation in the BBGKY hierarchy to the
BV equation.
If we next retain terms to order ND~l, while neglecting those of
still lower order, we have, from the first two equations in (6.5.2),

/<«(a,d) = /<»(a)/(1>(|8)( 1 + ga?) 5

/(3)(a, 18,7) = /(1)(aO/(1)Q3)/(1)( y)(l + gat3 + gp-1 + gay)

If we replace the g’s in these equations by pair correlation functions

as given by Eq. (6.5.3), it is evident that these are now the equations
which we used in the last section to truncate the/(1) and/(2) equations.
With no further assumption, the reader will recall [from Eqs. (6.4.4) to
(6.4.10)] that the use of Eq. (6.5.6) in the BBGKY equations leads to
the equations

sr (6.5.7)
/(1) + Var
dX ai
Y f a“^<2)(a’T) d*Xy d3Vy

and

P+j
L,
L~+a.fi^
dXai OVai
-/“‘(ft 4- I a„eir(c,fi)d*x,d>ve
OVai J
a=1 '
P 7^a

+ f J- Jf aayi[fllK«)P(fi,y)
dVai
+f“Ky)P(c<,P)}d3Xyd* V-y] = 0
7 =3
(6.5.8)

There is one important difference between the present approach

and the ad hoc method used in the last section to obtain Eqs. (6.5.7)
and (6.5.8), namely, that we now have specifically assumed that

Nd = imr\D3 » 1 (6-5,9)

and we have eliminated by assuming that it is of a lower order than

Nd-h Hence to be consistent we should also examine the order of each
312 THE BBGKY EQUATIONS FOR A PLASMA

term in Eqs. (6.5.7) and (6.5.8) to make certain that we eliminate any
other terms that are of lower order than iV#-1. For this purpose we
shall apply some simple dimensional analysis to examine the relative
magnitudes of the various terms in these equations.
As is always the case, the choice of a “characteristic” dimensional
quantity is rather arbitrary. However, for the electron gas to which
we generally apply these equations, it is at least plausible to use
the inverse of the electron plasma frequency, as a characteristic time
and Ad, the Debye length, as a characteristic distance over which /(1)
or P might change. (Recall that we have, in fact, already used Ad as
a characteristic length when we assumed gals was of order No-1.) To
obtain a characteristic velocity, we recall that, as shown in Eq. (2.4.20),

(6.5.10)

which is of the order of the average thermal velocity for the electrons.
We emphasize again that for some problems (involving beams, for
example, where the beam velocity greatly exceeds vth, or involving
structures with dimensions smaller than Ad) we might well use a different
set of characteristic values. However, the choice we are using here is
the most reasonable one for large relatively homogeneous stationary
plasmas.
With these characteristic values for time, length, and velocity, we
can now estimate the order of each term in Eqs. (6.5.7) and (6.5.8).
In the /(1) equation, for the first two terms we have

(6.5.11)

so that these terms are of comparable importance. Next, for the inter¬
particle force terms, note that for a coulomb force (assuming the dis¬
tance between particles is of order Ad) we have

<9/(2) g2«2)

aayiW~- ~ -7 2- (6.5.12)
dVai 4:irme0\D2Vih

To put this into a form which can be more easily compared with
the other terms, we use Eq. (6.5.10) to eliminate vlh and recall that
we2 = ne2/me0, so that Eq. (6.5.12) reduces to

(6.5.13)
 6.5 FORMAL PERTURBATION TECHNIQUE FOR WEAK CORRELATIONS 313

This shows that, a single interaction term is of order Nd~1 compared to

the other terms in the equation. However, as the term on the right
side of Eq. (6.5.7) involves a summation over the interactions with all
the other particles, it is obvious that the entire sum might still be of
order or even greater. Hence there is no reason to omit any term in
the/(1) equation (6.5.7).
The situation is a little different in the equation for P. Here,
since P(l,2) is of order ND~l times/(1)(1)/(1)(2), we have, for the first
two terms in Eq. (6.5.8),

dP Vth
P '—' coeP P CO ,p (6.5.14)
’ dXai

so that these two terms are of the same order, but both are of order
ND~l compared to such terms as coe/(1)(a)/(1)(/3). Of the remaining
terms in Eq. (6.5.8) all involve the operator

d We (6.5.15)
dafU
dV0 3 ND

Hence when this operates on P, it produces a term of order ue/3Nz>2,

which we should neglect (unless, as in the last term, we have a summation
of N such terms). As a consequence, if we neglect the terms of order
Nd~2 in Eq. (6.5.8), as indeed we must if we want to be consistent with
our omission of the gapy term, then Eq. (6.5.8) reduces to

^+1 a—1 1
P
{■ dXai
+ /(1)(£) d~^~- (a«* - <a^»
OVai

N 1
= 0 (6.5.16)
+ I aT / a“^(1)(a)p^’T) +f(1)(i)PM]d3xyd3v^
7=3

where

(aapi) = d3xpd3v^ (6.5.17)

is the average acceleration on particle a due to particle ft.

Equation (6.5.16) differs slightly, for nonhomogeneous media, from
any of the equations for P discussed in the last section. However,
when we go to the limit of an infinite homogeneous plasma, we have
seen [in Eq. (6.4.15)] that the term (a«#) goes to zero, as does the similar
integral in the last term of the y summation. Thus we recapture Eq.
314 THE BBGKY EQUATIONS FOR A PLASMA

(6.4.17), the equation used by Lenard and Balescu (as well as many
other workers). In summary, what we have seen in this section is that
an ordering of terms with respect to Ncan be used to obtain a pair
of equations for/(1) and P, rather than merely relying upon the ad hoc
procedures of the last sections. Although modifications of this sort
provide a somewhat more satisfactory basis for solving the BBGKY
hierarchy of equations, it is clear that they are still far from being
rigorous or complete. Hence these kinetic equations must be treated
as tentative until they have either been tested against more experi¬
mental results or put onto a more sound theoretical base.
appendix

MATHEMATICAL
APPENDIX
appendix

A.l DELTA FUNCTION AND FOURIER TRANSFORM THEORY

The Dirac delta function

In the problems treated in this text we shall find it useful to employ

the Dirac delta function, 8(x — x'), which can be any function having
the following properties:

( 0 for X ^ X'
5
1. 8(x — x') = (A.1.1)
j 00 for x = x'

2. [X+€
Jx-t
S(x - x') dx' = 1 (« > 0) (A.1.2)

rh 7(*) a < x < b

8. fa f(x')8(x — x') dx' = • a < b is assumed (A. 1.3)
x < a or x > b

Although there are many possible representations for the 5 function,

probably the most useful form is to let

8(x — x') = lim^- f °° e-cc\k\eik(x-x’) dfr (A. 1.4)

a—>0 27T J ~“

This integral can be easily evaluated by setting

eik(x-x’) _ cos _ x') -(- { sin k(x _ x(A. 1.5)

The part of the integrand proportional to the cosine is even, while the
term proportional to the sine is odd; hence we have

. 1 /* 00
8(x — x') = lim - / e~ak cos k(x — x') dk
a—*0 TT

(A.1.6)
a-*0 7T «2 + (x — x')2
 A.l DELTA FUNCTION AND FOURIER TRANSFORM THEORY 317

Verification of the delta-function properties

We next need to verify that this representation actually satisfies the

three properties of a delta function. The first property is clearly satis¬
fied simply by taking the limit indicated in Ecp (A. 1.6), first for x 9^ x'
and then for x = x'. To verify the second property note that, with
the change of variable z = x' — x, we have

a fx+t dx' __ a fe dz _ 2 1
(A.l.7)
TV Jx-< a2 + (x - x'Y ~ TV J-e a2 + z2 “ tv 1111 a

Now, for e > 0,

lim tan 1 - = ^ (A. 1.8)

so that the integral of this delta function does equal unity.

To verify the third delta-function property we note, first, that

}(x')a (A. 1.9)

F{x') = lim 0 X 9^ X
0 Tv[a2 + (X — X')2]

provided f(x') has no singularities. Hence the integral of this function

is zero if the point x' = x is not included in the domain of integration.
For the case when we do integrate over the point x1 = x, we can first
separate the integral into three parts

[x+e F(x') dx' + fb F{xf) dx’

[* € F(x') dx' + Jx-e Jx+t
(A.l.10)

where F(x') is given in Eq. (A.l .9) and e is a very small positive number
which is larger than a but which also approaches zero. F rom Eq.
(A.l.9) the first and third integrals vanish, while in the second integral
we again let z = x' — x, to have

af(z + x) (A.l.11)
lim dz
0 7r(a2 + z2)

We now integrate by parts, letting fU dV = UV\ - JT dU, where

a dz (A.l.12)
jj = f(z + x) and dV
tv (a2 + z2)
318 MATHEMATICAL APPENDIX

so that

dU = f'(z + x) dz V = - tan-1 — (A.1.13)

7r a:

With this integration we have

J* Fix') di

since tan_1(2/a) equals +(71-/2) in the limit as a—> 0, depending on

whether z is positive or negative. Hence, taking the limit now as
e —> 0, we obtain the desired result

(A. 1.15)

Application to Fourier transform theory

Having analyzed the <5 function representation shown in Eq. (A. 1.4)
with some care, we shall henceforth follow the customary practice of
taking the limit inside the integral and setting

(A. 1.16)

Although this procedure is not strictly valid, it does simplify the nota¬
tion a great deal, and for all problems of physical interest it leads to the
same results that one would obtain using the limit shown in Eq. (A. 1.4)
or (A.1.9).
As an example of the usefulness of this <5 function representation,
we can easily prove the following basic theorem in Fourier transform
theory. Given

(A.1.17)

where/(/c) is the Fourier transform of f(x), then we assert that

(A.1.18)
 A.l DELTA FUNCTION AND FOURIER TRANSFORM THEORY 319

To prove Eq. (A.l. 18) we need only compute the integral on the right.
Using Eq. (A. 1.17) for/(re), we have

-1 P f(x)e~*'* dx = ~ r r f(k)ei(k-k')x dk dx (A.l.19)

2-ir J — °° 2ir J 00 J -00 -

Now, from Eq. (A.1.16)

1 f eiOc-k>)z dx = 5(/c _ k') (A.1.20)

Z7T J

Substituting this result into Eq. (A. 1.19), which means, in essence, that
we have interchanged the order of integration, and applying condition 3
[in Eq. (A.1.3)], we have

f “ f(x)e-^ dx = p 5(fc - fc')/(*0 ^ = /(*') (A.1.21)

2ir J — “ y — oo

which completes the proof.

In similar fashion we can easily extend this proof to three dimen¬
sions (see Prob. A.l) to prove the following theorem. Givenf

/(x) = fimk1}kMe*~ d>k (A.1.22)

where /(k) is the three-dimensional Fourier transform of /(x), and

k • x = k\X -p k%y T k%z (or k\X\ -p /C2X2 ~P k3X3), then

/(V) = ~ Jff J{x)e~^dV (A.1.23)

with dV the usual volume element, dx dy dz = d3x.

Problem
A.l Prove the three-dimensional Fourier transform theorem stated in
Eqs. (A.1.22) and (A. 1.23).

Representation of delta function as — V2(47rf?)_1

We have noted earlier that there are many functions that satisfy the
basic delta-function properties given in Eqs. (A.1.1) to (A.l.3). One
representation which is particularly useful in three-dimensional prob-
f Note that here and in the next section it is convenient to let x be the position
vector with components x, y, z, so that R can be used to denote the diilerence in
position between two particles.
320 MATHEMATICAL APPENDIX

lems (see, for example, Sec. 6.3) is the identity

5(R) = -V2(4tt R)-i (A.1.24)

In this equation 5(R) is a shortened notation for the product of three

delta functions, 8(x — x'), 8(y — y'), b(z — z'), and R is given by

R = X — x' = (x — x’)ax + (y ~ y')av + (z — z')a2 (A. 1.25)

One way to prove Eq. (A. 1.24) would be to verify directly that
the function on the right satisfies all the delta-function properties.
However it is just as easy to instead show that the differential equation

V24> = — 5(R) (A. 1.26)

can be solved to find 4> = (4irR)-1. To solve Eq. (A. 1.26), we use
Fourier transform theory to first find $(k).
If we replace 4>(x) by its Fourier transform,

4>(x) = J$(k)eik‘x dzk (A.1.27)

then we find by direct calculation that

V2<F(x) = dzk (A.1.28)

Conversely, on the right-hand side of Eq. (A. 1.26) we can replace each
of the delta functions by an expression like Eq. (A. 1.16), to have

-5(R) = - ~ f eik-R d3k (A. 1.29)

Equating Eqs. (A.1.28) and (A.1.29), we find that the Fourier trans¬
form of 4>(x) is given by

e-ik-x'

$(k) = (A.1.30)
8tt3/c2

Now, to obtain 4>(x) we simply substitute this expression for <f>(k)

into Eq. (A. 1.22) and carry out the integration over all values of k.
We have

J r gik-R
4>(x) = f 4>(k)e,k'x d3k
8V* J ~Wd3k (A.1.31)
 A.2 EVALUATION OF SCATTERING ANGLE x 321

Since the choice of coordinate system in the integral is arbitrary, we

shall assume that R lies in the z direction (as shown in Fig. A.l). In
addition, if we use spherical coordinates, d3k = k2 sin 9 dk dd d<t>, so
that Eq. (A. 1.31) reduces to

<t>(x) = —f~v [* f eikRcoa 6 sin 9 dk dd d(f> (A.1.32)

v ' 8tr3 Jo Jo Jo

In the integral, we let cos 9 = u, so that — sin 9 dd = du; then integrat¬

ing over both u and <t>, we easily find that

(A.l.33)

In summary this calculation shows that (4irA) 1 is a solution to

the equation V2<h = — 5(R) which proves the identity [stated in Eq.
(A. 1.24) and used in Eq. (6.3.11) of the text]

5(R) = -V2(47rE)-x (A. 1.34)

A.2 EVALUATION OF SCATTERING ANGLE %

The equivalent one-body problem

In Sec. 5.1 we have shown that in a two-body elastic collision the
center of mass velocity c0 is unchanged, while the relative velocity g

Fig. A.l Coordinate system with K lying in the z direction.

322 MATHEMATICAL APPENDIX

changes only in direction but not in magnitude. Hence the collision is

completely described by the scattering angle x between g and the relative
velocity after the collision g'. In this section we shall show how we
actually calculate x hi terms of g, the impact parameter b, and the
interparticle force between the colliding particles.
Suppose we let x and xx be the positions of the colliding molecules
with masses m and mx. We shall assume that the interparticle force
acts in the direction R = xx — x and that the magnitude of the force
depends only on R, the distance between the particles. Since we know
that the force on the two particles must be equal in magnitude and
opposite in direction, we can setf

mx = —F(R) aR m xxx = F(R) aR (A.2.1)

where is a unit vector in the direction of R and F(R) is related to

$(/?), the potential energy of the interaction, by the condition

nm = - 1| (a.2.2)

The relative motion of the two particles is found by noting, from

Eq. (A.2.1), that

F
R = *i - * = - (A.2.3)

with n = mmi/(m + rax) the reduced mass. Note that this is the equa¬
tion of motion for a particle of mass g, whose position and velocity are
that of particle mx relative to m, and which is acted on by the force
F = — d$/dR. The geometry of this problem is shown in Fig. A.2,
where we choose the origin of the coordinate system to be the point
where R = 0, and where we set a„ perpendicular to the plane of the
orbit. Note that the impact parameter b would be the distance of
closest approach Rm if the two particles did not interact. In addition,
R (which always points in the direction of the path of the orbit) is the
relative velocity of the two particles. Hence, the scattering angle x

f Note that we neglect here any effect on the particles of external forces.
This is due to the fact that the interparticle forces generally act over just a very
short range but, in this close range, the forces are very large compared to typical
external forces. In addition, as a practical matter, the solution to the collision
problem where there are large external forces is so complicated that one would
probably not bother treating such problems via the Boltzmann equation but would
instead use the BBGKY equations of Chap. 6.
 A.2 EVALUATION OF SCATTERING ANGLE x 323

Fig. A.2 The path of particle mx relative to particle m

(assumed to be at the origin) for a collision.

between g and g' is the angle between the two asymptotes of the path
and can be easily found if we know R(6).

The equation for the orbit

The path is most easily found by using the conservation laws for energy
and angular momentum.f First, setting the kinetic and potential ener¬
gies at any point equal to the initial kinetic energy (since the initial
potential energy is zero), we have

TgR • R + = iug2 (A-2-4)

f Note that the angular momentum L = /t(R X R). Hence

L = m(R X R + R X R)

which is zero, since, from Eq. (A.2.3), R and R are parallel, so that their cross
product (like R X R) vanishes. Hence, for this case L is constant, and the path
lies in a single plane.
324 MATHEMATICAL APPENDIX

Now, from Fig. A.2,

x = R sin 9 z = R cos 9
(A.2.5)
x = R sin 9 + RO cos 9 z = R cos 6 — Rd sin 9

Hence, we find

R • R = x2 + i2 = R2 + R2 A2 (A.2.6)

so that, in Eq. (A.2.4),

}v(R2 + R292) + *(R) = ^ (A.2.7)

In similar fashion, we obtain a second relationship between R and

9 b}^ setting the angular momentum L at any point equal to its initial
value. In terms of the coordinate system used in Fig. A.2, the initial
angular momentum is

(L)init = a yfj.bg (A.2.8)

For any other point on the orbit

Ly = pl(zx — xz) — fiR-d (A.2.9)

so that, from the conservation of L,

m = b9 (A.2.10)

From Eqs. (A.2.7) and (A.2.10) we can easily obtain a differential

equation for R(9) by the following simple trick. Divide through Eq.
(A.2.7) by d2 and use the fact that R/d = dR/dd to eliminate the term
with R~. Then use Eq. (A.2.10) to eliminate the remaining terms with
9. The resulting equation for the orbit is then found to be

dR\2 _ R*X2 2<t>

XHR) = 1 - ^ (A.2.11)
do) ~ b2~ M02

so that

bdR
dd (A.2.12)
± R2X
 A.2 EVALUATION OF SCATTERING ANGLE 325

The choice of signs here is made on physical grounds. We note that R

is infinite both before and after the collision, so that as 9 increases from
its initial value of zero, R first decreases (until it reaches the distance
of closest approach Rm) and then increases again.f Hence the minus
sign is used in Eq. (A.2.12) for 9 < 9m and the plus sign for d > dm.
At R = Rm, dR/dd must be zero; hence from Eq. (A.2.11) we find that
the distance of closest approach is given by X = 0 or

A.=b (i - <A-2-i3>

Evaluation of x

From Eq. (A.2.12) we can compute the orbit and the scattering angle x-
First, however, we note that the orbit is symmetric about the angle dm,
since if we integrate Eq. (A.2.12) from 9m to some other angle 9, we have

rR b dR' (A.2.14)
6 ~ 9m = ± ]Rm Rnxm

where the plus sign is used for 9 > 9m and the minus sign for 9 < 9m.
Note that when R —> °o , —> 0 while 9+ > 2(?m. However, from Fig.
A.2, the latter asymptote is also tv — x) hence we have

(A.2.15)
X = tt ~ 2 9m

or

of-__ (A.2.16)
x _ T “ Jr, 7f2[l - (b/R)2 - 2(4>/w72)]1/2

FTom this equation we can finally compute x provided we know the

impact parameter b, the magnitude of the initial relative velocity g,
and the interparticle potential <f>.
As a simple illustrative calculation, consider the collision of two
perfectly elastic spheres. If we let a be the sum of the radii of the
colliding particles, then

0 R > a (A.2.17)
HR) = 00 R < a

f Note that the monotonic increase of 0 with time follows from Eq. (A.2.10),
since R, b, and g are all positive.
 326 MATHEMATICAL APPENDIX

Hence, in Eq. (A.2.16), if we let b/R = sin 8, we find

(b/Rm)
= *-2 /„* (Id 7T 2 sin —i (A.2.18)
Rr

If b exceeds a, it is clear from Fig. A.2 that the particles never collide,
and Rm = b, whereas, for b < a there is a collision, so that Rm = a.
Applying this to Eq. (A.2.18), we have

0 ■ _tb
7T — 2 sm 1 - b < a
X a (A.2.19)
o b > a

Problem

A.2 Prove that for a coulomb interaction force

F = ^
47re0E2

the scattering angle x is given by

x = 2 sin"1 (1 + ^02)-1/2 vo = — M%2

qq i

A.3 DERIVATION OF BOLTZMANN COLLISION TERM

Evaluation of AN

In Sec. 4.1 we derived the equation for the rate of change of the velocity
distribution function /

dfa
dt + Vifv
oXi + a^
dVi (A.3.1)

where the term on the right is the rate of change of fa due to collisions.
More specifically if we let AN denote the net change (due to collisions
during time At) of the number of particles of type «, /„ dzx dh>, in the
volume element d3x at x with velocities in the velocity-space volume
element d3v at v, then as in Eq. (4.1.15)

AN = d3v A<
(A.3.2)
 A.3 DERIVATION OF BOLTZMANN COLLISION TERM 327

Fig. A.3 Volume element with base b db de and height

9 At.

To obtain the Boltzmann collision term we separately calculate the

loss term AN-, due to collisions in which one of the a particles at x
initiallyf has velocity v, and the gain term AN+, due to collisions in
which the final velocity for an a particle at x is v.
To calculate AN- consider first a single a particle at x with
velocity v. Before taking account of all possible collisions involving
this particle during time At, let us first consider just the average
number of collisions with particles of some specific type (3 with velocity
vi which approach with an impact parameter between b and b + db
and with the collision plane lying between the angles e and e + de.
If we assume that the time of interaction between the colliding particles
is very short compared to At, then any such collision partners must lie
initially in the volume element (shown in Fig. A.3) with height g At
and cross section area b db de. The number of 0 particles in this volume
with velocity vi is, from our definition of fp,

fp(x,\i,t)gb db de At d3vi (A-3-3)

Now to take account of all possible collision partners we must integrate

this over all possible values of b, e, and vi, and we must sum t he result
over all species 0. In addition, since the total number of a particles
at x with velocity v is fa(x,\,t) d3x d3v, the total value for AN- is given

f Here and throughout, by particles at x with velocity v or vi, we really mean

particles in the volume element d3x at x with velocities in the velocity-space element
d3v at v or vi.
328 MATHEMATICAL APPENDIX

by the product

AA_ = fa(x,y,t) d3x d3v HI

0
fp(x,vi,t)gb db de At d3Vi (A.3.4)

In similar fashion, to calculate AAr+ we consider the inverse colli¬

sions, in which an a particle with initial velocity v' collides with some
other particle with initial velocity vj to produce an a particle with
velocity v. Again, for a single a particle having an initial velocity v'
the average number of collisions with particles of some specific type /3
with velocity v( which approach with an impact parameter between b
and b + db and with the collision plane lying between e and e + df
is given by

Mx,\'ut)g'b db de At d3v[ (A.3.5)

To take account of all collisions, this must be integrated over all b, e,

and v( as well as summed over /3. Since the total number of a particles
with velocity v' at x is fa(x,v',t) d3xd3v', the total value for AN+ is
given by the product

AN+ = fa(x,\',t) d3x d3v' ^ Iff fff(x,\[,t)g'b db de At d3v\ (A.3.6)

e

These expressions for AN+ and AN- lead directly to Boltzmann’s

collision term. We recall, first, that from Eqs. (5.1.9) and (5.1.22)

g' d3v' d3v\ = g d3v d3vi (A.3.7)

If we use this identity in the expression for AJV+, we have, from Eqs
(A.3.2), (A.3.4), and (A.3.6),

MA = AN+ - AN-
\ dt Jc d3x d3v At

= J / (JLfpi ~ f*fm)gb db d,e d3vx (A.3.8)

0

which is the Boltzmann collision term used in Chap. 5.

Discussion of Boltzmann’s assumptions

The calculation of (dfa/dt)c which we have just performed involves

four basic assumptions. First, in the calculation we assumed that all
 A.3 DERIVATION OF BOLTZMANN COLLISION TERM 329

the distribution functions were evaluated at x, so that the collisions

were with “nearby” particles. This assumption is a relatively mild
one, but it requires that / be essentially constant over distances at
least on the order of the range of the interparticle force law. Next,
in our treatment of the two-body problem we neglected any external
forces. In reality the collision process would be quite different if there
were external forces comparable in magnitude to the short-range
interparticle forces. However, the treatment of the collision integral
for this case has never really been examined in a systematic fashion.
The third assumption concerns our use of only two-body collisions.
For ordinary gas molecules which have short-range forces (and thus
interact something like a collection of billiard balls) it is legitimate to
assume that each molecule drifts along unaffected by the other par¬
ticles except when it suffers an abrupt, almost instantaneous collision
with some other molecule (see Fig. A.4). For this case multibody
collisions, when more than one particle happen to collide at the same
time, are unimportant unless the gas is very dense. 1 he use of only
two-body collisions is much more difficult to justify for a plasma, how¬
ever. The coulomb force between the charged particles is a long-
range force, and, as we have seen in our discussion of the Debye
potential problem, each charged particle interacts to some extent with
particles as much as one Debye length away. Since the number of
particles No in this Debye sphere has been shown to be large, it is
clear that a charged particle in a plasma does not really drift along
freely most of the time the way a neutral molecule does. Instead, it
is interacting with many other particles nearly all the time. The
strength of these individual interactions, however, is very small. This
can be seen by recalling from Prob. A.2 that, for a coulomb interaction,
the scattering angle for a two-body collision is given by

• X (A.3.9)
sin
2 (1 + no2)1'2

Fig. A 4 Path of a particle in a rarefied gas with

short-range forces.
330 MATHEMATICAL APPENDIX

If the impact parameter b is taken to be the Debye length \D, and if

the average value for g2 is used, we shall show [in Eq. (A.4.22)] that
Vo = 9Nd, which is generally extremely large. Hence the scattering
angle x is typically very small for such a long-range collision. Of
course we have just been emphasizing that we do not have individual
long-range two-body collisions; instead the particle interacts simul¬
taneously with all the charges in the Debye sphere. However, when
the individual interactions involved are sufficiently weak, it is argued
that the collective effect of many simultaneous interactions on the
path of a particle is the same as a succession of individual two-body
collisions. Nevertheless, there is no question but that the Boltzmann
collision term must be used cautiously for charged-particle interactions
in a plasma.
The final (and, in some ways, most crucial) assumption used in
the derivation occurred when we set the joint probability of having
both an a particle with velocity v and a /3 particle with velocity vi
at x at time t proportional to fa(x,y,t)f^(x,yht). This is known as the
molecular-chaos assumption. More generally, this joint probability is
proportional to

fa(^,y,t)Mx,yht)[l + ^(v,vi,x,<)] (A.3.10)

where 4>aff is known as the correlation function. Physically this correla¬

tion term arises because, in the true dynamics of a system of interacting
particles, the presence of one particle at a point either enhances or
diminishes the probability that another particle will be in the same
neighborhood. Thus the molecular-chaos assumption (which is
responsible for the irreversible behavior of the Boltzmann equation
described in Sec. 5.2) overlooks these (hopefully small) correlation
effects among the particles. If one wants to avoid this assumption,
the only present alternative is to work with the reversible BBGKY
equations of Chap. 6.

A.4 CALCULATION OF CROSS SECTIONS

In ( hap. 5 we have seen that calculations of collision integrals

using the Boltzmann collision term involve a cross section

(A.4.1)

where b is the impact parameter and x is the scattering angle for a

collision. In this section we shall evaluate Sa0(l), first for hard-sphere
collisions, next for a general inverse-power interparticle force law, and,
finally, for the special case of coulomb collisions.
 A.4 CALCULATION OF CROSS SECTIONS 331

Hard-sphere cross sections

For elastic collisions between hard spheres we have seen in Eq.

(A.2.19) that the scattering angle is given by

2 sin-1 - ~b < a (A.4.2)

b>a

with o- the sum of the radii of the colliding spheres. Hence the
integrand in Eq. (A.4.1) is zero for b > cr, and we have for hard-sphere
collisions,

Sap® = 2tt f’ (1 - cos' X)b db (A.4.3)

To complete the calculation, note [from Eq. (A.4.2)] that over this
range of integration

b = a sin [£(tt — x)] = ° cos ix (A.4.4)

so that

db = -fr sinixdx

Substituting these expressions for b db into Eq. (A.4.3), we easily find

(letting 2 = cos x)

l odd
lira2
(A.4.6)
l even
J+ 1

Thus, for odd values of l this cross section is just the geometric
cross section ira2, while for even values of l this cross section is slightly
smaller.

Cross section for inverse-power force laws

For a general inverse-power interparticle force K/Rv, the potential
is given by

K (A.4.7)
(p — 1 )RP~1
332 MATHEMATICAL APPENDIX

Using this potential in the formula for the scattering angle x [Eq.
(A.2.16)] and introducing two nondimensional variables /3 and Vo,
we find

2 r R™ dV-1 -1/2

d/3
X = 7T 1 - /32 (A.4.8)
JO p — 1 \v0/

where (3 = b/R [so that d/3 = — (b/R2) dA], and w0 is chosen so that

(A.4.9)

From Eqs. (A.4.7) and (A.4.9), we easily find that

so that in Eq. (A.4.1) we have after a change in variable from b to the

nondimensional impact parameter v0

/jcyi(p-i)
SaP^ = 2ir Mv) (A.4.10)
U2/

where Ai(p) is the dimensionless integral

Mv) = fj (1 - cos' x)v0dv0 (A.4.11)

For most cases At(p) must be computed by numerical integration.

Some tabulated values are available, however,! and Ai(p) is found
to be of order unity provided the power in the force law is greater
than 2. For coulomb collisions (p = 2), however, this integral
diverges. Hence the calculation of the coulomb cross section must
be treated as a special case.

Coulomb cross sections

In treating the interaction of two charged particles in a plasma, we

recall from the Debye potential problem (discussed in Sec. 4.8) that
the potential about a charge in a plasma is smaller than the coulomb
potential because of the shielding effects of the other charges. There
are a number of ways to take account of this shielding. For example,

f See for example S. Chapman and T. G. Cowling, “The Mathematical Theory

of Non-uniform Gases,” p. 172, Cambridge University Press, New York, 1952.
 A.4 CALCULATION OF CROSS SECTIONS 333

we could replace the coulomb interparticle potential!

qaqfs (A.4.12)
=
4ireoR

by the Debye potential

O/.% 2 yyt
4>d = <t>c exp ( — Icd'R) kD>2 = (A.4.13)

in calculating the scattering angle x in Eq. (A.2.16). This can be

done, but the analysis required to calculate x is excessive, especially
in view of the fact that the Debye potential is only an approximation
and is not valid for all values of R anyhow.
For this reason we shall adopt here an alternative approach, which
is much simpler to use but leads to essentially the same result. We
note that the shielding effect reduces the potential very markedly for
distances large compared to the Debye length, whereas it has almost
no effect at all for R « \D- Hence, as a simple approximation, we
assume that x is given by the simple result for a coulomb interaction
(seeProb. A.2)

X = 2 sin- (1 + W *■ - (^Y (A.4.14)

provided the impact parameter is less than Ad, while if b exceeds Ad,
we shall assume that x = 0. In other words we neglect the screening
for b < \D and assume perfect screening for 6 > \D- With this
assumption

Vo2 ~ 1 b < Ad
2 sin2 ^x = | v02 + 1 (A.4.15)
cos x =
b > Ad

so that, from Eq. (A.4.1),

Vo2 - IV
(A.4.16)
Sa^ = 2tr b db
Ho2 + 1/

f Note that this is the total potential energy for the two charges in contrast
with the coulomb potential per unit charge used in Sec. 4.8, but also denoted by
334 MATHEMATICAL APPENDIX

The evaluation of this integral is now straightforward. We

introduce the new variable

4nreoMfi^Y
(A.4.17)

in terms of which Sm3(0 becomes

-it r>C2\d
Saf)v = 1 - (A.4.18)
= C2 Jo

Since l is an integer, we can easily expand this integrand in a finite

series of the form

21 21(1 - 1) , 41(1 - l)(l - 2) ,

z + 1 (z + l)2 ^ 3(0 + 1)3 + ' ' ‘

(A.4.19)

Each term in this sum is easily integrated to find

= 2tvC-H In (z + 1) + t-i
(i - m - 2) C*\d
0 + 1 3(0 + l)2 0
(A.4.20)

which does not diverge as a result of our having cut off the integration
at 6 = \D. (Otherwise, the upper limit here would be infinity and the
log term would diverge.)
In view of the approximate nature of this calculation (due to the
uncertainty in the actual interparticle potential), it is customary to
simplify this result still further by replacing the quantity in the upper
limit (C\D)2 (which is a function of g2) by its average value. If we
assume that the a and gases have no drift velocity and a common
temperature, then we find by direct calculation that

(f) = db / (vi - v>! i>v d■»! = 2. / u (~+dh

3 kT
(A.4.21)
 A.5 THE CHAPMAN-ENSKOG TECHNIQUE 335

Applying this to the expression for CXD, we find that for singly charged
particles

12™ Ten
A = (C\D) = / = 127rneXn3 = 9 ND (A.4.22)
e2

with Nd the number of electrons in a Debye sphere.

Since A, like Nd, is generally a very large number, the cross section
in Eq. (A.4.20) reduces to

(i - m - 2)
3
(A.4.23)

This is the cross section commonly used for problems involving coulomb
collisions. Although it is possible to make calculations which are
much more sophisticated than this, the final results are always virtually
identical to this, except possibly for differences in the slowly varying
function In A.

Problem
A.3 Find the average value of g2 and calculate A for the case when
species a and /3 have arbitrary distribution functions and where
Ta 9^ Tp but ua = up = 0.

A.5 THE CHAPMAN-ENSKOG TECHNIQUE

Lowest-order results
As noted in Chap. 5, the original technique developed for solving the
Boltzmann equation and calculating the transport coefficients for a
gas is due to Chapman and Enskog. Their method is an expansion
procedure based upon the limit when the gas is close to equilibrium and
the collision frequency is very high. In order to discuss the method
as simply as possible, we shall restrict our attention to the Boltzmann
equation for a simple electron gas and use the Krook collision model,
so that we have

(A.5.1)

In this equation / is the electron distribution function, v (for the

reasons indicated in Sec. 4.3) is the electron-electron collision frequency,
336 MATHEMATICAL APPENDIX

and a is the Lorentz acceleration — (e/m)(E + v x B). In addition

we again let /o be the local maxwellian velocity distribution function

ne-cVa2
2 kT
c = v — u (A.5.2)
7r3/2a3 m

so that (as in Secs. 4.3 and 5.7) the collision integrals for mass, momen¬
tum, and energy transfer are all zero.
The basic expansion procedure used in the Chapman-Enskog
technique is to assume that the self-collision frequency v is very large
and that / can be expanded in a series of the form

/ = Ao + V~1A\ + r~2A2 + • ■ •

= /(0) +/(1) +/<2) + • • • (A.5.3)

In this expansion all the A/s are assumed to be of the same order, so
that

/<”> » |/<»1 » |/<»| ■ • • (A.5.4)

In addition, it is assumed that all derivatives of an A,- are of the same

order as A, itself. Hence, in Eq. (A.5.1), if we keep only terms pro¬
portional to v, we have

0 = —v(A0 - f0) or Ao =/0 (A.5.5)

1 hus, as might have been expected, our lowest-order solution for / is

simply the maxwellian f0.

The Euler equations

Before proceeding to calculate/(1>, it is useful to note the form obtained

for the conservation equations at this level of approximation. As
given in Eqs. (4.2.12), (4.2.16), and (4.2.29) the conservation equa¬
tions (with all collision terms zero) are

Dn
+ nV • u = 0 (A.5.6)
Dt

Duk 1 dpik
~Dt - {ak) ~ p (A.5.7)

§i VP~ilZ = -1p-5/3 (v.q + gA (A.5.8)

 A.5 THE CHAPMAN-ENSKOG TECHNIQUE 337

These conservation equations are derived with no assumption about /

but involve the pressure tensor and heat-flow vector (in addition to
n, u, and T). However, at this lowest level of approximation, where
/ = /o, we easily findf [from (Eq. (A.5.2)] that

qk(0) = ±mff0c2ck dzc = 0 (A.5.9)

and

p«/n) = mffoCiCj d3c = riK T8t] (A.5.10)

If we substitute these expressions for qk(-0) and pt-/0) into the conserva¬
tions, we obtain the Euler equations of hydrodynamics

Dn
— nV • u (A.5.11)
JTt
Duk . , 1 dp
(A.5.12)
Dt ~ p dxk

and

D — 5/3
0 (A.5.13)
DtVU

There are several alternate forms for the last of these equations (the
energy equation). For example, if we replace p by ukT and use Eq.
(A.5.11) for Dn/Dt, we find

DT
-fTV•u (A.5.14)
Dt

In summary, to lowest order the Chapman-Enskog procedure

leads to the result that / is a local maxwellian velocity distribution,
with n, u, and T functions of x and t that are determined by the Euler
equations of hydrodynamics. As we go to higher orders of approxi¬
mation we shall find that /(1), /(2), etc., are also functions of just n,

t Note that Eq. (A.5.10) has been calculated earlier [see Eq. (5.7.4)], while
Eq. (A.5.9) follows from the fact that the integrand is odd in ck.
338 MATHEMATICAL APPENDIX

u, and T] however, the conservation equations used to determine these

quantities become more complicated than the Euler equations.

Calculation of /(1)
To calculate the next term in/ we consider the terms in the Boltzmann
equation (A.5.1) that are independent of v

dAo . i dAo dA o
+ U; — A\ (A.5.15)
+ Vi ~dxl dVi

Since A0 = f0 is known (in terms of n, u, and T), this equation enables

us to calculate A i simply by taking the derivatives of A0 that are
indicated on the left. This calculation, while straightforward, is
rather lengthy and will be treated in two stages.
To start, let the entire operator on the left be denoted by D/SDt.
Then, since A0 = /0 is a function of n, c, and /3 = a~2 = m/2/cT, we
have

£>A0 _ dfo Dn dfo DCi dfQ D/3

(A.5.16)
SDt ~ dn Dt + dCi &t + </3

From our expression for/0 we easily find that

dfo _ fo dfo dfo

dn n
= -^@Cifo = /o(-c2 + fd-1) (A.5.17)
dCi d(3

Hence, to complete the calculation we need only evaluate 2Dn/2D<,

SDCi/SDfi and SD|9/2Db
By direct calculation we find

S)n _ Dn dn
iDt ~ Dt + Ci dXi
£>Cj /Dm dui _
SD t ~ \Dt + Cj dxj " (A.5.18)

£>/? = (DT dT\

Dt T\Dt ' Ci dxi)

The hydrodynamic derivatives here are eliminated by using the Euler

equations (A.5.11) to (A.5.14). (This is permissible since we are cal¬
culating Df0/Dt, and we have seen that n, u, and T, when / = /„, are
related by the Euler equations.) In addition, we use the Lorentz
 A.5 THE CHAPMAN-ENSKOG TECHNIQUE 339

acceleration term for a to find

2D n dn
■nV • u + c,
dxi

2D Ci 1 dp
- (e, X B); (A.5.19)
p dXi m
2D<3
2D t

Using these expressions and the partial derivatives in Eq. (A.5.17)

in Eqs. (A.5.15) and (A.5.16), we find (after a certain amount of
algebra) that

1 2DAc _ _/oA
(A.5.20)
r 2D t v

with

A = | W - I) ^ - icHti) || (A.5.21)

The Navier-Stokes equations

As noted earlier, /(1) is again just a function of n, u, and T (and their
spatial derivatives). However, the conservation equations governing
these quantities are changed, since the heat-flow vector and pressure
tensor are computed using f0 + /(1), instead of just /0. To find the
new conservation equations we let

<7fc(1) = ^mj/(1)c2c* d3c pun) = mjf(1)CkCi d3c (A.5.22)

When the expression for/(1) in Eqs. (A.5.20) and (A.5.21) is substituted

into these integrals and the integration table in Sec. 5.3 [Eqs. (5.7.3)
to (5.7.9)] is used, we find that

dT 5 up
5*C1) = — Xo (A.5.23)
dxk 2 mv

dUk
Pki(l) = Vo (%hiV ■ u (A.5.24)
dxi

These results agree exactly with the expressions obtained in Eqs.

(5.8.6) and (5.8.7) using the 13-moment equations and the expansion
 340 MATHEMATICAL APPENDIX

procedure in powers of v~l (with p«(1) equal to the traceless pressure

tensor PH). Hence, as in that section, if we set pik = Pik(0) + p;*(1) and
Qk — qk<'0) + qk(1) (and assume that rj0 is constant), the momentum
equation (A.5.7) reduces to the Navier-Stokes equation

Duk
Dt
£m ia + (u X B)„] p dXk
+ mp V2uk + l
6 dXk
(V • u)

(A.5.25)

and the energy equation (A.5.8) becomes

p5/3 R_
r -FT,
Fw PP-5/3 (A.5.26)

These equations are commonly used in hydrodynamics for viscid

heat-conducting fluids. However, the viscosity and thermal-conduc¬
tivity coefficients, which are empirical constants in the macroscopic
theory, are now defined in terms of the self-collision frequency of the
particles. In addition, it is evident that further corrections to the
heat-flow vector and pressure tensor can be calculated if we want to
solve for / (and pkj and qk) to next order in i^-1.

A.6 EVALUATION OF (df/dt)c FOR ELECTRON-NEUTRAL COLLISIONS

Lowest-order analysis

In this section we want to consider in some detail the calculation of

(df/dt)c for the case of a very weakly ionized plasma in an electric
field. If we let /(v) and F(V) be the velocity distribution functions
for the electrons and for the neutral molecules, respectively, then for a
weakly ionized plasma, where the electron collisions are nearly always
with neutral molecules, the collision term for / is given by

(|)c = / (J'F> - fF)qb db de dW (A.6.1)

As shown in Sec. 5.9, when there is a strong electric field, it is cus¬

tomary to approximate /(v) by the first two terms in a spherical-
harmonic expansion, which is equivalent to setting

/(V) — fo(v) + ~ fi(V) fi(v) «fo(v) (A 6.2)

 A.6 EVALUATION OF (df/dt)c FOR ELECTRON-NEUTRAL COLLISIONS 341

Hence, to this level of approximation,

(I). - r» + r' <A-6-3>

where

To = f (f'oF' - f0F)gbdbded3V (A.6.4)

and

Tx = f (^f'F' - VjfiF\ gb db de d3V (A.6.5)

As discussed in Sec. 5.9, in an electron-neutral collision the neutral

molecule can be taken in lowest order to be infinitely massive and
stationary, so that v = v' = g and V = V' = 0. Hence /o = fo’,
fi — fi> and F' = F, so that, as in Eq. (5.9.17),

To = 0 Ti — -Mn„Sen^(v) (A.6.6)

Thus, to obtain a nonzero value for T0 it is necessary to take some

account of both the motion and the finite mass of the ions.

Next-order treatment of T0
The evaluation of T0 to next order is a difficult calculation which is
very similar to the calculation of the Fokker-Planck coefficients in
Sec. 5.10. As in that section, the analysis can be shortened somewhat
by introducing a new form for T0. Suppose we multiply both sides
of Eq. (A.6.4) by some arbitrary function Q(v) and integrate both
sides over all values of v. Then, upon interchanging primed and
unprimed quantities in the first term on the right and recalling that
g' = g and d3v' d3V' = d3v d3V, we have

JQTo d3v = f(Q' - Q)f0Fgb db de d3v d3V (A.6.7)

We now note from Eq. (A.6.4) that although g, v', and 1 ' all
depend on the angle between v and V, there is no term in To that
depends on the orientation of v relative to the laboratory coordinate
system. Hence, after we integrate over all values of V, both 7’0 and
the integrand on the right side of Eq. (A.6.7) must, by symmetry,
depend only on the magnitude of v. For this reason, it we write d3v in
spherical coordinates and integrate both sides of Eq. (A.6.7) over the
342 MATHEMATICAL APPENDIX

two angles, we have (after canceling the factor 4-ir on each side)

f0“ QT0v2dv = J” f0v2dv f (Q' - Q)Fgb db de dzV (A.6.8)

To lowest order, we have noted already that v' = v. To next

order we shall show that

v' = v + Ay (A.6.9)

where, for virtually all collisions, Av is much less than v. Hence, in

Eq. (A.6.8) we can set

Q' = QW) ^ Q(v) + ~ Av (A^2 (A.6.10)

so that we have

Jo QT0v2 dv d2Q p / u
(A.6.11)
^2 G^V) dv

where

Gi(v) = /0f2J Ay Fgb db de dzV

(A.6.12)
Gi(v) = ^foV2J(Av)2Fgb db de d3V

lo obtain the new expression for T0 we integrate by parts, once

in the first integral and twice in the second integral on the right side of
Eq. (A.6.11), to find

/»" <?7V * = /»" 0 (- W + Tf) * (A-6-‘3)

Since this equality must hold for any function Q(v), it follows that

1/ rfGj d2GA
T
J o
v2 \ dv dv2 J (A.6.14)

Hence, to find T0 we need only calculate Gy and Go as given in Eq

(A.6.12).

Determination of Ay and (Ay)2

4 he expressions for Gi(y) and G%(v') involve Av and (Ay)2, where

Ay = v' - v is the change in speed of the electron. To calculate these
 A.6 EVALUATION OF (df/dt)c FOR ELECTRON-NEUTRAL COLLISIONS 343

quantities we recall first that, from conservation of energy,

v'2 = v2 + — (V2 - V'2) (A. 6.15)

m

We can eliminate V in this equation by recalling [from the difference

between Eqs. (5.1.11) and (5.1.12)] that

m
m0 = m + M (A.6.16)
v' = v + ^'-*>

Now, from Eqs. (5.1.13) and (5.1.14) we also have

(g' - g) • (g' - g) = a2 = 4g2 sin2 ^x = 2^2(1 - cos x) (A.6.17)

Hence, if we square both sides of Eq. (A.6.16), we have

2w?
y2 = v2 + —
m0
V-(g'-g)+^ 02(1 - cos x) (A.6.18)
Trio

Upon using this expression for V'2 in Eq. (A.6.15), we have with
no approximation,

v'2 = v2 + — [M\ • (g - g') - p#2(l - cos x)] (A.6.19)

m0

where g = mM/m0 is the reduced mass. We now note that if the

electrons and neutrals have the same temperature, then, on the
average, \mv2 = V2. Hence, for most collisions, V is of order
(m/M)ll2v and g = |v — V| is of order v (as is also g'). For this reason,
we can use a double expansion, in which we calculate Av up to terms
of order m/M but in which we also assumet thatE is of order (m/M)ll2v.
To start, note that, to order m/M,

m0 1 + m/M 1 M
(A.6.20)
m ( 1 V^ m
m0 M \1 + m/M) ~~ M

f An objection can be raised against this argument since the integrations in

Eq. (A.6.12) are over all possible values of V from zero to infinity. However, for
larger values of V (where Av may be comparable to v) F(V) is extremely small, so
that the entire integrand in these expressions for Gi and G2 is negligibly small, and
any errors in Av or (At>)2 are unimportant.
 344 MATHEMATICAL APPENDIX

Hence, with V of order (m/M)1/2, we can rewrite Eq. (A.6.19) (to

order m/M) as

2m
./2
i +4v (g - g')
M v2
g2(l - cos x) (A.6.21)

where the first underlined term here is of order (m/M)112, and the
second is of order m/M. Therefore, if we take the square root of
both sides of Eq. (A.6.21), using the formula

(1 + e)i/2 = i + ie - ±e2 + ' • •

we find

v' — v + Av (A.6.22)

where, to terms of order m/M,

Av
— v~ {v- (g - g') -j„'( 1 - cos x) - [V' (S2[J g,)p
v (A.6.23)

It is this expression which we use for Av in evaluating Gfi.

Since our calculations of G\ and G2 are to be only up to terms of
order m/M, the approximation for (Ar)2 can be still simpler. If we
square both sides of Eq. (A.6.23) and remember that V/v is of order
(m/M)112, then we have

/ \ 3/2
—J = y-4[y . (g _ g')]2 + terms of order (A.6.24)

Therefore, in evaluating G2 we use the expression

(Av)2 = v~2ViVj(gi — g\)(gj — g'/) (A.6.25)

Evaluation of G\(v) and G2(v)

Now that expressions (to order m/M) have been obtained for Av and
(Av)2, we can proceed with the calculation of G^v) and G2(v) [as given
in Eq. (A.6.12)]. To perform the integrations over b and e we recall
that in the text we show that [see Eqs. (5.4.25), (5.4.30), and (5.10.17)]

/(0» ~ g'/)b db d,t = giSeM^g) (A.6.26)

/(I — cos x)b db de = S,e„(1)(gr) (A.6.27)

fbi - 9'i)(gj - g')b db de = 2glg3Sen11)(g) + ^Vra(2)(?) (A.6.28)

 A.6 EVALUATION OF (df/dt)c FOR ELECTRON-NEUTRAL COLLISIONS 345

with

^ij 'V/i(lj
Uij (A.6.29)
g gz

Therefore, when we use Eqs. (A.6.23) and (A.6.25) for Av and (Ar)2
in the expressions for Gi(v) and Gi{v) [Eq. (A.6.12)] and integrate over
b and e, we find

G2(v) = ifof[2gigjSenW(g) + W^enW(g)]ViVjF(y)g dW (A.6.30)

and

G1(v) = f0v f (viQi - ^ Sen^(g)F(V)g dW - v~'G, (A.6.31)

To complete the calculation we need only integrate these expres¬

sions over all possible values for V. Although an exact treatment of
these integrals is extremely difficult, we can obtain results to order m/M
fairly easily. We note first that, for V of order (m/M)ll2v,

gi - —’Vi + Vi

(A.6.32)

Next, taking the square root of g2, using the formula

(1 +e)i/2 = 1 +*e-

we have

vkV k m\
(A.6.33)
« = t'{1- — + eM)

Finally, if we expand Se„(i)(sO in a Taylor’s series about g = v, we have

S„«>(g) ‘ S.„<‘>(v) + (S -!>)+•• ■

(A.6.34)

Since the term Gt(v) contains the factor VlVJ- (which is of order
mv2/M), the expansion to order m/M in this term involves the single
346 MATHEMATICAL APPENDIX

term where each g is replaced by — v, Therefore, using the identity

f ViVjF(V) d*V = ^ijTlnK. Tn

M
8ijnnA 2
2 (A.6.35)

and letting

X (TinSen) (A.6.36)

we easily find, from Eq. (A.6.30),

fov3A2
G2(v) (A.6.37)

To evaluate Gi(v) we note [from Eqs. (A.6.31) to (A.6.34)] that in

the integrand of the first term, we have, to order m/M,

(A/. - fi-'j gN„.'“(g) = -

+ «„<»(■.) (vh + Mm -”„■) + F.F,W» (A.6.38)

')
If we substitute this result into Eq. (A.6.31) and integrate over all
values of V, using Eq. (A.6.35) plus the additional identities,

JE(F) dW = nn $F(V)Vi d3V = 0 (A.6.39)

together with Eq. (A.6.37) for G2(v), we find, finally, that

A2 mv*
Gt(v) = fo
2 MX
, /A2 d, v3 m z>4\
h \Y dv \ ~ M X ) (A.6.40)

Final result

From these expressions for G^v) and G2(v) we can now calculate
T0(v) = f(f'0F' - f0F)gb db de d3V. From Eqs. (A.6.14) and (A.6.37)
we have

1/_ dGi (PGA

To
v>2 \ dv dv2 J

]_d_ n +A2(f d 1,3 _i »3 dfA

v2 dv G, + T(Asx +x *) (A.6.41)
 A.7 IMPORTANT FORMULAS AND DEFINITIONS 347

The underlined term here exactly cancels the first term in Gi(v) [note
Eq. (A.6.40)], so that we find

m A2 d_ /v3 df0\
(A.6.42)
Mv2 dv X 2v2dv\\ dv)

It is this result which is used in the calculation of the distribution

function for a strong E field in Sec. 5.9, to find [on the right side of
Eq. ,(5.9.30)]

3m d v4f0 _ 3A2 d (v3 dfQ\

To sin 9 dd d</> (A.6.43)
M dv A 2 dv \A dv J

A.7 IMPORTANT FORMULAS AND DEFINITIONS

Physical constants

Charge of a proton, e. E602 X 10-19 coul

Mass of an electron, me. 9.109 X 10-31 kg
Mass of a proton. 1.673 X 10~27 kg
Boltzmann’s constant, k. 1.380 X 10-23 joule/°K
Speed of light in vacuum, c. 2.998 X 108 m/sec
Planck’s constant, h. 6.626 X 10-34 joule-sec
10-9
Permittivity of free space, e0 8.854 X 10“12 ^ farad/m
367T

Permeability of free space, /x0.47r X 10-7 henry/m

Avogadro’s number, A.6.0224 X 1023 particles/mole
Loschmidt’s number (number
density of an ideal gas at 1 atm
and 273°K), NL.2.69 X 1028 particles/m3
Number density at 1 mm II g and
273°Iv.3.54 X 1022 particles/m3

Conversion factors

1 weber/m2 = 104 gauss

1 coul = 2.998 X 109 esu units of charge
1 volt/m = (2.998 X 104)-1 esu of electric-field intensity
1 joule = 107 ergs
1 ev = 1.602 X 10-19 joule = kT0, when T0 = 1.160 X 104°Iv
1 atm = 760 mm Hg = 1.013 X 10s newtons/m2
1 torr = 1 mm Hg
348 MATHEMATICAL APPENDIX

Plasma parameters

Electron Cyclotron Frequency, |ojc| = eB/me

|wc| = 1.76 X 10nZ? rad/sec (B in webers/m2)
|Wc|
= 2.80 X 10SB cps (B in gauss)
2tt
/nee2\1/2
Electron Plasma Frequency, we = [ —— )
\ 77le6o/
CO, = 56.5 y/ne rad/sec (ne in electrons/m3)

2^ = 8.99 y/ne cps (ne in electrons/m3)

Debye Length, \D = (kT/me)1/2/u>e

\d = 69.0(7'/ne)1/2 m (T in °K, ne in electrons/m3)

Electron Thermal Velocity, vth = \oue = g) 1/2

Vth — 3.90 X 1037V/2 m/sec (T in °K)

Number of Electrons in Debye Sphere, No = |7rXD3ne

1.37 X 106T312
Nd = 1/2 (T in °K, ne in electrons/m3)
n

Coulomb Cutoff Parameter, A = 9Nd

1.23 X 107T312
A = 1/2 (T in °K, ne in electrons/m3)
n

Electron Collision Frequencies for Momentum Transfer

vei = 3.62 X 10~6Wi7\r3/2 In A sec-1
ven = 2.60 X 10V2w„T’e1/2sec_1 (T in °K, n in particles/m3;
In A is typically about 10; a is of order 10-10 m)

Other formulas and definitions

Lorentz force on charged particle: F = q(E + v x B)

Adiabatic invariants: p = A
Ah oi
= 2 ir
\vx\L
B<

Electron mobility: ge —-—

mv
71
Dc conductivity: <r0 = —!—
17leV e

Free-electron-diffusion coefficient: De
ITleV e

x(Te + Td
Ambipolar diffusion coefficient: DA
meven + mtvi
CTe + Ti)
miVi
 A.7 IMPORTANT FORMULAS AND DEFINITIONS 349

Alfven velocity: Va = 7-vet?

Magnetic pressure: 77—

Magnetic viscosity: vm = (mo<to)_1

uL
Magnetic Reynolds number: Rm = —
Vm
Cross sections for general inverse-power interparticle force laws:
BIBLIOGRAPHY

Alfven, H., and C. G. Falthammar: “Cosmical Electrodynamics,” 2d ed.,

Oxford University Press, Fair Lawn, N.J., 1963.
Allis, W. P.: Motion of Ions and Electrons, in “Handbuch der Physik,” vol. 21,
Springer-Yerlag OHG, Berlin, 1956.
-: “Nuclear Fusion,” D. Van Nostrand Company, Inc., Princeton, N.J.,
1960.
— -, S. J. Buchsbaum, and A. Bers: “Waves in Anisotropic Plasmas,” The
M.I.T. Press, Cambridge, Mass., 1963.
Arzimovich, L. A.: “Controlled Thermonuclear Reaction,” Gordon and
Breach, Science Publishers, Inc., New York, 1964.
— --: “Elementary Plasma Physics,” Blaisdell Publishing Company, New
York, 1965.
Balescu, R.: “The Statistical Mechanics of Charged Particles,” Interscience
Publishers, Inc., New York, 1964.
Bates, D. R.: “Atomic and Molecular Processes,” Academic Press Inc.,
New York, 1962.
Binder, R. C.: “Fluid Mechanics,” 4th ed., Prentice-Hall, Inc., Englewood
Cliffs, N.J., 1962.
Bishop, A. S.: “Project Sherwood: The U.S. Program in Controlled Fusion,”
Addison-Wesley Publishing Company, Inc., Reading, Mass., 1958.
Bogoliubov, N. N.: Problems of a Dynamical Theory in Statistical Physics, in
J. de Boer and G. E. Uhlenbeck (eds.), “Studies in Statistical Mechanics,”
vol. 1, North Holland Publishing Company, Amsterdam, 1962.
Boltzmann, L.: “Lectures on Gas Theory,” University of California Press,
Berkeley, Calif., 1964.
Born, M., and H. S, Green: “A General Kinetic Theory of Liquids,” Cambridge
University Press, New York, 1949.
Brandstatter, J. J.: “An Introduction to Waves, Rays, and Radiation in
Plasma Media,” McGraw-Hill Book Company, New \ork, 1963.
Briggs, R. J.: “Electron Stream Interaction with Plasmas,” The M.I.T. Press,
Cambridge, Mass., 1964.
Brown, S. C.: “Basic Data of Plasma Physics,” John Wiley & Sons, Inc.,
New York, 1959.
352 BIBLIOGRAPHY

Budden, Iv. G.: “Radio Waves in the Ionosphere,” Cambridge University

Press, New York, 1961.
Cambel, A. B.: “Plasma Physics and Magnetofluidmechanics,” McGraw-Hill
Book Company, New York, 1963.
Chandrasekhar, S.: “Hydrodynamic and Hydromagnetic Stability,” Oxford
University Press, Fair Lawn, N.J., 1961.
-: “Plasma Physics,” The University of Chicago Press, Chicago, 1960.
Chapman, S., and T. G. Cowling: “The Mathematical Theory of Non-uniform
Gases,” 2d ed., Cambridge University Press, New York, 1952.
Clauser, F. H. (eel.): “Symposium of Plasma Dynamics,” Addison-Wesley
Publishing Company, Inc., Reading, Mass., 1960.
Collin, R. E.: “Field Theory of Guided Waves,” McGraw-Hill Book Companv,
New York, 1960.
Cowling, T. G.: “Magnetohydrodynamics,” Interscience Publishers, Inc.,
New York, 1957.
Delcroix, J. L.: “Introduction to the Theory of Ionized Gases,” Interscience
Publishers, Inc., New York, 1960.
-: “Plasma Physics,” John Wiley & Sons, Inc., New York, 1965.
Denisse, J. F., and J. L. Delcroix: “Plasma Waves,” Interscience Publishers,
Inc., New York, 1963.
Desloge, E. A.: “Statistical Physics,” Holt, Rinehart and Winston, Inc.,
New York, 1966.
Drummond, J. E. (ed.): “Plasma Physics,” McGraw-Hill Book Company,
New York, 1961.
Dungey, J. W.: “Cosmic Electrodynamics,” Cambridge University Press,
New York, 1958.
Ferraro, V. C. A., and C. Plumpton: “An Introduction to Magneto Fluid
Mechanics,” Oxford University Press, Fair Lawn, N.J., 1961.
Francis, G.: “Ionization Phenomena in Gases,” Butterworth Scientific Publica¬
tions, London, 1960.
Fried, B. D., and S. D. Conte: “The Plasma Dispersion Function,” Academic
Press Inc., New York, 1961.
Gartenhaus, S.: “Elements of Plasma Physics,” Holt, Rinehart and Winston,
Inc., New York, 1964.
Ginzburg, V. L.: “Propagation of Electromagnetic Waves in Plasmas,”
Gordon and Breach, Science Publishers, Inc., New York, 1963.
Glasstone, S., and R. Lovberg: “Controlled Thermonuclear Reactions,”
D. Van Nostrand Company, Inc., Princeton, N.J., 1960.
Grad, H.: Principles of the Kinetic Theory of Gases, in “Handbuch der
Physik,” vol. 12, Springer-Verlag OHG, Berlin, 1957.
Green, H. S.: “The Molecular Theory of Fluids,” Interscience Publishers, Inc.
New York, 1952.
Guthrie, A., and R. K. Wakerling (eds.): “The Characteristics of Electrical
Discharges in Magnetic Fields,” McGraw-Hill Book Company, New
York, 1949.
Harman, W. W.: “Fundamentals of Electronic Motion,” McGraw-Hill Book
Company, New York, 1953.
 BIBLIOGRAPHY 353

Hasted, J. B.: “The Theory of Atomic Collisions,” Butterworth Scientific

Publications, London, 1964.
Heald, M. A., and C. B. Wharton: “Plasma Diagnostics with Microwaves,”
John Wiley & Sons, Inc., New York, 1964.
Hellund, E. J.: “The Plasma State,” Reinhold Publishing Corporation, New
York, 1961.
Hirschfelder, J. O., C. R. Curtiss, and R. B. Bird: “The Molecular Theory of
Gases and Liquids,” John Wiley & Sons, Inc., New York, 1964.
Holt, E. H., and R. W. Haskell: “Foundations of Plasma Dynamics,” The
Macmillan Company, New York, 1965.
Huang, K.: “Statistical Mechanics,” John Wiley & Sons, Inc., New York, 1963.
Huddlestone, R. H., and S. L. Leonard: “Plasma Diagnostics,” Academic Press
Inc., New York, 1965.
Jackson, J. D.: “Classical Electrodynamics,” John Wiley & Sons, Inc., New
York, 1962.
Kadomtsev, B. B.: “Plasma Turbulence,” Academic Press Inc., New York,
1965.
Kennard, E. H.: “Kinetic Theory of Gases,” McGraw-Hill Book Company,
New York, 1938.
Kunkel, W. B. (ed.): “Plasma Physics in Theory and Application,” McGraw-
Hill Book Company, New York, 1966.
Landau, L. D., and E. M. Lifshitz: “Fluid Mechanics,” Addison-Wesley
Publishing Company, Inc., Reading, Mass., 1959.
Langmuir, I.: “The Collected Works of Irving Langmuir,” vol. 4, “Electrical
Discharge,” vol. 5, “Plasma and Oscillations,” Pergamon Press, New
York, 1961.
Linhart, J. G.: “Plasma Physics,” North Holland Publishing Company,
Amsterdam, 1960.
Llewellyn-Jones, F.: “Ionization and Breakdown in Gases,” John Wiley &
Sons, Inc., New York, 1957.
Loeb, L. B.: “Basic Processes of Gaseous Electronics,” University of California
Press, Berkeley, Calif., 1964.
Longmire, C. L.: “Elementary Plasma Physics,” Interscience Publishers, Inc.,
New York, 1963.
Margenau, H., and G. M. Murphy: “The Mathematics of Physics and Chem¬
istry,” 2d ed., D. Van Nostrand Company, Inc., Princeton, N.J., 1956.
Massey, H. S. W., and E. H. S. Burhop: “Electronic and Ionic Impact Phe¬
nomena,” Oxford University Press, Fair Lawn, N.J., 1952.
McDaniel, E. W.: “Collision Phenomena in Ionized Gases,” John Wiley &
Sons, Inc., New York, 1964.
Meek, J. M., and J. D. Craggs, “Electrical Breakdown in Gases,” Oxford
University Press, Fair Lawn, N.J., 1953.
Mintzer, D.: Transport Theory of Gases, chap. 1 in H. Margenau and G. M.
Murphy (eds.), “Mathematics of Physics and Chemistry” vol. 2, D. Van
Nostrand Company, Inc., Princeton, N.J., 1964.
Mitra, S. K.: “The Upper Atmosphere,” Royal Asiatic Society of Bengal,
Calcutta, 1947.
354 BIBLIOGRAPHY

Montgomery, D. C., and D. A. Tidman: “Plasma Kinetic Theory,” McGraw-

Hill Book Company, New York, 1964.
Northrop, T. G.: “The Adiabatic Motion of Charged Particles,” Interscience
Publishers, Inc., New York, 1963.
Pai, Shih-I: “Magnetogasdynamics and Plasma Dynamics,” Prentice-Hall,
Inc., Englewood Cliffs, N.J., 1962.
Panofsky, W. K. H., and M. Phillips: “Classical Electricity and Magnetism,”
2d ed., Addison-Wesley Publishing Company, Inc., Reading, Mass., 1962.
“Physics of the Ionosphere: Report of the Physical Society Conference on the
Physics of the Ionosphere,” Physical Society, London, 1955.
Pierce, J. R.: “Traveling-wave Tubes,” D. Van Nostrand Company, Inc.,
Princeton, N.J., 1950.
Plonsey, R., and R. E. Collin: “Principles and Applications of Electromagnetic
Fields,” McGraw-Hill Book Company, New York, 1961.
Present, R. D.: “Kinetic Theory of Gases,” McGraw-Hill Book Company,
New York, 1958.
“Proceedings of the International Conference on the Ionosphere,” The
Institute of Physics and the Physical Society, London, 1963.
Ramo, S., and J. R. Whinnery: “Fields and Waves in Modern Radio,” John
Wiley & Sons, Inc., New York, 1953.
Ratcliffe, J. A.: “The Magneto-ionic Theory and Its Application to the
Ionosphere,” Cambridge University Press, New York, 1959.
Reif, F.: “Fundamentals of Statistical and Thermal Physics,” McGraw-Hill
Book Company, New York, 1965.
Rose, D. J., and M. Clark, Jr.: “Plasmas and Controlled Fusion,” The M.I.T.
Press, Cambridge, Mass., and John Wiley & Sons, Inc., New York, 1961.
Schmidt, G.: “Physics of High Temperature Plasmas,” Academic Press Inc.,
New York, 1966.
Shkarofsky, I. P., T. W. Johnston, and M. P. Bachynski: “The Particle
Kinetics of Plasmas, ’ Addison-V esley Publishing Company, Inc., Read¬
ing, Mass., 1966.
Simon, A. L.. An Introduction to Thermonuclear Research,” Pergamon Press,
New York, 1960.
Spitzer, L., Jr.. Physics of Fully Ionized Gases,” 2d ed., Interscience Pub¬
lishers, Inc., New York, 1962.
Stix, T. H.: “The Theory of Plasma Waves,” McGraw-Hill Book Company,
New York, 1962.
Stratton, J. A.: “Electromagnetic Theory,” McGraw-Hill Book Company
New York, 1941.
Sutton, G. W., and A. Sherman: “Engineering Magnetohydrodynamics,”
McGraw-Hill Book Company, New York, 1965.
Thompson, W. B.: “An Introduction to Plasma Physics,” Addison-Wesley
Publishing Company, Inc., Reading, Mass., 1962.
Uman, M. A.: “Introduction to Plasma Physics,” McGraw-Hill Book Com¬
pany, New York, 1964.
Wu, T. Y.: “Kinetic Equations of Gases and Plasmas,” Addison-Wesley Pub¬
lishing Company, Inc., Reading, Mass., 1966.
 INDEX

Ac conductivity, 85 Bernstein, I. B., 197?i.

Acceleration of cosmic rays, 59 Bers, A., lOOn.
Adiabatic compression, 17 Bessel functions, 197
Adiabatic invariants, 52-55, 57-59, 348 /3, for a fusion device, 145
Adiabatic law, 108, 109, 124 /? versus o> diagram, 98, 99
conditions for, 170, 171 Birnbaum, G., 29?i.
derivation of, in kinetic theory, 336-338 Bishop, A. S., 15, 22
(See also Energy-conservation equa¬ Bogoliubov, N. N., 292
tion) Bohm, D., 218w.
Adiabatic sound speed, 109, 143 Bohm criterion, 217-219
for electrons, 122 Bohm-Gross dispersion relation, 180, 186
Alfven velocity, 137, 151-154, 349 Boltzmann collision term, derivation of,
Alfven waves, 136, 137, 153, 154 326-330
Allis, W. P., 100 and H theorem, 232-234
Ambipolar diffusion, 117-121 reduction of, to Fokker-Planck equa¬
coefficient for, 118, 348 tion, 282-289
Angrist, S. W., 29n. statement of, 230
Angular momentum conservation, in Boltzmann equation, compared to'
two-body collision, 323, 324 BBGKY equation, 295, 298, 304
Anisotropic media, definition, 35 compared to BV equation, 172-175
Anomalous diffusion, 121 derived, 164-166
Appleton-Hartree equation, 86-89, 98, equilibrium solutions, 232, 233, 235-238
100, 195 solved by Chapman-Enskog technique,
for arbitrary angle, 96 335-340
compared to warm plasma, 123 solved by moment method, 240-241
for k parallel to Bo, 92, 93 Boltzmann’s constant, numerical value
for k perpendicular to B0, 95, 96 of, 347
Arcs, 12, 13, 30 Born, M., 292
Arzimovich, L., 19n. Bremsstrahlung radiation, 20-23
Avalanche breakdown, 8, 13 Briggs, R. J., 201n.
Averaging with velocity distribution, 161 BV equation, 172, 173, 292, 298
Avogadro’s number, 347 applied to Debye potential problem,
203-209
applied to longitudinal waves, 177-187
Balescu, R., 308n. applied to plasma sheath, 212, 213
BBGKY equation, 234, 292, 330 applied to transverse waves, 188-198
for /W, 295-298 applied to two-stream instability, 199-
for/<»>, 299 201
Beam-plasma amplifier, 202, 203 derived from BBGKY equation, 299-
Beams, 7, 29, 198, 201, 202, 250 302
356 INDEX

Carrier frequency, 74- Continuum model, 104

Center of mass velocity, conservation of, Converging B fields, 50, 51
225 Conversion factors, 347
definition, 224 Coordinate systems, 31
Cesium plasma, 12, 26, 29 Correlation effects, 234, 302, 304, 308, 330
Chapman-Enskog procedure, 261, 335-340 (See also Pair correlation function)
Chapman, S., 332n. Coulomb barrier, 15
Charge of proton, 347 Coulomb cut-off parameter, A, 334, 335,
Charge conservation law, 34, 113 348
Charge density, 34, 139, 140, 155 Coulomb force, cross section, 252, 253,
Clark, M., Jr., 17n., 21 n. 256, 332-335
Clemmow, P. C., 100 definition, 245, 246, 251
Cluster expansion, 309, 310 interaction, 284, 288
CMA diagram, 100, 101 scattering angle, 326, 329
Cold plasma, 79, 113, 123 Coulomb potential, 208, 209, 258, 333
fully ionized, 137 Cowling, T. G., 332n.
Collision cross section (see Cross sections) Critical field for runaway, 259
Collision frequency, for electrons, 62, 69 Critical frequency, 80, 81
as expansion parameter, 267, 268, 336 Cross sections, for collisional transfer,
for Maxwell molecules, 279 245, 277
numerical values for, 348 for fusion, 16-19
relation to viscosity and thermal con¬ for various force laws, 330-335, 349
ductivity, 339, 340 Crossed E and H fields, 26, 27, 38-40
for transfer of momentum or energy, Curl, 32
109, 110, 113, 246-253 Current density of plasma, 34, 63, 140,
as used in Krook model, 176 155
(See also Damping; Dc conductivity) Cut-off frequency (see Reflection points)
Collision integral, definition, 167 Cyclotron frequency, 39, 42, 46, 48, 348
for electron-neutral collisions, 340-347 Cyclotron heating, 17
for energy transfer, 252, 253 Cyclotron motion, 41
for heat-flow equation, 267 Cyclotron radius, definition, 40, 41
for momentum transfer, 169, 243-246, in an inhomogeneous B field, 52, 56
256-258 plotted versus temperature, 144, 145
for spherical harmonic expansion,
275-277
Collision plane, 225 Damping, of Alfv6n waves, 154
Collisions, effect on transverse waves, 82-86 collisionless, 186, 192
force due to, 109, 110 versus growth criteria, 83, 84
Complex plasma frequency, 82 of transverse waves, 84-86
Concentration, 297 by viscosity and thermal conductivity,
(See also Density for gas mixture) 272, 273
Conductivity, ac, 85 Damping factor, 85, 86
dc, 65, 278, 279, 348 Dc conductivity, 65, 278, 279, 348
electron, 65-70 Debye length, as cut-off for scattering,
for fully ionized gas, 71, 255 329, 330, 333
Hall, 66, 67, 70 definition, 3, 348
perpendicular, 66, 67, 70 for different temperatures, 210
for solid, 71 in diffusion, 116, 117, 119
tensor, 69, 71 in dimensional analysis, 312
Confinement schemes, 16-20 reciprocal, 182, 206
Confinement time, 19 relation to plasma frequency, 76, 77
Confining fields for plasma, 145, 146 in sheath, 218, 219
Constitutive equations, 35 Debye potential, 203-210, 233, 333
Constraint equations for a gas, 236-239 Debye shielding, 2-4, 209, 308
Conte, S. D., 185n. Debye sphere, particles in, 4, 209, 306,
Continuity equation (see Mass-conserva¬ 329, 335, 348
tion equation) Deflection angle (see Scattering angle)
 INDEX 357

Degeneracy factor, 9, 11 Distribution function, with a strong

Degrees of freedom, 108, 180 E field, 273-282
Del operator, 32 (See also Maxwellian velocity distribu¬
Delta function, applied to Fourier trans¬ tion)
form theory, 318, 319 Divergence, 32
in Debye potential problem, 199, 202, Drift, in E and B fields, 27, 39, 41—43, 66
208 in gravitational field, 41-43
definition and representations, 316-321 in inhomogeneous B fields, 43-52
in deriving BV equation, 301, 302 toward a wall, 219, 220
Density for gas mixture, 139, 155 Druyvesteyn velocity distribution, 281
Derivative of an integral, 197, 257 Dynamical friction coefficient, 282, 284
Deuterium plasma, 15, 22, 23
Diatomic gases, 108
Dielectric tensor, 35, 36, 71, 72 Earth’s magnetic field, 43, 50, 89
Diffusion in velocity coefficient, 282, 284, Einstein relation, 120
286, 287 Electron diffusion, 113-117
Diffusion equation, for magnetic field, 147, with magnetic field, 120, 121
148 Electron free-diffusion coefficient, 113, 348
for particles, 113-122 Electron runaway effect, 246, 254-260,
for temperature, 171, 269 274
Diffusion Mach number, 249, 250 Electron volt, definition, 9, 347
Diffusion time, for electrons, 115 Energy-conservation equation, as a con¬
for magnetic field lines, 147, 148, 150 straint, 236
for temperature, 172 for a fluid, 108, 109, 170, 171, 241
Diffusion velocity, 156 with viscosity and thermal conduc¬
Dimensional analysis, of BBGKY equa¬ tivity, 269
tions, 312, 313 in a two-body collision, 225, 323, 324
for diffusion equation, 114, 116, 119, (See also Adiabatic law)
120 Energy exchange in a collision, 343
for magnetic Reynolds number, 146 Entropy, 231, 234
in mhd approximations, 142 Entropy-maximum principle, 235, 238-
Dirac, P. A. M., 316 240
{See also Delta function) Equilibration, of temperatures, 112, 176
Direction cosines, 257 of velocities, 110, 111, 176
Dispersion equation, for cold plasma, 89 Equilibration processes, 176, 177
definition, 73 Equivalent current loop for particle, 53
for fully ionized gas, 130-138 Error function, 257
for longitudinal electron waves, 177- Euler equations, 336-338
187 Evanescent waves, 78, 79, 133
for mhd waves. 150-154
for simple electron fluid, 77-79
for transverse waves, 188-198 Fermi mechanism, 59
for two-stream instability, 199-201 Figure of merit for fusion, 20
for warm electron gas, 124-129 Fluid velocity for a mixture, 111
Dispersion function, 185 Fluorescent tubes, 30
Dispersive medium, 75 Flux, 162, 163
Distance of closest approach, 322, 325 Flux conservation, 56
Distribution function, in Boltzmann Flux equation for electrons, 120
equation, 230 Fokker-Planck equation, applied to elec¬
definition, 160 tron-ion collisions, 286-289
for nonequilibrium problems, 238-240 derived from Boltzmann equation,
relation to macroscopic variables, 160- 282-286
164 relation to BBGKY equations, 292,
relation to one-particle probability, 294 298, 308
in spherical harmonic expansion, 340 Forces {see Interparticle force laws)
for steady state with potential, 204, Fourier transform theory, 318-320
205, 232, 233 Francis, G., 14n.
 358 INDEX

Free-diffusion coefficient, 113, 348 Ideal gas law, 164

Fresnel problem, 79-81 Ignitron, 30
Fried, B. D., 185n.
Impact parameter, definition, 226, 228, 322
Frozen field lines, 148-150 Impurities, 23
Fusion, cross section for, 16, 19 Index of refraction, for cold plasma, 89
figure of merit for, 20 definition, 87, 190
probability for, 20 Inertial term in hydrodynamics, 113
Fusion devices, 137, 145, 150 Inhomogeneous magnetic fields, 43-51
Fusion processes, 14-23, 69 Instability, in confinement problems, 18,
Fusion reactions, 15, 22, 23 19
definition, 198
examples, 198-203
Gz coordinate system, 226, 243, 284, 285
Integral table, 261, 262
Gas discharges, 13, 68
Interaction volume, 19
Gauss’ theorem, 32, 33, 105, 149, 207
Interparticle force laws, 225
Generalized Ohm’s law, 23, 65-67, MO¬
effect on collision frequency, 251
HS, 157, 158
(See also Coulomb force; Hard-sphere
Grad, H., 261
collisions; Inverse power force
Gradient, definition, 32
laws; Maxwell molecules)
Gravitational drift, 41-43
Inverse collisions, 227-230, 328
Grazing collisions, 282
Inverse power force laws, as applied to
Green, H. S., 292
cross sections, 245, 331, 332
Group velocity, definition, 75
effect on collision frequency, 251, 254
for whistlers, 97, 98
in runaway effect, 260
Growing waves, criterion for, 84
Ion cyclotron heating, 17, 137
Gi'owth rate, for beam-plasma amplifier,
Ion plasma frequency, 130
203
Ion sound speed, 131, 132
for two-stream instability, 201
Ion waves, 132
Guiding center, 42
Ionization, energy for, 8, 9
Guthrie, A., 218w.
optical or r-f, 14
Gyrofrequency, 39
percentage, 9, 11, 12, 26
photo-, 13
H theorem, 230-235 x-ray, 14
Hall effect in mhd problems, 142n. Ionization chamber, 14
Hall electrical conductivity, 66, 67, 70 Ionosondes, 81
Hall thermal conductivity, 270 Ionosphere, 7, 14, 80, 81, 89
Hall viscosity, 271 Irreversibility, 230-235, 330
Hard-sphere collisions, cross section, 245 Isothermal sound speed, 108, 109, 143,
251-253, 281, 331 188
interaction, 282, 284
scattering angle, 325, 326
Jacobians, 165, 229, 230, 249, 295
Harris, E. G., 197n.
Joint probability function (see Proba¬
Heat-conduction equation, 171, 269 bility function)
Heat-flow equation, in general, 266, 267
273
with magnetic field, 269 Kinematic viscosity, 147

Heat-flow vector, definition, 163, 164 Kinetic-energy density, 110, 111, 249
Kirkwood, J. G., 292
in Euler approximation, 242, 243, 337
Kronecker delta, 168
in magnetic field, 269, 270
Krook collision model, applied to Chap-
in Navier-Stokes approximation, 339
man-Enskog technique, 335-340
340
applied to Landau damping, 181, 189,
related to 13-moment coefficients, 260
197
263
description of, 173-177, 224
Holes, 1, 8, 92, 133
in stability problems, 200
Hydrodynamic derivative, 105, 155, 169
with 13-moment equations, 260, 261,
Hydrodynamic viscosity (see Viscosity) 265, 268, 273.
 INDEX 359

Lagrange multipliers, 236, 238 Maxwell equations, 33-36

Landau damping, of longitudinal waves, Maxwell molecules, 246-248, 258, 267,
181-188 278, 279, 282, 289
of transverse waves, 188-198 Maxwellian velocity distribution, deriva¬
Landau damping factor, 186, 187, 192 tion of, 235-238
Langevin equation, applications of, Mean free path, 116, 277, 281
70-72, 76, 87, 89, 104, 124, 195, Mhd approximations, 141-145, 158
273, 278, 279 Mhd equations, linearized, 138-145
introduction of, 62, 63 nonlinear, 155-158
linearization of, 67, 68 Mhd generator, 23-26, 150
Langmuir, I., 2 Mhd waves, 150-154
Lasers, gas, 29 Mirrors (see Magnetic mirrors)
Left circularly polarized (l.c.p.) waves, Mobility, 67, 120, 348
94, 95, 135, 136, 198 Molecular chaos, 330
Left-hand rule, 94 Moment method, 240, 241
Lenard, A., 308n. Momentum-conservation equation, in a
Lenard-Balescu equation, 308, 314 collision, 225
Linear differential elements, 30, 31 as a constraint, 236
Linear discharge, 68 derivation of, 105-107, 168, 169
Liouville equation, 292-294, 299 for special cases, 120, 122, 129, 139-
Longitudinal adiabatic invariant, 57-59 141, 155-157, 211, 216, 217, 241,
Long-range forces, 329 258, 259
(See also Coulomb force) Monatomic gases, 180
Lorentz acceleration, 326, 339 Montgomery, D. C., 293?i., 308n.
Lorentz equation, 45, 62 Mullaly, R. F., 100
Lorentz force, 36, 37, 64, 65, 106, 139,
169, 348
Lorentz gas, 173, 274 Navier-Stokes equation, 147, 267, 269,
Loschmidt’s number, 347 272, 339
Newton's law, 28, 37
Nuclear fusion devices (see Fusion)
Magnetic bottle, 18 Number density, 160, 161
Magnetic equipotentials, 44, 45
Magnetic field lines, 44, 45, 48
Magnetic fields in a plasma, 145-150 Ohm’s law, 24, 26
Magnetic flux, 56, 148 (See also Generalized Ohm’s law)
Magnetic force term, 95 Orbits, in crossed E and B fields, 40, 41
Magnetic heating, 17, 55 in inhomogeneous B fields, 49
Magnetic mirrors, 17, 18, 51, 52, 56-59 in oscillating E field, 188
Magnetic moment, 52-55, 145, 348 in two-body collisions, 323-326
Magnetic pressure, 145, 146, 349 Orthogonal functions, 275
Magnetic Reynolds number, 146, 147, 349 Orthonormal coordinate systems, 30, 31
Magnetic scalar potential, 43
Magnetic viscosity, 146-148, 349
Magnetohydrodynamics (see under Mhd) Pair correlation function, definition, 303
Margenau, H., 280n. equilibrium value, 306
Mass of electron, 347 governing equation, 304-307
Mass of proton, 347 Panofsky, W. K. H., 20n., 54n.
Mass-conservation equation, as con¬ Period, definition, 73
straint, 236 Permeability of free space, 347
derivation of, 104, 105, 167, 168 Permittivity of free space, 347
for electron gas, 122 Perpendicular electrical conductivity, 66,
for fully ionized gas, 129 67
for gas mixture, 138, 139, 155, 241 Perpendicular thermal conductivity, 270
for plasma sheath, 215-217 Perpendicular viscosity, 271
Maximum-entropy principle, 235, 238- Phase velocity, definition, 73
240 Phillips, M., 20n., 54n.
360 INDEX

Photoionization, 13 Reflection by magnetic mirror, 51, 52,

Pinch effect, 17, 18, 67-69 56, 57
Pitch angle, 57 Reflection from plasma, 79-81
Planck’s constant, 347 Reflection points, cold-plasma, 86, 89,
Plane waves, 72-75 91, 92, 95, 98, 99, 101
Plasma, confinement of, 16-18 definition, 81
definition, 2-7 fully ionized gas, 137
heating of, 16, 17 warm-plasma, 127-129
liquid, 8 Relative velocity, definition, 224
naturally occurring, 7 Relative-velocity change in collision,
solid, 5, 8, 71, 92, 133 225, 322
Plasma beams (see Beams) Relaxation model (see Krook collision
Plasma frequency, complex, 82 model)
definition, 76, 348 Relaxation time, 173, 174, 176, 177
as reflection point, 92 Residues in contour integration, 184
Plasma jets, 17, 23, 26 Resonance conditions, 65
Plasma propulsion, 26-29 Resonances, cold-plasma, 86, 89-91, 95,
Plasma sheaths, 210-221 96, 99, 101
Plasma sound speed, 133 fully ionized gas, 137, 138
Poisson’s equation, definition, 204 upper-hybrid, 91, 92, 127
Polarization of waves, 93-95 warm-plasma, 127-129
Poles in contour integration, 183 Reversibility paradox, 233-235
Polyatomic gases, 108 Reynolds number, 147
Potential, electric, 35, 38 Right circularly polarized (r c.p.) waves,
magnetic, 35, 43 94, 95, 98, 135, 136, 198
(See also Coulomb potential; Debye Rose, D. J., 17n., 21 n.
potential) Runaway effect, 246, 254-260, 274
Potential at a wall, 213-215, 221
Poynting vector, definition, 36
Poynting’s theorem, 36 Saha equation, 8-12
Pressure, scalar, 108, 164 Scalar potential, 35, 38
Pressure force, 106 Scalar pressure, 108, 164
Pressure tensor, 163, 164, 169, 260, 263, Scalar product, 31
337, 339, 340 Scattering by plasma, 81
(See also Traceless pressure tensor) Scattering angle, for central forces, 321—
Probability function, for multiparticles, 326
294 definition, 225
for N particles, 293 for particular force laws, 326
for one particle, 294 relation to cross section, 245
for two particles, 297 Screening, 333
Propagation bands, 78 (See also Shielding)
Propagation vector, definition, 73 Second law of thermodynamics, 233, 234
Seeded plasmas, 26
Sherman, A., 23».
Quantum effects, 4, 5, 11 Shielding, 2-4, 209, 308
Shock heating, 17
Short-range forces, 329
Radiation, bremsstrahlung, 20-23 (See also Hard-sphere collisions)
cyclotron, 21 Solar wind, 7
Radii of colliding molecules, 252, 253, Solid-state plasma, 5-8, 71, 92, 133
325, 348 Speed of light, 347
Radius of gyration, 40, 41 Spherical harmonic expansion, 274, 275,
(See also Cyclotron radius) 340
Random velocity, definition, 162, 164 Stellarator, 18
Ratio of specific heats, 108, 180 Stix, T. H„ 187n.
Reactor, fusion, 14-23 Stokes’ theorem, 32, 33, 148
Recombination at a surface, 211, 220 Stratton, J. A., 44n.
 INDEX 361

Sturrock, P. A., 201n. Tunneling effect, 81, 82

Summation convention, 31 Two-body collisions, 224-230, 321-326
Sutton, G. W., 23n. Two-stream instability, 198-202

Tchen, C. M., 308n.

Ultraviolet radiation, 13
Temperature, definition, 162, 164
Uncertainty principle, 4, 5
in different directions, 240
Uncompensated solids, 8
effect of, on collision frequency, 251,
Upper hybrid resonance, 91, 92, 127
252
Ursell-Mayer expansion, 309
on runaway, 259
of gas in E field, 281
Test particle, 203
Thermal conductivity, 171, 260, 267, 268, Van Allen belts, 7
339, 340 Variational technique, 236
Thermal-conductivity tensor, 270 Vector potential, 34, 35
Thermal-energy density, 112, 206, 249 Vector product, 32
Thermal velocity, 41, 50, 348 Vectors, 31-33
Thermionic converter, 29 Velocity of entire fluid, 139, 155
Thermodynamic relations, 238 Velocity distributions (see Distribution
Thermonuclear processes (see Fusion) function)
Thirteen-moment approximation, 260-267
Viscosity, 147, 169, 260, 267, 268, 271,
Thompson, W. B., 11 n.
339, 340
Thyratron, 30
(See also Magnetic viscosity)
Tidman, D. A., 293n., 308n.
Viscous-stress equation, 264-266, 273
Topside soundings, 81
Torr, 347
Traceless pressure tensor, 169, 242, 243,
263, 270-272 Wakerling, R. K., 218n.
Transmission through plasma slab, 82 Wall conditions, 210-221
Transmission at vacuum-plasma inter¬ Warm plasma, definition, 123
face, 98 Wave length, definition, 73
Transport coefficients (see Dc conduc¬ Wave number, definition, 72
tivity; Diffusion equation; Thermal
Wave packets, 73-75
conductivity; Viscosity)
Whistler mode, 97, 98
Transport equation, 166, 167, 264
Wu, T. Y„ 308n.
Truncation, of BBGKY equations, 299,
303, 310
of moment equations, 172
of series, 240 Yvon, J., 292
 Date Due

4 » 'gf

y 4

iiOV ££

& CAT. NO. 23 233 PRINTED IN U.'

 QC 718 .T23
Tanenbaum, B. Samuel.
Plasma phys cs. by B. Samuel 010101 000

0 1163 0148533 4
TRENT UNIVERSITY

QC718 .T23
Tanenbaum, B Samuel
Plasma physics

113306
■v