The Adiabatic Motion of Charged Particles (Interscience Tracts On Physics Astronomical) (Theodore G. Northrop)

THEODORE G.
NORTHROP
The Adiabatic Motion
of Charged Particles
Number 21
Interscience Tracts on Physics and Astronomy

NUNC COCNOSCO EX PARTE
TRENT UNIVERSITY
LIBRARY
Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation
https://archive.org/details/trent_0116301278317
INTERSCIENCE TRACTS ON PHYSICS AND ASTRONOMY
Edited by R. E. MARSHAK
University of Rochester
1. D. J. Hughes
NEUTRON OPTICS
2. M. S. Livingston
HIGH-ENERGY ACCELERATORS
3. L. Spitzer, Jr.
PHYSICS OF FULLY IONIZED GASES, second edition
4. T. G. Cowling
MAGNETOHYDROD YNAMICS
5. D. ter Haar
INTRODUCTION TO THE PHYSICS OF MANY-BODY
SYSTEMS
6. E. K. Opik
PHYSICS OF METEOR FLIGHT IN THE ATMOSPHERE
7. K. Mendelssohn
CRYOPHYSICS
8. J. L. Delcroix
INTRODUCTION TO THE THEORY OF IONIZED GASES
'an introduction to celestial mechanics

10 • J»
'general relativity and gravitational waves
11. R. E. Marshak and E. C. G. Sudarshan
INTRODUCTION TO ELEMENTARY PARTICLE PHYSICS
12. J. L. Olsen
ELECTRON TRANSPORT IN METALS
13. M. Fran§on
MODERN APPLICATIONS OF PHYSICAL OPTICS
14. P. B. Jones
THE OPTICAL MODEL IN NUCLEAR AND PARTICLE
PHYSICS
15. R. K. Adair and E. C. Fowler
STRANGE PARTICLES
16. R. Wilson
THE NUCLEON-NUCLEON INTERACTION:
EXPERIMENTAL AND PHENOMENOLOGICAL ASPECTS
17. J. F. Denisse and J. L. Delcroix
PLASMA WAVES
18. W. F. Brown, Jr.
MICROMAGNETICS
19« .A.» Rose
CONCEPTS IN PHOTOCONDUCTIVITY AND ALLIED
PROBLEMS
20. A. Guinier and D. L. Dexter
X-RAY STUDIES OF MATERIALS
21. T. G. Northrop
THE ADIABATIC MOTION OF CHARGED PARTICLES
22. G. Barton
INTRODUCTION TO ADVANCED FIELD THEORY
23. C. D. Jeffries
DYNAMIC NUCLEAR ORIENTATION
24. G. B. Benedek
MAGNETIC RESONANCE AT HIGH PRESSURE
Additional volumes in preparation

THE ADIABATIC MOTION
OF CHARGED PARTICLES
THEODORE G. NORTHROP
Lawrence Radiation Laboratory

University of California, Berkeley, California
Published under the auspices of the Division of Technical Information,

U. S. Atomic Energy Commission
INTERSCIENCE PUBLISHERS 1963
a division of John Wiley & Sons, New York • London • Sydney

Qfr J2.0 ■ M
Copyright© 1963 by John Wiley & Sons, Inc.
All Rights Reserved
This book or any part thereof must not be reproduced in any form without the
written permission of the publisher and the assignee.
Library of Congress Catalog Card Number 63-22462
This copyright has been assigned and is held

by the General Manager of the United States Atomic Energy Commission.
All royalties from the sale of this book accrue to
the United States Government.
Printed in the United States of America

Foreword
Some experts like to use for the subject of plasma physics, the
fully descriptive and complicated name of magnetohydrodynamics.
The complication is proper. It signifies the all but simple marriage
of the disciplines of fluid mechanics with electromagnetism.
This subject has in recent years attained great impetus from its
hoped-for applications to the production of controlled thermo¬
nuclear power. The main stumbling block in this application has
been the fact that magnetic fields are not perfect means of con¬
tainment of the hot ionized plasma. As we like to say, the magnetic
bottle is leaking.
Controlled thermonuclear power is only one of the many
important applications of hydromagnetics (to use still another of
the several customary designations). Other practical applications
are coming to the fore, and an even wider field has been opened in
planetary, solar, and astrophysical science. The interplay of
magnetism with charged particles has to be studied if one wants to
understand earth magnetism, the Van Allen belts encircling our
globe, the magnetic storms that accompany the sun spot cycle,
and the extensive magnetic fields that occur in quiescent or stormy
forms throughout our Milky Way system and other galaxies.
A point that seldom receives proper emphasis is the limited
effectiveness of instabilities. One should perhaps be less surprised
about the lack of complete confinement; it may be more proper to
draw attention to the fact that the confinement of the plasmas
is so good. The existence of the Van Allen belts is itself due to the
exceedingly long duration that an electron or ion once caught in the
earth’s magnetic field will spend in that field before finding its
way into the absorbing atmosphere or before escaping into space
FOREWORD
VI
If the magnetic field were not present, a fast electron would get
lost in a few milliseconds. From the behavior of the added belts
induced by nuclear explosions, we now know that the actual con¬
tainment time is at least a few years. For some electron orbits,
it might be very much longer.
The point of view used in this book is directly applicable to the
theory of these periods of long containment. The focus is on the
orbits of individual charged particles. This treatment is in fact
essentially sufficient for the understanding of magnetohydrody¬
namics, provided the plasma is very tenuous. If the number of
collisions between particles is low enough and if the energy car¬
ried by the particles is negligible compared to the energy of the
electromagnetic fields, then it becomes permissible to consider the
plasma from the point of view of a collection of noninteracting
particles. It is in this case that exceedingly long containment times
have been observed.
The approach taken in the book corresponds to the explanation
of the behavior of gases from the point of view of statistical me¬
chanics. This comparison makes the involved nature of magneto¬
hydrodynamics particularly clear. The basic orbit considered in
statistical mechanics is the movement of an atom with constant
velocity along a straight line. In the subject of the book, this
simple orbit has to be replaced by a helix suffering reflections and
systematic drifts. The fact that in spite of such an involved motion
long confinement times are observed and can be understood is due
to the presence of adiabatic invariants. The treatment of these
invariants is the backbone of the following discussions.
The understanding of the individual orbits is not the same as
complete understanding of plasma physics. The science, or the art,
of treating cooperative phenomena must be added to the thorough
discussion of the motion of the individual particles. But it is a
basic step in the development of plasma physics to lay the founda¬
tions by discussing individual particle motions. The result is, as
always when meaningful progress is made in physics, that a seem¬
ingly involved subject is beginning to show signs of regularity and
therefore of simplicity.
FOREWORD Vll
If the author had wanted to give a properly involved title to the

subject, he might have called it, “Prolegomena to Magneto¬
hydrodynamics.” This may have scared off many readers. It
also might have attracted the most thorough ones because this
title clearly implies that a complete understanding of plasma
physics presupposes the mastery of the subject of this book.
Edward Teller
Department of Applied Science
University of California
Livermore, California
July 1963
Preface
The origins of the adiabatic theory of charged particle motion

can be traced back to Alfven in the 1940s. Since, in a formal sense,
the problem is one in classical perturbation theory, the mathe¬
matical formalism really existed much before then. However, the
theory for the charged particle seems to have been developed
from physical insight initially, and only afterward was the con¬
nection with more formal analysis made. Such an evolutionary
history is very common in the physical sciences.
In this book I have retained the more physical approach. So
long as only first order effects in gyration radius are considered,
this approach is adequate. It also leads to a good comprehension
of the subject. The danger in a physical approach is not so much
that effects will be predicted that do not exist, but that real effects
will be completely overlooked. A deductive analysis from first
principles reduces this danger, and this is the method that I have
adopted. To investigate higher order effects, the more formal
methods that start from a canonical formulation are most useful.
Recent work along these lines is reviewed in the text.
In this work I have attempted to be explicit about matters that
are frequently misunderstood. An example is the fact that each
adiabatic invariant is the sum of action integrals over the degrees
of freedom, and that the number of adiabatic invariants may be
less than the number of degrees of freedom. Confusion about this
point arises because systems with one degree of freedom are often
used for illustrative purposes.
A second major objective is to point out where extensions and
applications of the theory would be profitable. Some of these
extensions are to higher order terms and may be tedious to cal-
IX
X PREFACE
culate; but manifestations of higher order motions can be im¬

portant. In contrast, there seem to be times when the particle
motion rather completely ignores the adiabatic predictions, even
to low order. Under these circumstances, higher order terms are of
no help in explaining the observations. Such nonadiabatic be¬
havior should not be surprising, since adiabatic theory is really an
asymptotic theory, rather than a convergent one.
I wish to thank the many scientists whose work is reviewed in
this book; without their work there could be no book. I am in¬
debted also to friends who have read the manuscript and made
suggestions and contributions. In particular, I thank Drs. Edward
Teller, Conrad Longmire, Martin Kruskal, and Clifford Gardner
for their help.
T. G. Northrop
Berkeley, California
June 1963
Table of Contents
Introduction. xiii
1. The Guiding Center Motion. 1
A. The Equation of Motion of a Particle. 1

B. Derivation of the Guiding Center Equation of Motion .... 3
C. The Drift Velocity and the Longitudinal Equation of Motion . 6
D. Energy Gain. 10
E. Integration of the Energy Equation: Fermi Acceleration;
Rotating Plasmas. 12
F. The Geometric Interpretation of the Drift Equation and of the
Longitudinal Equation of Motion. 23
G. The Relativistic Case. 27
2. A More Formal Derivation of the Nonrelativistic Guiding Cen¬

ter Equation. 36
3. The Adiabatic Invariants of the Motion. 41
A. The Magnetic Moment. 42

B. The Longitudinal Adiabatic Invariant. 47
C. The Third or Flux Adiabatic Invariant 4> 61
D. Adiabatic Invariants to Higher Order. 68
E. The Adiabatic Invariants of Singly Periodic Systems .... 71
4. Additional Applications of Adiabatic Theory. 78
A. The Current Density in a Plasma. 78

B. Loss-Free Geometries. 31
C. The Geomagnetic Case . 34
D. Plasma Stability. 35
5. Nonadiabatic Behavior. 96
A. Nonadiabatic Effects in Mirror Machines—Numerical .... 96

B. Nonadiabatic Effects—Experimental .101
C. Another Problem.102
References.193
Index.19^
xi
Introduction
Knowledge of the trajectories of charged particles in electro¬

magnetic fields is important to many branches of science—for
example, to space physics, to plasma physics, and to the design
of accelerators. A particle trajectory is generally very complicated
and must be obtained by a numerical integration of the equation
of motion. In special cases, analytic solutions can be obtained,
such as in the trivial case of the uniform static magnetic field,
where the particle gyrates in a helix about the field. Usually such
solutions are possible only because of a symmetry.
In a general field, solution of the equation of motion in a Taylor
expansion about the initial conditions is practical only for short
times. Suppose the particle is in an approximately uniform mag¬
netic field (one that varies slowly in space and time—that is,
slowly compared with the gyration radius and period); the par¬
ticle motion is approximately helical. A Taylor expansion would
be practical only over a fraction of a gyration period. A different
approximation is needed if one wishes to follow the particle over
many gyration periods by other than numerical methods. The
gyration and motion parallel to the field line should be introduced
explicitly into the expansion and deviations from strict helical
motion treated as the perturbation. This is the so-called “guiding
center” or “adiabatic” approximation. The term “guiding center”
arises because in a slowly varying field the particle moves ap¬
proximately in a circle whose center drifts slowly across the lines
of force and moves rapidly along the lines. Not only the equations
governing the guiding center motion, but also the approximate
constants (“adiabatic invariants”) of the particle motion, are
useful in applications.
Xlll
X1Y INTRODUCTION
In spite of the limitation to slowly varying fields, the adiabatic

approximation is of utmost importance, especially to plasma
physics and to the study of particle motion in space and in the
terrestrial (Van Allen) radiation belts. Adiabatic theory explains
why long-term containment of charged particles is to be ex¬
pected in the earth’s magnetic field, in spite of the effects of asym¬
metries in the magnetic field pattern, earth rotation, and the like.
The adiabatic approximation is also the basis of much plasma
theory used in controlled thermonuclear research.
The theory of the guiding center motion of the adiabatic invari¬
ants will be developed fully in this volume and the utility of the
theory illustrated by examples.
CHAPTER 1
The Guiding Center Motion
In this chapter we will first give an informal derivation of the

guiding center motion. Although it is informal, it makes good
physical sense and is a good one for classroom use. It, of course,
gives the correct results, but may leave the very exacting reader
wondering why, since there are points where objections can be
raised. These objections cannot be met will)out the more formal
mathematical proofs, which will be outlined in Chapter 2.
The derivation does not proceed by considering separately
special fields in which one guiding center drift or another appears
alone, and then superposing the various drifts. This method has
been used frequently (1,2) but becomes lengthy if all possible drifts
are considered. Instead, the drifts will be obtained deductively at
once by starting with a field of general geometry and doing a small
amount of vector algebra. The method is similar to Hellwig’s (3),
but is carried out for a more general situation. It will be assumed
initially that the particle has nonrelativistic energy, and the non-
relativistic equation of motion will be used as a starting point.
After understanding the geometric reason for each drift, it is easy
to write the modified equations for a particle with relativistic
energy, at least for the case in which the electric field is small.
A. The Equation of Motion of a Particle
The equation of motion of a charged particle is
dp/dt = (e/c)dr/dt X B(r, t) + eE(r, t) + m0g(r, t)

2 ADIABATIC MOTION OF CHARGED PARTICLES
where r is the particle position, p is its relativistic momentum

moyr, B and E are the magnetic and electric fields, m0 is the rest
mass, e is the charge, y is m/mo, m being the total mass, and g
is the total nonelectromagnetic force per unit mass. The radiation
reaction force terms are omitted. In many practical cases their
effect on the particle motion is negligible.
It is worthwhile writing in dimensionless form the nonrelativistic
equation of motion:
(mc/e)[r — g(r, {)] = r X B(r, t) + cE{r, t). (1.1)
Let
B = Br, (t)/B0(t) E = (c/v0Bo)E(r, t) T = (v0t/L)
R = r/L G = (L/v02)g
where v0 is the initial particle velocity, B0(t) is the magnetic field
at a typical point at time t, and L is a characteristic dimension or
distance over which the fields change. Then the equation of motion
becomes
mcv0 T d'2R dR
~G(R, T) X B(R, T) + E(R, f) (1.2)
eBoL \_df2 dT
with the initial conditions that at T = 0, R = r0/L and dR/dT = C0,
where r0 is the initial position of the particle and v0 equals v0v0.
Equation (1.2) is formally identical with (1.1) by the substitutions
mcv o me
R —*■ r T -*■ t G-+g B-+B
eB0L e
and E —*■ cE.
Thus any solution of (1.2) gives a solution of (1.1) by these substi¬

tutions.
Now mcvo/eBoL is the ratio of the radius of gyration to the char¬
acteristic distance over which fields change and therefore is the
quantity which one expects must be made smaller in order for the
adiabatic approximation to become more valid. But because Eq.
(1.1) is formally identical with (1.2), we can work with (1.1) and
THE GUIDING CENTER MOTION 3
just use the dimensional quantity m/e = e as the smallness parame¬

ter. It obviously is impossible to make the m/e of, say, an electron
smaller in a series of experiments in order to make its behavior
more adiabatic. But it really is only mcv0/eB0L that must be made
smaller, and this is possible by changing v0, B0, or L. The advantage
of using m/e instead of mcv0/eB0L is that one does not have to work
with dimensionless equations, and results are obtained directly in
dimensional form. A more complete discussion of the scaling of the
equation of motion has been given in reference 4.
B. Derivation of the Guiding Center Equation of Motion
To derive the equation of motion of the guiding center, let r =

jR + 9, where r is the instantaneous position of the particle, R is
the position of the guiding center, and 9 is a vector from the guiding
center to the particle (Fig. 1.1). The vector 9 can be given a precise
definition by the equation 9 = (mc/eB2)B X (v — cE X B/B2),
where E and B are evaluated at r. This, combined with r = R +
9, gives a precise definition of R. To lowest order in e the fields
can, of course, be evaluated at either r or R, the difference being of

order e2 in the equation for 9. Now substitute r = R + 9 into Eq.
(1.1). Since the radius of gyration is proportional to e, terms con¬
taining 92 can be neglected compared with those in 9.
The result of substituting r = R + 9 into Eq. (1.1) and expand¬
ing the fields in a Taylor series about R is
R + 9 = g + (e/m){E(R) + q-VE(R)+ (1 /c)(R + 9)
X [B(R) + 9-VB(R)]} + 0(e) (1.3)
where 0(e) means terms of order e. The term (9/c) X q-VB(R)

must be retained in Eq. (1.3); as will become apparent shortly,
this term is not of order e2 but is of order e.
Now define three orthogonal unit vectors: let ei equal B/B, let
e2 be a unit vector perpendicular to the line of force, and let
e3 be ex X e2, a unit vector perpendicular to ex and e2. In
order to correspond to the picture of the particle moving about
a circle of radius p, let 9 = p(e2 sin 6 + e3 cos 0), where 0 = f wdt,
w being the gyro frequency eB(R)/mc. Then 9 = cop (62 cos 0 — e3

sin 0) + (pe2)‘ sin 0 + (pe3)' cos 0. The first term contains cop and
is of zero order in e, since co ~1/e and p ~ e. The second and third
terms contain p or p and are of order e. The reason for retaining
9 X (lo-'V)B in Eq. (1.3) is now formally apparent, since it is of
order e, whereas a term such as ^(qq:W)E in the Taylor expansion
is of order e2. A second differentiation gives
9 = — [co2p(e2 sin 0 -f- e3 cos 0)] + cbp[e2 cos 0 — e3 sin 0]
T 2co[(pe2)‘ cos 0 — (pe3)' sin 0] + [(pe2)“ sin 0 + (pe3)“ cos 0]
the four terms being of order 1/e, 1, 1, and e, respectively. These

expressions for 9, 9, and 9 are now substituted into Eq. (1.3) and
the resulting equation time-averaged over a gyration period, by
taking f02ir(.. .)dd and considering coefficients, such as (pe2)‘, to
be constants. Then (9) = (9) = (9) = 0, where the angular brackets
denote the average. The result of the time-averaging is
R = g(R) + -
m
+ - X B(R)
c
+
^ [e2 X (Ss-V)B - e3 x (e2-V)B] + 0(e) (1.4)

TflC £
since
(e X (e-V)B) - (p2co/2) [e2 X (e3-V)B - 63 X (&-V)B]. (1.5)
The coefficient p2a>/2 is an approximate invariant of the motion

and is Mc/e, where M is the well-known magnetic moment. The
adiabatic invariants will be treated in Chapter 3. The invariance
of M is not used in deriving the guiding center equations of motion
and therefore does not have to be assumed now.
The right-hand side of Eq. (1.4) can be simplified as follows:
e2 X (es-V)B = (63 X ex) X (es-V)B = 61 teu(e3-V)B]

- 63[e!-(e3-V)B]. (1.6)
Now
er (e3*V)JB = ei • (e3 • V) (^B) = (B/2)(63-V)^2 + 6 3-VB = e3-VB

(1.7)
since ei2 = 1. Therefore Eq. (1.6) becomes
e2 X (e3-V)B = ei[^3-(e3-V)B] - e3(63-V)£. (1.8)

Similarly
e3 X (e2-V)B = —e![e2-(e2*V)B] + ( -V)B.
62 62 (1.9)
The fact that, V • B = 0 must now be used. The operator V can be

expressed as ei(erV) + e2(e2-V) + e3(e3-V), so that
V B = er (er V)B + e2-(e2-V)fi + e3-(e3-V)B = 0. (1.10)
But er(erV)B = ei-dB/ds = dB/ds, where s is the distance

along the line of force. Therefore, by subtracting (1.9) from (1.8)
and using V • B = 0, one obtains
e2 X (e3-V)fi - e3 X (e2-V)S = -e^dB/ds)

- e2(e2* V)B - e3(e3-V)B - -VB. (1.11)
The time average of Eq. (1.3) is then
K = g(B) + (e/m)[E(R) + (A/e) X B(R)]

- (M/m)VB(B) + 0(e) (1.12)
with an initial velocity R(0) equal to [eiei-v + (cE X B/B-)]t=o

+ 0(e). Equation (1.12) is the basic differential equation for the
guiding center motion. See the Addendum on page 35.
C. The Drift Velocity and the Longitudinal Equation of

Motion
The guiding center Eq. (1.12) will now be solved by iteration to

obtain equations for the guiding center motion parallel and per¬
pendicular to B.
Crossing the equation with ex(R) gives the perpendicular com¬
ponent of the vector equation as
R - (6i K)ei = R±
cE X ei , Me eiX VB , me (g - R) X ei , „ 1DN
= ---1---1-5-1" ) (L13)
B e B e B
where f?j_ is the component of R perpendicular to ei (/?). It is the
drift velocity. The first term is the usual “E X B” drift. The second
term is the “gradient B” drift, and the third is the “acceleration
drift.” By dotting Eq. (1.12) with effR) one obtains the scalar
parallel component as
= + + ew. (1.14)
e e e os
In Eq. (1.13) the guiding center acceleration R is needed to calcu¬
late the drift velocity; but because the term in which it occurs
already contains e as a coefficient, R is needed only to zero order
in e. It is assumed that R is not of negative order, such as 1/e. If
it were of negative order, the fields would change by a large amount
in a gyration period when e is small, and the guiding center picture
would not be valid.
The acceleration R equals (d/di)(R± + eji-ei), and dR^/dt can
be obtained to zero order in e from (1.13) as dR^/dt — (d/dt){cE
X ei/B) + 0(e). Only the first term in the drift velocity (1.13) is
needed, since the third term is ~e and the second term contains
M/e — m(pw)2/2eB ~ e. If the perpendicular electric field happens
to be of order e instead of zero order, the retention of cE X ex/B

= uE will be unnecessary in the calculation of R. The acceleration
is
(1.15)
^ = dt R = It ^ ^ + Ue> + G<^
a dv\\ , . duE
R - ©(e) =
ei * + H + -M
dv i dei
A I I
+ (eiW|| + rt#)-VeiJ
= ei*+”•
[c)uE
+ (eit)|| + u^-Vue
dv|
- dt + rll + »ll2 ^ + VnUE-Ve!
duB buE . .
+ 4- »n + uE-VuE (1.16)
where means R-ei(R). Other “parallel” velocities can be de¬

fined, such as v-ei(r) or v-ei(R), but here V\\ always stands for
R •ei(/?). The first term is the tangential acceleration, the third is
the centripetal acceleration, the second occurs in nonstatic fields,
where the direction of the line of force changes with time, while the
last four terms occur in the presence of a zero-order electric field.
The presence of a zero-order electric field means that the electric
field is held constant in a series of experiments in which m/e is
successively reduced, with vQ and r0 held constant. The electric
field is of order e if it is reduced in proportion to m/e. Whether the
dei/dt term need be retained or not depends on how the time in
which fields vary is to be scaled in the series of experiments. If the
time scale is held constant, then d/dt is of zero order and dei/dt
contributes a first-order drift. If the time scale is increased in pro¬
portion to 1/e, then d/dt is of order e and de3/dt is not needed.*
* In a specified field it is never wrong in principle to make the more general
assumption that d/dt and uE are of zero order and to retain all terms. Some of
them may turn out to be small.
Of course, because m/e is fixed, it is really Bo, v0, or L that must be

varied in the series of experiments. Appropriate modifications of
the actual fields and time scale must be made to keep the dimen¬
sionless fields B, E, and G unchanged at a given R and T.
With expression (1.16) for R, Eq. (1.13) for the drift becomes
where uE = cE X ei/B.
The longitudinal Eq. (1.14) shows that E\\ = E-ei must be of
order e if R is to be of nonnegative order. Thus, in contrast to E L,
which may be of zero order, E\\ must be of order e. If this were not
so, the parallel acceleration would be ^1/e and the fields would
change by a large amount in a gyration period.
Equation (1.14) can be put in a form more useful for obtaining
an energy integral by rewriting R ei as
R ex = (d/dt)(R-ei) - R 4 = (dvl{/dt) - R & (1.18)
and noting that
R-ei = [eii)|| + uE + 0(e) ]-ei = uE-e i + 0(e)
= uE-[(dei/di) + (eir|| + us)-Vei] + 0(e). (1.19)
In Eqs. (1.18) and (1.19) we consider only the contribution of the

zero order motion to d/dt. This is all that is required, since R • ei
has e for a coefficient in Eq. (1.14). The longitudinal Eq. (1.14)
then becomes
m dv\\ m MSB m dei

g|| + E\\ + — uE • ~77 + 0^2)
e dt e e ds e dt
m MSB
g|| + E\\ +
e e ds
m / dei dei A \ oN
-UE[ + — + uE-VeiJ + 0(e2).
dt
(1.20)
Let us now introduce a true curvilinear coordinate system (a,
/3, s) such that a(r, t) and /3(r, t) are two parameters specifying a
line of force and therefore constant on it; s is the distance along
the line, as previously. For a divergence-free field such as B, a and
/3 can be chosen so that the vector potential A is aV/3 and B —
Va X V/3. To prove that this is possible, let Ci(r, t) and c2(r, t) be
two parameters which are constant on a line of force, but such that
A is not CiVc2. The lines of force are at time t the intersections of
the two families of surfaces given by Ci(r, t) = constant and c2(r, t)
= constant. Now ei — Vci X Vc2/|Vci X Vc2|, and since V • B
vanishes and equals V-(eiB), we have V-[fiVcj X Vc2/|Vci X
Vc2| ] = Vci X Vc2-V(B/|Vci X Vc2|) = 0. Consequently,
B/\ Vci X Vc2| is constant along a line of force, since Vci X Vc2 =
| Vci X Vc2|ei and ei-V = d/ds. Thus a general clf c2 coordinate
system has the property that B/\ Vci X Vc2| is constant along a
given line, but varies from line to line. Starting with a Ci, c2 system
for naming the lines of force, an a, (3 system can be obtained as
follows: Since B/|Vcj X Vc2| is independent of s, let it equal
^ (ci, c2). Then B equals |Vcj X Vc2. Let (3 = c2 and a = a(ci, c2),
where the functional form is to be determined. Because B is to
equal Va X V/3 it must be true that £Vci X Vc2 = Va X V/3, or that
£Vcj X Vc2 = [(da/dci)Vci + (da/bc2)Vc2] X Vc2.
Then £ = da/dci, and finally a = fn £(ci, c2)dci. It is therefore

possible to go from a general cu c2 system to an a, /3 system. Be¬
cause a is the integral of £ with respect to cu it is clear that the a, (3
system for a given field is not unique; an arbitrary function of c2

can be added to a(ci, c2). In an a, /3 system | Va X V/3| /B is
constant not only on a line of force but everywhere, being equal to
unity. This fact often simplifies the algebra, especially in connec¬
tion with the adiabatic invariants.
In Eq. 1.20 the parallel electric field is given by
B< = e,'(- C St - v*) “ _ + (1-21)
where </>(r, t) is the scalar potential for E, and \p is (a/c)(dj3/dt).

Multiplication of Eq. (1.20) by r|| and substitution of (1.21) for E%
gives (the g field will be omitted from here on, but could be retained
if desired)
d_ m dei d (MB \
dl - e' Tt ~ v‘ dS YT + * + V + ®( )-
(1.22)
Furthermore
d (MB\ _ 5 (MB\
+ e(e2).
dt \ e ) dt\e)
(1.23)
Equation (1.22) can then be written as
d_
dt
(m
2e
V+ ~ ds
+ HE • (1.24)
D. Energy Gain
The rate of change of kinetic energy of the particle, averaged

over a gyration period, can now be calculated. The average kinetic
energy is (mr,|2/2) + MB + (mizB2/2), since MB is the energy of
rotation about the guiding center; (mvl{2/2) + (muB2/2) is the
kinetic energy of the guiding center motion, since R equals ei«||

+ uE + 0(e). To give a more formal proof, start with v = R +
q = R + pco(e2 cos cot — e3 sin cct) + 0(e). Then
/mv2\ m . mp2eo2 TTL MB

— > = — R2 + —-— y (Ue* + «||2) + - + 0(e2).
v 2e / 2e 2e
From Eq. (1.24)

dB
dt
M _ , m dei m d ,,
+ uE ' — VB -\-uE-v|| — + — J + 0(e2). (1.25)
e e at 2e dt
Furthermore, in Eq. (1.25),
M „ m dei m d
• — VB UE ■ W|| + uE2
e e dt 2e dt
cE X ei . I(M ^ m dei , m duE\
B •(tvb
K e + 7’1 *+7 dt)
(Me me dei
X VB +
\ e e dt
ei Me „ me dei , tnc duE

= E
B
X
(- cE + —- VB d-+
e e ' dt e
,,
dt )]
= E-R±. (1.26)
Finally, therefore,
Id 5 M dB
- — (kinetic energy) = — ®ll ^ ^ ^ ^
e dt
+ 0(e2) (1.27)
12 ADIABATIC MOTION OF CHAKGED PARTICLES
- j (kinetic energy) = R E(R, t) H-—^ ^ + 0(e2).

e at e ot
(1.28)
This is not a surprising result and probably could have been guessed
initially. The term eR E is the rate of increase of energy due to
work done by the electric field on the guiding center, while M
(bB/dt) is the induction effect of a time-dependent field and is due
to the curl of E acting about the circle of gyration. One might
wonder why the second term does not have the total time deriva¬
tive dB/dt = (dB/dt) + »u(dB/ds) + uE-VB rather than the
partial derivative. The reason is that a magnetic field gradient [as
represented by ay (dB/ds) + uE-VB] does not change the total
kinetic energy, but merely interchanges energy between the per¬
pendicular and parallel components.
We now have the three fundamental Eqs. (1.17), (1.20), and
(1.28) for the guiding center motion and rate of change of energy.
E. Integration of the Energy Equation: Fermi

Acceleration; Rotating Plasmas
If d/dt and E (i.e., uE, \f/, and 4>) are 0(e) instead of 0(1), Eq.
(1.27) can be integrated. In that case
d(\f/ + 4>)/dt = V||d(^ -f- <£)/ds + 0(e2)
and Eq. (1.27) can be written as
d fmv||2 MB \
it h/ + v + * + *) -0 + <L29>
since Rj_-E = 0(e2). Thus mvn 2/2 + MB 4- + <£) is a constant
of the zero-order motion along a line of force. In a static field, \p =
0, and (1.29) is just conservation of energy.
Fermi Acceleration
Although it does not appear generally possible to integrate

Eq. (1.27) or (1.28) when E± is 0(1), there are special cases for
which it can be done. A very important example is Fermi accelera¬

tion (5-9). Fermi suggested that repeated collisions in space be¬
tween a charged particle and moving clumps of magnetized plasma
could accelerate some charged particles to the very high energies
observed in cosmic rays. Fermi had in mind special adiabatic

situations—namely, those where there is a frame of reference in
which the magnetic field is static. This “static” frame will in gen¬
eral differ from the observer’s frame, so that the particle may gain
energy in the observer’s frame, somewhat in analogy to a ball
struck by a baseball bat. A particle will lose energy in a collision
if the static frame is moving in the same direction as the particle
but more slowly, so that the particle overtakes it.
The magnetic field pattern of the static frame is carried along by
a clump of plasma containing a large number of low-energy par¬
ticles. These clumps may be considered as single massive particles
with which the high-energy particles attempt to establish thermal
equilibrium by collisions. A clump has large kinetic energy due to
its large mass, so that the high-energy particles will tend toward
very high energy at thermal equilibrium. The statistics of these re¬
peated collisions will not be discussed further here (see reference
7). Instead, application of the adiabatic theory developed in the
preceding sections will be made to single collisions of the type
Fermi envisaged.
The cosmic problem is relativistic, but the nonrelativistic theory
is adequate for illustrative purposes. Equation (1.28) applies, of
course, to any adiabatic situation, and the purpose here is to apply
this general equation to Fermi’s cases. He visualized two types of
collision. Type A occurs when a particle moves along a straight
line of force and is reflected by a moving magnetic mirror. In the
static frame the guiding center velocity is merely reversed by the
collision, as sketched in Figure 1.2A. With this fact in mind, the
energy gain seen by the observer is simply computed to be 2mu||
[Du (co) — «|| ], where u is the velocity of the static frame as seen
by the observer, ti|| is u ■ ei, and tty (°°) is the parallel velocity of
the guiding center as seen by the observer when the particle has
reflected and returned to infinity. There is no energy gain in the
static frame. Acceleration type B occurs when the lines of force are
curved but the magnitude of B is constant along the guiding center
trajectory (Fig. 1.2B). Again the guiding center velocity is reversed
in the static frame, and the energy gain seen by the observer is
2m«n(°°) [»n(°°) — M|| (00) ], where U||(°°) is the component of u

parallel to B at infinity. It will be shown how these results follow
from Eq. (1.28) applied to the observer’s frame.
As Eq. (1.28) stands, the two terms (M/e) (dB/dt) and R-E do
not correspond to types A and B collisions; but when there is a
static frame, Eq. (1.28) can be rewritten in such a way that one
term is type A and the other is type B. The electric field in the ob¬
server’s frame of reference is given by
E = -(u/c) X B* = -(u/c) X B. (1.30)
The magnetic fields B and B* are equal through order u/c—i.e.,

nonrelativistically. The rate of increase of kinetic energy (K.E.) is
given by
1 d K.E. „ M
V\\E\\ + R±E + — — + 0(e2) vÊ | + mb’VJB
e dt
m dei m duE MdB

-)-VnUE * ~7T + — Ue ' -jr + - -T7 + 0(e2) (1.31)
e dt e dt e dt
the latter form via Eq. (1.26). The quantities in (1.31) must now
be expressed in terms of u. Most of the following steps are self-
evident. E and d/c)t are taken as 0(1).
uE = c{E X ei/B) = u — (u-ei)ei = u±

(1.32)
B(r, t) - B*[r*(r, t)]
where r* — r — ut. Therefore
Vr£ =
and
dr*
f - V"B* ~dt
-u-VB. (1.33)
de i dei
+ Vn —-h uE-Vei + 0(e)
ht ds
— u-Vei + An ~ b Uj_-Vei
as
. dei dei
(1.34)
(”. - “i) ys = ”i aJ-
Similarly
duE du±
(®| - «|) ^ + 6(0
dt dt
~(®o ~ un) (gi“ ‘ ^ ^ M^) + ®(0
since du/ds = 0.
duE
uE ■ “(»B - «|)U1UX- (1.35)
dt
The relation between M and M* will be needed. By definition (see

Chapter 3), M ism(vx - uE)2/2B. But v± - uEis v± - u±, which
is Pj_*- Thus M equals M*
With the substitution of Eqs. (1.32) to (1.35), the rate of energy
gain (1.31) becomes
1 d K.E. M dB m dei
— »] r~ H-(»i - «|) Vl + e(e2) (1.36)
e dt 6 05 6 ds
for this Fermi case, in which there is a static frame of reference-

The first term on the right-hand side is now indeed Fermi type A
and the second term is type B, as will be demonstrated in more
detail presently by exhibiting the two special cases in which one
term or the other vanishes. In general both types occur simul¬
taneously.
An interesting observation can be made here: “betatron accelera¬
tion” and Fermi acceleration are not distinct processes. In Eq.
(1.28) the (M/e)(dB/dt) term is the betatron-type acceleration.
This term is indeed distinct from the R E term, which is accelera¬
tion due to guiding center motion. However, in the transition from
(1.31) to (1.36), the (M/e)(dB/dt) term loses its identity and com¬
bines with the (M/e)uE-'VB term to give — (M/e)u^(dB/ds),
which is Fermi type A. The remainder of the terms in (1.31) yield
Fermi type B. Thus, the betatron acceleration is part of type A and
is not a process distinct from the Fermi accelerations. However,
betatron acceleration is a process distinct from acceleration due to
guiding center motion in the electric field.
Type A acceleration occurs in Figure 1.2A. In the observer’s
frame dei/ds is zero at all times during the guiding center motion
into the mirror from infinity and back out again. The guiding
center drift velocity uE is just u±, which is also the velocity of the
line of force perpendicular to itself, so the guiding center remains
on the straight line. In other words, the guiding center motion in
the static frame is inward along the mirror axis and back out again,
with no drifts at right angles. The total energy gain by the reflec¬
tion is given by
where the integration extends over the guiding center motion from
infinity to the reflection time and back out to infinity. In a manner
similar to that in Eq. (1.34),
dB/dt = (flu — Uj) (dB/ds) (1.38)
so that
(dB/ds)dt = dB/{v^ - u„) = dB/v(1.39)
Now let K* be the constant kinetic energy existing in the static

frame:
K* = -^ + M*B* = ^-2 + MB. (1.40)

2 2
Then
(1.41)
and
r dB
r2
- (K* - MB)
-1/2
i AK.E. = -
e e J
-^mirror _rn
= (2m/e)u1i»1|*(®)
= (2m/e)uu[»u(°o) — 3 + 0(e2). (1-42)
Type B acceleration occurs in Figure 1.2B. Since dB/ds is zero,
- AK.E. = —
e e J - oo
feft(»n - U|)2ux •
Os
+ ®(£2)- (1-43)
Substitute for dei/ds from (1.34) to obtain, since V\* is constant,
- AK.E. = — Vn* f wx-c?ei (1.44)

e e J
By some simple geometry
f u±-dei = — /a“0_,r u sin a da = 2u cos (tt — a0) = 2uu(c°)

(1.45)
where a is the angle between ex and u, ao is the value of a at t —

— co, and uy( oo) is the component of u parallel to B when the line
of force has returned to infinity.
J AK.E. = y uu(co)[^(co) - «„(«)] + 0(e2). (1.46)
After rather cumbersome calculation, the expected result has

been obtained for both types A and B, which differ only in the
mechanism by which the guiding center velocity is reversed. As
pointed out previously, AK.E. is more easily computed by ob¬
serving that the parallel velocity is merely reversed in the static
frame. However, the object here has been to apply Eq. (1.28) in
the frame of reference in which there is an energy change and to
verify that the Fermi mechanisms follows as special cases of it.
It is relatively simple to integrate (1.36) without resorting to the
special cases in which one term or the other vanishes. Let the field
look somewhat as it does in Figure 1.3, where neither dB/ds nor

dei/ds vanishes.
1 Tr _ Mu r r2 ~] -1/2
- AK.E. -- 1 dB cosa - (K* - MB)

e e _m
1/2
mu
da since - (K* - MB) + 0(e2). (1.47)
e f m
Integration by parts of the second term on the right gives

1/2
1 __ —, m
- AK.E. = - - (K* MB) u cos a
e e m t = — a>
m
= ~ [«|*(oo)u|(c°) - »y*(— oo)Bu(— oo)]
= - { [®|(») “ «i(°°)]hb(®)
— [»]|(— °o) — U|](— co)]ua(— co) } + 0(e2). (1.48)
In the special case of type A, «y(— °°) = U|(°°) and «!*(— ro) =
—"i/|*(°°). For type B, U|(— «>) = —uy(co) and «j*(—°°) =
i'l*(°°). In either case, (l/e)AK.E. is (2m/e)uy(r| — U|).
If E± and d/dt are assumed of 0(e), then it turns out that [see
Eq. (3.48)] (l/e)(dK.E./dt) is the same as in (1.36), except that
it is plus 0(e3) instead of 0(e2); u± is now 0(e), since it is propor¬
tional to Ex, and Fermi acceleration is of 0(e2) instead of 0(e).
Rotating Plasma
Another special case in which Eq. (1.27) or (1.28) can be

integrated is that of a static magnetic field with rotational sym¬
metry (such as a mirror machine), and a static E, where Ex has no
azimuthal component and — 0 (Fig. 1.4). The potential 4> is
thus a constant on a line of force. Such a mirror machine has been
discussed by Longmire, Nagle, and Ribe (10). The zero-order drift
uE is in the azimuthal direction; the component parallel to B of
the resulting radial centrifugal force muE2/r has the desirable
property of making it more difficult for the particle to escape at the
ends. The effect is just that which would be observed if a bead
were placed on a smooth wire bent in the shape of the line of force,
and the wire then rotated about the z-axis. This analogy will be¬
come apparent in the following analysis.
Under the specified restriction on the E and B fields, all terms on
the right side of Eq. (1.27) vanish except R^ E, which reduces to
Figure 1.4. Mirror machine with large electric field perpendicular to B.

(mv$/e)uB-(uE-V)ei in this special case, and equals (mvg/e)

(cE/B)2e3-(e3-V)ei. Because e3-ei = 0, the factor e3-(e3-V)ei is
— er (e3-V)e3. But (e3-V)e3 = — er/r, where er is a unit vector
in the radial direction. Therefore — er(e3-V)e3 = ere,/r.
In order to integrate (cE/B)2(ei-er/r) over the zero-order
motion on a flux surface (defined as the surface formed by revolv¬
ing a line of force about z), the variation of cE/B and er er/r with
longitudinal position must be known. The following is a proof that
cE/rB is independent of position on a flux surface. Let SEr(r, z) be
the stream function (11) for the magnetic field. The stream function
has the properties that = constant is the equation of a line of
force and that Bz = (l/r)b^/dr and Br = — (l/^d^/dz. Since E
is perpendicular to B, flux surfaces are also equipotentials and c/> is
therefore a function of 4r. The components of electric field are
Er = —d(f)/dr = —(d(f)/d\J/)d'I'/dr and Ez = — (rf^/d'kjd'k/dz.
Thus E — [(d'k/dr)2 -f- (d'k/dz)2]1'2 dtp/d'V = rB(d<f>/d'ty), and
cE/rB = cdtji/d'f', which is constant on a flux surface. The quantity
cE/rB is the angular velocity of the uE drift about z and will be
denoted by Q. Therefore (rnv^/e)uE-(uE-V)ei is (m/e)vp,2rei-eT,
which equals, (m/e)(d/dt)(tt2r2/2) because
1 d Q2 /7/*2 r/p
- — (fl2r2) = — — = S22r — = S22r(e iV\ + uE)-Vr + 0(e)
2 at 2 dt dt
= Q,2Vf ^ = A2tf|jr(ei• er) + 0(e). (1.49)
The integral of Eq. (1.27) is then
mt’|2/2 + MB — m22r2/2 = a constant (of the zero-order motion
on the flux surface). (1.50)
If the subscript c denotes quantities at the median plane of

Figure 1.4 and e at the mirror (i.e., at the location of maximum
magnetic field on the flux surface), Eq. (1.50) becomes
v\e = v2le + {2MBc/m)(\ - Be/Bc) - G2re2(l - r2/r2). (1.51)

Therefore v2$e < 0, i.e., the particle is contained, if
< v\c[(Be/Bc) - 1] + uEc\ 1 - r2/r2). (1.52)
If M is set equal to zero in Eq. (1.51) the change in parallel

kinetic energy between the median plane and the mirror is (m/2)S22
(rc2 — re2), which is exactly the work done against the centrifugal
force. Thus when M = 0, the problem is that of the bead sliding
on the wire described previously.
Terms containing uE in the drift Eq. (1.17) give a small (order
e) motion in or normal to a flux surface, the zero-order velocity
being R = eirj + uE in the surface. When crossed with ei/B, the
third term in the square brackets is in the azimuthal direction. If
E is outward as in Fig. 1.4, the fourth term is
cE b&i
— e3 Vei - -vtfl — =
where & is the azimuthal angle in cylindrical coordinates and e2 is

a unit vector in the z-direction. When crossed with ei, this fourth
term gives a drift normal to the flux surface. The sixth term in
brackets is v0 (duE/ds) = -»|(d/ds)(flre3) = -vê^dr/bs) =
— r|fle3(er-ei); hence it is the same in this geometry as the fourth
term. The last term in the square brackets is fire3-V(fire3) =
fl2r2(e3-V)e3 = — S22rer. When crossed with ei this last term gives
another order e drift in the surface, in addition to the VB and line
curvature drifts.
Because of the two order e drift terms perpendicular to the flux
surface, there is an order e change in 0 (and therefore of kinetic
energy) as the particle traverses the surface. This change in <j> can
be calculated directly from the product of the drift velocity normal
to the surface and the electric field:
me
2- X -— D|fie3(er-ei) (~E)
dt ZB X
2mc dr
2me dr „ 2m dr
(1.53)
IF ”,si a«E ” “ T *’,nvS'
On integrating
A(e</>) - — mfl2A(r2) = — mA(uE2). (1-54)

The change in e</> caused by the first-order drift off the surface
equals twice the change in muE2/2 as the particle moves in zero
order on the surface. This result can also be obtained by energy
conservation. The total average energy associated with the per¬
pendicular motion is MB + muE2/2. Therefore (mvf/2) + MB +
(muE2/2) + e</> is a constant of the zero plus first-order motion.
But from Eq. (1.50), mvj2/2 + MB — muE2/2 is a constant of
the zero-order motion. By subtraction A(e</>) = —2A(muE2/2).
The drift normal to the flux surface is not cumulative, since the
sign of »[] reverses when the particle reflects near the mirror.
F. The Geometric Interpretation of the Drift Equation

and of the Longitudinal Equation of Motion
To get some physical insight into Eqs. (1.17) and (1.20), it is

instructive to look at the geometric reason for the occurrence of
some of the terms. All the drift terms in Eq. (1.17) arise because of a
variation at the gyration frequency of the curvature of the particle
trajectory. This variation results in a cycloid-like motion at right
angles to B. The reason for the variation at the gyrofrequency is
different for each drift. The familiar drifts cE X eJB, (mc/eB)g
X ei, and (Mc/eB)ei X VB have often been illustrated in the
literature (1) and will not be diagrammed here. For example, the VB
drift occurs because B varies during a gyration period, and there¬
fore the radius of curvature does also. The remaining drift terms
in Eq. (1.17) come from the it X ei term in Eq. (1.13), and are
usually described as the result of a d’Alembertian force due to the
guiding center acceleration. However, such a description does not
help one’s geometric understanding.
Consider, for example, a physical situation in which the dei/dt
drift appears. Let a magnet with large parallel pole faces as shown
in Figure 1.5 be rotated about the z-axis to give a dei/dt = i20y,
where ^ is a unit vector along the y-axis and is the magnet’s
angular velocity (Q0 co). Because there is a dB/dt there will in
Figure 1.5. Rotating magnet gives a dei/dt drift in the z direction.
general be an E and therefore the drift uE = -cex X E/B will

also occur. The two terms in Eq. (1.17) that are proportional to
ei X duE/dt and to ei X uE-VuE are not obviously zero, although
this will turn out to be the case. From cV X E = -dB/dt one
finds dEJdx = tt0B/c at zero time, if we take Ex and Ev zero.
Then Ez = c + Ez(x = 0). Now Ez(x = 0) also equals Ez
everywhere for £20 = 0. Let us assume there is no E in the absence
of rotation, so that Ez(x = 0) is zero. The drift is
uE = cE X ex/B = x£l0y at zero time. (1.55)
Since uE is independent of y, uEVuE = 0. Also, duE/dt is paral-

Figure 1.6. Geometric explanation of the d&i/dt drift.
lei to ei, so that ei X duB/dt = 0. Thus, both of these drift terms

vanish, leaving
R± = cE X (ei/B) + v$(mc/eB)ei X dei/dl = yfl0x + zv^l0/w.
(1.56)
The dei/dt drift is perpendicular to the page and of magnitude

Vyflo/W.
The parallel equation of motion (1.20) becomes
dv^/dt = u^-dei/d/ = (1-57)
This is just the centrifugal acceleration at a distance x from the

axis of rotation.
The geometric reason that the curvature of the trajectory varies
in the presence of dei/d/ is that the perpendicular velocity | v X ei |
varies as ei changes direction. The drift velocity can be derived
(except possibly for a numerical factor) by holding ei fixed for
half a gyration period, then changing its direction for the next half
period, and so on. A view along the x-axis of Figure 1.5 will appear
as shown in Figure 1.6.
Let 8 be the angle between v and ei, so that v± is v sin 8 and V\ is

v cos 8. At the end of the first half-period (y > 0) let ei change by
Aei in the y-direction. For the second half-period (y < 0) v± will be
changed by = v cos 8 A8 = rj A8. The drift velocity equals the
difference in the diameters of the two semicircles divided by the
gyration period, or w(p2 — pi)/tv. Since p equals vx/a, Ap is Ar±/w
or WjAS/co, and A8 = Q,qtv/w. Thus the drift velocity equals rjfJo/co,
which in this case happens to be correct even to numerical factors.
Similar geometric derivations can be given of the other drifts
containing »q and uE in Eq. (1.17).
In Eq. (1.20) the parallel acceleration of the guiding center
comes from several sources. It is obvious that the parallel com¬
ponents of g and E will accelerate the particle along the line of
force. The third term — (M/e)(dB/ds), is the well-known mirror
effect. It can be understood by considering the special case of a
particle gyrating about the axis of symmetry of a mirror-type
field (Fig. 1.7). The force on the particle is (e/c)v X B(r) =
(e/c)(ei + vx) X (Br + Bz) = (e/c)vx X (Br + Bz). The
(e/c) v'± X Bz term is a radial force and gives the centripetal ac¬
celeration. The (e/c)v± X Br term is directed to the left and
equals — M(dB/ds), since the Br at a gyration radius from the z-
axis is related to dR/ds by V B — 0.
The last term in Eq. (1.20) is (m/e)R-ei and obviously arises
because the parallel velocity z?y is altered not only by an accelera¬
tion R of the guiding center, but also by a change in direction of
the B field without a change in R. A specific example of this term
was seen in Eq. (1.57).
G. The Relativistic Case
The drifts and longitudinal equation of motion for a particle of

relativistic energy can now be written, at least for the case in
which Ex and d/d^ are of 0(e). To begin with, let us consider the
case in which E — 0; the relativistic equation of motion is
moydv
(1.58)
dt
since y is constant in the absence of E. Thus the trajectory is cor¬

rectly given in detail by the nonrelativistic equation for a particle
of mass m0y. Therefore the guiding center drifts are given by Eq.
(1.17) with E set equal to zero and m replaced by moy.
ei m0yvx2c m0yc
Ri = X -—— vn + - (1.59)
B 2 eB e
or, in terms of momenta,
Ri =
ei
X ^ VB + ii a! ^1 (1.60)
B ye ye m o os J
where MT is p±2/2maB, the relativistic magnetic moment.* To

* It is simple to demonstate that Mr is the adiabatic invariant for the special
case of a uniform, azimuthally symmetric, time-dependent field with the par¬
ticle gyrating about the axis of symmetry. It is also true (see reference 13) that
PxV2m0B is the adiabatic invariant in a general geometry so long as E is
0(e). and is the flux of B enclosed by the circle of gyration.
understand the relativistic effect physically, consider the VB drift

term in Eq. (1.59). It is y times as large as the nonrelativistic ex¬
pression for the same v±. Relativistically the particle has y times
the mass, hence y times the radius of gyration, and therefore
samples y times as much of the B field inhomogeneity as it would
nonrelativistically. Since the VB drift comes from the small dif¬
ference in radii of curvature on opposite sides of the “circle” of
gyration, the larger the radius, the larger the difference, and
therefore the larger the drift velocity will be. The relativistic effect
will be clarified by an order of magnitude calculation similar to
that for the dei/d/ drift in Figure 1.6. The radius of gyration is
p = moycv^/eB. (1.61)
The difference in radii on the two sides is
Ap . - AS = - 55Si pVB. (1.62)

eB‘ eB2
The drift velocity due to VJ5 = —^.Pî = w| Ap| = m°y^± upVB

period eB2
pVB
= »x (1.63)
B
Since p is y times larger relativistically than nonrelativistically,
the drift velocity is also y times larger.
A similar argument holds for the term in Eq. (1.59) containing
dex/ds. Nonrelativistically, U|,2(dei/ds) is v^dei/dt, and the ex¬
planation for the drift is similar to that given in Section I.F for
the dei/dt drift. Relativistically, the increase in mass multiplies
the gyration period by y. The difference in radii of curvature be¬
tween one half-period and the next is (from Eq. 1.61)
Ap = (m0yc/eB)Avx. (1.64)
The drift velocity due to dei/d$ is
| Ap|/period ^ co| Ap| = Av±

= («|/wXdSx/ds) (1.65)
and 1/co is 7 times larger relativistically than nonrelativistically.
The parallel equation of motion (1.20) in the absence of E

becomes
m0y(efo|/G?0 = — (moyv±2/2B)(dB/ds) (1.66)
or
dpi/dt = —(Mr/y)(dB/ds). (1.67)
From Eq. (1.66), the parallel force is y times larger than it would
be nonrelativistically. This can be understood from Figure 1.7 and
the nonrelativistic explanation of the mirror effect. Relativisti-
cally, the radius of gyration is larger by y, and therefore BT is
larger by y at the position of the particle because of the convergence
of the field lines.
If E is different from zero and if the fields are nonstatic, the drift
and parallel equations cannot logically be obtained from the non¬
relativistic ones, since y is no longer constant. One might surmise
that the drift Eq. (1.60) would be modified by the addition of the
term cE X ei/B, and that the longitudinal Eq. (1.67) would have
the term eE^ added. This surmise is correct for the case in which
E± is 0(e), and the relativistic equations become
(1.68)
dPD pF Mr dB
(1.69)
Tt eE' ~ T a?
- = MT = constant, where p .is the
2 m0B
perpendicular relativistic momentum. (1.70)
Relativistic: E'j, E±, and d/dt ~ e.
The total guiding center velocity ei«| + R± can be written in a

rather unusual fashion (12) when the electric and magnetic fields are
static and ei • V X B vanishes. Let K be the constant total particle
energy (p2c2 -j- mo2c4)1/2 + e<p. Consider the quantity (cv§/eB)V X

(eiP|).
V X (eip0) -
eB
K — e<j> e
eB
PyV X ei +
PS \
E X ei + m0Mrei X V5
)]
CP|'
cE X ei , Mrcei X VB
m0yeB
V X ei +
B
+ yeB
cpf $*A
R± +
m0yeB ( VXei-exX—)•
ds /
(1.71)
A little algebra with unit vectors shows that V X ei — ei X dei/ds

equals (ei/B)er V X B. Since er V X B vanishes,
R = + Rl = (vt/B)V X A' (1.72)
where A' = A + cpj/e. The guiding center consequently has as

trajectory the magnetic lines of force derived from the vector
potential A'(r) = A(r) + (c/e)p^(r).
The more general relativistic case in which E± is 0(1) has been
studied by Vandervoort (13). The method is necessarily covariant
and is not restricted to cases in which E0 is of 0(e). The analysis will
not be reviewed in detail here; the method will merely be indicated
and the resulting guiding center equations of motion given [for
the case in which is 0(e)]. The starting point is the relativistic
four-dimensional equation of motion (14)
d2x{ dxk
= (1.73)
dr2 k dr
where
0 Bz — By - iEx\
-B2 0 Bx -iEv \
E oc — (1.74)
m0c 1 B
Bv -Bx 0 — iEz Y
iEx iEv iEz 0 /
and Xi is the four-vector (x, y, z, id) and r is the proper time. The
first three components of Eq. (1.73) are the relativistic vector
equation dp/dt — (e/c){dr/dt) X B + eE, and the fourth compo¬
nent is the rate of change of energy d(moyc2)/dt = ev-E, where v —
(x y, z). If Fik is independent of xt (i.e., fields independent of position
and time), the solution of Eq. (1.73) exhibits a gyration in four¬
dimensional space at the frequency
E2 1 ) 1/2
— + - [(B2 - E2)2 + (1.75)
where coT is the angular frequency in radians per unit proper time.
The actual gyration frequency is to = uT(dr/dt) = wT/y radians
per unit real time. When E = 0, the frequency « reduces to the
usual relativistic frequency eB/m0yc.
The guiding center equations of motion are the equations of
motion with the gyration at frequency oj averaged out. There are
three equations, corresponding to the first three components of
Eq. (1.73), which give the actual guiding center motion in three-
space. Corresponding to the fourth component, there is an equa¬
tion for the average (over a gyration period) rate of increase of the
total particle energy. Written in the present notation, these equa¬
tions are given in (1.76^-1.79).
Here p±* is the perpendicular momentum the particle has when
observed from the frame of reference moving at uE, and B* is the
magnetic field observed in that frame. It is given by B* = B( 1 —
Ej_2/B2)i/2 + 0(e). Mr is actually proportional to the flux through
the circle of gyration, as observed in the frame of reference moving
at uE■ When E± is 0(e), p±*2/2m0B* equals p±2/2moB to lowest
order in e, and Mr is as defined previously for that case.
In Eqs. (1.76) to (1.79), y oscillates at the gyrofrequency. How¬
ever, this oscillation can be averaged out to give yavg = y*(\ —
E±2/B2)-112.
Equations (1.76) and (1.77) are the guiding center equations of
motion in three-space, while (1.79) is the average rate of energy
increase. Because of the denominator 1 — Ex2/B2, it is apparent
that Ex must be less than B for the equations to be valid.
ADIABATIC MOTION OF CHARGED PARTICLES
32
ft
x
= ei_ X
B(1 - Ex2/B2) {- 0 - ¥) cE
1/2" m0cy( de i duE\

+ v +
ye “TV Ht+~di )
/ j? 2\ i/2“n
Mr 5
+ û«+—J}
c ye c 5/
+ 6(e!)UE
(1.76)
m0c?(7W||) dp{ | rfei

__ = - = mo7u, dt
+ etf, -
Mr
7 5s
B
0 - ¥)1
1 (1.77)
*2
Px = Mr = constant (1.78)
2m0B*
— (m0c27) = e(Rx + ei v^-E

dt
Mr5
+ 7 &
B
0 - f)1 (1.79)
Relativistic: Ea = 0(e), Ex, and 5/5/ = 0 (1).
Equations (1.76), (1.77), and (1.78) are, respectively, the gen¬

eralizations of the relativistic Eqs. (1.68), (1.69), and (1.70) to the
case in which Ex is 0(1) instead of 0(e). If Ex is 0(e), then so are
uE and Ex/R, and Eqs. (1.76) and (1.77) reduce to (1.68) and
(1.69) upon dropping terms of order e2.
Equations (1.76), (1.77), and (1.79) are, respectively, the rela¬
tivistic forms of (1.17), (1.20), and (1.28). A comparison of (1.76)
and (1.17) shows that relativistic effects not only modify the
existing terms in (1.17), but also introduce two new drift terms in
the direction of ei X uE, i.e., in the direction of Ex. It is possible
to prove directly that these two new drift terms are of order r2/c2
smaller than the others, and therefore are indeed purely relativistic
effects. To show this, the relativistic equation of motion dp/dt =
(<e/c)(dr/dt) X B + eE must be written in dimensionless form.

The scaling is similar to that for the nonrelativistic equation. Let
B = B(r, i)/B0(t), E = (m0c/p0B0)E(r, t), T = p0t/m0L, and R =
r/L, where p0 is the initial momentum m0t>o/(l — 18o2)1/2 and the
other symbols are the same as they would be nonrelativistically.
In terms of these dimensionless quantities, the equation of motion
becomes
p0c d jdR ( po_ dky — 1/2
dft
-[f XB + E (1.80)
eBoL df \df \m0c df)
with the initial conditions that at f = 0, R = r0/L, dR/dT —
vo[1 + (po/m0c)2]~1/2 = vo X rest energy/initial total energy.
The problem now contains the two dimensionless parameters
poc/eBoL and p0/m0c, whereas the nonrelativistic problem con¬
tained just the first one. The new parameter p0/m0c is 70v0/c =
v0/c + O(v03/cs). An order of magnitude comparison of, say,
Ex 2\i/2_
— — - ("b (1
ye c dt L \ B2 )
with the term
gives
^ H* - [B( 1 - E±2/B2y2] /— V [B( 1 - E±2/B2)112]

ye c M / ye
~ Lue/cH ~ LE/cBt. (1.81)
But LE/cBt = (p0/m0c)2 E/BT = (v02/c2)E/BT and therefore is

of order v2/c2. Similarly, (viEl/c)uE is of order v2/c2.
It may seem strange that drifts in the direction of E± occur.
The origin of the drift which is proportional to i\E^ is easily under¬
stood. Because of the parallel electric field, the magnetic field is
not in the same direction when viewed from the frame of reference
moving at uE as when viewed from the laboratory frame. Conse-
quently a velocity which is parallel to B* (asterisk refers to uE

frame) will have components both parallel and perpendicular to B
when observed from the laboratory frame. Suppose that in the uE
frame there is a uniform static field B* and an electric field E*
parallel to it, as shown in Figure 1.8(A). The guiding center
velocity R* will consist of pj* only. When a Lorentz transforma¬
tion along x* is made to the laboratory frame, the fields and
/
/
/
Figure 1.8. The explanation of the drift proportional to Ej_.
guiding center velocity appear as shown in Figure 1.8(B). The B

and E vectors lie in the y-z plane, but B does not lie along z and E
is not parallel to y. The angle between B and the z-axis is propor¬
tional to E$. As shown, R± can be resolved into two components;
one is uE and the other is the drift (vÊÊL)/B2(l - Ex*/B2) in
the direction of E±.
The origin of the last drift term in Eq. (1.76), which is also in
the direction of Ex, is less obvious. One can start with a magnetic
field having straight lines of force and a gradient of B at right
angles to B. There is a drift velocity due to VB. If one now makes
a Lorentz transformation to a frame (denoted by an asterisk)
moving along VB, an Ex* and a dB*/dt* will appear. The VB drift
velocity, when transformed to this new frame, equals the sum of
several drifts as calculated for the new frame from Eq. (1.76),
including this second drift term in the direction of Ex*. Making the
transformation does not really make the origin of that drift more
transparent geometrically and therefore it will not be done here.
Finally, a comparison of Eq. (1.77) with (1.20) and Eq. (1.79)
with (1.28) shows no new terms, only a modification of existing
terms.
Addendum
The last paragraph under SectionB,page6,is supplemented as fol¬

lows. Without the order e term it is the same as the equation of mo¬
tion of a particle in a magnetic field B and an electric field E — MvB/
e, and therefore is more complicated than was the original equation
of motion (1.1). A solution of Eq. (1.12) (without the order e term)
would show that R travels in roughly a helix about the field line,
just as the particle does. However, it can be shown that the
radius of this helix is one order of e smaller than the radius of the
particle helix. Consequently, this smaller helix can be ignored in
the present first order theory. The vector R0 in Chapter 2 has a
more complicated definition, and is a guiding center that shows
no gyration in any order.
CHAPTER 2
A More Formal Derivation of the

Nonrelativistic Guiding Center
Equation
The derivation of the guiding center equations of motion for

nonrelativistic particles presented in Chapter 1 requires rigorous
justification. The work of Kruskal (15) and of Berkowitz and Gard¬
ner (16) provides the justification for Eq. (1.4). Kruskal derives
equations for the Rn appearing in a series of the form
r (2.1)
by equating coefficients of equal powers of exp [(i/ec) f B dt] after

substituting the series into Eq. (1.1). The fields are expanded in
Taylor series about R0. Each Rn is itself a power series in e, of the
form Rn(t) = Rno(t) + eRni(t) + etc. J?_„ must equal the complex
conjugate of Rn, in order for r to be real. It is not immediately ob¬
vious that equating the coefficient of each ein9 [where 9 = f w(R0)
dt] to zero is justified; the Rn are functions of time, so that the
series is not simply a Fourier series. However, Berkowitz and
Gardner prove that the series obtained by this process is actually
an asymptotic expansion of r for small e. Their proof is necessarily
a formal mathematical one and will not be repeated here. But it is
worthwhile discussing what is usually meant by an asymptotic
expansion and what it is they proved.
The usual definition (17) of an asymptotic power series expan¬
sion can be stated as follows. Given: the power series S(e) = «o +
36
NONRELATIVISTIC GUIDING CENTER EQUATION 37
die + a2e2 + .... It is called the asymptotic expansion in

e of a function/(e) if for all n lim^dê) — <S»(e)|/en) is zero,
where Sn(e) is the sum of n + 1 terms (i.e., including anen) of the
series. By the definition of a limit, for every number Q there is a
number e0(Q) such that |/(e) — Sn(e)\ /en is < Q for e < e0. (We
take e as positive only.) If this is true for all n, one can show that,
given a positive number A, there is a value of e, say eh such that
|/(e) — Sn(e) | < Aen+1 if 0 < e < ei(A). This is the precise meaning
of the statement that/(e) — Sn(e) is 0(en+1). Now the series in Eq.
(2.1) is not a simple power series in e. Each term is itself an infinite
series in e, multiplied by exp (infBdt/ec); and this exponential
cannot be expanded in a power series in e. Nevertheless, Berkowitz
and Gardner prove that if is the sum of terms from R-n to Rn
inclusive, then there is an A{tx) for which | r(e, t) — £„| < A(t1)en~1
if e < d and t is < /. Therefore, r — = 0(en_1) and £„ is an
asymptotic series for r, but is of a more general form than a power
series (18). A en_1 serves as an estimate of the accuracy of E„. With
n equal to 2, it follows that r — S2 = 6(e). But S2 = Ro + e{Rxe%d
+ R-ie-*) + e2(R2e2ie + R-2e~2ie), so that S2 - R0 = 0(e). By
subtraction, (r — S2) + (S2 — Ro) = r — R0 = 0(e). This means
that the difference between the actual particle position and R0 is of
first order in the radius of gyration, and therefore R0 is a suitable
definition* of the guiding center position.
Following Kruskal, we will now derive equations (some alge¬
braic, some differential) for the Rn. It will be found that the equa¬
tion for Ro, the guiding center position, is Eq. (1.4). Substituting
Eq. (2.1) into Eq. (1.1) and collecting coefficients of em6 gives,
for n = 0:
eRo = E(Ro, t) + RoX R(Ro, t) + ieB(R0, t) [Rx X (R-i-V)B(Ro, t)
- R_! X (RrV)R(Ro, t)] + 0(e2) (2.2)
* R, as defined at the beginning of Chapter 1, and R0 do not agree to all

orders in e. In fact, R — R0 = 0(e2). However, this is of no significance where
we are considering only effects which are first order in gyration radius.
for n = 1:
B2Ri + iBRi X B = -e[RvVE + Ri X B +
Ro X (RvV)B - iBRi - 2iBRi] + 0(€2) (2.3)
for n — 2:
-4B2/?2 - 2iBR2 X B = X (Ki-V)B + 0(e) (2.4)
for n £ 2:
— n2B2Rn — inBRn X B = Gn (2.5)
where Gn is a function of the R’s. The velocity of light c has been
taken equal to 1. It can be reintroduced by dividing B and B by
c. In all four equations the fields are evaluated at Ro and t. In
Eq. (2.5) the lowest order of Gn, denoted by Gn0, will always con¬
tain only Rpo (where 1 < p < n — 1). For example, from Eq. (2.4),
G2o = iB(R0, t)Rio X (Rio-V)B(Ro, 0- general Gnl will contain
Rn. It is thus possible to solve algebraically for Rn0 in terms of the
Rio, R2o,--., Rn-i,o. To solve Eq. (2.5) for Rn0, take its scalar
product with ei = B/B to get
Rno-ei = -(Gn0-^/B2n2) (2.6)
and its cross product with ei
n2Rno X ei -f in(eiRno-&i — Rno) = — (Gno X ei/B2). (2.7)
Substitute Rno-ex = —(Gn0-ei/n2B2) from Eq. (2.6) and Rn0 X e1
= (i/n) [n2Rno + Gn0/B2] from Eq. (2.5) into Eq. (2.7); the result¬
ing equation can be solved for Rn0:
— n2Gnp -|- (ei-Gnp)ei -f- in Gnp X ei

RnO (2.8)
B2n2{n2 — 1)
where the fields and ei are evaluated at (Ro, t) as usual.
i,
To find Rn Eq. (2.5) is written to next order as
-in2BRnl - inBRni X B = Gnl (2.9)
where Gnl will contain Rp (p < n — 1) and Rn0. Thus a recursion

scheme exists for R2 and higher, provided R{) and Ri are known so
that the recursion scheme can be initiated.
NONRELATIVISTIC GUIDING CENTER EQUATION 39
Equations (2.2) and (2.3) determine R0 and Ru The zero-order

equation (2.3) is
Rio + iRio X ei = 0. (2.10)
Dotting with ei(R0) gives R10 • ei = 0, so that R10 is a vector per¬
pendicular to the magnetic field. Therefore, let Rw equal (a +
ib)e2 + (c + id)63, where e2(Ro) and e3(/?o) are perpendicular to
each other and to ei(Ro) as previously, and a and b are real. Sub¬
stitution into Eq. (2.10) gives c = —b and a = d. As a result, Rw
is of the form
Rio — (a + ib)(e2 + ie 3). (2.11)
Equation (2.11) contains all the information present in Eq. (2.10).
If Rw from (2.11) is now substituted into the square brackets in
Eq. (2.2), the result is
eRo = E + Ro X B + 2e£(a2 + 62)[62 X (e3-V)B -

e3 X (e2-V)jB] + 0(e2) = E + R0 X B
+ eBRw■ Rio*[e2 X (e3-V)B - ^3 X (e2-V)B] + 0(e2). (2.12)
The coelficient BRW- Rw* will now be identified as the magnetic

moment by differentiating Eq. (2.1) with respect to time. Since
Rn = R-n*, differentiation yields.
v = r = Ro + iB{Rweie - Rio*e~ie) + 0(e) (2.13)
(v - Ro)x = iB(Rweie - Rio*e~iB) + 0(e) (2.14)
since Rw is perpendicular to B by Eq. (2.11). Squaring (2.14)
gives
(v - Ro)x2 = 2B2Rw Rw* + 0(e) (2.15)
because Rio Rio = 0. However, (v — R0) j_ is the perpendicular or
gyration velocity in the frame of reference moving at the guiding
center velocity, and this is pw. Therefore, eBRw Rw* — ep2co2/2B
= p2w/2c = M/e, and Eq. (2.2) becomes
eRo = E(R0, t) + (Ro/c) X B(R0, t)

+ (p2co/2c) [e2 X (^-V)B - e3 X (e2-V)B] + 0(e2) (2.16)
which is the same as Eq. (1.4) for R. We have thus justified the
averaging process used to get Eq. (1.4) in Chapter 1.
In the next chapter, Eq. (2.3) will be used to show that (d/dt)
(BR10-Rio*) = 0 + 0(e), and this will be a proof of the adiabatic
invariance of the magnetic moment in the general case in which
E± and d/dt are of 0(1).
CHAPTER 3
The Adiabatic Invariants

of the Motion
Since the solution of the equation of motion has been obtained

as an asymptotic series in e in the previous section, it is reasonable
that any approximate constants (adiabatic invariants) of the par¬
ticle motion should be obtainable as asymptotic series in e. In
analytical dynamics exact constants of the motion are usually ob¬
tained by the canonical formulation: if the Hamiltonian is inde¬
pendent of a given coordinate, the conjugate momentum is an in¬
variant. By analogy, we expect that an expansion of the Hamil¬
tonian in an asymptotic series in e should reveal the adiabatic
invariants, provided that at each step of the expansion, variables
can be found which make the Hamiltonian independent of one of
the coordinates. (The remainder term may contain the coordi¬
nate.) The systematic procedure for finding the proper variables
has been given for the nonrelativistic case by Gardner (19).
It is similarly true that to find exact invariants the Hamiltonian
must be expressed in terms of the proper variables. As an example,
if the Hamiltonian for a charged particle in a magnetic field having
azimuthal symmetry is written in rectangular coordinates, it is
not at all evident that the canonical momentum P# = mr24 +
(,e/c)rA# is an exact invariant of the motion. It only becomes ap¬
parent when H is written in cylindrical coordinates.
A general theory of asymptotic solutions and adiabatic invari¬
ants of coupled first-order differential equations of a certain type
exists; the equation of motion of a charged particle is a special
41
case. The most complete form of the theory is due to Kruskal (20)
and will be discussed in some detail later in this chapter.
Although the adiabatic invariants are asymptotic series of the
form: constant = a0 + eai + e2a2 + ..., it is customary to speak
of the lowest-order invariant ao as “the” adiabatic invariant. For the
charged particle there are as many as three such series, one for the
magnetic moment, one for the “longitudinal” invariant, and one
for the “flux” invariant. These three series will be designated by
M + eM' + ..., J + eJ' + • •and $ + e4>' + ..., respec¬
tively. Proofs of the invariance of the lowest orders M, J, and <t>
will be given below.
The number of adiabatic invariants is less than or equal to the
number of degrees of freedom of the system. The charged particle,
which has three degrees of freedom, will have M, but may or may
not have J and <t>, depending on the field geometry. The number of
adiabatic invariants is determined by the number of periodicities.
To illustrate, suppose that B is nowhere large enough to reflect
the particle. The particle motion is nearly periodic because of the
gyration about the line of force, but there is no semblance of
periodicity in the motion along the line of force. There would only
be one adiabatic invariant series, the one for the magnetic moment,
even though there are three degrees of freedom. If the field is such
that a particle is always trapped and oscillating between two
mirrors, there will be a second or longitudinal invariant J asso¬
ciated with the parallel motion. Finally, if the drift from line to
line as the particle oscillates between mirrors with constant M and
J carries the particle repeatedly around a closed surface, there is a
third periodicity associated with this motion and a third adiabatic
invariant <f> will exist. The charged particles which comprise the
Van Allen radiation possess all three periodicities and invari¬
ants (21). The periodicities are the gyration about the geomagnetic
field lines, the north-south oscillation, and the precession about the
earth.
A. The Magnetic Moment
In this section two proofs of the invariance of the magnetic

moment will be presented. The first proof assumes static magnetic
THE ADIABATIC INVARIANTS OF THE MOTION 43
and electric fields, so that conservation of energy can be used.

The second proof holds for the most general case of time-dependent
fields. '
Before giving the proofs, it should be emphasized that the mag¬
netic moment is not always mvx2/2B, with v± the perpendicular
velocity in the laboratory frame of reference. If there is an electric
field, the perpendicular velocity must be that observed in the frame
of reference moving at the guiding center velocity—i.e., (v —
R0)± = (v — uE)x -f- 0(e). This is almost obvious geometrically,
since when E± is large the particle trajectory in the laboratory
frame may not be at all circular, but may be a prolate cycloid. It
does not matter whether the component of v — uE perpendicular
to B(r) or to B(R0) is used; the difference is of 0(e) and therefore
appears in the eM' term of the magnetic moment series.
Proof for static fields:
From Eq. (2.13)
= R0± + iB(R10ei9 - R10*e~ie) + 0(e) (3.1)
(^2> = R0± 2 + 2B2Rw-Rio* + 0(e) (3.2)
where ( ) means averaged over 9 [one can equally well use here the
9 = p(e2 sin 9 + e3 cos 9) notation of Chapter 1]. By Eq. (2.15),
2B2Rw-Rio* is 2MB/m. Furthermore, R0±2 is uE2 + 0(e).
(v±2) = uE2 + (2MB/m) + 0(e). (3.3)
By conservation of energy
Id / m(vx2) mv^
(3.4)
e dt V 2 + T~
or, by Eq. (3.3),
Id f muE2
+ MB (3.5)
e dt V 2
On the right-hand side, Eq. (1.20) is used for (rri/e) (dv{]/dt). More¬
over,
= v„^ + Ro±-V<j> + 0(e2)

dt ds
d<?!>
= W[i —-h uE- V<t> +
os
ei (Me , me dei me duE\
- X — VB d-«n rrr 4--jr) ■ v</> + G(«2)
B \ e e dt e dt J
via Eq. (1.17). The v^/ds) is —vÊf and uE-V<t> is zero.
m de i duE\
m f/uj
— VB H— »j -77 + + ©(62).
dt = (te e dt e ~dt)
(3.6)
Equation (3.5) becomes
1 d(MB)
= |j+ ue-Vb) + 0(e2)
dt
M dB
= - -17 + e(€ ) (3.7)
e ai
or
(l/e)(dM/cft) = 0 + 0(*2). (3.8)
The second and general proof is longer because an energy con¬

servation law is not available. It will be shown via Eq. (2.3) and
Maxwell’s equations that (m/e) (d/ dt) (BI?10 • Rio*), which is
(1 /e)(dM/dt), equals zero plus 0(e2).
Equation (2.3) is of the form
R1 + iRi X ex = -(eF/B2). (3.9)
If this is first dotted and then crossed with ei in an attempt to

solve for JRj in a fashion similar to the solution for Rn0 [Eq. (2.8) ],
the result is [by setting n = 1 in Eq. (2.8) ]
-F + (e!'F)e: + iF X ej = 0
or
Fj. - iFx X ej = 0. (3.10)
This is a condition on F±, which to ( ) is a condition on F0±,

0 1
where
Fo = Rio’VE(Ro, t) -f- Rio X B(Ro, t) T-
Ro X R10 VB(R0, t) - OKio - 2iB(«„, flJftio. (3.11)
Equations (3.10) and (3.11) constitute a differential equation for

Rio. If the terms with Rw are separated on the left-hand side, the
differential equation is
(Rio)± - i(iho)x X ej = (1 /B)(L± - iL± X ei) (3.12)
where Lx = [-iRw-VE — iR00 X (K10-V)B — 5Kio]jl and R00

is the zero-order motion êi + uE. One might conclude that
(Rio)± must equal EJB. Such is not the case, however; any
complex vector quantity of the form W± — iW± X ei can be
written as (g + ih)(e2 + ie3), where Wx = ge2 + he3, and g and h
are complex. By collecting all terms on one side of Eq. (3.12),
one finds that (Ri0) ± — (L±/B) is also of this form and is not neces¬
sarily zero. Equation (3.12) must be used as it stands to prove that
(d/dt){BRio-Rio*) equals zero.
(d/dt)(BRw-Rio*) = B [Rio-Rio* + Rio-Rio* + {8/B)Rw-Rio*].
(3.13)
Equation (3.12) contains only (Rw)±; but this is all that is re¬
quired in Eq. (3.13) because of the fact that Rw* and Ri0 have no
parallel components. Now Eq. (2.11) gives, when differentiated
with respect to time,
Rio — (a + ib)(e2 + ies) + (a + ib)(e 2 + ie 3). (3.14)
Substitution of this into Eq. (3.12) gives (via either the e2 or e3

component)
a + ib — —i(a + ib)(e 2-63) + (g + ih)/2B (3.15)
where (g + ih) (e2 + ie3) now stands for L± — iLx X ei. From
Eqs. (3.14) and (3.15) it follows that
Rio — 4" ib)(e2-63) + (g + ih)/2B](e2 + ie3)

-(- (a + ib) (e2 "I- 163) = — ie2 • e&Rio
+ [((/ + ih)/2B(a + ib) ] Rio + (a + ib) (62 + 463) (3.16)
RwRio* — {— ie2-e3 + [(</ + ih)/2B{a + i'6)]}i?io-Kio*

+ i (a2 + 62)(e2-e3 — e3-e2). (3.17)
The sum of Eq. (3.17) with its complex conjugate is
2(Re(R10-Rw*) -
Rio-Rio*
B
(Re (\a + if)
g-+-
£6/
= 0. (3.18)
To prove that the right-hand side of Eq. (3.13) vanishes, it

remains only to show that (Re (<7 + ih)/{a + ib) equals — (dB/dt).
Now g + ih is defined by L± — iL± X ei = (g + ih) (62 + ie3).
This can be solved for g + ih by dotting with e2 or e3
g + ih — Lj_*(e2 — ie3). (3.19)
The explicit expression for L± given immediately following Eq.

(3.12) must now be used; L± contains Rw, which is to be replaced
by (a + ib) (e2 + ie3).
g H~ ih _ • (e2 — ie3)
[ —1 (e2 + ie3) • VE
a + 16 a + 16
— il?oo X (e2 + ie3)-VU — Zi(e2 + ie3)]-(e2 — te3). (3.20)
The first term on the right-hand side is
~ (63 + ie2) • [(e2 + ie3) • V ]E

= e2-(e3-V)E -e3-(e2-V)E + £(...)
= ~er [e3 X (e3-V)E + e2 X (e2-V)E] + i(...)
= -erV X E + £(...). (3.21)
The imaginary part does not have to be evaluated because only the
real part of (3.20) is required. The second term on the right side of
Eq. (3.20) is
Roo' [(©3 X (e2 — 1*63)

= Roo'{(ies — 62) X [(e3 — 162) • V](fiei)}
A
= R00-B[es X (e2-V)ei - e2 X (£,-V)&]

+ Roo- [e3(e3-V)B + e2(e2-V)B] + l'(. • •)
= jRoo • Bei [ — e2 • (e2 • V) ei — e3'(e3-V)ei]

+ Koo-V±B + i(.. .)
= Roo■ [-BV-ei + V±B] + £(. . .)• (3.22)
Because
0 = V B = V-(e i£) = ei-VB + BV-ei (3.23)
it follows that — BVe 1 = eier VB, and the real part of the right
side of Eq. (3.22) is simply R00-VB.
The last term on the right side of (3.20) is
— B(e2 + ie3) • (e2 — te3) = —2B (3.24)
and
(Re = -ei-V X E + R00-VB - 2B. (3.25)

\a + 10/
By Maxwell’s equation, V X E = — (dB/dt) and er V X E =

— (dB/dt), since er(dei,/dt) = 0.
(Re (-26= -B (3.26)

\a + 16/ di
since B = (dB/dt) + Rq0-VB + 0(e). Thus it is proved that

(d/dt)(BR10-R10*) = 0 + 0(e) or that BRWR10* equals a constant
+ 0(e).
B. The Longitudinal Adiabatic Invariant
The next adiabatic invariant to be studied is the longitudinal

invariant
J = £ Puds (3.27)
Figure 3.1. Guiding center oscillates along a line of force and drifts slowly at
right angles to it.
where is the guiding center momentum parallel to the line of

force, and the integral is taken over a complete oscillation from one
mirror point to the other and back again. As stated previously, the
longitudinal motion must be periodic for J to exist. The procedure
thus far has been to start with the equation of motion (1.1) of a
charged particle and to average over the gyration. The resulting
guiding center Eqs. (1.17) and (1.20) are new equations of motion.
The next step is to average over the longitudinal oscillation and
obtain a third set of equations of motion governing the average
drift from line to line, and then to show that this average drift
conserves J.
As the guiding center moves along a line of force in accord with
Eq. (1.20), it drifts at right angles to the line at a rate given by
Eq. (1.17). See Figure 3.1. If the drift at right angles is slow
compared with the longitudinal motion, i.e., if E± and d/bt are of
0(e), one can calculate the average drift rate at right angles to the
line during a longitudinal oscillation as if the guiding center did not
deviate from the line of force, the error being of order e2 since the
drift rate R± is 0(e). If Ex is 0(1), contains the 0(1) term uE
and the guiding center moves a long way from a given line of force
in one oscillation. It is then no longer possible to ignore the devia¬
tion from the line of force; the guiding center does not remain
even approximately on the line and it will be found that <f p^ds
is not conserved.
A proof has been given in reference (21) that the average drift
from line to line conserves J; the proof is for relativistic particles.
Rather than repeating that proof here we will give a somewhat dif-
ferent proof. It will be for nonrelativistic energies, but the modi¬

fication to relativistic energies is easy with the relativistic expres¬
sions (1.68)j, (1.69), and (1.70) available.
To formulate the problem explicitly, the a, (3, s curvilinear
coordinate system introduced in Chapter 1 will be used. Let K —
(mv\j2/2) -f MB + e(</> + where \p = (a/c)(dl3/dt). By Eq.
(1.29), K is a constant of the longitudinal motion. Then J is
given by*
J(a, (3, K, M,t) = f {2m[K - e{$ + <£) - MB]}1,2ds (3.28)
where \p, </>, and B are all functions of (a, (3, s, t). The rate of change
of J is
dJ _dJda dJ dj3 d£dK dJ

(3.29)
dt da dt d/3 dt dK dt dt
where
dJ r ds
da~ ~ me J {2m[K - e{$ + </>)- MB]}112'
M n
+ d + \p + $ -\-B (3.30)
da e )
dJ
d(3
= — me <j) ds
{2m[K - e($ + </>)- MB]} 1/2
( ?) (3.31)
dJ
dK = m/ ds
{2 m[K — e(ip + </>) — MB]} 1/2
- /*-
J®H
(3.32)
and T is the period of the longitudinal oscillation.

* It is best to define J by Eq. (3.28) rather than by Eq. (3.27), and to define
K as particle kinetic energy plus e(\p + 4>).
dJ ds
— = — me 1/2
dt {2m[K - e{\p + <t>) - MB]}
dt (* + * + v)--‘/JI(* + * + v)' (3'33)

In Eqs. (3.30) to (3.33), i/s <t>, and B are to be considered as func¬
tions of (a, (3, 5, t) and not of the guiding center position R and of
t. R is itself a function R(a, (3, s, i). It must be remembered that
d\p(a, fl, s, t) = dt(R, t) dR(a, ft, s, t) _ R q
dt dt dt
(3.34)
dxfs(a, (3, S, t) dR (a, ft, S, f)
--- = --- V\l/{R,t),
Oa Oct
etc., for ft and s.

Since a(R, t) and (3(R, t) are constants on a line of force, then-
time derivatives contain R± and not i'|.
da da(R, t)
+ R±-Va(R, t) + 0(e2) (3.35)
dt ~~dt
dft
^ + R±-V(3 + 0(e2). (3.36)
dt
The expression for Rx (with uE and d/dt ~ e) must now be sub¬
stituted into Eqs. (3.35) and (3.36). The procedure will be carried
out in detail only for da/dt. It is the same for d[3/dt (except for a
sign).
da da ei „ Me „ me
— = —■ -L — v -cE -|-VB H-w,|2 ■ Va + 0(e2).
dt dt B e e
(3.37)
Now a = a[R(a, ft, s, t), t], (3 — (3[R(a, ft, s, t), t], and s = s[R(a,
(3, s, t), t]. By implicit differentiation of a with respect to a, ft, 5,
and t, and of (3 and of s with respect to the same four variables, one
obtains the following equations:
1 = Va(R, t)-dR(a, j8, s, t)/da
0 = Vo;' (di?/d/3)
0 = Va - (dR/ds)
0 = (da/dt) + Vet- (dJR/5/)
0 = V/3(R, t)-dR(a, /3, s, t)/ba
1-W-VRM
0 = V/3- (d/i/ds)
0 = (dfi/dt) + V/3 - (djR/d/)

0 - Vs(R, t)-dR(a, (3, s, t)/da
0 - Vs - (di?/d/3)
1 = Vs-(dl?/ds)
0 = (ds/ht) + Vs- (dR/dt).

In Eq. (3.37) Vet can therefore be replaced by (dR/d/3) X B, since
B(R) = Va X V/3 and
by Eq. (3.38). Also, in Eq. (3.37),
(3.40)
With these substitutions, Eq. (3.37) becomes, after interchang¬

ing the dot and cross and expanding the triple vector product,
[(dfl/dj8) X ej X ei:
da da / A dR dR\ / dp
cV \p + cV0 — — Va
dt = w + le,ei ‘a? “aJ'V
da Me me „ 5eA .
+ * VIS + T VB + T **) + e(t )
a« +, cei
, • aR —
d ((,^ +, 5A +, 'MB
d<
7-7-
5/3 ds )
5B F\ 5a
+
5R me „ dR 5ei , .
^+5+ — ®u2^r*3T + 0(£)- (3-41)
2>|S •v( v) e u dp ds
By Eq. (3.38), and because (5R(a, P, s, t)/dp) • V = 5/5/3, Eq. (3.41)

becomes
da A 5JR (a, /3, s, f) 5

eft CCl 5/3 5s ( i + 5 + ?)-
MB\ me dR 5ei (a, /3, s, f)
+ 0(e2)
5/3 ( 5s
(3.42)
where ^ + 0 + (.MB/e) is to be considered a function of (a, P, s, t).

By Eq. (1.22), (5/5s)[\p + </> + (MB/e)] = — (m/e)(dv\\/dt)
+ 0(e2). In the last term of Eq. (3.42),
dR 5ei & ,(e 5 R\ 5 dR (a, P, s, t)

— ex ■
A
dp ds 5s I"1 dp) 5s dp
& ,(c, 5 R\ 5 dR
— ei ■
A
5s dp) dp ds
since ei = dR/ds. The last term, then, is
me dR ,dei me d / dR\
.
eVi Vs = 7 Si \ei '
mev n d dR\
ei • — + 0(e2). (3.44)
e (*l
dt \ dp)
The first and last terms on the right side of (3.42) combine to
me d f dR\
“7 dt V>e' WJ + • 0(e!)
and
da MB\ me d dR\
dt -_el( >/' + </> +
e )
-
dt ( ’ dp) + 0(e ^
(3.45)
Similarly,
dp d MB\ me d (
dt da ( -— ) +
7 dt (’** f5)
da /
+ 8(«!).
(3.46)
An analogous procedure is needed for dK/dt in Eq. (3.29). The

quantity K was defined as the particle kinetic energy plus e(\p + 4>).
From Eq. (1.29), (1 /e)(dK/dt) = 0 + 0(e2) when Ex andd/dfare
0(e), as here. However, for present purposes it is necessary to know
the 0(e2) term of (1/e)(dK/dt). The time rate of change of + <£
correct through order 0(e2) is
**7 0) = | im,t) +
+ A-V[#(R, 0 + *(R, 0] + S(es). (3.47)
If E± and d/dt are 0(e), both ip and <f> are themselves 0(e) and their
partial time derivatives are 0(e2). From Eq. (1.28) the rate of
change of kinetic energy is the sum of the drift velocity in the di¬
rection of E, plus the induction effect (M/e)(dB/dt), plus terms
of 0(e2), at least when E and d/di are 0(1). It is also true that if
Ex and d/d/ axe of 0(e),
d (kinetic energy)
R-E + - ^ (R,t) + 0(e3). (3.48)
e dt e dt
This can be verified by starting with Eq. (2.1) for r and calculating
d(kinetic energy)/dt = ev-E(r, /). Upon expanding E(r, t) about
E(R0, t) and time averaging over a gyration (over d, that is),
Eq. (3.48) results. The Maxwell equation V X E = —dB/dt must
also be used.
Addition of Eqs. (3.47) and (3.48), with E from Eq. (3.40),
yields
K
— ^ + <i> d-— ^ + jR- [E + V(^ + </>)] + 0(e5)
e
d / MB\
s(* + * + v) +
d/3 da
+ 0(e3) (3.49)
di Va - d/
where \p, <j>, B, a, and /3 are all to be considered functions of R and

/. The vector (d/3/d/)Va — (da/d/)V/3 appears frequently and will
be denoted by w. The drift R± is now replaced by
Ci Me me „ dei
■cE +
B X e V1 dsj
where E is — + <j>) + (w/c).
R\-w ei Me me w
= R X VB H- U||2
c B e e ds c
MB' deU tv X ei
v l ^ + </> +
( ds B
(3.50)
tv X ei d/3 Va X ei da V/3 X ex
B dt B ~ dt B
/dR d(3 dRda\

(3.51)
\dp dt da dt)±
where Va has been replaced by (dR(a, (3, s, t)/dp) X B from Eq.

(3.39), and similarly for V/3. Now R = R(a, (3, s, t) = R[a(R, t),
P(R, t), s(R, t), t]. By implicit differentiation with respect to time,
dR da dR dp dR ds dR
(3.52)
da dt dp dt ds dt dt
From (3.51) and (3.52), and the fact that dR/ds = ei, it follows
that
tv X Cl \ ds (R, t) , dR (a, p, s, t)'

B ei at + Dt
(DR\ DR _ aR
= Uj± = ¥“e‘e‘'¥ (3'53)
and
R±-tv
.[»(„,+m?)+=.,.I],
MB\ dR (a, p, s, t)
* + <t> +
9
MB\ dR ,
--( ^ + <t> H-—) ei • — +
ds \ dt
m „ dJR dei
— V\\2- (3.54)
e di ds
The last term is (rnv^/e)(d/ds)(ei• dR/dt), since er (d/ds) (dR/dt)

equals er(d/dt)(dR/ds), which is er(dei/dt) and vanishes. In
addition,
m b / bR\ m d / bR\ .
e ”,S Ss (ei' 3() 7 “ dt l*1 ¥) ®(<
' + ( 5 ’
Replacing -(d/ds) [* + <j> + (MB/e)] by -(m/e)(dv[l/dt) + 0(e2)
finally gives
Rx-w MB\ bR md ( A bR\

-v( \[/ + </> +
-7)-U+-edt\V^bt)+^ e3)
(3.56)
and
lrfK 5 r x MB (R, t)
-
e dt
= bt
dL *(*, o +«(«. o +-7^
e
MB\ dR , m d / c)R\
+ V ( </' + $ +
( —) ‘ * + 7 dt l”*e‘' W + 0(e >' (3'57)
As indicated, 4>, and B are functions of R and l, while R =

R(a, /8, 5, f). If \p, </>, and B are now regarded as functions of a, p,
s, t as in Eq. (3.33), then by Eq. (3.34),
1 dK _ d MB
\p(a, ft, S, t) + (j) +
e dt bt
+ 71 + 0(e!)- (3’58)
We are finally prepared to evaluate dJ/dt. Comparison of da/dt
in Eq. (3.45) with bJ/bp from Eq. (3.31) reveals a similarity be¬
tween da/dt and the integrand of bJ/bp. In fact,
e
c
(3.59)
But ds/v|| equals dt, the time for the guiding center to traverse ds,
so that the last integral in (3.59) is the net change in vfa-(bR/bp)
over a period of the longitudinal oscillation. This change is zero.
As a special case, if one takes the period from one mirror reflection
to the next reflection at the same end of the line of force, »n vanishes
both times'and therefore the integral vanishes. Equation (3.59)
can be written as
(e/c)T(a) = dJ(a, (3, K, M, t)/b(3. (3.60)
where (a) is the average rate of change of a over a longitudinal
oscillation, the rate of change of a being caused by the time-
dependent fields and the guiding center drift. The period of the
longitudinal oscillation is T. Similarly,
(e/c)T ((3) = — dJ/da (3.61)
T(K) = -bJ/bt (3.62)
and, by Eq. (3.32),
1 = (1 /T)bJ/bK. (3.63)
Equations (3.60) to (3.63) are the equations of motion that govern

the average motion from one line of force to another (22).
Equation (3.29) for dJ/dt when the guiding center is at some
point (a, (3, s) can, by virtue of these equations of motion, be
written as
(S = ? [<“^(s) - (®“(s)1 + nK(s) ~
y [fSW*(»0 - *(s)/3(s')] + K(s) - H(s')
(3.64)
where is the parallel velocity the guiding center has at s'.

There is no general reason that dJ/dt should vanish. However,
/ dsdJ _
dt J J »n »o'
tf(s') + -
c
= 0. (3.65)
Figure 3.2. A line of force and the drift velocity in a Cartesian a, §, s space.
The double integral vanishes because of the antisymmetry of the

integrand in 5 and s'. Equation (3.65) means that although the
instantaneous rate of change of J is not zero, the change averaged
over a complete oscillation is zero.*
The above proof of the invariance of J has been carried out non-
relativistically. The relativistic modifications are (21)
K = [p2c2 + moV]1'2 + g(* + <f>) (3.66)
IdK Mr dB d
— RE -\-— + f?-V(i/' + 0) + — (</' + <^>)
e dt ye ot ot
Rx-to MT dB d
+ ;-(* + *) (3.67)
c ye ot ot
and the relativistic guiding center Eqs. (1.68) to (1.70). With
these, a proof similar to the preceding one can be carried
out, or the proof in reference 21 can be used. In the relativistic
case exactly the same equations of motion (3.60) to (3.63) result.
A convenient space in which to illustrate the guiding center
motion is a Cartesian (a, /3, s) space, as shown in Figure 3.2A.
In this space a line of force appears as a straight line parallel to s.
* If the orders of e are followed through, it is found that (1 /m){dJ/dt) =
0 + 0(e2) and (I/mXdJ/dt) = 0 + 0(e).
At any instant of time the guiding center is drifting in the a, (3

plane with velocity components a and (3, as given in (3.45) and
(3.46) anddllustrated in Figure 3.2B. As the guiding center moves
along s, the direction and magnitude of the drift vector in Figure
3.2B change, since a and /3 are functions of s. The guiding center
therefore does not always drift toward the same adjacent line of
force during its rapid motion along 5. In a special case it could,
however, always drift toward the same adjacent line. Consider, for
example, a static field, so that K = 0. If the drift is toward the
same adjacent line at all s, then a/(3 must be constant and the
constant consequently is (a)/(0). Under these circumstances d,J/dt
is instantaneously zero, by Eq. (3.64). Such would be the case in
an azimuthally symmetric mirror machine, where the drift is al¬
ways in the azimuthal direction.
Equations (3.60) and (3.61) for (a) and (0) may appear to be
canonical, with J as the Hamiltonian. But they are not, because T
is also a function of (a, (3, K, M, t). If J = J(a, /3, K, M, t) is
solved as K = K(a, /3, J, M, t), then by implicit differentiation
dJ/d/3 = —(jdK/d0)/(jbK/dJ), etc., and
(a) = —(c/e)dK(a, 0 J, M, t)/b(3

(j8> = (c/e)bK/da
(3.68)
(R) = dK/dt
1 = TdK/dJ.
These are canonical in form; consequently a Liouville theorem
exists in (a, 0 J, M) space. Let Q(ot, 13, J, M, t) dadp be the number
of particles at time t with invariants J and M in the flux tube of
flux da d(3. The equation of continuity in a, /3 space is
^
Ot
~
Oa 0(3
(om = o. (3.69)
With (3.68) this becomes
(3.70)
Figure 3.3. Cancellation effects of ds and ds' on J.
and Q is constant along a guiding center trajectory in a, /3 space.

An interesting theorem is a consequence of the Liouville theorem:
In the steady state and in the absence of an E field, constant B
contours are also constant particle density contours on a longi¬
tudinal invariant surface. Details of the proof are given in reference
21.
The antisymmetry of the integrand in Eq. (3.65) has a physical
interpretation. By Eq. (3.64) the contribution of ds' to the rate
of change of J when the guiding center is at s can be (but does not
have to be, as will become apparent below) considered to be the
integrand (ds'/v$'){.. .s, s'...}. The contribution of ds' to the
change in J while the guiding center traverses ds is dt = ds/vy
times this rate, or (ds/0||)(ds'/fl[|'){.. .s, s'...}. See Figure 3.3.
At a later time, when the guiding center has actually arrived at
ds', the contribution of ds to the change in J while the guiding
center traverses ds' is (ds'/v\\')(ds/v^) { .. .s', s...}, which is just
the negative of the contribution of ds' to the change in J when the
guiding center traverses ds. This cancellation is an interpreta¬
tion of the antisymmetry of the integrand in Eq. (3.65) and holds
for all pairs of arc elements ds and ds'. Such a cancellation is
somewhat remarkable, especially if one recalls that the guiding
center does not even drift toward the same adjacent line of force at
s and s'. The remarkability disappears to some extent if one
realizes that Eq. (3.64) can also be written as (for simplicity let
E = b/bt = 0)
(3.71)
where the prime means evaluated at s'. It can be written this way
because the difference between /3(s') and d(MB')/da is m(d/dt)

[flyer (dR/da)]', which integrates to zero when the s' integration
is performed; and similarly for a (s'). In truth, anything can be
added to d(s') and 13(s') in (3.64) which integrates to zero over s'.
The integrand in (3.71) is no longer antisymmetric in s and s', and
the nice physical interpretation is no longer present. The interpre¬
tation of the antisymmetry is therefore not unique. It nevertheless
remains true that
is zero.
C. The Third or Flux Adiabatic Invariant <t>
As a particle oscillates between mirror points it drifts across

lines of force on which J is constant. These lines form a surface in
space. It may happen that the surface is closed. Such is the case in
laboratory-type mirror machines and probably in the geomagnetic
field. A longitudinal invariant surface is illustrated in Figure 3.4
for the earth’s field.
In the Cartesian a, /3, s space of Figure 3.2A longitudinal invari¬
ant surfaces are curved in one direction only, the elements of the sur¬
faces being straight lines parallel to s. If the invariant surfaces are
closed, they are represented by cylinders in a, 0, s space, as shown
Figure 3.4. A longitudinal invariant surface in the geomagnetic field.

Figure 3.5. A longitudinal invariant surface in a, P, s space.
in Figure 3.5A. The elements of a cylinder are not all of equal

length because the distance between reflection points is not a con¬
stant of the guiding center motion. The intersection of a cylinder
with the a, (3 plane appears as a closed curve in Figure 3.5B.
The intersections of the cylinders with the a, (3 plane form a
three-parameter family of curves. The longitudinal invariant J is a
function J(a, (3, K, M, t), as can be seen from the defining integral
in (3.28). The equation J = J(a, (3, K, M, t) can be solved as a =
a(/3, J, M, K, t), which at any time defines a family of curves with
J, M, and K as the parameters.
Suppose that at a given instant the time dependence of the
fields were turned off; K as well as J and M would be constant
and the guiding center would precess about the corresponding
(J, M, K) surface. The third adiabatic invariant $ is the flux of
B through this surface. That $ is a constant is a trivial statement
if the fields are static, for the guiding center repeatedly precesses
around the same surface and the surface does not change with
time. An analogous trivial statement would be that the magnetic
moment is constant in a uniform static magnetic field. If, on the
contrary, the fields are time dependent, the time dependence being
slow compared with the precession time once around the surface,
the third invariant applies and yields nontrivial information.
Since K is no longer constant, the guiding center gradually moves
from one J, M, K surface to another as it rapidly precesses, and is
at all times on a surface with the same J, M, and F. Thus, although
one constant of the motion K has been lost, another one, <f>, re¬
places it.
The proof of the invariance of <t> is fortunately much simpler
than that for M or J. The flux through the invariant surface is
<f>(J, M, K,t) = £ A di = f aVI3-dl = f a(ft J, M, K, f)dp

(3.72)
where the contour is any closed curve lying on the surface (in real
space—not in the space of Figure 3.5) and going once around it.
The quantity ad(3 is also the area enclosed by the curve in
Figure 3.5B on which the guiding center is located.
dHK,t) /T>, , dHK,t)

= ~bir(K) ~^r + ' (3-73)
The J and M dependence of $ will not be exhibited in the next few

equations, since J and M are constants. The average, (R), has
been used in place of R because we are not interested in fluctua¬
tions of <t> over a longitudinal oscillation. Now
d<t> SapS, K, t)
di3. (3.74)
dK dK
By implicit differentiation of K = K[a((3, K, t), (3, t] we obtain
dK (a, t) da (ft K, t)
1 = (3.75)
da dK
dK da dK
0 = (3.76)
da dt + dt
d<t>
£ m c i’ d/3
(3.77)
dK ‘ J dK/da ef 0)
the last equality being via Eq. (3.68). But J" d(3/0) equals
f dt, which is the time Tv for the guiding center to precess once
around the surface. Therefore d^/dK = (c/e)Tp, and this is the
analog of Eq. (3.63). The last term in (3.73) is
a* £ da 03, K, Odi3 £ dK/dt Ja

dt= f -dt- =~ ? dKfdad?'
Replacing dK/da by 0) again and dK/dt by (it), we have
d<t> _ c C (K)d(3
- <f {K)dt = - ^ «tf» (3.79)
dt e J 0) e J e
where ((it)) is the average of (it) over a precession period. Equa¬

tion (3.79) is the analog of Eq. (3.62). The instantaneous rate of
change of <t> is therefore
^dt = —e 1 [<*> - «*»]. (3.80)
As with dJ/dl, we observe that d$/dt 9^ 0, but that (d<t>/dt), the

average rate of change over a precession period, does vanish, again
by the antisymmetry of the integrand in
[<it)ff - {K)A (3.81)
where da' is the element of arc length and va0 is the velocity at
a' about the closed curve in Figure 3.5B.
The new and final set of equations of motion is
_e_ d$(J, M, K, t)
«*» = - cTp (3.82)
dt
e 5<t>
(3.83)
cTp dK'
The averaging processes used to establish the three adiabatic
invariants and sets of equations of motion are summarized in
Figure 3.6.
A few facts about the family of longitudinal invariant surfaces
should be made clear now. In the first place, at any instant of time
they are not simply nested but intersect in a very complex fashion;
this is to be expected, because there are three parameters, not one.
For example, two particles with the same K which are oscillating
j Equation of motion / Average over /Guiding center equation of'j

/ of a particle (1.1) (
-:— > < motion (1.17) and (1.20), and>n
the gyration
' the adiabatic invariant M)
Average over
the longitudinal
oscillation
{Equations of motion}
Average over the
{Average guiding center
(3.82) and (3.83), ( <-;--- equations of motion
<■
and the adiabatic in-1 precessional motion (3.68), and the adiabatic
variant 4> invariant J
Figure 3.6. The averaging process used to obtain the three adiabatic
invariants and equations of motion.
Figure 3.7. Longitudinal invariant surfaces are not simply nested and may
intersect.
along the same line of force, but which have different mirror points,
have different M and consequently, by Eq. (3.28), different J.
There is no reason for these two particles to traverse the same line
of force anywhere else. Figure 3.7 illustrates how their two in¬
variant surfaces might appear in the a, (3 plane. (Of course, they
may intersect elsewhere; there is generally nothing to prohibit it.)
An infinite number of surfaces therefore intersect along any line
of force; if particles are injected on one line of force with a distribu¬
tion of mirror points, they spread into a layer of finite thickness
elsewhere.
Secondly, in the a, (3 plane a surface denoted by fixed values of

J, M, K will move in time as indicated schematically in Figure
3.8 by curves I and II. It is clear that the surface must move, be¬
cause a equals a(/3, J, M, K, t) and therefore the value of a for a
given /3 varies with t at fixed J, M, K. The fact that a constant
J, M, K surface moves with time does not mean that the lines of
force move in a, 13, s space. In fact, they can be considered fixed,
even though they move in real space. The velocity of a magnetic
Surface on which guiding
-—-0
Figure 3.8. Longitudinal invariant surfaces at different times.
line of force is not a physically observable quantity, and therefore

must be defined. The definition should be a “flux-preserving” one
(23): let an arbitrary closed loop be drawn in space and let the veloc¬
ity u of each arc element of the loop be such that the flux of B
through the loop is constant in time. The velocity of the line of
force at each point is defined as u. It is easy to show that u must
satisfy V X [E + (u/c) X B] = 0. The velocity (w X ex/B) is
such a velocity, where w = (dp/dt)Va - (da/bt)VP, as defined
previously.
E + (w X ei/cB) X B = E - (w/c). (3.84)
By Eq. (3.40), E equals — V(^ + </>) + (w/c), so that
E + (w X ei/cB) X B = -v(^ + tf>) (3.85)
and indeed the curl vanishes. It has thus been established that
(w X ei/B) can be taken as the velocity of a line of force. It will
be proven that if an observer moves at this velocity, the total rate
of change of a (and /3) he observes is zero. In other words, the
(a, /3) label on a line of force is not changed by the motion of the
line and, consequently, all lines are fixed in a, ft, s space. The rate
of change of a under the time dependence and the velocity of the
line of force is
da ( w X ei da /d/3 da ei
Va
dt H B~ Va = dt
^ + ) XB
da da V/3 X ej-Va
dt dt B
= 0 (3.86)
because B = ei-B = er(Va X V/3). It is therefore clear that al¬

though lines of force themselves do not move in (a, (3, s) space,
the locus of lines which form an invariant surface with constant
J, M, K does move. In Figure 3.8, curves I and II are the pro¬
jections on the a, /3 plane of two surfaces having the same J, M,
and K at different times t0 and U.
A third point is that the actual surface in a, /3, s space on which
the guiding center precesses at time U is neither curve I or II, but
some third curve represented schematically by curve III. This is
necessarily so because the particle’s K has changed and is no
longer K0. Curve III has the same enclosed area as curve I because
the flux invariant $ is this area. Although it is true that if the
guiding center had remained on curve I at time h, it would have
conserved its <t>, its J would have been different.
I>. Adiabatic Invariants to Higher Order
It is apparent from the foregoing extensive analysis that to

proceed to higher orders in the three adiabatic invariant series by
use of these direct methods would be laborious. Indeed, even to
guess what eM', eJ', and e$' are would tax one’s imagination.
Proofs of the type presented for M, J, and are valuable in pro¬
ducing a physical picture of what is happening, but must be
abandoned in favor of a systematic, canonical method like that of
Gardner (19) or Kruskal (20) to go to the next higher order. Probably
nothing is lost in the way of physical understanding either, since
effects which are second order in the gyration radius are difficult
to visualize anyway.
In Gardner’s procedure, to obtain each new term in one of the
adiabatic invariant series a canonical transformation must be made
from the variables used in the previous order. A prescription is
available for obtaining the generating function of each successive
transformation. At any order, all the preceding canonical trans¬
formations must be inverted to express the adiabatic invariant
series in terms of the original variables (velocity and position of the
particle). In practice this may be very laborious, but at least a
deductive method is available and no guessing of the higher order
terms is required.
The eM' term is the only higher term in any of the three series
that has been worked out in general. Gardner (24) has obtained the
e2M" term for the special case of an azimuthally symmetric mirror
geometry. This second-order correction will be introduced in Chap¬
ter 5 in connection with nonadiabatic effects in mirror machines.
In the present section the eM' will be used to illustrate the various
forms in which these correction terms may be expressed.
The first-order corrections eJ' and e<t>' to the longitudinal and
flux invariant series have not been calculated. They would be useful
in studies of particle motion in the geomagnetic field.
The eM' term is given in reference 15 in a slightly different (and
more useful, it will turn out) form than one might expect. The
series, correct through order e, is (when E — 0)
vx e
constant = («2ei + OT-ei)-[(» X ei)-V]B
B(rj ~~ IP
+ (ei-t>)(V X B)
(¥ ei + 2vx&i-v\\ + 0(e2) (3.87)
where vx is the instantaneous particle velocity perpendicular to

ei(r), not ei(i?o), r being the instantaneous particle position and
R0 the guiding center position. Also note that in the first term B
is evaluated at r and not R0. In the 0(e) term, it does not matter
whether the field and ei are at r or R0, nor does it matter whether
is perpendicular to ei(r) or ei(i?o), since the difference is 0(e2).
Form (3.87) of the series is useful for comparison with a numerical
integration of the equations of motion because the result of such a
computation would most likely be the particle velocity and position
as functions of time. The fields at the particle position would there¬
fore already be present in the code, whereas the fields at R0 would
require an auxiliary computation.
The first term of Eq. (3.87) can be converted to velocities per¬
pendicular to ei(I?o) and to fields at the guiding center R0 as follows:
®L(r) = « — ei(r)ei(r)• v = v - ei(r)va(r)
= v - [Sx(Ro) + 9-Ve1(R0)][ei(R0) + q-V^(Ro)]-v
= v±(Ro) - %Ro) p-Vei — ei«-(e-V)ei + 0(e2) (3.88)
«2±(r) = v2xw - 2u(|(Ro)t;• (p- V)ei + 0(e2) (3.89)
where the r or R0 following the perpendicular or parallel subscript
signifies the direction to which the velocity component is perpen¬
dicular or parallel. The vector p is t{Rvpd + Rio*e~'e) = ei X vx/co
+ 0(€2). In addition, 1/J5(r) must be transformed:
B(r) B(R0)
^
B
+ 0(6').
2
(3.90)
With the substitution of (3.89) and (3.90) and some vector alge¬
bra, Eq. (3.87) becomes
constant = - e ^ vx- [(ea X vx) -Vei] + e -~
(ei X vx) ■ VB - ^ wu(V X B) ■ + 2®x»,^ + 0(e2). (3.91)

It may be observed that if Vy is zero at all times, then v2_l(r0)/B(Ro)

is constant to one higher order in e.
There is a third form in which the series may be written. The
particle velocity may be eliminated from (3.89) and the Rn of
series (2.1) used instead. Differentiation of the asymptotic series
(2.1) for r gives v and therefore v±. In Eq. (2.13) v was obtained
correct through zero order in e. Now, however, vx is needed correct
through 0(e), because the first term v2xm/B(R0) will then yield
more 0(e) terms. It should not be overlooked in differentiating the
series for r that e2R2oe2ie terms contribute 0(e) terms to v, as a re¬
sult of differentiating the exponential. When v is substituted into
(3.91), the result is (E = 0, still)
constant = 2B(RW + eRn) • {Rw* + eRn*)
(l + + 0(6’) (3.92)
where Rn is the second coefficient in Ri — Rio + eRu + .... This

is certainly the least useful form for comparison with numerical
computation.
Because v contains exn6 it may seem surprising that there are no
exponentials of this type left in (3.92). The reason is that the
adiabatic invariant series is an integral over 9 and therefore 6 can¬
not appear. This will become apparent in the discussion in the next
section of adiabatic invariants of systems of coupled differential
equations.
The statement is often made that the magnetic moment mv±2/2B
is “constant to all orders.” This means that if a particle goes from
one region in space and time in which E and B are constant to
another such region via time- and space-dependent fields, the
change in v^/B between the initial and final states vanishes faster
than any power of e, even though the change at intermediate times
may not. This conclusion follows from the fact that 0(e) and higher
terms in the magnetic moment series vanish in uniform fields.
That the 0(e) term vanishes when B is constant can be seen in Eq.
(3.87). Higher terms always contain field gradients and vanish in
uniform fields. If the magnetic moment series were convergent

instead of asymptotic, the change in vAJ/B would be rigorously
zero, but because of the asymptotic property, all we know is that
the change goes to zero faster than any power of e. It is fre¬
quently suggested that the change in v±2/B is proportional to
exp (—constant/ e), which does indeed vanish faster than any power
of e, but the change certainly could be some other function of e
that has no power series expansion in e.
E. The Adiabatic Invariants of Singly Periodic Systems
A more general theory of asymptotic solutions and adiabatic

invariants has been given by Kruskal (20) for coupled differential
equations of a certain type. Let
dxi/ds = /i(xi, x2, .. -, xN, e)

(3.93)
dxN/ds fi, ... ,3^) c)

be a set of coupled first-order differential equations in which the
independent variable s does not appear explicitly on the right-hand
side; e is a parameter. The set of equations may be written vectori-
ally as
dx/ds = f(x, e) (3.94)
where * is the vector (xu x2,... ,xN). Distinguish between the inde¬
pendent variable s used here and the distance along the line of
force.
Let the system (3.94) have the property that solutions of dx/ds
= f(x, 0) are simple closed curves in x-space, as illustrated in
Figure 3.9 for two dimensions. All the components of x are periodic
with the same fundamental frequency. It is also assumed that /
possesses a power series expansion in e. Under these conditions
there exists a transformation x = x(z, 6, e), where x is periodic in
9, such that the transformed equations (3.94) assume the form
dz/ds = eh(z, e)
(3.95)
dd/ds = co (a, e).
xi
The vector z has only N — 1 components. It is not immediately

obvious that such a transformation even exists, which makes h and
co independent of 6. The first part of reference (20) is devoted to
proving this and to developing the step-by-step recursion method
which gives the transformation x(z, 6, e) and the functions h and
co as series in e. It is possible to prove that if the solution z(s, e),
0(s, e) of (3.95) is substituted into x — x(z, 6, e), the result is an
asymptotic (in e) approximation to the exact solution of (3.93).
The equation of motion of a charged particle may be written
in the required form dx/ds = f(x, e). If x were the six-component
vector (v, r), the equation of motion would not be of the required
form because the independent variable t would appear or the right-
hand side when the fields were time dependent. But if time is
treated as a seventh dependent variable by the substitution t —
es (where e = m/e), seven equations of motion of the required
form result. They are
dv/ds = E(r, t) + v/c X B(r, t)
dr/ds = ev (3.96)
When e = 0, this set reduces to
dv/ds = E + (v/c) X B
dr/ds = 0 (3.97)
dt/ds = 0.
Because dr/ds and dt/ds vanish, r is a constant = r0, and t is a

constant = to, so dv/ds = E(r0, to) + (v/c) X B(r0, to). Because E
and B are constants when their arguments are constant, the zero-
order motion is that of a particle in a uniform field. In v space, it is
a circle with center at [cE(r0, to) X B(r0, U)]/B2, provided En is
assumed of 0(e) as usual, and therefore the motion in the complete
seven-dimensional space is periodic when e = 0. Consequently,
the theory applies, and for r would give the asymptotic series in
Eq. (2.1) and for v its derivative.
Thus far the theory has produced no adiabatic invariants, only
asymptotic solutions. If, however, Eqs. (3.93) are of canonical
form, one or more adiabatic invariants exists in addition to the
asymptotic solutions. Let the vector x be (/>, q) and s be t. The
canonical equations are
Pi = — dH(p, q, e)/dqi
(3.98)
qi = &H(p, q, e)/dp{
and these are of the required form. If the Hamiltonian is time

dependent, time and energy can be used as conjugate variables and
the number of degrees of freedom increased by one. Given the
necessary conditions on the periodicity of the zero-order solutions,
these exist the transformations p = p(z, 9, e) and q = q(z, 9, e)
which are periodic in 9 and are such that the transformed canonical
equations are
dz/dt = eh(z, e)
(3.99)
dd/dt = co(z, e)
The solutions * = z(t, e) and 9 = 9(t, e) of (3.99) can be substituted

to give p = p[s(t, e), 9(t, e), e] as a series in e, and the same holds
for q. These are not exact solutions of the canonical equations;

only asymptotic solutions.
The adiabatic invariant is
i(z, e) — f
Jo
de p(s, e, 0
5q (z, 0, e)dd
50
pdq (3.100)
constant
where, to be specific, the angle variable 0 is assumed to have the

period 0 to 1. The invariant / will in fact be obtained as a series in
e, since p and q are themselves series in e. The proof of the adiabatic
invariance of I consists of showing that dl/dt is zero, or, more
precisely, that (d/dt)(IQ + eh + . - -enIn) = 0 + G(«n+1).
dl (t, e) dijit, e) 5/ (z, e)

dt dt 5 z}
/5pft 5<7* 52qk\

(3.101)
\5z^ 50 505z^/
where sums are to be taken over repeated indices. The factor

dzj/dl is not a function of 0 and therefore has been put under the
Figure 3.10. The unperturbed path in phase space differs from the constant s
curve.
integral. The second term in the integrand may be integrated by

parts to give
<U dzj (bpk bcp bqk bpA
(3.102)
dt dt bO bz3 bO )
where the term /,/ d0(b/b0) [pk(bqk/bz}) ] vanished because p and
q are periodic in 6. Let us now show that the integrand in (3.102)
is simply — (bH/bO).
bH[p(z, 6, e), q(z, 0, e)e]

b0
bH bjh bH btp _ dqk bpk dpt bcp
bpk bO bqk b0 dt bO dt bO
bqk
fbqjc (foj bqk dd\ bpu
\bz3
bZj dt
+ bO dt) bO
fbpk dzj bpk d0\ bqk
\dz^ dt bO dt) bO
f bgk bpk _ bjh bqA dz , (3.103)
\bzj bO bz3 bo) dt
Therefore
dl , bH n
do -r- = 0. (3.104)
dt dO
The integral vanishes due to the periodicity of p and q in 0. It

should be emphasized that the adiabatic invariant is the integral
oip dq around a closed curve on which s is constant in p, q phase
space, and not about the closed curve representing the zero-order
(i.e., e = 0) periodic motion. The difference between the two curves
is shown schematically in Figure 3.10 for a system with one degree
of freedom. To lowest order in e it does turn out, however, that the
s = constant curve is the same as the unperturbed curve. It is in
calculating higher-order terms in the series for / that the difference
appears. The constant z curves are known to be closed due to the
periodicity of x = x(z, 0, e) in 0.
In the present situation, in which the coordinates all have the

same fundamental frequency in the unperturbed state, the adia¬
batic invariant (3.100) has turned out to be the sum of action inte¬
grals over the degrees of freedom, rather than a component action
integral alone. Of course, for a system with one degree of freedom
there is no distinction. The integral for / can be written
I = SS dp-dq (3.105)
where the double integral is over the area of the z = constant

curve; this is one of the Poincare invariants. It is not surprising
that / should be a Poincare invariant: the value of the adiabatic
invariant for a given system should be independent of the canonical
variables used in (3.98) and (3.100), and indeed the Poincare
integrals are invariant under canonical transformations, whereas
each term of f/ do p- (dq/dd) is not. It is now clear that the num¬
ber of degrees of freedom is not necessarily the number of adiabatic
invariants. A system with, say, two degrees of freedom might have
just one adiabatic invariant.
On the other hand, there may be additional adiabatic invariants.
It is possible to prove that the Poisson bracket [0,1] equals unity.
The proof is a little lengthy to be included here, but it is given
in reference (20) in detail. The fact that [6, I] = 1 means
that 6 and / can be used as new conjugate variables in a canonical
transformation of the form (px. . .pN; qx.. ,qN) —► (Px. . PN_u I;
Qi- ■ -Qn-u 0). Let H' (Q, P, I, e) be the transformed Hamiltonian.
As indicated, the new Hamiltonian is not a function of 0 because
I = — dH'/dO and / is zero. Thus, IP will have one less degree of
freedom than H, and / will be merely a parameter. If, now, the
new canonical equations
Qi = dtf'(Pi.. .PN_U Qx. ..Qn_u e)/dP<

(3.106)
Pi = —dH'/dQi
have solutions which are periodic in P, Q space when e or some

other small parameter is zero, the entire process from Eq.
(3.98) on can be repeated and a second adiabatic invariant <£P • dQ
obtained. For the case of the charged particle this would be the
longitudinal invariant tfp{](ls. If the new canonical equations
again happen to have periodic solutions, there will be a third
adiabatic invariant, which will be the flux invariant <i> in the case
of the charged particle. Since the number of degrees of freedom is
reduced by one for each adiabatic invariant, it is now clear why
for singly periodic systems the number of adiabatic invariants is at
most equal to the number of degrees of freedom, and will be less
if at any step the new canonical equations corresponding to (3.106)
fail to have periodic solutions in the unperturbed state. In connec¬
tion with the charged particle, this would be the case if the motion
along the line of force were not periodic.
By use of the four-dimensional relativistic Hamiltonian (25) //(t)
= — (l/2mc2)r?lK[p‘ — (e/c)<f>l]\pK — (e/c)4>K], the method of
this section would furnish an alternate (to reference 13) way to
study the adiabatic motion of relativistic particles.
The preceding proof that dl/dt is zero would establish I as a
rigorous (not adiabatic) invariant of the motion provided the
series for p and q were rigorous solutions of the equations of motion
instead of nonconvergent asymptotic solutions. It is the asymp¬
totic property of the series for p and q that makes / an adiabatic in¬
variant. The theorem establishing a rigorous invariant would be
as follows: Given (1) two functions p(z, 9) and q(z, 9), which are
periodic in 9, and (2) that 6 = 6(t) and z = z(t) are functions of
time such that p(z(t), 9{t)) and q(z(t), 9(t)) are solutions of the
equations of motion pt = — [dH(p, q)/dqt], = dli/dpt derived
from a Hamiltonian. Then
dq(z(t), 6)
de
dd
is independent of t. Such a theorem is not of much use unless one

knows how to find functions p, q, z, and 9 with the required proper¬
ties. The theory discussed in the present section supplies the func¬
tions p(z, 9), q(z, 9), z, and 9 as asymptotic series in e.
CHAPTER 4
Additional Applications of
Adiabatic Theory
Several applications of adiabatic theory, such as its application

to Fermi acceleration, have been made at appropriate places in the
preceding chapters. A few more are presented in this chapter.
A. The Current Density in a Plasma
In electricity and magnetism texts (26) it is usually proved that the

equivalent current density due to magnetization of a material is
given by cV X M, where M is the magnetic moment per unit vol¬
ume. Under adiabatic conditions the extension to a collisionless
plasma would be expected to add to cV X M the current of the guid¬
ing center motion. This is indeed correct, but is not so easy to
prove rigorously for a general B field geometry. The author is
aware of no published proof and therefore will outline one here,
with many of the details omitted. The complete proof is lengthy.
We wish to prove that the current density at r is
j(r, t) = Ne(R0) + cV X M (4.1)
where N = number of guiding centers per unit volume at r and t,

(R0) = average guiding center velocity of particles with guiding
centers at r, and M = total magnetic moment per unit volume of
particles with guiding centers at r. There will be an equation of
this type for each charged component, for example, for ions and
electrons.
There does exist a reasonably simple demonstration of the per¬
pendicular component, jx = Ne(R0X) + c(V X M)x. The proof
78
ADDITIONAL APPLICATIONS OF ADIABATIC THEORY 79
starts with the second moment (27) of the collisionless Boltzmann

(Vlasov) equation. The second moment is
d(v) (v)
nm-^- = -V P + ne — X B + ne E (4.2)
at c
where (v) is the average particle velocity, n is the particle density,
<->
andP = nm((v — (v))(v — (v))). The current density perpendicular
to B is jx = ne(v)±. Consequently, jx can be obtained from (4.2)
by crossing it with B, but jg cannot. The analysis has been given
in detail in reference (4).
To derive the parallel component of (4.1), it is necessary to work
from the asymptotic series (2.1) and the collisionless Boltzmann
equation itself rather than from its second moment. Equation (4.1)
complete, and not merely its parallel component, is obtained in
this way. Let/(r, v, t) be the particle distribution function satis¬
fying the collisionless Boltzmann equation: (df/dt) + r• V/ +
(e/m) [E + (v/c) X B] • Vvf = 0. The current density is
j(r, t) = ne(v) = efvf(r, v, t)d3v. (4.3)
The procedure is first to differentiate series (2.1) with respect to
time and to obtain v correct through terms of 0(e). Next, the “guid¬
ing center variables” (28) R0 and V are introduced, where the com¬
ponents of V along the directions of ei, e2, and 63 are Vi = v$, V2 =
pco cos d, and V3 = — pco sin d. Vx is the gyration velocity in the
frame of reference moving at the guiding center velocity R0. The
particle velocity v from the asymptotic series can then be expressed
in terms of R0 and V:
V — (Eo)± + Eiei + y2e2 + Fses + ~ e2 — e2^
V2 f & A ; 1 \
Je2 /n K *)V
-( — e3 — e3 1 d-(«n — Rn ) V2
00 \2B / p
+ - (Ru + Rn*)V, + — (R20 - R2o*)(V22 - Es2)

p w
4f2
+ — (R2o + R2o*)F2F3 + 0(62) (4.4)
OJ
where all vectors and fields are evaluated at Ro. It will be noticed
that the R2o term, which was of order e2 in r, gives an order e con¬
s
tribution to v due to differentiation of exp (2 i udt) with respect
to time. Terms of 0(e) are needed in v if / vfd3v is to be correct
through 0(e), and this is necessary since (c/e)V X M is of order e.
The next step is to transform from the particle distribution func¬
tion to the guiding center distribution function F defined by:
dn = fir, v, t)d3rd3v = F(R0, V, t)d3R0d3V. (4.5)
Thus F(R0, V, t)dsR0d3V is the number of guiding centers in d3R0

of particles with parallel velocity in dV\ and gyration velocity in
dV2dV3. Division by d3r gives
fir, v, t)d3v = FiR0, V, t)3iR0/r)d3V (4.6)
where $ is the Jacobian of R0 with respect to r at constant V. Be¬

cause R0 = r + (F X ei/u) + 0(e2), the Jacobian is
3iR0/r) = 1 + (F2/co)e3-[(62-V)62 + (erV)^ + (VB/B)]
- (T3/co)e2- [(e3-V)e3 + (erV)ei + (VB/B)] + 0(e2). (4.7)
Zero-order terms in the expression for v, such as e2(Ro) V2, must be

expanded about r because we want jr'(r) bi terms of guiding center
velocities at r and in terms of the magnetic moment of particles
with guiding centers at r, not R0. For the same reason F(V, Ro)
must be expanded about F(F, r). In addition, F{V, r) can be ex¬
panded as FoiV, r) + eFfV, r) + .... After these expansions have
been made, Eqs. (4.4), (4.6), and (4.7) are substituted into f vfd3v.
Even after terms of 0(e2) are dropped, many terms still remain in
the integrand. A large percentage of these integrate to zero, how¬
ever, by virtue of the following facts: (1) F0 and Fx are functions of
V<2 and F3 only through the combination V22 + F32; this is not ob¬
vious and must be demonstrated by use of the Boltzmann equation
which F satisfies. (2) Rn is a function of Vi, E22 + E32, R0, t. (3)
JR20 is a function of F22 + F32, R0, t. The last two statements again
are not obvious, but can be verified by examining the equations Rn
and R20 satisfy.
The expression for f vfdsv simplifies to
J vf(r, v, t)dh = J d3VR0(r)F(r, V, t) + J d3F6x

'F22 + F32 F22 + F32
X v F(r, F, 0 d*V
2co -/ 2co
[(e3-V)e2 — (e2-V)e3 — e2e3-(e2-V)e2 — e2e3-(ex-V)ex
+ e3e2- (e3-V)e3 + e3e2- (ex- V)ex]F(r, V, t) (4.8)
where j?0(r) is the velocity of guiding centers at r, not R0. The unit
vectors and o> are at r too. The long vector expression in the last
integral of (4.8) turns out to be V X ex; this is by no means obvious.
Therefore,
£ vf(r, v, t)d3v = f\d*VR0F(r, V, t)
r y 2 . y2
-v XJ ex(r) F(r, F, t)d3V. (4.9)
The magnetic moment per unit volume of particles with guiding

centers at r is
e r Fa2 + F32
M{r) = - - ex(r) ^ ~ F(r, F, f)d3F (4.10)
c J 2co(r)
and therefore
J = J d3VR0(r)F(r, V, t) + -e V X M(r) + 0(e2) (4.11)
j(r, t) = efvfd3v = Ne(R0(r)) + cV X M(r) (4.12)
and the theorem is proved.
B. Loss-Free Geometries
The ordinary laboratory mirror machine with a magnetic field

like that of Figure (1.4) has a “loss cone.” If at any point the
velocity vector makes too small an angle with the field line, the
82 ADIABATIC MOTION OF CHAKGED PARTICLES
particle will escape through the mirror [by Eq. (1.29)]. In the ab¬
sence of diffusion in velocity space due to collisions with other par¬
ticles, there would be no way adiabatically for a particle to get into
this loss cone if it did not start out in it. But in practice such colli¬
sions do occur. The question therefore arises of whether there are
any static field configurations from which a p rticle cannot escape
so long as it behaves adiabatically. (Nonadiabatic behavior is
another possibility that must be considered.) Although the adia¬
batic invariants do not seem to tell how to find such a configura¬
tion, they do provide a means for testing a proposed one. The
criterion is that all the longitudinal invariant surfaces passing

through an arbitrary point in the containing volume must at no
other place intercept a physical obstacle, such as a vacuum chamber
wall or a current-carrying wire. Since the surfaces are not generally
simply nested and different velocity vector directions at the same
point in space correspond to different surfaces, considerable effort
may be needed to calculate where the surfaces lie. One loss-free
configuration is that of a circular current loop (Fig. 4.1). For that
matter an ordinary mirror machine is loss-free if the external
return flux is included in the vacuum system. The guiding center
of a particle remains on a flux surface and repeatedly transverses
it. The problem is to supply the current to the coil and to support
it mechanically without introducing wires and other obstacles into

the path of the lines of force.
Another geometry which has been extensively studied (29,30) for
loss-free properties is the “bumpy torus.” It consists of several dis¬
crete circular windings arranged around a toroidal vacuum cham¬
ber (Fig. 4.2) muc' as if several mirror machines were arranged
end to end and then bent into a circle. It does, indeed, have regions
in which a particle will not be lost regardless of the direction of its
velocity vector, and the coils can be supported in such a way that
the supports do not interfere with the containment region. How¬
ever, losses still occur, due to a rather novel type of diffusion,
which is a consequence of the existence of many invariant surfaces
through any one point. If the velocity vector of a particle is
changed by scattering at a point, the particle changes invariant
surfaces and may change to one that lies nearer an obstacle at
some other point in the system. When the particle reaches this
second point, it may again scatter onto a third invariant surface
which lies still further out, etc. Thus, because of the presence of
many invariant surfaces at any one point, the particle may still
work its way out of the system but more slowly than by scattering
into the loss cone of an ordinary mirror machine. Of course, a
geometry that is loss-free in the present sense may still lose parti¬
cles by ordinary diffusion across the lines of force. If the loss due
to the new mechanism in the bumpy torus is less than that due to
ordinary diffusion across the lines, the torus is for practical pur¬
poses loss-free. The stability of a plasma in the bumpy torus is
another subject for study.
Loss-free geometries are of particular interest in connection with
the change of e/m injection methods which have been proposed (31,
32) for building up a thermonuclear plasma. In these methods par¬
ticles such as H2+ or neutral H are injected into a magnetic field at
energies of many kilovolts or even a million electron volts. Some of
these particles undergo a change in e/m by means of collisions with
neutral background gas or with previously injected particles, and
thereby become trapped if the geometry of the field is proper.
The time scale for such systems to build up has been calculated to
be as long as several minutes. The final steady-state density is

limited by the loss rate of particles already trapped. One loss
mechanism is scattering into a loss region of velocity space, so that
a loss-free geometry would increase the final density.
Another advantage that a loss-free geometry may possess is
that of greater plasma stability. Sufficiently anisotropic velocity
distributions are known theoretically to be unstable, (33-35) and the
instability may have been observed experimentally (36)in an ordi¬
nary mirror machine. If a field had no loss region in velocity space,
the distribution function could be nearly isotropic.
C. The Geomagnetic Case
The behavior to be expected of adiabatic particles in an unper¬

turbed geomagnetic field has been studied in some detail (21). The
geomagnetic field approximates that of a current loop, but the
system is not loss-cone-free because of the presence of the earth,
nor is the field exactly azimuthally symmetric about any axis.
Nevertheless, so long as the three adiabatic invariants are con¬
served, a trapped charged particle cannot escape, but continues to
precess about its longitudinal invariant surface as described in
Chapter 3. The third adiabatic invariant predicts the effect of the
earth’s rotation: the longitudinal invariant surfaces rotate rigidly
with the earth, while each particle precesses rapidly about its
slowly rotating surface. The azimuthal asymmetry of the earth’s

field, coupled with the earth’s rotation, produces an electric field
in an inertial frame. Very likely the ambient plasma conductivity
prohibits an Ef however, the remaining E± produces drifts with
components in a generally radial direction. The third invariant
tells us that the integrated effect of these drifts does not constitute
a loss mechanism, for if the particle remains on its rigidly rotating
surface, its M, J, and <t> are indeed conserved.
An unsolved problem in connection with the Van Allen radia¬
tion is why the density of the inner or proton radiation belt falls off
so rapidly with radius. It is observed (37) that the density of protons
with energy greater than 75 Mev falls from a maximum at about
10,000 km from the center of the earth to practically zero at 12,000
km in the equatorial plane. A current theory (38) holds that the pro¬
tons are the product of /3-decay of cosmic-ray-produced neutrons.
If this is the only source of protons, it is necessary to explain the
large decrease in density beyond 10,000 km. The decrease in the
source strength is not this rapid. One possible explanation is in
terms of a large loss rate due to nonadiabatic effects. If, for ex¬
ample, the magnetic moment decreased during many longitudinal
oscillations, the particle would eventually be absorbed by the
earth’s atmosphere. Such a process might be described as scatter¬
ing into the loss cone by the magnetic field, rather than by other
particles. Possibly the gyration radius of a proton of, say, 100 Mev
energy is so large at 12,000 km that nonadiabatic effects are im¬
portant even in a static dipole field (39). It is also possible that rapid
variation of the field due to Alfven waves affects the magnetic
moment (40-42), or that fluctuations in the geomagnetic field affect
the longitudinal or flux invariants (43) in a fashion deleterious to
long particle containment.
An excellent summary of present knowledge regarding the
trapped radiation has been given by Van Allen (44).
D. Plasma Stability
Plasma instabilities are frequently observed in experimental

plasma devices, and the instabilities usually affect the plasma con-
Figure 4.3. The linear diffuse pinch.
tainment adversely. Consequently, the theoretical study of plasma

stability has received much attention. Three models for the plasma
are commonly used—the hydromagnetic model (45), the collisionless
model, and the Chew, Goldberger, and Low (C. G. L.) or “double
adiabatic” model (46). Each model has a range of validity or utility.
The hydromagnetic model assumes that the plasma behaves as an
ideal adiabatic gas (Pp-7 = constant) with scalar pressure and
without resistivity or viscosity. The collisionless model uses the
Vlasov equation as the basic equation, and hence assumes that
close collisions between particles are negligible. The C. G. L.
model also assumes no collisions, but makes the additional as¬
sumption that heat flow along the lines of force vanishes both in
the equilibrium state and in the perturbed state. The zero heat
flow assumption is not correct, but, as will be described below, the
C. G. L. model does supply a necessary condition for stability.
Since the adiabatic particle theory appbes when there are no
close collisions to change the particle’s momentum suddenly, it is
logical to suppose that adiabatic theory could be useful in working
with the collisionless model. This is indeed the case, and the tech¬
nique has been developed to an advanced state by Newcomb (33),
who uses the magnetic moment and longitudinal invariant to de¬
termine stability criteria. The method will be illustrated here with
the linear pinch. One can use the very general expressions in refer¬
ence 33, but it is more instructive to derive the stability criterion
from first principles.
Let the equilibrium state be as illustrated in Figure 4.3, where
the magnetic field is azimuthal, and both B and the pressure of the
plasma vary with radius, but not with z. The pressure is assumed
nonscalar. Let/(w±2, wj, r)dv±2dv[\dzr be the number of particles in
d?,r in the equilibrium state having parallel velocity in dv^ and v±2
in dv±2. The pressure tensor of the ions or electrons is
P — eieiPy + (e2e2 + 6363) Pj_ (4.13)
where
F*U = m f fv^dv^dv^
(4.14)
P± = ~ J fv^dv^dv^
and m is the appropriate mass (ion or electron). There will be an /

for ions and one for electrons, and the total pressure is the sum of
the ion and electron pressures. From here on the problem will be
treated as if the plasma had one only component (say electrons
with a neutralizing positive background). This assumption merely
simplifies some of the expressions by eliminating sums over ions
and electrons. The ultimate stability criterion obtained becomes
correct for the two-component plasma by regarding P± and P0 as
total pressures rather than electron pressures.
The equilibrium state is specified by the condition that there be

no net force on an element of volume. Thus, at equilibrium,
-V-P + (j/c) X B = 0. (4.15)
The Maxwell equation V X B = 4irj/c is used for the current

density j. In the cylindrical coordinate system of Figure 4.3, the
operator V is
V = - +ez- + (4.16)
T br z bz r b&
where the unit vectors have obvious definitions. For the problem*
ei is etf, e2 is — e„ and e3 is e2. Thus
VP = + (erer + e2e2)Pj_].
(4.17)
With due attention to derivatives of the unit vectors, one finds
(4.18)
Similarly,
(V X B) X B = (rB0) (4.19)
r br
and the equilibrium condition (4.15) reduces to
b P ± + BjV4t
(4.20)
br r
To test the stability of this equilibrium, a virtual displacement

£(r) is given to the particles with guiding centers in d3r. The change
8W in the energy of the system is calculated. If this change is nega¬

tive for any £(r) the system is unstable.* The object is to find from
among all the equilibria that satisfy (4.20) the ones for which 8W
is positive for all l(r).
The change in energy is composed of the change in magnetic
energy and the change in particle kinetic energy. Consider first the
magnetic energy: [B2(r)/8ir]d3r is the energy in the volume d'r
before the displacement. After the displacement the magnetic field
is B(r + 0 and the element of volume has become dzr[${r + £)/r],
where $ is the Jacobian of the transformation from r to r +
All quantities are needed to second order in ?, since to first order
the change in energy should vanish if the unperturbed state is an
equilibrium state. The Jacobian is: determinant |v[r + £(r)]|,
which turns out to be
$(r + ?)/r = 1 + V-5 + |[(V-e2 - V£:V?] + 0(e) (4.21)
where means (Pî/^ri)(P^j/^rd- T° obtain B(r + ?)>

i,j
more information about the £ displacement is needed. During the
displacement the guiding centers drift at velocity uE + 0(e), in
accord with Eq. (1.17). Since E + (uB/c) X Bis zero, Maxwell’s
equation V X E = — (1 /c)(dB/dt) becomes
V X (u, X B) = 5B/dt (4.22)
where the arguments of uE and B are at the same point in space

and time. Because the displacement from equilibrium proceeds as
a function of time, let £(r, t) be this function; ?(r, fiinai) is the quantity
* There are several questions which may occur to the reader in connection
with this extension to a plasma of the principle of virtual work from mechan¬
ics. For example, one expects that, for the plasma, 8W must depend only on
the final value of £(r) in order to employ this energy principle. If SW depended
on the time history with which the displacement away from equilibrium were
carried out, 8W would not even be unique for a given £(r). This and other
questions are answered by Newcomb.
denoted by «£(r) previously. Likewise, what has been called

B(r + ?) is B[r + 5(r, tnn*i), f«n.i]. Consequently,*
uE[r + £(r, 0, C = [d£(r, 0/<^]_l- (4.23)
The field B(r + Q is determined by Eqs. (4.22) and (4.23). They

could be solved directly to second order in l. However, a much
easier method for obtaining B(r + £) to second order uses the fact
that the uE drift is flux preserving (Section 3.C). The problem is
thereby converted to the geometric one of how elements of surface
area change under the displacement L The result is (48) B(r + £) =
(B + B-VQ/g, which is correct to all orders in L Expanding l/g
in powers of £ gives
B(r + 0 = B + B-V? - BV 3; + |B(V-02
+ IBVCVC - (V-0B-V5 + 6(e) (4.24)
where B on the right side stands for the unperturbed value B(r, 0).
The change in magnetic energy of the system is
5W,
iC_Jr 8tt
magnetic
— BKr + $$ B2(r) (4.25)
The pinch stability problem will now be limited to azimuthally

symmetric perturbations only, for which £r and are independent
* The reader who is familiar with the particle drift explanations which
have been given for some instabilities (4,47) may ask why the 0(e) drifts have
been ignored and d%/di equated to ue only, for in these explanations the first-
order drifts are all important in providing charge separation and the necessary
regenerative action. The 0(e) drifts are different for ions and electrons and
therefore lead to charge separation, whereas the ue drift does not produce
charge separation. The resolution of the paradox lies in the fact that the 0(e)
drifts are inherently present when the energy principle is employed. In deriving
and justifying the use of an energy integral the macroscopic equation (4.2) is
used. As shown in Section 4.A, this macroscopic equation yields a jj_ equal
to the guiding center current plus the curl of the magnetic moment per unit
volume. Consequently, the first-order drifts have not been ignored, but are
implicitly accounted for by use of the energy integral. Reference 33 contains a
discussion of this matter. It should also be noted that if the 0(e) drifts had
been required, it would have been impossible to consider only one sign of
particle (electrons), as has been done here.
of azimuth, and can be considered zero. The perturbation thus

corresponds to expanding or contracting the radii of annular ele¬
ments of volume and displacing them in the z-direction, as in
Figure 4.3. With the substitution of (4.21) and (4.24), 5TFmagnetic
becomes
8W,magnetic
- J 8tt
BAr) -V-? + 2Mi:VC + - (V-02
+ l V?:V£ - 2(V-?)e1e1:V£ + ^ (4.26)
where B#(r) is the unperturbed value.

The change in particle kinetic energy consists of the change in
perpendicular energy and the change in parallel energy. Because
the magnetic moment M of each particle is conserved* during the
displacement,
bWx
= fd’rf dv±2dvnf M[B(r + 0 - B(r)]
B(r + Q
dzr P i l (4.27)
-f B(r)
where the magnitude of the magnetic field now appears and P± is
the unperturbed value defined by (4.14). Equation (4.24) must
therefore be squared and the square root taken through 0(£2).
The result gives
8W,
-f d3r P i -V-5 + l (V-02 + \ V?:V?
(V-ptAWi - - (Mi:V6* + (4.28)
The change in parallel energy produced by the £ displacement

is obtained from the longitudinal invariant. Because the lines of
* Since only the neutrally stable or “marginal stability” condition is to be
found, the £ displacement can be considered slow enough for the adiabatic
invariants to be conserved.
force are circles both before and after the perturbation, Dj is con¬
stant along the line of force in both the unperturbed and perturbed
state. The longitudinal invariant is simply times the circum¬
ference of the circle. Therefore rug (initial) equals (r + ^r)»j(final),
or
fll(final) = »[ (initial) r/(r + £r). (4.29)
The change in parallel energy of a particle is
^ A(»s2) = ^ fly2(initial) ^ — + 0(£3)- (4-30)
The total change in parallel energy becomes
8Wi = j db J dv^dvtf^ »,* y +
+ (431)
where Py is the unperturbed value.

The total energy change is
5W = J db (-V-? + 2eiei:V0 + P±(e- V-©
-H*g 2 (V-02 + 2
- 2(V^)eiei:V? + ^] + P ^(V-02 + ^V^:V?
- (V-*)6A:V* “ ^ (e^rVO2 + |^2 + PB (4.32)
For this problem eiei: V £ = £r/r. After an integration by parts,

the first integral vanishes by virtue of the equilibrium condition,
as it should. The second integral from which the marginal stability
condition will be obtained simplifies to
8W = \ / d*r {tT P\\ + (p± + [(v-?)2 + V£:V^]
• (4.33)
For the purpose of finding the marginal stability condition, it will

be convenient to have i appear only in the form of V • £ or £r/r.
It is therefore necessary to express the term containing V£:V£ in
this form. Let P± + B#2/8ir = S(r), and use the identities
V£:V* = (V-$2 - V-K(V-e) - (C-V)ei (4.34)
SV-K(V-*) - tf-V)?] = V-KV-ftS) - Stf-V)*]

- 2(V-?)(C-V)S - {.(C-V)VS. (4.35)
The latter is not obvious, but is readily verified; essentially a double

integration by parts is being performed. The divergence on the
right side of (4.35) will lead to a surface integral upon integration
over the volume of the pinch. This surface integral vanishes via
the boundary condition = 0 at the wall and because periodic
boundary conditions in z are assumed. The energy change now be¬
comes
-iJ* 2(p-+ff)
1,rl
r7 r i sr
- £) - ("i +to]
v K ~r
r 2(P_l + 8tt) _
.
& ( , B**\ 1
+
L3p«+1 +r**= + «7)J
r d ( b#2\ ( BJW 2
J dr (p-+1) ~ (p- + 7)J (£rV r2) (4.36)
2(Pj_ +
where the latter form is obtained by completing the square in

V • For stability against all azimuthally symmetric displacements,
the equilibrium state must not only satisfy the equilibrium condi¬
tion (4.20) but must also be such that 8W is positive no matter
how £;(r) is chosen. The first term in (4.36) is positive or zero for
any £(r). By properly relating V • £ and it can be made to vanish.
Suppose now that the coefficient of £r2/r2 from the second and
third terms is negative at some radius. Then £r can be chosen non¬
zero in a small interval about that radius and zero elsewhere; 8\V
will be negative. Consequently, for stability it is necessary for the
coefficient of £r2/r2 to be positive. It is also sufficient.
If the equilibrium condition is employed, the necessary and
sufficient condition for stability is found to be that at all radii the
following must hold:
For isotropic pressure, Py = Px = P, and the pinch is stable if
Bl 5 P2
(4.38)
4tt + 2 B//4*-’
The last may be compared with the result (49) from the hydromag-
netic model: the system is stable if
( 7P + 4'7r / B# dr 4<t
(4.39)
The utility of the C. G. L. theory lies in an inequality (33,50): 8W

(adiabatic invariant) ^ 8W (C. G. L.). For the present example
the equality holds and the C. G. L. model yields Eq. (4.37). This
is physically plausible because of the symmetry. No heat flow
would be expected in the azimuthal direction even when the system
is perturbed, since the lines of force remain circular. In general,
when the inequality holds, it is apparent that the C. G. L. model is
useful because it supplies a necessary condition for stability.
The hydromagnetic model supplies another limit (50) on 8 W.
If 7 = 5/3, 5lF(hydromagnetic) ^ 5lF(adiabatic invariant, with
scalar pressure). The hydromagnetic model therefore furnishes a
sufficient condition for stability. If Eq. (4.39) holds for y = 5/3,
then Eq. (4.38) also holds.
CHAPTER 5
Nonadiabatic Behavior
The practical importance of nonadiabatic behavior has been indi¬

cated in connection with the geomagnetic case in the preceding
chapter. Knowledge of nonadiabatic behavior may be crucial in
explaining what is observed in the space around the earth. Non¬
adiabatic effects can be studied in both static and nonstatic fields,
the static case being perhaps simpler. In this chapter some of the
present knowledge and questions concerning nonadiabatic effects
in static fields will be reviewed. There is also some very good work
with nonstatic fields (references 40 to 43, for example), and much
more is needed. A simple nonstatic case is that of the flux invariant
in the rigidly rotating, nonsymmetric field of the earth.
What are often called nonadiabatic effects really fall into two
classes. The first are not true nonadiabatic effects, but are merely
manifestations of higher terms in the asymptotic series for the
invariants. These terms may indeed be important. The second are
true nonadiabatic effects and cannot be represented by higher terms
in an asymptotic series. Both types will be seen below in the mirror
machine.
A. Nonadiabatic Effects in Mirror Machines—Numerical
A principal unresolved question is the extent to which particles

obey adiabatic predictions. The question probably does not have
a general answer, and each situation must be examined inde¬
pendently. It is possible to follow particle trajectories by numeri¬
cal methods and to compare the results with the adiabatic predic¬
tions. This has been done (51) for the laboratory type of mirror ma-
96
NONADIABATIC BEHAVIOR 97
Figure 5.1. A periodic mirror geometry for which numerical calculations have
been compared to the adiabatic predictions.
chine, or, more precisely, for many mirror machines end to end in a
straight line (fields taken as periodic functions of z. Fig. 5.1).
The somewhat surprising observation was made that as the par¬
ticle oscillated between mirrors the magnetic moment could vary
by large amounts but that the variations were highly self-canceling
from one oscillation to the next. To be more specific, the particle
was started olf at the plane midway between mirrors, as shown in
Figure 5.1. The angle between v and B(r) is 8, while X is the angle
between v± and the plane of the page. Each time the particle re¬
turned to the median plane, new 8 and X were computed. A cumu¬
lative drift of 8 would imply a gradual change in the magnetic
moment; in an azimuthally symmetric field, conservation of the
canonical angular momentum p6 prohibits a radial drift, and there¬
fore B(R0) cannot change. Thus a change in v±2/B(R0) must come
from a change in v± = v sin 8 and therefore from a change in 8,
since v is constant. Phase plots of the type shown qualitatively in
Figure 5.2 were obtained. The numbered points represent succes¬
sive traversals of the median plane, zero being the initial point.
There are two types of particle behavior evident. The first might
be termed stable (curve A or B) and the second type is unstable,
as exemplified by points x. The 5 of a stable particle may undergo
severe variations, and the larger the gyration radius, the larger
the variation. However, the variation is highly self-canceling; the
Figure 5.2. Phase plots for a particle in the mirror geometry of Figure 5.1.
phase points always appear to fall on a well-defined curve. On a

large-scale plot the points appear to fall on a smooth curve to
within less than 1()~3 degrees in 8, and that much scatter can be
attributed to numerical errors. Thus it appears that the stable
particles are very stable indeed as a result of a “memory” from one
median plane traversal to the next.
By contrast, the unstable particles show no such memory and
form no regular pattern. They usually escape within 10 or so
mirror reflections. The line of demarcation between the stable and
unstable regions also is quite sharp, although just how sudden the
transition is has not been carefully studied. It would be interesting
to investigate the transition region more minutely. The unstable
region, which begins at the loss cone and extends upward in 8 to
the line of demarcation, does not appear to cover the entire range of
X from 0 to 2 it. Near X = tt/2 there are stable curves of type B even
below the line of demarcation. The fixed points at the center of
type B patterns are rigorously fixed by virtue of the symmetry
about the median plane. Because the unstable particles escape in
a few reflections, the demarcation between the stable and unstable
regions is the loss cone for practical purposes. It has been found
that the unstable region vanishes when the ratio of gyration radius
to the distance between mirrors is less than about 0.03. Particle
trajectories have also been studied by the author for cases in
which there is an electric field which is radial or perpendicular to
B. In both cases, patterns similar to those in Figure 5.2 occur.
The behavior of particles within the loss cone as they go from
one mirror section to the next in this periodic machine has not
been studied. It would be worthwhile studying, since these parti¬
cles are analogous to those in the bumpy torus that are not trapped
in one section but go completely around the torus.
The type A stable curves are predicted qualitatively by the first
two terms of the magnetic moment series [Eq. (3.87)]. From the
fact that the canonical angular momentum p# is a constant of the
motion, it is possible to write that equation in the form f(b, X) =
constant. The field used in the numerical calculations of Figure
5.2 was derived (in cylindrical coordinates) from the vector poten¬
tial:
LBo
Aa — + a COS (5.1)
2tt 2
where L is the distance between mirrors, p is 2xr/L, f is 2-kzIL, a is

0.2, B0 is the field halfway from the mirror to the median plane,
and h is the Bessel function. With this field
/(5, X) = sin 5(sin b + a cos2 b sin X) = constant (5.2)
where
4>iraIi(po) mcv
a (5.3)
[1 - alo(po) ]2 eBoL
and po is the solution of
4tT2cp#
(5.4)
eB0L2
The agreement between (5.2) and the numerical computations

improves as the gyration radius is reduced.
Gardner (24) has obtained the magnetic moment series including
the 0(e2) term for the special case of the particle at the median
plane. In terms of the instantaneous quantities at the particle
position [as in Eq. (3.87)], it is:
D±2 ecB'
constant = ^:-—
777 (d2 + vf)v#
B(r) B3
e2c2B"
2 Bi
v»\2 + \
(v + .p) (V2 + vf)
e2c2B'2
+ B5
^ (3d;,2 + vn2) • (V2 + I)D2) + ^ r±4
t2c2B'
+ 2 rBi
V# V 2 2 — — - v±2(v2 + Du2)
4
(5.5)
The velocity v# is rd — v sin 5 sin X, etc., for Dj_ and Dy; B' is dB/dr
and B" is d2B/dr2 at the particle position. With this expression,
/(5, X) of Eq. (5.2) could be extended to 0(e2). This and even
higher terms of the magnetic moment series might give more de¬
tails of the stable orbits. But the unstable behavior could not be
predicted from the series.
A theory of the behavior observed in Figure 5.2 has been given
by Chirikov (52). According to the theory, particles are unstable
which exhibit a resonance between the fundamental longitudinal
oscillation frequency (or one of its harmonics) and the gyration
frequency. The theory has been given only in the approximation of
straight lines of force. This is equivalent to assuming V • B ^ 0.
The importance in this theory of line curvature would be worth
investigating.
Another theory of the charged particle in a mirror geometry has

been developed recently by Moser (53) and Gardner (24). The ex¬
istence has been demonstrated of closed curves in a two-dimensional
space which are invariant under an area-preserving mapping in an
annular region of the space, provided the mapping functions satisfy
certain conditions. The theorem applies to the charged particle
in a mirror geometry if the radius of gyration is sufficiently small.
When applied to phase plots like those in Fig. 5.2, the theorem
proves that some type A curves are invariant—which means that
any point on the curve becomes another point on the same curve
after one reflection at the mirror. Since particles cannot cross an
invariant curve, any particle above an invariant curve in Fig. 5.2
would be permanently trapped,* even though it might not itself
be moving on an invariant curve. It is probable that among curves
of type A, all of which appear numerically to be invariant, there
are some that are not really invariant, but that the deviation is
much less than the numerical errors.
The theorem also proves that the class of invariant curves is of
nonzero measure and therefore covers a finite fraction of the area of
the phase plot. Moreover, the measure approaches unity as the gyra¬
tion radius approaches zero. For a nonazimuthally symmetric
mirror geometry the topology is such that an invariant curve
cannot “seal off” noninvariant curves from the loss cone. But the
measure of permanently trapped particles is still nonzero, and
therefore the theorem is of practical importance.
B. Nonadiabatic Effects—Experimental
The containment time of relativistic positrons from Ne19 in

laboratory mirror machines has been measured by Lauer, Gibson,
and Jordan (54-56). Exponential decay of the positron density is
* The reader may be aware that certain particles which encircle the mirror
axis are trapped forever. This fact is easily demonstrated from the conserva¬
tion of energy and canonical angular momentum p$. However, we are here
concerned with particles whose trapping is not insured by the conservation
laws, at least in so simple a fashion—for example, particles that do not encircle
the axis.
observed, with time constants as long as many seconds. The observed

loss rate is quantitatively attributable to scattering from the back¬
ground neutral gas, and therefore is not a nonadiabatic effect,
even though a particle has made of the order of 1010 mirror re¬
flections in the decay time. The accuracy of the self-canceling ef¬
fect in the stable orbits A and B of Fig. 5.2 therefore is quite
phenomenal. It would be impossible to test for such long contain¬
ment by numerical methods because of the computer time required
and the numerical errors that would accumulate.
If in the experiment the gyration radius is increased so that the
ratio of gyration radius to L is > 0.03, evidence of the unstable
region and enhanced loss cone appears.
Introduction of considerable azimuthal asymmetry (56) does not
produce markedly different results. The enhanced diffusion dis¬
cussed in Chapter 4 does become prominent as a result of the
asymmetry. This diffusion process serves to increase greatly the
mobility of particles in a generally radial direction. However, there
is no evidence that radial particle motion occurs because the longi¬
tudinal invariant is violated. Similar containment experiments
with a field resembling the earth’s would be worthwhile, and the
possibility that enhanced diffusion occurs in the geomagnetically
trapped radiation should be investigated.
C. Another Problem
In closing, an unresolved problem in connection with the longi¬
tudinal invariant will be posed. Suppose that in a static field the
guiding center oscillates along a line of force and on the average
drifts at right angles to it in accord with Eq. (3.68). Moreover, let
B initially have a single minimum within the range of oscillation
[Fig. 5.3A]. Suppose the drift carries the particle to lines of force
where B has a maximum within the range of oscillation and that
the maximum becomes successively larger and larger as the par¬
ticle drifts further. The magnitude of the maximum will approach
the value required for reflection [Fig. 5.3B], and the time of
oscillation will become so long that the longitudinal invariant J
will not be conserved. Eventually, when Bmax is large enough, the
Figure 5.3. J is not conserved.
particle may end up on one side or the other of the maximum. Is

there a general relation between the initial J and the two new pos¬
sible J’s, such as the initial J being equal to the sum of the two
possible final ones? If there is a general relation, it would be obtain¬
able from the guiding center drifts.
In a more general form the same problem arises in the bumpy
torus (Fig. 4.2), where because of their drifts particles can make
transitions between oscillating in one small section and moving
clear around the torus.
References
1. H. Alfven, Cosmical Electrodynamics, Clarendon Press, Oxford, 1950.

2. J. Linhart, Plamsa Physics, Interscience, New York, 1960.
3. G. Hellwig, Z. Naturforsch. 10a, 508 (1955).
4. T. Northrop, Ann. Phys. N. Y. 15, 79 (1961).
5. E. Fermi, Phys. Rev. 75, 1169 (1949).
6. E. Fermi, Astrophys. J. 119, 1 (1954).
7. E. Teller, Rept. Progr. Phys. 17, 154 (1954).
8. L. Davis, Jr., Phys. Rev. 101, 351 (1956).
9. E. Parker, Phys. Rev. 109, 1328 (1958).
10. C. Longmire, D. Nagle, F. Ribe, Phys. Rev. 114, 1187 (1959).
11. H. Lamb, Hydrodynamics, 6th ed., Dover, New York, 1945, p. 373.
12. A. Morozov and L. Solov’ev, Dokl. Akad. Nauk. SSSR 128, 506 (1959);
Soviet Phys. “Doklady” (English Transl.) 4, 1031 (1959).
13. P. Vandervoort, Ann. Phys. 10, 401 (1960).
14. P. G. Bergmann, Introduction to the Theory of Relativity, Prentice-Hall,
Englewood Cliffs, N. J., 1942, Eq. (7.49). Bergmann uses a different
metric tensor from the one used here.
15. M. Kruskal, “The Gyration of a Charged Particle,” Princeton University,

Project Matterhorn Rept. PM-S-33 {NY0-7903), March, 1958.
16. J. Berkowitz and C. Gardner, Communs. Pure and Appl. Math. 12, 501
(1959).
17. E. Whittaker and G. Watson, A Course of Modern Analysis, 4th ed., Cam¬
bridge University Press, New York, 1945, Ch. VIII.
18. A. Erdelyi, Asymptotic Expansions, Dover, New York, 1956.
19. C. Gardner, Phys. Rev. 115, 791 (1959).
20. M. Kruskal, The Theory of Neutral Ionized Gases, John Wiley, New York,
1960. For a more detailed version, see M. Kruskal, J. Math. Phys. 3,
806(1962).
21. T. Northrop and E. Teller, Phys. Rev. 117, 215 (1960).
22. B. Kadomtsev, Plasma Physics and the Problem of Controlled Thermo¬
nuclear Reactions, Vol. Ill, Akad. Nauk SSSB, Moscow, 1958; this
volume contains a similar proof for the case of static fields and E = 0.
23. W. Newcomb, Ann. Phys. N. Y. 3, 347 (1958).
24. C. Gardner (private communication).
25. Reference 14, Eq. (7.53).
26. For example, see R. Becker, Theorie der Elektrizitdt, Vol. II, Edwards
Brothers, Ann Arbor, 1945, Paragraph 21.
27. L. Spitzer, Physics of Fully Ionized Gases, Interscience, New York, 1956,
p. 94.
28. K. Watson, Phys. Rev. 102, 12 (1956); he has used guiding center variables
to prove the perpendicular component of Eq. (4.1) for the case in
which the magnetic field is static, torsionless, and curl-free.
29. G. Gibson, W. Jordan, and E. Lauer, Phys. Rev. Letters 4, 217 (1960).
30. A. Morozov and L. Solov’ev, Zh. Tekhn. Fiz. 30, 261 (1960).
31. G. Gibson, W. Lamb, and E. Lauer, Phys. Rev. 114, 937 (1959).
32. A. Simon, Phys. Fluids 1, 495 (1958).
33. W. Newcomb, “Stability of a Collisionless Plasma, I and II,” Ann. Phys.
(to be published).
34. M. Rosenbluth, Proc. 2nd Intern. Conf. Peaceful Uses At. Energy, Vol. 31,
p. 89 (1958).
35. S. Chandrasekhar, A. Kaufman, and K. Watson, Proc. Roy. Soc. (London)
A245, 435 (1958).
36. W. Perkins and R. Post, Bull. Am. Phys. Soc. 5, 353 (1960).
37. C. Fan, P. Meyer, J. Simpson, Proc. Intern. Space Sci. Symp., 1st, Nice,
1960, p. 951.
38. S. Singer, Phys. Rev. Letters 1, 181 (1958).
39. S. Singer, Phys. Rev. Letters 3, 188 (1959).
40. J. Welch, Jr., and W. Whitaker, J. Geophys. Res. 64, 909 (1959).
41. A. Dragt, J. Geophys. Res. 66, 1641 (1961).
42. D. Wentzel, J. Geophys. Res. 66, 359, 363 (1961).
43. L. Davis, Jr. and D. Chang, J. Geophys. Res. 67, 2169 (1962).
44. J. Van Allen, “Dynamics, Composition and Origin of the Geomagnetically
Trapped Corpuscular Radiation,” State University of Iowa Rept.
S(JI 61-19 (1961).
45. I. Bernstein, E. Frieman, M. Kruskal, and R. Kulsrud, Proc. Roy. Soc.
(London) A244, 17 (1958).
46. G. Chew, M. Goldberger, and F. Low, Proc. Roy. Soc. (London) A236, 112
(1956).
47. M. Rosenbluth and C. Longmire, Ann. Phys. N. Y. 1, 120 (1957).
48. W. Newcomb, Nucl. Fusion 3(2), p. 451 (1962).
49. Yu. A. Tserkovnikov, “Convective Instability of a Rarefied Plasma,”
Division of Theoretical Physics, Mathematical Inst, of V. A. Steklov,
Academy of Sciences of U.S.S.R., Moscow, Rept. No. T-2, 1960.
50. M. Rosenbluth and N. Rostoker, Phys. Fluids 2, 23 (1959).
51. A. Garren et at., Proc. 2nd Intern. Conf. Peaceful Uses At. Energy, Vol. 31,
p. 65 (1958).
52. B. Chirikov, At. Energ. (USSR) 6, 630 (1959).
53. J. Moser, Nachr. Akad. Wiss. Goettingen, II, Math. Physik. Kl., No. 1
(1962).
54. G. Gibson, W. Jordan, E. Lauer, Phys. Rev. Letters 5, 141 (1960).
55. G. Gibson, W. Jordan, and E. Lauer, “Particle Behavior in Static, Axially
Symmetric, Magnetic Mirror and Cusp Geometries,” University of
California, Lawrence Radiation Laboratory Rept. UCRL-6771, March,
1962.
56. G. Gibson, W. Jordan, and E. Lauer, “Particle Behavior in a Static,
Asymmetric, Magnetic Mirror Geometry,” University of California,
Lawrence Radiation Laboratory Rept. UCRL-6856, March, 1962.
Index
Adiabatic invariants, 41-77 d’Alembertian force, 24

defined for a charged particle, 5, 27, Diffusion, enhanced, 83, 102
41-43, 47, 61-62 ordinary, 83
in higher order, 68-71, 96-100
number of, 77 Drift velocity, 8, 27, 29, 32
of singly periodic systems, 71 averaged over a longitudinal
to all orders, 70-71 oscillation, 57
Alfven waves, 85 canonical form of, 59
Asymptotic series, 36-37 in (a, |8) space, 58
and order symbol, 37
as adiabatic invariants, 42 Energy change, in Fermi type A and
for charged particle motion, 36-40 B collisions, 18-19
rate of for adiabatic particle, 12, 32
Berkowitz, J., 36, 37
rate of for Fermi acceleration, 16
Beta ray experiment, 101-102
when E is small, 12
Boltzmann (Vlasov) equation for
Energy of adiabatic particle, 10
guiding centers, 80
Enhanced diffusion, 83,102
Bumpy torus, 83, 99
Canonical angular momentum, 41, 99, Fermi, E., 13-14

101 Flux invariant, 61-64
Charged particle motion, applications of, 84-85
asymptotic series for, 36 definition of, 61-62
dimensionless equation of, 2, 33 Flux surface, 21
equation of, 1
nonrelativistic equation of, 2 Gardner, C., 36, 37, 41, 68,100,101
Chew G., 86 Geometry of drift motion, 23-27
Chirikov, B., 100 Gibson, G., 101
Coordinate systems, Cartesian, 58 Goldberger, M., 86
curvilinear, 9 Guiding center, acceleration of, 7
local Cartesian, 3 definition of, 3, 35, 37
Current density, 78-81 definition of parallel velocity of, 7
107
108 INDEX
drift, 8, 27, 29, 32 Magnetic moment, nonrelativistic

equation of motion of, 5 definition, 5, 43
longitudinal equation of motion, 9, per unit volume, 78, 81
29, 32 proof of invariance in general,
variables, 79 44-47
Guiding center motion, geometric proof of invariance in static fields,
interpretation of, 23-27 43-44
relativistic, 27-35 relativistic, 27, 29, 32
Gyration frequency, 4 Mapping, area preserving, 101
radius, 2, 3 Marginal stability of pinch, 91, 93-95
Mirror force, nonrelativistic, 26-27
Hamiltonian, 41, 59, 73, 77 relativistic, 29, 32
Hell wig, G., 1 Moser, J., 101
Higher order effects, 68-71, 96-100
Hydromagnetic stability of the pinch,
95 Nagle, D., 20
Ne19, experiment with positrons from,
Injection of particles, 83
101
Newcomb, W., 87, 89
Invariants, adiabatic 41-77
exact, 41, 77 Nonadiabatic particle motion, 82, 85,
96-103
Jordan, W., 101 Numerical orbit computations, 96-99
Kruskal, M., 36, 37, 42, 68, 71

Period of oscillation between mirrors,
49
Lauer, E., 101
Phase plot for particle motion, 98,101
Lines of force, velocity of, 66-67
Pinch, equilibrium condition for, 88
Liouville theorem, 59
stability conditions for, 94-95
Longitudinal adiabatic invariant,
Plasma stability, 84-95
applications of, 61, 84, 87, 91
definition of, 47, 49 Poincare invariant, 76
Poisson brackets, 76
for geomagnetic field, 61
Pressure, in a collisionless plasma, 79,
instantaneous rate of change of, 57
87
proof of invariance, 49-58
tensor, 79
properties of surfaces of constant,
64
relativistic proof of invariance, 58 Radiation reaction force, 2
Longmire, C., 20 Relativistic guiding center motion, for
Lorentz transformation, 34, 35 large E, 32
Loss cone, 81, 98-99 for small E, 29
enhanced, 98-99 for zero E, 27, 29
Loss-free geometries, 81-84 Relativistic Hamiltonian, 77
Low, F., 86 Ribe, F., 20
INDEX
109
Rotating plasma, 20 definition of, 61
20-22 ,
and enhanced particle containment, proof of invariance, 63-64
Stability of a plasma, 84-95 Van Allen, J., 85

Stability of particle motion, 97-101 Van Allen radiation 42, 84-85
Stream function, 21 Vandervoort, P., 30
Velocity of a line of force, 66-67
Third (flux) adiabatic invariant, Virtual work, principle of, and plasma
application of, 84-85 stability, 89
Date Due
*
4.
CAT. NO. 23 233 PRINTED IN U.S.A.

QA 920,N6 c.2
Northrop, Theodore G. 010101 000
The ad abat c motion of charge
63 27831 7
TRENT UNIVERSITY
QA920 .N6 cop.2

Northrop, Theodore G
The adiabatic motion of
charged particles.
DATE ISSUED
949S0

The Adiabatic Motion of Charged Particles (Interscience Tracts On Physics Astronomical) (Theodore G. Northrop)

Uploaded by

Copyright:

Available Formats

The Adiabatic Motion of Charged Particles (Interscience Tracts On Physics Astronomical) (Theodore G. Northrop)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Adiabatic Motion of Charged Particles (Interscience Tracts On Physics Astronomical) (Theodore G. Northrop)

Uploaded by

Copyright:

Available Formats

THEODORE G.

The Adiabatic Motion

Interscience Tracts on Physics and Astronomy

'an introduction to celestial mechanics

Additional volumes in preparation

Lawrence Radiation Laboratory

Published under the auspices of the Division of Technical Information,

INTERSCIENCE PUBLISHERS 1963

a division of John Wiley & Sons, New York • London • Sydney

Copyright© 1963 by John Wiley & Sons, Inc.

All Rights Reserved

Library of Congress Catalog Card Number 63-22462

This copyright has been assigned and is held

Printed in the United States of America

If the author had wanted to give a properly involved title to the

The origins of the adiabatic theory of charged particle motion

culate; but manifestations of higher order motions can be im¬

1. The Guiding Center Motion. 1

A. The Equation of Motion of a Particle. 1

2. A More Formal Derivation of the Nonrelativistic Guiding Cen¬

3. The Adiabatic Invariants of the Motion. 41

A. The Magnetic Moment. 42

4. Additional Applications of Adiabatic Theory. 78

A. The Current Density in a Plasma. 78

A. Nonadiabatic Effects in Mirror Machines—Numerical .... 96

Knowledge of the trajectories of charged particles in electro¬

In spite of the limitation to slowly varying fields, the adiabatic

The Guiding Center Motion

In this chapter we will first give an informal derivation of the

A. The Equation of Motion of a Particle

The equation of motion of a charged particle is

dp/dt = (e/c)dr/dt X B(r, t) + eE(r, t) + m0g(r, t)

where r is the particle position, p is its relativistic momentum

(mc/e)[r — g(r, {)] = r X B(r, t) + cE{r, t). (1.1)

Thus any solution of (1.2) gives a solution of (1.1) by these substi¬

just use the dimensional quantity m/e = e as the smallness parame¬

B. Derivation of the Guiding Center Equation of Motion

To derive the equation of motion of the guiding center, let r =

can, of course, be evaluated at either r or R, the difference being of

R + 9 = g + (e/m){E(R) + q-VE(R)+ (1 /c)(R + 9)

X [B(R) + 9-VB(R)]} + 0(e) (1.3)

where 0(e) means terms of order e. The term (9/c) X q-VB(R)

w being the gyro frequency eB(R)/mc. Then 9 = cop (62 cos 0 — e3

9 = — [co2p(e2 sin 0 -f- e3 cos 0)] + cbp[e2 cos 0 — e3 sin 0]

T 2co[(pe2)‘ cos 0 — (pe3)' sin 0] + [(pe2)“ sin 0 + (pe3)“ cos 0]

the four terms being of order 1/e, 1, 1, and e, respectively. These

^ [e2 X (Ss-V)B - e3 x (e2-V)B] + 0(e) (1.4)

(e X (e-V)B) - (p2co/2) [e2 X (e3-V)B - 63 X (&-V)B]. (1.5)

The coefficient p2a>/2 is an approximate invariant of the motion

e2 X (es-V)B = (63 X ex) X (es-V)B = 61 teu(e3-V)B]

er (e3*V)JB = ei • (e3 • V) (^B) = (B/2)(63-V)^2 + 6 3-VB = e3-VB

e2 X (e3-V)B = ei[^3-(e3-V)B] - e3(63-V)£. (1.8)

The fact that, V • B = 0 must now be used. The operator V can be

V B = er (er V)B + e2-(e2-V)fi + e3-(e3-V)B = 0. (1.10)

But er(erV)B = ei-dB/ds = dB/ds, where s is the distance

e2 X (e3-V)fi - e3 X (e2-V)S = -e^dB/ds)

The time average of Eq. (1.3) is then

K = g(B) + (e/m)[E(R) + (A/e) X B(R)]

with an initial velocity R(0) equal to [eiei-v + (cE X B/B-)]t=o

K* = -^ + MB = ^-2 + MB. (1.40)