fl.0
U
,+......_£ li S " "'" / t/'}H,,....v'J
'Lf
..;_.''.·''
IRA B. BERNSTEIN
Yale University, New Haven, Connecticut lntrodu~tion , . . . . , . . . . . . . . . . .
I. The Gyl'ating Particle and the Magnetic Moment Adiabatic Invariant
II. The SecOnd or Longitudinal Adiabatic Invariant
III. The Third or Flux Invariant
References
311
312
321
328
333
Abstract
The reduced description in terms of drifts and adiabatic invariants of the motion of a charged particle in a strong magnetic field is derived. The demon~ strati on employs systematically two time scales and an iteration scheme for each quasiperiodicity. This leads to a particularly expeditious derivation, as well as the details of the rapid oscillations at each stage. Moreover the motivation of each part is clear, as is the relation to simple problems in dynamics. The small parameters, the existence of which underlines the method, are displayed explicitly.
Introduction
A central problem in plasma physics is the derivation of a tractable description of the motion of a charged particle in a strong magnetic field.
This task was initiated by Alfven (I) on a physical basis, and carried to a high degree of mathematical sophistication by Kruskal (2). The formal considerations of the latter author provide a constructive technique for the development of a so.,.called "reduced description" of the motion in powers of an appropriate small parameter. The method applies to all dynamical system that exhibit one or more, almost periodic motions. A partial summary
311
312 THE MOTION OF A CHARGED PARTICLE of both points of view has been given by Northrup (3), who combines several points of view. The basic notion underlying all treatments is that of the existence for each almost periodic motion of two time scales, one of which describes the rapid periodic aspect, and the other any slow perturbation of this. When this notion is applied systematically to the problem of the motion of a charged particle in a strong magnetic field, coupled with an appropriate iteration scheme, it is possible to derive, in a very efficient manner, all of the wellknown results and, in addition, indicate explicitly what are the small parameters and the details of the reduced description.
Section 1 is devoted to developing the guidingcenter description and the associated approximate constant of the motion or adiabatic invariant, the
~magnetic moment, to the lowest significant order. Section II is concerned with
'.
''the derivation of the reduced description when the motion of the guiding center along the lines of force is periodic and the particle does not move much perpendicular to the line in one period. The second adiabatic invariant, the socalled longitudinal invariant, is found to the lowest significant order in this second small parameter, as well as a description of the rapid oscillation. In
Section Ill, a final reduction is affected in the description when the energy of the guidingcenter particle changes but little in the time required to circulate once on a magnetic surface. Here, in addition to a third adiabatic invariant the magnetic flux, the details of the motion in the constant flux surface are found.
The derivations presented here have the virtue of considerable analytical simplicity and conceptual unity. It is also clear from them explicitly what the small parameter in question is, and also how to proceed tq the next order.
Moreover, at all stages the treatment is simply related to a familiar problem in classical mechanics.
I. The Gyrating Particle and the Magnetic Moment Adiabatic Invariant
The equation of motion of a particle of charge q and mass m acted on by an electric field E(r,t), a magnetic field B(r,t) and a gravitational potential
G(r,t) is l
I
I
I where r=anxr a(r,t) = qE(r,t)/mVG(r,t)
O(r,t) = qB(r,t)fmc
(11)
(12)
(13) j i
:j
'l i j
I. GYRATING PARTICLE AND MAGNETIC MOMENT INVARIANT 313
When a and n are constant in space and time, the solution of eq. 11 can be written (4): r=R+ p (14) where
R(t) = R(O) + bb · [R(O)t + .J: at 2
]
+ b x a t/f! p(t) = [e
2 cos (r:!t + </>) + e
1 sin (nt + ¢)]wjn
(15)
(16) q, are constants, and we have introduced n
= qBfmc (17) b = B/B (18) and the orthonormal righthanded set of Cartesian unit vectors, e
1 e
3
=b. Clearly,
, e
2
, and
R(t) = bb • [R(O) + at] + b x a;n p
= w[ e
2 sin (nt + ¢) + e
1 cos (nt + ¢)]
=!1 X p
(19)
(110)
The solution is readily verified by substitution in eq. 11. The vector R(t) describes the trajectory of the socalled guiding center; the term in brackets in cq. 15 arises from the accelerated motion in the direction of the magnetic field; the term b x ajn is designated the drift velocity perpendicular to the magnetic field.
Consider the case in which a and n depend on space and time but do not change much in a distance wjn or a time 1/f!, where w is the magnitude of the component of the velocity of the particle orthogonal to the magnetic field measured relative to the drift velocity, and n is the value of the gyration frequency that prevails at the point in question on the trajectory of the particle. It then seems plausible that the solution of the equation of motion will be very much like that given in eqs. l4 to l6. If this is so, one is led to seek a solution effectively in powers of the small parameter
B
R. v)
In aBI
+ j(wfr:!)V lnaBJ (111)
This is the program adopted by Kruskal (2) and leads to an asymptotic representation.
An alternative method that is more expeditious for obtaining lowest significant order results consists in the introduction of an auxiliary variable
THE MOTION OF A CHARGED PARTICLE
0 contrived to describe the rapid gyration indicated in eq. 16 for the constant field case and a suitable iteration scheme. We shall require periodicity in 8 and choose the period to be unity, so that 8 has the character of an angle variable in Hamilton Jacobi theory. For the case of constant fields one sees from eq. 16 that 0 = ntj2n is an appropriate choice. For the general case we write r = R(t) + p(8,t) whence if we denote partial derivatives by subscripts and set il = v(t),
(112) r
= R(t)
+ v(t) p,(O,t) + p,(O,t) (113)
Presumably,
V
2
Po
2 ~ p/ and we expect that in the order of magnitude v ~ 0/2rr. w'/0
In dealing with the equation of motion, since we anticipate that p 2 =
2
, we are led to expand r!(r,l) and a(r,t) in powers of p because this is effectively an expansion in powers of the small parameter e of eq. Ill. Thus one writes a(R + p,l) = a(R,t) + p · Va(R,t) + 1PP: VVa(R,t) + · · · (I14) and a parallel expansion for n. When these expansions and the time derivative of eq. l!3 are employed in the equation of motion 11, we obtain the result,
•. 2 .
R + v p
60
+ 2vp., + vp
8
+ p, =a+ p · Va + 1pp: VVa + · · ·
 n x il.p · (V!l.) x il. 1pp:(VVn) x il.
 vr!
X
Po vp · (V!l.)
X
Po tv pp; (VV!l.) x Po
 0 x p,  p · (V!l.)
X p, !pp;(VV!l.)
X p,
(115)
Note that in eq. 115 8 occurs only in p and its derivative, a= a(R,I), !l. =
!l.(R,I), and V denotes the gradient with respect toR. We shall require that p be periodic in 0 with period unity and the average of p over one period in 8 vanish. That is, following Kruskal, one can write the Fourier series: p(O,t) =
00
L
[p<•l(t)e 12
'"
0 + p<•>'(t)e12" 0]
11"'1
Thus, if we integrate eq. 115 over one period of 8, we obtain
R = a + v
1 J dO p
0
0 x [p · (V!l.)]  !l. x R + · · ·
(116)
(117) j
I. GYRATING PARTICLE AND MAGNETIC MOMENT INVARIANT
1 i
I l i
I
As will be shown later, the dependence of p on 8 to the lowest significant order is given by eq. 16 with 2n0 replacing 01. Thns, if one carries out integrations over one period in fJ, the integral of any quantity cubic in p or its derivatives will vanish to the lowest significant order. In particular, it follows from this observation that the terms indicated by dots above are smaller by a factor of the order e
2 than the largest term retained explicitly.
When eq. 117 is subtracted from eq. 115, on regrouping terms and recognizing that R is not a function of R but depends only on t, we find that
(v 2 p
0
+ v!l. xp)
0
= 2vp,
0
+ p · V(a + R x !l.) +p, x !l.
+
1 vp,
X ((l' V!l.)
Jo dO ''Pox (p · V!l.)
+ ···
(118)
3I5
The terms indicated by dots on the righthand side of eq. 118 are smaller by a factor of order e than these indicated explicitly on the right. These latter in turn are smaller by a factor of the order ethan those written on the lefthand side of eq. 118. Thus, to the lowest significant order, we require that the lefthand side above vanish, whence on integration in 0 v
2 p, + vn
X p = W) (119) where ~(t) is the constant of integration. If one Integrates eq. 1~19 over one period in 0, it follows that the lefthand side vanishes because ofeq. 116. Thus
~ = 0, and if we resolve eq. 119 in the Cartesian coordinate system associated with the unit vectors introduced prior to eq. 119, we obtain the result:
(120) n p,+p,=O v
(121) p,= 0 (122)
It follows from eq. 122 and the requirement that p have no part constant in 0 that p
3
= 0. Moreover, if one adds i times eq. 121 to eq. 120,
(p, n
+ ip,)o = i (p1 + ip,) v
(123) whence p
1
+ ip
2
= ip(t)e[i(~
0 + cp(t))]
112
(124) where we have introduced the real constants of integration, p(t) and cjJ(t).
In order that p, as determined from eq. 124, be periodic in 8 with period unity, we must require that v(t) = 0(R(t),t)/2rr (125)
,,
,,,
16 THE MOTION OF A CHARGED PARTICLE
'hus, on rewriting these results in vector form, to the lowest significant order
1 E, we obtain p(B,t) = p[e
2 cos (2n0 + ¢) + e
1 sin (2n0 + ¢)] (126) vhere p, ¢, e
1 , and e
2 all depend on time. Clearly eq. 126 reduces to eq. 16
'or the case of constant fields. The time dependence of 0 is now determined
'rom
O(t) = f dt D(R(t),t)f2n (127)
\late, however, that to this order in n, p, and </> arc not yet determined as
·unctions of time.
Jf one wishes to calculate to the next order in e, it is adequate to drop the
:erms indicated by dots in eq. 118. Rather than solve the resulting equation
;ompletely, we shall be content to derive an approximate constant of motion, oorrect to that order in e corresponding to dropping the dots in eq. 118. The derivation proceeds by forming the scalar product ofeq. 118 with p
0 , deleting the terms indicated by dots: after
G
2
Po')
0
+ (vpo'), =PoP: V(a +
R
X
D)+ Po
X
p, ·!1 Po. f:dOvpo
X
(p· vn)
(128)
If we integrate eq. 128 over one period in 0, we obtain
(
J dOvpo'
)
=
J d0p
0 t 0
0 p:V(a + :R x n) +
J d0p
0
0 x p, · n (129)
The integration over 0 has removed the nominally large terms in eq. 128, and it is adequate to use the lowest significant order approximation eq. 126 in eq. 129. Thus
1 I
Jo dOpo' = 4n
2 p
2
Jo d0[cos
2
(2n0 + ¢) + sin
2
(2n0 + ¢)] and
= 4nlpl
J
0
1 d0p
0 p = 2np
2
II dO[e
2
0 sin
'
(2n0 + ¢) + e
1 cos (2n0 + ¢)]
· [e
2 cos(2n0 + ¢) + e sin (2n0 + ¢)]
= 2np 2
1 fo dO{(e
1 e
1 e
2 e
2) sin (2n0 +¢)cos (2n0
+
¢)
(130)
= np
2
(e

1 e
2 
+ e
1 e
2 cos
2
(2n0
+
¢) e
2 e
1 sin
2
(2n0 + ¢)) e
2 e
1)
(131)
.. GYRATING PARTICLE AND MAGNETIC MOMENT INVARIANT !7
In order conveniently to reduce the remaining integral, we note that on combining eqs. 119 and 125 we can write
(132) p
0
= 2n b x p
But substituting bin eq, 132, and using the result eq. of 126 that b · p = 0, we find that b x p
0
= 2np (133)
Thus
1 J d0p
0
0 x p, · n
= n
I J dOb x Po · p,
0
(134)
= 2nD
1 f
dOp · p,
0
=
J
1
1
2nDdOp
8t o 2
2
=
8 p
2
81 since p 2 = p 2 is independent ofO. These results permit one to write eq. 129 as
(2nDp 2
),
= np 2 (e
1 e
2 •
V e
2 e
1 •
V) ·(a+
R x n) + nD(p
2
) 1
=
np2[(el X e,) X V] '(a+
R
X n) + (nDp
2
), np'D, (135)
On using eqs. 12 and 13 on the righthand side above, after transposing the term that is a multiple of that on the lefthand side, we obtain the result, since e
1 x e
2
= b,
(n11p 2
),
= np 2 {(b XV). (a+ R
X
!1) + n,)
= np
2
{b ·V
X
(a+ R xn) +D,}
= np
2
; ,
(b ·
V x
[E+ ~
R x
B]
+
~ B,)
(136)
In eq. 136, B, is to be interpreted as a time derivative holding 0 fixed, i.e., a convective derivative following the guidingcenter motion characterized by
R, namely,
8B
B =+R·VB
' 8t
(137)
But from the Maxwell equation,
8B cV x E=  at
.(138)
J
318
THE MOTION OF A CHARGED PARTICLE it follows, since V · B = 0 and VR = 0, that an
B =+R·VB
' at
= cV x [ E
+
~
R x B
J
+
B · VR
+
RV · B  BV · R
= cV x [
E ~
R x
B]
(139)
Thus
1 c
J
1 1 1 c c t c t
= 
~ cB
B · B
1
+
~ (B') c 2
1
=0
(140)
Note that E + 1/c (R x B) is just the electric field seen by an observer moving with the guiding center.
We can now conclude from eq. 136 that
(nnp'),
=
whence
J1 q n
=np c 2n
2 q
= vnp c
2
""" the socalled magnetic moment, is an approximate constant of the motion.
Such an approximate constant is conventionally termed an adiabatic invariant.
Let us now return to eq.J17. Note that the term therein involving Vn, on using eq. 131, can be approximated by v
1 J d8 p,
X
[p · (Vfi)] = vnp
2
(e
0
1 e
2 •
Ve
2 e
1
•
V) x (qBjmc) p
=  m
[(e
1 X e
2) X
Y')
X
B p
=(bxV)xB m p
=  m
[(VB) · b  bY' ·B) p
=VB m
(143) l
I. GYRATING PARTICLE AND MAGNETIC MOMENT INVARIANT
Thus on dropping the terms indicated by dots, after dotting and substituting b, eq. 117 yields b·R=b·a~b·VB m
(144) and
R~
= b
X
(R
X b)
=
2 a X b + .£_ b
X VB+ _1. b
X
1!.
= c
ExB 1 p 1 ..
+ Q b x VG + mQ b
X
VB+ Q b
X R (145)
If we define u=b·R (146) we can write eq. 144 in the form
= b · [qE mVG  pV B] + mb · R
(147) since b · b
= 0. Clearly, the acceleration along the magnetic field should not be so large as to change the magnetic field in a time comparable with n•; otherwise, the theory here developed is invalid.
Equation 145 can be solved by iteration, assuming that the acceleration a dominates, namely, to the lowest order,
(148)
~<;.J<P ~ and to the next order
In order to iterate once again and preserve accuracy, one would have to restore the terms indicated by dots in eq. 117 and also evaluate p to the next order in e.
The details of the gyration can be fixed, e.g., by choosing
(150) in which event
(151)
320
THE MOTION OF A CHARGED PARTICLE
The derivation Of the equation governing the evolution in time of the slowly varying phase function ¢(t) can be found by returning to eq. 118 and viewing it as an inhomogeneous equation for the second approximation. That is, one writes eq. 116 in the form
[p, + (fl.jv) x p], = f
(152) where v'f is given by the righthand side of eq. 118 with the terms indicated by dots deleted, and p given by eq. 126. It is readily seen that f involves only the first and second harmonics of 2n0, namely, f =
2 L.
(f(u)e2trfn0
+ f(n)•e2nfn0) n= 1
( 153)
In order that the solution of the homogeneous equation associated with eq. 152 reproduce eq. 126, we must as before select v = 2nj0.. If then, as before, we resolve eq. 152 in a Cartesian coordinate system defined by the orthornormal vectors e
1 , e
2 , the onecomponent of eq. e
3
= b, on adding i times the twocomponent to
152, we obtain
(p
1
+ ip
2 ) 00
+
2ni(p
1
+ ip
2 ) 0
= /
1
+ if
2
(154)
We require that p be represented by cq. 116, namely, that it be periodic in 0 with period one and have no part constant in 0. That is, if we write
Pl
+ ipz
00 =
L en elnnlO n= co
J, + if,
=
11=
2
I: d, e
2
"'
18
2
(155)
(156) where c
0 and d
0 are both zero, then the insertion of these expressions in eq. 154 and the equating of the coefficients of like Fourier factors e 2
'" yields
18
4n
2 n(n
+ 1)c, = d,
Clearly, when n # 1, one has c, = d,/4n
2 n(n
+ 1)
(157)
(158) and the
C
11 vanish for n = ±3, ±4, .... In order that a solution exist for n = 1, one must have d.
on multiplication by e 2 n
10
1
= 0, or equivalently as follows from eq. and integration over one period in 0,
156 d
1
= l J dG(/
0
1
+ if
2
)e"" = 0 (159)
~II. THE SECOND OR LONGITUDINAL ADIABATIC INVARIANT 121
Equation ]. 59 is equivalent to two real conditions resulting from taking the real and imaginary parts. These are effective equations for ¢ and p. On judicious combination they yield eq. 141. We shall not develop them in detail.
If one is not interested in analyzing the details of the gyration, it suffices to consider the equations governing the guiding center R(t): eq. 147 which gives the time rate of change of the component of the guiding center velocity along the magnetic field at the location of the guiding center and eq. 149 which gives the velocity of the guiding center perpendicular to the magnetic field at the location of the guiding center. The advantages of these equations over eq. 11 are twofold: first, they exhibit no fast gyrations on the scale of the gyration frequency; second, they constitute a fourthorder system of ordinary differential equations as opposed to eq. 11 which is a sixthorder system. These features are useful both for purposes of numerical calculation, and also for analytic work and qualitative analysis.
Higher approximations can be found by iterating the results just found, but in general the results are so complicated that the virtues of the reduced description are lost.
It has been shown (6) that these lowest significant order results represent the leading term in an asymptotic expansion of the trajectory of the particle in powers of the small parameter
E of eq. 111. That is, if one writes the partial sum,
SN(t) = r
0
(t)
+ er
1
(1)
+ e2 r
2(t)
+
···eNrN(t) then, for any fixed time t,
I
. tm e+0
I r(t)SN(t)
I
N =
E
0
This is distinct from what would prevail were the procedure convergent, namely, lim lr(t) SN(r)l = 0
N~oo
II. The Second or Longitudinal Adiabatic Invariant
A further reduction of the preceding guidingcenter description can be made when the motion along the lines of force is quasiperiodic and much more rapid than the motion associated with the drift. The demonstration is assisted by writing the magnetic f1eld in terms of two scalar fields a(r,l) and
{J(r,t) via
B = (Va) x (V{J) (21)
,.,. i'·
. 322 THE MOTION OF A CHARGED PARTICLE which clearly satisfies V · B = 0. To show that eq. 21 is possible, recall that one can define lines of force by the equation dr x B(r,t) = 0
Let S be some surface nowhere tangent to the lines of force. In this surface choose a family of lines. The set of all lines of force through one of the lines
. of this family defines a surface. Let a(r,t) = canst be the equation of such
"magnetic surfaces." Now choose a second family of lines in S nowhere tangent to the first, and in a parallel manner associate with them a family of magnetic surfaces y(r,t) = canst. By construction
B · Va = 0 B · Vy =0 (22) and as follows directly from the above, since Va, Vy, and (Va) x (Vy) are noncoplanar on writing B = (Va) x (Vy)j).
+
(Va)Jl
+
(Vy)v,
(Va) x (Vy) = ).B
Jf one takes the divergence of the above equation and uses V · B = 0,
B ·VA =0
I
I
I
~
•
~
' and A must be a function of a andy. We shall now introduce a new variable
{J(a,y). Clearly, B · V {J = 0. lf we viewy as a function of a and {J, and denote partial derivatives by the subscripts,
(Va) x (Vy) = (Va) x [y. Va + Yp V{J]
= Yp(Va) x V{J
We choose Yp =A. This yields the desired result
B = (Va) x (V{J) = V x (aV{J) (23)
Therefore, the intersection of any two surfaces a = canst and {J = const is a line of force, and one can interpret the associated pair of values a, {J as the coordinates of the line of force. Even though the pattern oflines of force may change in time, we shall identify that line labeled by a given pair a, {J as the same line of force. The functions a and {J need not be singlevalued. See
Figures I, 2, and 3.
U. THE SECOND OR LONGITUDINAL ADIABATIC INVARIANT whence and
Fig. 1. Diagram illustrating the construction of surfaces. oc =con st.
B
Fig. 2. Diagram illustrating the conStructFig. 3. Diagram illustrating the use of
IX, y ion of surfaces.y =canst. coordinates to label a line of force.
Now one can write the Maxwell equation
1 oa 1 o{J
E= V¢!V{JaVc at c at
b · E = b · V</!
~ c
ab · V ofJ at
= b · v(q,
+
~ ap) c at
(24)
(25)
324
THE MOTION OF A CHARGED PART!CLF
~ since both b ·\'a  0 and b Vfl 0 Th us, tf one defines the potential
V(r,t) = q</J + qa ap c at
+
J1B
+ mG
(H) t~e e'!,uBation of motion for the parallel velocity, assuming that rna x b and
J1 x v are of the same order of consistent order
't d magm u e can be written to the lowest mil= b · VV
+
mub · (Vb). R.c since to this order we make the parallel assumption that b = abj at +
(ub
+ R. .cl . Vb
~ ub . Vb
The associated expression for R' can be expressed as
_.. t..tK
(27)
R. _ b [ mn x VV
+;;
(aa at
ap )
Vfla;Va
+ mu'b· Vb] (28)
'bl Obse;lve that, in th? expression for mti, the term involving R is ostenst y sma compared With b. VV. We assume in a sense that we shall make precise later.
' .
' moreover, that V,
IS small the character of a potential well, as indicated schematic~fy 0 ;": ~:::: ~~~~:~
II. THE SECOND OR LONGITUDINAL AD!ABAT!C IN VARIAN 1
If one solves this for s, it is easy to show that t = r
l/
2
Clearly, the motion is periodic with a period
"(£) = f ds {2[£V]/m} 
112
The orbit of the particle in the s, s phase plane is the closed curve E = const.
See Figure 5.
; s
Fig. 5. A representativeS, s phase plane diagram for
'he case of constant E.
   V • E s
Fig. 4. A typical effective potential energy curve V
~s. s. both
R and V, are zero, since u =S there is a first integral of the equation of motion, fms' +
V = const = E
When it and V, do not vanish, the energy E will be a function of I.
Suppose, however, that we extend the definition of the period <(E) by means of the integral above to this case and assume that
(29)
It seems plausible in this circumstance that the motion should be almost periodic. Let us assume so and seek a solution of the equation of motion via the introduction of an auxiliary variable 0(1) such that 0 accounts for the rapid oscillation of period of the order <, and any explicit dependence on I is associated with the slow time variation. That is, we write* s = s(O,I)
(21 0) whence if we define v(l)
=
~
(211)
*The symbols 8 and v are distinct from the quantities so labeled in the Introduction.
We use the same symbols to illustrate the parallelism of the development.
326
THE MOTION OF A CHARGED PART!CLt . where the subscripts denote partial derivatives. The equation of motion reads v
2 s
00
+
V, = 2vs
01 v,s
0 
(vs
0
+ s,)b · (VR,) · b where we have used the fact that b · R
= 0 to write
(2 i2)
(Vb) ·li., = V(b · RJ(VR,) · b = (VIi.,)· b
The terms on the lefthand side above are presumably larger by a factor
1/r. than those on the righthand side. Thus to the lowest order we require that the lefthand side above vanish. This requirement on multiplication by s
0 leads to
(~v 2 s/+v),=o whence on integration m
v
2 s
0
2 + V = E(t)
The constant of integration E(t) is as yet unknown as a function of t. When one solves. for s
0
'from the above a further integration is possible, namely e
J' ds{l[E(t) V(s,a,fl,t)]/mr 112
(214)
In the integrand we have indicated explicitly that the potential V depends on the points on the line afforce labeled by a and /3, and by the timet. We have not indicated explicitly that it also depends on p.
Let us pick v = v(E,a,fJ,t), so that 8 is an angle variable; i.e., when s goes through one period of its motion for fixed a,f3,t, we require that 8 change by unity. Therefore,
~
= f
ds{2[E V]/m) 1
/
2
=
and
8(1) = f dt v (216)
In order to determine E(t), we revert to the equation of motion 213 and note that, if we retain terms to the next order in e beyond that part which
Jed to eq. 213, we find that v
2 s
00
+
V,
+ 2vs,. + v,s,
+ vs
0 b · (VR,) · b
=
0
If we multiply this equation by s,, the result can be written
(1v's/ +
V),
+
(vs/)
0
+ vs,'b · (VR,) · b =
0
,1
i ll. THE SECOND OR LONGITUDiNAL ADIABATIC INYA j
J j lf we integrate this equation with respect to 0 from zero to one, and recall that s(O,t) is presumably periodic in 8 with period one, we obtain a
J,
at o d8vs/
J
1 + dOb · (VR,) · bvs,' o
= 0 (217)
Let us in the above eqnation use s as the variable of integration and recognize that to the lowest significant order we may use eq. 213 to express s, in terms of E and V. The equation then reads on multiplication by m:
I ~ f ds {2m[E(1)V(s,a,/3,1]
J'''
+
T ds b · ('YR,) · b(2m[E(t)V(s,a,fJ,t)]} 1
1
2
= 0 (218)
327
Note that afat acting on the first integral above means a time derivative holding the line of force fixed. We shall now show that dsb · ('YR") · b is jusj the time rate of change of the element of arc length ds due to the velocity li.,.
See Figure 6.
~s+oc\S line.genero1ed ol 1+81
Fig. 6. Schematic diagram illustrating the calculation of the time rate of change of arc length along a line of force due to R.1.
'i;'
Let us consider a vector ds = dsb
In an infinitesimal time Jt the end of ds, as indicated in Figure 6, is carried a distance R,(R,t) bt by the guidingcenter motion. The tip of ds is carried into
The net change in ds is to lowest order
Jds = ds · (VR") Jt
328 . THE MOTION OF A CHARGED PARTICLE whence the square of the element of arc length is carried into
(ds + ods) 2 =
=
=
(ds) 2 + 2ds. (ods) + ...
(ds)
2
(ds)
2
+
2dsb · [dsb · (VR J.l iit]
+ · · ·
[1
+
2b · (VRJ.l· bot+ · · ·]
Thus ds + &Is= ds [1 + b · (VR 1_) · bot + · ·] and in the limit 01> 0, ods

.It
= b • (VR ) · b ds j_
Equation 218 is then to be interpreted as a time derivative of the integral following the guidingcenter motion, and
J = f ds(2m[E(r) V(s,a,{J,t)])
1
' '
(219)
·is an approximate constant of the motion. For a given value J and known potential V this expression is an implicit equation for E. The constant J is conventionally termed the second or longitudinal adiabatic invariant.
To recapitulate then, the motion along the line is determined by eq. 214 withE given by eq. 219. The motion perpendicular to the line is then given by
Ji.J.
(see eq. 28), associated with where we may replace mu 2 by 2(£V). To find the trajectory
RJ. requires only the solution of a secondorder system of ordinary differential equations.
It is interesting to note that, if the technique of this section is applied to the equation, ·
+ w(t)
2 x = 0 corresponding to
V =
2 x
2 and
(ciJ) ~ w' then it yields the wellknown, lowest order WKB results.
III. The Third or Flux Invariant
When the fields involved in
RJ. are changing sufficiently slowly, a notion that will be made more precise later, a further reduction in the description is possible. To demonstrate this, it is convenient to write equations for c(: and
[J, instead of dealing with li.. To this end v.:e view RJ. as a function of s,a,{J, and 1 and write li.
= SR, +ctR, + PRp + R, (31)
329
111. THE THIRD OR FLUX INVARIANT where the subscripts indicate partial derivatives. Moreover, by the chain rule for differentiation, if f denotes the unit dyadic,
VR
= f =
VsR, + VaR, + VpRp
(32) whence on taking the dot product on the lefthand side with b, one has b = (b · Vs)R, = R,
(33) since b · Vs = s, = I. If one takes the dot product of eq. 32 on the lefthand side with b x R, and b x R
1 ,
(34) b b x R, = V {JRp · b X
R,
• x R
1
= VaR, · b x Rp
(35)
The cross product of these two equations yields
 '\Ia x V{J(b · R, X R
1)
2 ~
(b X R
1) X
(b X R,)
=bR,. b
X Rp
But, since Va x V fJ
= B = Bb, one has b · R, x Rp = 1/B
Now the dot product ofeq. 31 with R, x R1
~ b X R, yields ciR,. b X
R, =
R . b X
R,  R, . b X
R,
(36) or on using eqs. 28, 35, and 36, ci/B = b x R 1 · (b/m!J.) x [VV + (q/c)(a, vp
{J,Va) + 2(£V)b,] + R< Va/B
Since, by the chain rule for differentiation, R, · Va =a,, R
1 ·
VV=
b · VV= R, · VV= V, Rp · V{l = pp = 1, R, · Va = ap =0, while b · Va=O, b . V {1 = O, and since b is a unit vector b · bs = 0, the above reduces to
= a.(R,  bb · R,) · (c/q)[V V
+
(q/c)(a, vpp,
Va)
+ 2(£V)b,]
=a,(c/q)[Rp · VV +
(qfc)a,R
1 ·
V{J (qfc){J,Rp · Va
+ 2(£V)Rp · b,  b · V V b · R,]
= (cfq)[Vp V, b · R, + 2(£V)b, · R
1]
But, on recognizing that Rps = (Rs)/J = bp, and b · b
11
= 0, this can be written
(c/q){V, + [2(£ V)/m]
112
([2m(E V)]'
12 b · Rp),)
)J j
330 THE MOTION OF A CHARGED PARTICJ
Finally, if we introduce the angle variable G in place of s, since d = ds = [2(£ V)fm] 1
1 2 r de,
(cfq)[Vp
+ {r1
[2m(EV)] 112 b · Rp)o]
If one integrates this expression over one period in 8,
1 1
J d8a = (cfq)
J dBVp
0 0
=
(cfq)r• f ds(2(EV)fm) t/
2 Vp (37)
The righthand side of eq. 37 can be related to the energy E, as de~ned implicitly by eq. 219, and considered to be a function of ~./3,1, and of course the constants of the motion J and fl.lfwe take the partial derivative ofeq. 219 with respect to /3, we find that since J is an independent parameter,
0 = fds{2(E V)/m)1
1
2 [Ep Vp] or on using eq. 215, f ds(2(E V)/m) t/
2 Vp =rEp
Thus, if we interpret
Jb de& as the time derivative of the average value of a associated with a particle over a period t, we can cast eq. 37 in the fonTI a= (cfq)Ep
In similar fashion we can show that fJ
= (cfq)E,
(38)
(39) where it is to be emphasized that a and {3 are the coordinates of the mean line of force on which the particle is gyrating and oscillating.
The equations of motion for o: and fJ are in Hamiltonian form with
E(a,/3,1) playing the role of a timedependent Hamiltonian. When E, = O,E is a constant of the motion, and the orbit in the a,{J phase plane is the c~rve
E
= canst. Suppose that this orbit is a closed curve, as shown schematically in Figure 7. Then the motion is periodic with period
T = (qfc) f d/3/E, = (qfc) f da/Ep
Suppose that
U;v<T>
a ln E/811 ~I c. L{t b~ !..1'•
t::
_L
~
IL~
\ f.l"~ ol.
01.

~
_L
OJf;:..
.~~lc'
.....t."l}~;t
~
[!. 5D
Ill. THE THIRD OR FLUX INVARIANT j3
_,.,..."/ E
= canst. a
Fig. 7. Schematic diagram of constant flux surface illustrating a particle trajectory therein.
We anticipate as in the former cases that the motion will be almost periodic we introduce an auxiliary variable x(l) and write a= a(x,t) f3 = f3(x,l) (311)
If we define w(l) = ;(, we can write the equations of motion as w~, +a,= (c/q)Ep
(312) . wfJ, + /3, = (cfq)E,
(3J3f{
,.•,•
To the lowest order we delete the ostensibly small term a
1 and then
[J, and note tha(;::·:
··
(cfq)E,
=
(cfq)[a,E, + fJ,Ep]
=
~, w/3, + /3/ wa,)
=0
Thus to this order
E = H(l)
(314)? where the constant of integration H(t) is as yet undermined. Let us choose.: w = lfT, where Tis defined by eq. 310 but with the integrals extended over . the closed curve E =H. This makes X an angle variable, and one can formally'· integrate the approximate equations of motion, eqs. 312 and 313 with a, and
/3, deleted, to obtain xT = (qfc) r da/Ep[a,f3(~,t),l]
= (qfc) ( d/3/
E,[~(/3,/),fJ,t] where f3(a,t) is determined from
E(~,/3,1)
= H(l), etc. '1
.,_
~ ~s:; it./) ,_ lc~ tP}~~J~ ~
('>'
~ y
I _£__
~
J fl.~ ~/1'1'
"' e 'i,_• t .,_
!
. J32 THE MOTION OF A CHARGED PARTICLE .
In order to determine H(t), we note that, without approximation of the equations of motion,
(cjq)E, = a,(cjq)Ea + J3,(cjq)Ep
= a,(w/3, + j3,)  J3,(wa, + a,)
= a,J3, J3,a,
= (a/3,), + (a/3,),
Thus, if we integrate this result with respect to
from zero to unity, we obtain and to the lowest significant order
(315) is an approxinmte constant of the motion, where the integral is extended o~er the closed curve E = H.
We shall now show that lj; is a magnetic ftux. To demonstrate this, we note that the Oux crossing any surface in x,y,z space is, on using Stokes theorem and B = (Va) x (V/3) = V x (o.Vj3),
I d
2
r · B =I d
2
r · '1 x (aVjJ)
=Idr·aVjJ
=I d{J a (316)
The line integral above is extended over any closed curve resulting from slicing the magnetic surface defined in x,y,z space by the equation E(a,[J,t) =
H(t), as shown schematically in Figure 8. particle path path of line Integral magnetic surface E=H
Fig. 8. Schematic diagram indicating particle path and line integral path in rnagrietic surface E =H.
REFERENCES
Jll
Clearly, lj; is independent of the choice of line as long as it is topological!y equivalent to that shown above. Equation 315 is then to be viewed as determining H(t) implicitly, given t/J. The approximate constant lj; is conventionally termed the third adiabatic invariant, or alternatively the flux invariant.
Acknowledgment
The author is indebted to Professor E. B. Hooper, Jr., and Peter Cato for their careful reading of the manuscript. The work was supported in part by Atomic Energy Commission contract AT 31 ~I ~3943.
References
1. H. Alfven, Cosmical Electrodynamics, Oxford University Press, New York, 1950.
2. M.D. Kruskat, J. Math. Phys., 3, 806 (1962).
3. T. G. Northrup, The Adiabatic Motion of Charged Particles, lnterscicnce, New York,
!963.
4. B. Lehnert, Dynamics a/Charged Particles, North Holland, Amsterdam, 1964, pp. 26 IT.
5. J. Berkowitz and C. Ga[dncr, Communs. Pure and Appl. Math., 12, 501 (1959).