Vous êtes sur la page 1sur 10

A curious and useful theorem in two-dimensional electrostatics

J. D. Jackson

Citation: Am. J. Phys. 67, 107 (1999); doi: 10.1119/1.19203


View online: http://dx.doi.org/10.1119/1.19203
View Table of Contents: http://ajp.aapt.org/resource/1/AJPIAS/v67/i2
Published by the American Association of Physics Teachers

Additional information on Am. J. Phys.


Journal Homepage: http://ajp.aapt.org/
Journal Information: http://ajp.aapt.org/about/about_the_journal
Top downloads: http://ajp.aapt.org/most_downloaded
Information for Authors: http://ajp.dickinson.edu/Contributors/contGenInfo.html

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
A curious and useful theorem in two-dimensional electrostatics
J. D. Jackson
University of California, Berkeley, California 94720-7300
~Received 5 October 1998; accepted 21 October 1998!
In 1956 metrologists Thompson and Lampard established a new theorem in two-dimensional
electrostatics that gives calculable capacitances for practical systems possessing a certain symmetry.
The theorem is well known among those concerned with electrical standards, but not so well known
beyond. This paper brings the theorem to a wider audience with a pedagogical discussion of it, first
by treating a simple example by standard methods of electrostatics, then proving the Riemann
mapping theorem on which the theorem is based. The ThompsonLampard theorem and its
generalization are then proved and its use in metrology described. Along the way, some curious
results about series and products are exhibited. 1999 American Association of Physics Teachers.

I. INTRODUCTION cylinder closely by another continuous cylinder that is also


grounded. Then in the limit of infinitesimal gaps, the only
Most of the significant theorems in electricity and magne- charge on the exterior of our chosen cylinder will be on the
tism are 100200 years old, from a time when giants in segment ab ~and on the inner side of the surrounding cylin-
mathematical physics were laying down the subject with pre- der in its immediate neighborhood!. See Sec. VI for a prac-
cision and elegance. It is surprising then to find that a curious tical realization.
new theorem in two-dimensional electrostatics only emerged The amazing aspects of the theorem are that the cross
in the second half of the 20th century, in 1956.1 The theorem capacitance is independent of ~1! the shape of the cylindrical
had its origins in a search for precision in electrical measure- surface ~apart from its symmetry!, and ~2! the location of the
ments, and while well known in metrological circles,2 seems plane bd perpendicular to the plane of symmetry, provided it
lesser known elsewhere. I confess to my ignorance until the intersects the cylinder. If we think of the plane bd at various
theorem was brought to my attention by the editor of this positions along the figure, we can imagine that there is some
journal in late 1997. His awareness of the theorem came compensation at workwhen the segment ab is small, the
from the paper in this journal on SI units by Zimmerman.3 segment over which the charge density cd is summed is large
There is a thorough if modest literature on the theorem and vice versa. But the curiosity is that the compensation is
and its extensions.46 The purpose of the present paper is to complete! This property and the resulting insensitivity to er-
give a pedagogical discussion of the theorem and to illustrate rors are what makes the theorem of practical importance to
its power with simple-seeming examples. metrologists.
What is this new theorem of Thompson and Lampard? In The plan of our discussion is to begin with some general
its original form,1 it reads ~adapted slightly!: aspects of cross capacitances in terms of Green functions in
Let the closed curve ~see Fig. 1! be the cross section two or three dimensions, then to examine in a straightfor-
of a conducting cylindrical shell, which cross section ward manner the example of a cylinder of square cross sec-
has one axis @plane# of symmetry ac but is otherwise tion ~with its two different symmetry axes!. The square and
arbitrary. Further, let this shell be divided into four parts its generalization, the rectangle, illustrate the lengthy route
by two planes at right angles, the line of intersection of one must take to the simple end result, if the Thompson
the planes being parallel to the generator and one of the
planes containing the symmetry axis ac. Then the direct
capacitance, per unit length of the cylinder, between
opposing parts of the shell ~for example, ab to cd!, due
to the field inside ~or outside! the shell, is a constant:
e0
C 05 ln 252.818 376 pF/m. ~1!
p
The electrostatic situation envisioned in the theorem is that
the surface in Fig. 1 between a and b is held at fixed poten-
tial V while the rest of the cylinder is at zero potential. The
direct capacitance between the opposing parts of the shell is
the ratio of the negative of the total charge per unit length on
the surface between c and d to the voltage difference V. Here
~as in later references cited above! we use the term cross Fig. 1. Cross section of a cylindrical conducting shell with plane of sym-
capacitance to denote the theorems direct capacitance metry, but otherwise arbitrary. The cylinder axis is perpendicular to the
page. The symmetry plane intersects the page in the line ~a,c!. A perpen-
per unit length, with the per unit length understood. We dicular plane, whose line of intersection with the first is parallel to the
consider here only the charge on the inner surface of the generator of the cylinder, intersects the page in the line ~b,d!. Infinitesimal
segment cd. In general, there will be charge on the outside, gaps occur at a, b, c, and d, so that the four electrodes can be placed at
too. In practice, this can be minimized by surrounding the various potentials.

107 Am. J. Phys. 67 ~2!, February 1999 1999 American Association of Physics Teachers 107

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
Lampard theorem is not at ones disposal. We follow these Since the Green function is symmetric in x and x8 , C 12
examples with a physicists proof of the Riemann mapping 5C 21 , the cross capacitance found if S 2 is held at fixed
theorem and then a derivation of the theorem in two ways, potential while the rest of S is at zero and the total charge on
one by solution of the two-dimensional potential problem S 1 is determined. This symmetry can be viewed as an illus-
with a circular boundary via a series expansion and the other tration of Greens reciprocation theorem. That is the reason
by Poissons integral solution. The latter permits easy access the dashed lines along the surfaces in Fig. 1 do not distin-
to a generalization of the theorem to unsymmetric situations. guish which segment is at fixed potential and which contains
Apart from the brief description in Sec. VI, we do not the charge of interest. Note also that in Fig. 1 with its axis of
consider the practicalities of construction of precision ca- symmetry the cross capacitance of ~a,b! and ~c,d! is the same
pacitors or the errors coming from lack of exact symmetry. as the cross capacitance of ~b,c! and ~d,a!.
For these, the reader may consult the literature.24,19 In the two-dimensional situation, the relevant Dirichlet
Green function features in a proof of the Riemann mapping
II. FORMAL SOLUTION FOR A CROSS theorem, which is at the heart of the ThompsonLampard
CAPACITANCE IN THREE OR TWO DIMENSIONS theoremsee Sec. IV below.
The electrostatic boundary value problem for the potential
within a charge-free volume enclosed by a surface S, of III. CYLINDER OF SQUARE CROSS SECTION
which a segment S 1 is held at constant potential F5V and As a first example of calculation of cross capacitance, we
the rest at zero potential ~as in Fig. 2! is a standard Dirichlet take a cylinder of unit square cross section. As shown in Fig.
problem. The solution is7 3, there are two types of symmetry for a square. In Fig. 3~a!,

F ~ x! 52
1
4p
E S
da 8 F ~ x8 !
] G ~ x,x8 !
]n8
the symmetry axis is chosen as the vertical line at x51/2,
with the orthogonal plane put at y5a. In Fig. 3~b!, the
symmetry axis is the diagonal line y5x, and the orthogonal,

52
4
V
p
E S1
da 8
] G ~ x,x8 !
]n8
, ~2!
y511a2x.
A. Square with symmetry axis parallel to sides
where the integrations are over the indicated surfaces, da 8 is We first consider the solution of the two-dimensional
an area element, n8 is an outwardly directed normal at da 8 , Laplace equation in the interior of the square, with the pe-
and G(x,x8 ) is the Dirichlet Green function for the problem rimeter of the square held at zero potential except for the
~G vanishes on the surface S and satisfies 2 G524 p d (x segment indicated at upper left in Fig. 3~a! by the dashed
2x8 ) in the interior of S!. lines ~x50, a,y,1! and ~0,x,1/2, y51!, which is held
The inwardly directed electric field at the surface S is at unit potential. We eventually want the charge on the
] F/ ] n; the surface charge density is s 5 e 0 ] F/ ] n. The total complementary segment of the boundary ~1/2,x,1, y50!
charge on the surface S 2 is given by the integral, and ~x51, 0,y,a!, but we first determine the potential

E E ]F F(x,y).
Q 25 da s 5 e 0 da . ~3! A series expansion for the appropriate Green function is
S2 S2 ]n not difficult to write down. We begin with the basic building
The cross capacitance C 12 is the ratio of the negative of the blocks, sin(npx), sinh(npy), and cosh(npy). In order to make
charge Q 2 to the potential V on S 1 . Thus we have the result, G(x,y:x 8 ,y 8 ) vanish on the boundary, but have the required
delta-function discontinuity, we use a completeness relation
C 125
e0
4p
E E
S2
da
S1
da 8
] 2 G ~ x,x8 !
]n ]n8
. ~4!
in x and x 8 , together with the correct combination of func-
tions in y, to obtain8
`
1
G ~ x,y;x 8 ,y 8 ! 58 (
n51 n sinh~ n p !
sin~ n p x ! sin~ n p x 8 !

3sinh~ n p y , ! sinh@ n p ~ 12y . !# , ~5!

Fig. 2. An arbitrary closed conducting surface S encloses a charge-free Fig. 3. Examples of a cylinder of square cross section as special cases of
volume V. The surface segment S 1 of S is held at fixed potential while the Fig. 1. The dashed lines indicate the two regions of interest, one held at
rest of S is held at zero potential. The cross capacitance C 12 is defined to be fixed potential and the other the region whose total charge ~per unit length!
the negative of the total charge on the noncontiguous surface segment S 2 of defines the cross capacitance. In ~a! the symmetry axis is chosen as the line
S divided by the fixed potential on S 1 . x51/2. In ~b!, the diagonal y5x is the symmetry axis.

108 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 108

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
where y , (y . ) is the smaller ~larger! of y and y 8 . We now b n 5 ~ cosh@ n p ~ 12a !# 2cos~ n p /2!!
construct the solution for the potential according to ~2!. We
need 2 ] G/ ] n 8 along the line x 8 50, a,y,1: 3 ~ cos~ n p /2! 2cos~ n p ! cosh~ n p a !! . ~12!

2
]G
]n8
5
]G
]x8 F G x 8 50
The summation in ~11! with the formidable coefficient
~12! is, according to the theorem, equal to ln A2, indepen-
dent of a! To demonstrate that independence is not trivial,
` although it does not involve magic. To impress upon you the
1
58 p (
n51 sinh~ n p !
sin~ n p x ! sinh~ n p y , ! power of the theorem, I give you an explicit demonstration.
First of all, we examine the asymptotic limit of b n for large
n:
3sinh@ n p ~ 12y . !# . ~6!
We also need 2 ] G/ ] n 8 along the line 0,x 8 ,1/2, y51: ~ 21 ! n21 n p
b n e

F G
4
]G ]G
2 52 so that
]n8 ]y8 y 8 51
` bn ~ 21 ! n21
1 .
58 p (
n51 sinh~ n p !
sin~ n p x ! sin~ n p x 8 ! n sinh~ n p ! 2n
Recall of the expansion of the logarithm ln(11z) shows that
3sinh~ n p y ! . ~7! the asymptotic form of the series, when extended over all n,
With ~6! and ~7! inserted into ~2!, the potential becomes sums to ln A2. This suggests the following maneuver: Add
`
and subtract the right multiple of sinh(np) to b n to give
1
F ~ x,y ! 52 (
n51 sinh~ n p !
sin~ n p x !
explicitly the desired multiple of ln 2, plus a correction term
that we hope we can show is identically zero. We thus write,
by adding zero to ~11!,
H
3 sinh~ n p y ! E0
1/2
dx 8 sin~ n p x 8 !
C5
e0
p
ln 21DC, ~13!

1 E 1

a
dy 8 sinh~ n p y , ! sinh@ n p ~ 12y . !# . J where the extra bit ~which is a rapidly convergent series
because the leading asymptotic behavior has been removed!
~8! is
`
Because we are interested in the potential and its derivatives 2e0 cn
only for y,a, we can put y , 5y and y . 5y 8 in ~8!. The DC5
p (
n51 n sinh~ n p !
~14!
integrals are then elementary; the result for the potential in
the lower region is with

H J
`
2 @ cosh@ n p ~ 12a !# 2cos~ n p /2!# ~ 21 ! n
F ~ x,y ! 5
p n51
( n sinh~ n p !
c n 5b n 1
2
sinh~ n p ! . ~15!

3sin~ n p x ! sinh~ n p y ! . ~9! Now our task is to show, first, that DC is independent of a,
and second, that it vanishes! Using the fact that cos(np/2)
To find the total charge on the complementary segment of
50 for n odd and (21) n/2 for n even, so that
the boundary, we need the normal electric field at ~y50,
cos(np)cos(np/2)5cos(np/2), and some identities among
1/2,x,1! and ~x51, 0, y,a!. The charge density at y
the sum and products of hyperbolic functions, c n can be
50 is s 1 52 e 0 @ ] F/ ] y # y50 and the charge density at x
expressed as the sum of two terms, the first dependent upon
51 is s 2 51 e 0 @ ] F/ ] x # x51 . If we denote the coefficient in
a and the second independent of it:
the curly brackets in ~9! by B n , we find the corresponding
charges to be c n 5c ~n1 ! 1c ~n2 ! ,

E
`
1 2e0 where
Q 15
1/2
dx s 1 5
p ( B n @ cos~ n p ! 2cos~ n p /2!# ,
n51 ~ 21 ! n
~10! c ~n1 ! 5 $ 4 cos~ n p /2! cosh~ n p /2!
E
` 2
a 2e0
Q 25
0
dx s 2 5
p (
n51
B n cos~ n p !@ cosh~ n p a ! 21 # .
3cosh@ n p ~ 122a ! /2# 2cosh@ n p ~ 122a !# % ~16!
The negative sum of the two charges is the cross capacitance and
for the complementary segments shown by the dashed lines
~ 21 ! n
in Fig. 3~a!: c ~n2 ! 5 $ sinh~ n p ! 2cosh~ n p ! 2 @ 11 ~ 21 ! n # % .
` 2
2e0 bn ~17!
C5
p ( n sinh~ n p !
, ~11!
n51
Consider now c (1)
n . The first term in the curly brackets van-
where ishes for odd n, and if we set n52m, cos(np/2)(21) n

109 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 109

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
5(21) m . This means that the part of DC that depends upon can be solved using product functions, sin(npx)sinh(npy),
a involves the sum, and the other, sinh@np(12x)#sin(npy). In each case the coef-
` ficients are simple Fourier series integrals,
c ~n1 !
(
n51 n sinh~ n p ! A n5
2
sinh~ n p !
E j2
sin~ n pj 8 ! d j 8 ~21!

H
j1
`
~ 21 ! m 4 cosh~ m p ! cosh@ m p ~ 122a !#
5 (
m51 2m 2 sinh~ 2m p !
with j 1 50, j 2 51/2 for the first situation and j 1 5a, j 2 51
in the second. When these two solutions are added and the

2
cosh@ m p ~ 122a !#
sinh~ m p !
. J ~18!
total charge on the complementary segment evaluated, the
result for the cross capacitance is of the form of ~11!, but
with ~12! replaced by
But since sinh(2mp)52 sinh(mp)cosh(mp), the first term is b 8n 5 ~ 12cos~ n p /2!!~~ 21 ! n cosh~ n p a ! 2cos~ n p /2!!
identically equal to the second, term by term. The contribu-
n to DC vanishes. We have therefore shown that
tion of c (1) 1 ~ cos~ n p a ! 2 ~ 21 ! n !~ cos~ n p a ! 2cosh~ n p /2!! .
DC is independent of a. ~22!
Next we tackle the remaining sum involving c (2)
n . Because Alas, in comparison to ~12! this expression does not seem an
(2)
of the last factor in c n , there are two sums, one over all n improvement. Its asymptotic form for large n still has depen-
and the other over just even n. The two can be combined to dence on a in both exponential and trigonometric ways.
give There seems to be no obvious ln 2 lurking in the woodwork.

F G
` `
c ~n2 ! 1 ~ 21 ! n21 e 2n p 1 Yet for 0,a,1, the series is rapidly convergentit must
( n sinh~ n p !
5
2 ( n sinh~ n p !
2
n sinh~ 2n p !
. give ( e 0 / p )ln 2 if the ThompsonLampard theorem is true!
n51 n51 The series ~11! with coefficients ~22! is not the first series
~19! thrown onto my Macintosh in the course of preparing this
If we denote a n 5e 2n p / @ 2n sinh(np)# and b n 51/@ 2n paper to verify the truth of the theorem numerically. I leave
3sinh(2np)#, then some simple algebra shows that a n 2 b n it to the inclined readers to check it out for themselves. @I
52 a 2n for any n. The series in ~19! is have not worked hard here to verify the necessary analyti-
cally, as I did above in showing DC50.#
a 12 b 12 a 22 b 21 a 32 b 32 a 42 b 4
1 a 52 b 52 a 62 b 61 a 7 . C. Square with diagonal symmetry axis
For clarity we group the odd n terms together, By now the reader should be persuaded that direct verifi-
cation of the ThompsonLampard theorem in simple-
~ a 12 b 1 !2 a 22 b 21~ a 32 b 3 !2 a 42 b 4 seeming situations is not always easy. Before leaving the
1 ~ a 5 2 b 5 ! 2 a 6 2 b 6 1( a 7 . square, however, we turn to the symmetric figure of Fig.
3~b!, for which the diagonal is the symmetry axis. With the
But a 1 2 b 1 52 a 2 . Hence the series becomes potential different from zero only on the top line (y51), we
a 22 b 21~ a 32 b 3 !2 a 42 b 41~ a 52 b 5 ! can solve the Laplace equation either by a simple series so-
lution with the products, sin(npx)sinh(npy), or with the
2 a 6 2 b 6 1( a 7 . Green function ~5!. The result is the same; we find the series
Now a 2 2 b 2 52 a 4 and the series becomes ~11!, but with the coefficients,

~ a 3 2 b 3 ! 1 a 4 2 b 4 1 ~ a 5 2 b 5 ! 2 a 6 2 b 6 1( a 7 . b 9n 5 ~ cos~ n p a ! 2 ~ 21 ! n !~ 12 ~ 21 ! n cosh~ n p a !! . ~23!

We use the relation, a n 2 b n 52 a 2n , repeatedly to change In contrast to b n given by ~12!, b 9n has both cos(npa) and
successively the signs of all the even n as so that the rela- cosh(npa) present. The process of showing the independence
tion then can be used on the even n terms as well as the odd. of the sum on a and then being a multiple of ln 2 does not
The result is that we keep pushing the first term in the series appear easy! It is perhaps amusing that the series for the
to larger and larger n. Since the coefficients vanish exponen- cross capacitance for Fig. 3~b! can be rewritten as

H(
tially with n, this rearrangement of the series leads in the
limit to zero for the sum in ~19!. We have thus proved that
e0 8 sin2 ~ n p a/2! sinh2 ~ n p a/2!
C5
p n even n sinh~ n p !
DC50 ~20!
for all a on the interval 0,a,1. The cross capacitance in
~13! reduces to the expected result, ~1!.
1 (
n odd
8 cos2 ~ n p a/2! cosh2 ~ n p a/2!
n sinh~ n p !
. J ~24!

The trigonometric and hyperbolic dependencies on a seem


B. An alternative solution inextricably tangled in ~24!, but, according to the theorem
The reader is justified in thinking there might be a simpler and numerical computation, the two sums add to ln 2, inde-
way to find the cross capacitance. What about linear super- pendent of a!
position? The situation in Fig. 3~a! can be thought of as the The special case of a50 has been discussed by Lampard.4
linear superposition of one boundary value problem with unit It describes the cross capacitance between the opposite sides
potential along the top ~y51, 0,x,1/2! and another with of a square. Utilizing the expansion of 1/sinh(np), the sur-
unit potential along the left side ~x50, a,y,1!. The first viving sum in ~24! can be written as

110 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 110

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
8e0 1 C ~ r ! 1C ~ 1/r ! 4p
C5
p (
n odd n sinh~ n p ! 2C 0
511 d 2
ln 2

5
8e0
p (
n odd
e 2n p
n
`

(
k50
2e 22kn p ~ a50 ! . ~25! 3 ( n
n odd
S 11cosh2 ~ n p !
sinh3 ~ n p !
2
cosh~ n p !
n p sinh2 ~ n p ! D
With an interchange of the order of summations and com- 1O ~ d 3 ! . ~29!
parison of the sum over n with the expansion of ln@(11z)/ Here C 0 is the magic result ~1!. A rapidly convergent expan-
(12z)#, we find sion for the series in powers of exp@2(2k11)p# can be writ-

F G
` ten down, but it is not too illuminating.12 Numerically, the
8e0 11e 2 ~ 2k11 ! p e0
C5
p (
k50
ln
12e 2 ~ 2k11 ! p 5
p
ln M, ~26! expansion is
C ~ r ! 1C ~ 1/r !
where 5113.454 727d 2 23.455d 3 1O ~ d 4 ! .

F G
`
2C 0
11e 2 ~ 2k11 ! p 8 ~30!
M5 )
k50 12e 2 ~ 2k11 ! p
. ~27!
The coefficients of the powers of d are not particularly small,
Could M52? Yes, it could! It turns out that ~27! is a special but the average is relatively insensitive to small departures
case of the ratio of two Jacobi theta functions9you have from r51, even though the individual values differ markedly
heard of them, I am sure.10 For those of us who are Jacobi- from C 0 . For example, for r50.8, we find C(r)/C 0
theta-function-challenged, it is sufficient to have the author- 51.888 62, C(1/r)/C 0 50.455 04, while the average is
ity of a formula in Gradshteyn and Ryzik11 to convince us 1.1706.
that the right-hand side of ~27! is a complicated way of writ- Having shown ~with difficulty! that the theorem seems
ing 2. For more depth, expertise in elliptic functions is re- true for the special geometry of the square, we turn to its
quired. Actually, once we have proved the Thompson proof in general. For that we need the Riemann mapping
Lampard theorem, M52 is proved without any recourse to theorem.
theta functions.
I have spent considerable time trying to show that ~24! is
equal to the canonical result, independent of a and have IV. RIEMANN MAPPING THEOREM
failed. Its proof ~for me at least! depends on the validity of
the theorem. Because analytic functions of a complex variable z5x
1iy satisfy the CauchyRiemann equations, the real and
imaginary parts of such functions separately satisfy the
D. Cross capacitance of sides of rectangular cylinder Laplace equation in two dimensions. Complex analysis
therefore is a useful tool in two-dimensional electrostatics.
Study of the two cross capacitances between opposite Furthermore, many complicated boundary value problems
sides of a rectangular cylinder allows us to comment briefly may be simplified by means of conformal transformations.
on issues of imprecision in construction of a capacitance Consider the one-to-one mapping w5 f (z), with f (z) ana-
standard. Lampard discusses this situation in detail.4 Let the lytic and f 8 (z)0. Analytic functions of z are mapped into
ratio of the sides of the rectangle be r. Let one side of unit
analytic functions of w. If two continuous curves in the z
length be held at fixed potential and the charge on the oppo-
plane intersect at z5z 0 , the angle between the curves is
site side be computed. Call the resulting cross capacitance
maintained for the corresponding curves in the w plane, even
C(r). Then let a side of length r be held at fixed potential;
though they are rotated by the angle arg f 8(z0) and are locally
the cross capacitance between it and its opposite side is
C(1/r). It is easy to see from the previous discussion that the magnified by u f 8 (z 0 ) u . Such transformations are called con-
formal transformations. Two-dimensional boundary value
only change in ~24! is to replace sinh(np) by sinh(npr) or
problems with various surfaces ~lines! or parts of surfaces
sinh(np/r): held at different potentials may be transformed into simpler
8e0 1 geometries in the w plane, amenable to solution. One well-
C~ r !5
p (
n odd n sinh~ n p r !
~28! known example is the SchwarzChristoffel transformation,
which maps the interior of a simple closed polygon in the z
and similarly for C(1/r). The dependence on r has the ex- plane into the upper half w plane.
pected consequence. For r.1, the sides of unit length are At the heart of the ThompsonLampard theorem is the
farther apart; the charge density induced on one side in re- Riemann mapping theorem.13,14 Paraphrasing Copson13 and
sponse to the voltage on the other is less than for r51 and so referring to Fig. 4~a! and ~b!, we state the theorem as fol-
is the cross capacitance. lows:
The issue of the effects of small departures from square If S is the simply connected domain in the z plane
cross section ~with its exactly known answer! can be ad- bounded by a simple closed curve C, there exists a
dressed by making a Taylor series expansion for C(r) unique analytic function f (z), regular in S, such that
around r51. Let d 5r21. The Taylor series expansions of w5 f (z) maps S conformally on u w u ,1 ~the unit circle!
both C(r) and C(1/r) will be of the form, A1B d 1D d 2 and also transforms the point z5z 0 within C into the
1 but the linear term in d will vanish in the average. We origin and a given direction at z5z 0 into the positive
find direction of the real axis.

111 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 111

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
Fig. 4. Conformal mapping from the z plane to the w plane. ~a! Arbitrary
closed curve C encloses a simply connected region S. Letters indicate arbi-
trary points along C. ~b! Unit circle with Greek letters indicating the w-plane
images of the four boundary points in the z plane.

Fig. 5. The unit circle that is the mapping of the contour of Fig. 1. The
The two parts of Fig. 4 give visual representation to the points labelled a, b, g, and d are the images of a, b, c, and d in Fig. 1.
Because of the axis of symmetry in Fig. 1, the points b and d are symmetri-
mapping.
cally located with respect to the line ~a,g!. The angle j depends upon the
We now give a physicists proof of the theorem, fol- positions of b and d in Fig. 1 and the specific conformal transformation. The
lowing Copson. The proof appeals because it is based on our cross capacitance C 1 is defined by the dashed arcs and is independent of j.
knowledge of electrostatics. First, assume that f (z) exists as
a unique analytic function, regular in S. Since z5z 0 maps to
w50, so that f (z 0 )50, it must be that f (z)/(z2z 0 ) is regu- finding of the function f (z) for any specific situation.
lar and nonzero in S. We may therefore write Finding f (z) may not be easy. For example, the square is
f ~ z ! 5 ~ z2z 0 ! e c ~ z ! , ~31! mapped into the circle by means of elliptic functions.17
Small wonder that our series solutions involved Jacobis
where c (z) is regular in S. Since u f (z) u 51 on the boundary elliptic theta functions!
C, c (z) must have a real part that satisfies
lnu z2z 0 u 1Re c ~ z ! 50 on C. ~32! V. PROOF OF THE THOMPSONLAMPARD
But both the real and imaginary parts of a regular analytic THEOREM AND ITS GENERALIZATION
function satisfy the two-dimensional Laplace equation. Also, With the Riemann mapping theorem under our belts, we
we recognize that lnuz2z0u is, up to a constant, the potential proceed to the easier problem of proving the Thompson
of a line source, l521/2, at z5z 0 . The real function, Lampard theorem for a cylinder of circular cross section. We
V ~ z,z 0 ! 5lnu z2z 0 u 1Re c ~ z ! , ~33! first consider the original theorem for a cross section with an
axis of symmetry, as in Fig. 1, and do it by an elementary
is equal to 2G(z,z 0 )/2, where G(z,z 0 ) is the two- series solution for the potential. We then consider the two
dimensional Dirichlet Green function for the region S cross capacitances for the more general situation of Fig. 4
bounded by the contour C. Alternatively, V is the electro- and establish the theorems generalization, using the Poisson
static potential ~in Gaussian units! of an axial line charge l integral solution for the Laplace equation in a unit circle.
521/2 located at (x 0 ,y 0 ) within a grounded conducting
right cylindrical shell C. The function Re c(z) is the solution A. Basic new theorem by series solution
of the Laplace equation ~within S! caused by the ~image!
sources external to S that make the Green function vanish on Any cross section with an axis of symmetry, such as
the boundary curve C.15 shown in Fig. 1, can be conformally mapped into the unit
We now appeal to the existence and uniqueness of the circle. The points ~a,b,c,d! on its perimeter curve C are
Green function or the line-charge potential.16 The real func- mapped into the points indicated by the points ~a,b,g,d! in
tion Re c(z) is therefore determined in principle. The Fig. 5. The axis of symmetry ~a,c! in Fig. 1 evidently maps
CauchyRiemann conditions enable us to determine Im c(z), into a diameter of the circle. The phase of the conformal
apart from an additive constant. This constant can be chosen transformation f (z) can be chosen to make the diameter
to make a given direction, say, the dashed line (z 0 ,a) in Fig. ~a,g! horizontal. The point b in Fig. 1 maps to the image
4~a!, map into the direction ~0,a!, the positive x axis, in Fig. point b somewhere on the upper semicircle. By symmetry
4~b!. The regular function c (z) determines the conformal the point d in Fig. 1 maps to the image point d. The confor-
mal transformation determines the value of the angles 6j
mapping, according to ~31!. This completes the physicists
that specify the locations of symmetric images a and d. All
proof of the existence of a unique conformal mapping that
that matters here is that the angles are symmetric on either
takes the interior of any simply connected region with
side of the x axis. Segments of the boundary C in Fig. 1 that
boundary curve C into the interior of the unit circle.
are held at fixed potentials are mapped into segments of the
Two comments:
unit circle at the same potentials. Cross capacitances and
~1! Further mappings can take the interior of the circle into other observables found easily for the circular boundary ap-
the upper half plane, but we use the circle in Sec. V to ply equally to the symmetric boundary. @Numerical values of
prove the ThompsonLampard theorem. some quantities will, of course, depend on the exact form of
~2! Although the Riemann mapping theorem establishes the the conformal transformation, but the cross capacitances do
existence of a unique mapping, it does not assist in the not.#

112 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 112

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
Consider the cross capacitance if the arc ~a,b! in Fig. 5 B. The ThompsonLampard theorem via Poissons
with subtended angle j is held at unit potential while the rest integral
of the unit circle is held at zero. The basic solution for the
potential F~r,f! inside the circle is an expansion in r n times The series solution ~34! and ~35! to the Dirichlet problem
sines and cosines of n f : inside the unit circle can be combined and summed formally
to yield Poissons integral solution18 for the potential,
`

F~ r,f !5 ( r n @ a n cos~ n f ! 1b n sin~ n f !# .


n51
~34! F~ r,f !5
1
2p
E 2p

0
F ~ 1,f 8 !
12 r 2
11 r 22 r cos~ f 8 2 f !
2 d f 8.
~42!
The coefficients are
Consider now the more general geometry of Fig. 4~b!, with
a n5
1
p
E 0
2p
F ~ 1,f ! cos~ n f ! d f ,
the arc segment ~a,b! held at unit potential while the rest of
the circle is at zero potential. We seek the cross capacitance
of the arc segment ~g,d!, namely, the negative of the total
~35!
E
1 2p charge on this arc. The surface charge density is
b n5 F ~ 1,f ! sin~ n f ! d f . e 0 ] F/ ]r u r 51 . The result for the cross capacitance is
p 0

Explicitly, we have C5
e0
2p
E fE
g
d
d
b

a
df8
1
12cos~ f 8 2 f !
, ~43!

1 1 where we now have specified the ends of the arcs by the


a n5 sin~ n j ! , b n5 @ 12cos~ n j !# . ~36!
np np corresponding angles.
Now
The negative of the total charge on the arc ~g,d! is propor-
tional to the integral of ( ] F/ ]r u r 51 ) from f 5 p to f
52 p 2 j . We thus have
E dx
1
12cos x
52cot~ x/2! ,

2Q
5
e 0 n51
`

( E 2p2j

p
n @ a n cos~ n f ! 1b n sin~ n f !# d f . E dx cot~ x/2! 52 ln@ sin~ x/2!# .

~37! Therefore the cross capacitance associated with the arcs


~a,b! and ~g,d! becomes

F S D S D
G
Combining the elementary integrals with the coefficients, we
find d2b
g2a
sin sin
e0 2 2

S D S D
`
2Q 1 1 C 1 5 ln . ~44!
e0
5
p (
n51 n
$ sin2 ~ n j ! 1 @ 12cos~ n j !# p
sin
d2a
sin
g2b
2 2
3 @~ 21 ! n21 1cos~ n j !# % . ~38! This expression is the general result in closed form for non-
contiguous segments with arbitrary angles.
The term in the series independent of j, @ (21) /n # , when n21
For the symmetric situation already considered, we have
summed, gives ln 2. We thus find the cross capacitance to be the angles in Fig. 5, namely, a 50, b 5 j , g 5 p , and d
52 j . The argument of the logarithm then becomes

F G
e0

S D S D
C 15 @ ln 21D # , ~39!
p d2b g2a
sin sin
2 2 sin~ 2 j ! sin~ p /2!

S D S D
where 5
d2a g2b sin~ 2 j /2! sin~ p /22 j /2!
` sin sin
1 2 2
D5 ( n
$ sin2 ~ n j ! 1 @ 12 ~ 21 ! n21 # cos~ n j !
n51 sin~ j !
5 52, ~45!
2cos ~ n j ! % .
2
~40! sin~ j /2! cos~ j /2!

An alternative form for D is independent of the angle j. The magical ln 2 emerges in ~44!,
as it must.
`
2 1
D5 (
n even n
cos~ n j ! 2 (
n51 n
cos~ 2n j ! . ~41!
C. Generalization of the theorem
Putting n52m in the first sum converts it into the second.
Therefore D50, independent of j; the result of the theorem There are two cross capacitances in Fig. 4~a! and ~b!. We
is demonstrated explicitly. have labelled C 1 the one involving the arcs ~a,b! and ~g,d!
We note in passing that the other cross capacitance C 2 , of Fig. 4~b!. We label as C 2 the one involving the arcs ~b,g!
involving the arcs ~b,g! and ~d,a!, is equal by symmetry to and ~d,a!. The result for C 2 can be read off ~44! by suitable
C 1 5C 0 . relabelling of the angles ~a b , b g , g d , d a !:

113 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 113

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
e0
C 2 5 ln
p
sin
a2g
2
a2b
2
Fsin

sin
d2b
sin
2
d2g
2
S D S D
S D S D G . ~46!

Now consider the following sum, created from ~44! and ~46!:
e 2 p C 1 / e 0 1e 2 p C 2 e 0
1

S D S D
5
g2a d2b
sin sin
2 2

F S D S D S D S DG
3 sin
g2b
2
sin
d2a
2
1sin
b2a
2
sin
d2g
2
.
Fig. 6. ~a! Ideal cylindrical boundary with cross section made up of four
quarter circles. ~b! Cross section of the NIST precision cross capacitor that
approximates to high accuracy the ideal of ~a!. The four cylindrical elec-
~47! trodes are 6.35 cm in diameter with clearances of 0.36 cm between elec-
trodes. The movable central electrode ~shield rod! is 2.72 cm in diameter
Straightforward trigonometry leads to the numerator being and only extends part way down the length of the other four. The grounded
equal to the denominator, with the result that the two adja- outer shell is 18.16 cm in inner diameter. The electrode at the upper right is
cent cross capacitances satisfy the general identity, at a fixed, nonzero potential and all the others are grounded. The change in
cross capacitance DC, upper right and lower left, is measured when the
e 2 p C 1 / e 0 1e 2 p C 2 e 0 51. ~48! shield rod is displaced by Dz: u DC/Dz u 5( e 0 ln 2)/ p .

This result holds for the arbitary set of ~two-dimensional!


surfaces of Fig. 4~a! that are conformally mapped into the
arcs of Fig. 4~b!. There is no requirement of axial symmetry the cross capacitance.19 According to the theorem, a change
for the initial set of electrodes. The general result ~48! was
Dz in the position of the shield rod produces a change in the
first published by van der Pauw,5 although special cases were
given by Lampard4 ~and the general result in an unpublished cross capacitance of DC52Dz( e 0 / p )ln 2.
report!. Of course, when the symmetry of Fig. 1 obtains, one The basic idea of the measurement is clear, but practical
implementation of the theorem involves may subtleties of
finds C 1 5C 2 and ~48! reduces to the answer of the original
design and sophisticated procedures. The interested reader
theorem.
may consult Refs. 2, 19, and the papers cited therein.

VI. PRACTICAL USE OF THE THEOREM

The origins of the theorem were in metrology, with prac- VII. CONCLUDING REMARKS
tical goals of precision in defining the absolute farad and
The curious ThompsonLampard theorem of electrostat-
ohm. As pointed out by Zimmerman,3 the National Institute
ics, far from being merely a curiosity, is of practical impor-
of Standards and Technology ~NIST! has implemented the
tance. It provides a calculable capacitance for geometries
concept of the theorem to design a precision capacitor, as
that are relatively easy to fabricate into real devices. Basic
have those responsible for metrology in other countries. The
standards of the farad and the ohm are more accurate, thanks
favored four-part cylindrical boundary is the interior inter-
to its use.
section of four circles of radius a inside a square of sides 4a, For the mathematical physicist, the theorem elicits wonder
as shown in Fig. 6~a!. The advantages of this geometry are and amusement. Unbelievable results such as
many. In particular, the manner in which the electrodes ap-
proach each other in cusplike fashion means that the errors 8 sin2 ~ n p a/2! sinh2 ~ n p a/2!
resulting from the presence of the necessary small gaps are (
n even n sinh~ n p !
extremely small. The practical realization of this boundary in
the NIST capacitor19 is shown in Fig. 6~b!. 8 cos2 ~ n p a/2! cosh2 ~ n p a/2!
The four long precision cylinders define the boundary. 1 (
n odd n sinh~ n p !
5ln 2,
With end effects neglected ~and the shield rod absent!, the
cross capacitance for cylinders of length L is just C(L) valid for any a (0,a,1), and its special case (a50),
5LC 0 . End fringing effects are present, of course, but the
8
measurement technique circumvents both these effects and
the need to make an absolute measurement of the length L.
(
n odd n sinh~ n p !
5ln 2,
The grounded shield rod is inserted on the axis a distance z
are causes for wonder, as is the equivalent product,
into the configuration. Its diameter is such as to give a close

F G
`
enough fit to prevent any significant charge accumulation on 11e 2 ~ 2k11 ! p 8
the cross electrode over the distance z. With the shield )
k50 12e 2 ~ 2k11 ! p
52.
rod in place, the cross capacitance can be written C(z,L)
5(L eff2z)C0 , where L eff takes into account the end effects. The physicist, if not the mathematician, may view these re-
The NIST apparatus is such that small changes in z can be lations as true curiosities, most easily proved by means of the
created and measured with great accuracy, as can changes in very practical ThompsonLampard theorem of electrostatics.

114 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 114

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission
ACKNOWLEDGMENTS thin dielectric films on their electrodes, Proc. IEEE 107C, 112119
~1960!.
I thank Editor Romer for bringing the Thompson
7
J. D. Jackson, Classical Electrodynamics ~Wiley, New York, 1998!, 3rd
Lampard theorem to my attention. If you enjoyed this paper ed., p. 39.
8
See, for example, Ref. 7, p. 89, Problem 2.15.
I should get the credit; otherwise Romer should get the 9
E. T. Whittaker and G. N. Watson, A Course in Modern Analysis ~Cam-
blame. He is the one who twisted my arm to write it. @A joke, bridge U. P., Cambridge, 1952!, 4th ed., reprinted, Chap. XXI.
in case anyone was in doubt.# I also thank Michael Berry for 10
It turns out that string theorists make use of Jacobis elliptic and theta
pointing me to the Jacobi theta function identity ~27!. At the functions. Consult your local string person for details.
time, I had not absorbed that it was all in Lampard.4 11
I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products
~Academic, New York, 1965!, 4th ed., p. 924, formula 8.197.4.
12
My expansion of the sum is S52e 2 p @ (121/p )14(321/p )e 22 p 16(5
1
A. M. Thompson and D. G. Lampard, A New Theorem in Electrostatics 21/p )e 24 p 1 # . The third term here disagrees with Lampards Eq.
and its Application to Calculable Standards of Capacitance, Nature ~Lon- ~43!Ref. 4. He has a coefficient 210 instead of 16.
don! 177, 888 ~1956!. 13
E. T. Copson, Theory of Functions of a Complex Variable ~Oxford U.P.,
2
B. W. Petley, The Fundamental Physical Constants and the Frontier of Oxford, 1935!, reprinted 1955, pp. 185187.
Measurement ~Hilger, Bristol, 1985!, pp. 142145. 14
C. E. Pearson, Handbook of Applied Mathematics ~Van Nostrand, New
3
N. M. Zimmerman, A primer on electrical units in the Syste`me Interna- York, 1983!, pp. 261263.
tional, Am. J. Phys. 66, 324331 ~1998!. 15
See, for example, Ref. 7, pp. 3840.
4 16
D. G. Lampard, A new theorem in electrostatics and its application to See Ref. 7, pp. 37, 38.
calculable standards of capacitance, Proc. IEEE 104C, 271280 ~1957!. 17
H. Kober, Dictionary of Conformal Representations ~Dover, New York,
5
L. J. van der Pauw, A method of measuring specific resistivity and Hall 1957!, 2nd ed., p. 182.
effect of discs of arbitrary shape, Philips Res. Rep. 13, 19 ~1958!. 18
See Ref. 7, p. 89, Problem 2.12.
6 19
D. G. Lampard and R. D. Cutkosky, Some results on the cross- R. D. Cutkosky, New NBS measurements of the absolute farad and
capacitances per unit length of cylindrical three-terminal capacitors with ohm, IEEE Trans Instrum. Meas. IM-23, 305309 ~1974!.

LESSONS FROM THE SSC


It has been said that experience is the name everyone gives to his or her mistakes. Experience
now tells us that big projects, whether in physics or biology, must have the support of the research
community, be managed soundly, and have some clearly articulated value to society. Perhaps
those lessons may be some small compensation for the money and effort that went into the SSC
and may help in the formulation of a more successful strategy when the next big project is
proposed.

David F. Voss and Daniel E. Koshland, Jr., The Lessons of the Super Collider, Science 262, 1799 ~1993!.

115 Am. J. Phys., Vol. 67, No. 2, February 1999 J. D. Jackson 115

Downloaded 14 Apr 2013 to 147.226.7.162. Redistribution subject to AAPT license or copyright; see http://ajp.aapt.org/authors/copyright_permission

Vous aimerez peut-être aussi