8

DISTRIBUTED CHARGE SYSTEMS

Introduction. We have recently been studying solutions of Maxwell’s equations

—solutions in the complete absence of sources (Chapter 5) and solutions in
the presence of but a single point source (Chapters 6 & 7). But in many
physical problems and most technological applications one has interest in the
fields generated by (static or dynamic) populations of charged particles; i.e., by
spatially distributed sources.
One might suppose that such problems could be solved by application of
the principle of superposition . . . but the “application” is more easily talked
about than done, and it is not at all straightforward: it inspired much of the
mathematical invention for which the period – is remembered. And
there are (as always) unexpected physical complications. For example: the
presence of conductive materials gives rise to “induced charges,” which join the
unknowns of the problem.
We will look first to the electrostatic problem—to the description of the
description of the electrostatic potential set up by an arbitrarily constructed
blob of charge. Information of the sort we now seek would comprise our point of
departure if se sought (say) to construct an account of the Bohr orbits around
a structured nucleus, or (in gravitational terms) of the motion of a satellite
around the inhomogenous earth.
x) describe a t-independent
1. Multipole representation of a static source. Let ρ(x
(or “static”) charge distribution. The resulting electromagnetic field has no
magnetic component (B B = 0 ), and its t-independent electric component (see
again page 25) can be described
E (x ∇ϕ(x
x) = −∇ x) 
1
x) =
ϕ(x 1
x)
ρ(x d3 x (506)
4π |x
x − x|
422 Distributed charge systems

R x

Figure 153: We use x to describe the constituent elements of a

distributed charge, and x to describe the location of a typical field
point. The vector R (xx , x) ≡ x − x stretches from the former to
the latter, and has length
√ x , x) = |x
R(x x − x |. We proceed in the
assumption that r ≡ x · x > a, where

radius of a mental sphere large enough
a≡
to enclose the entire distributed charge


The integral derives from, and expresses, the principle of superposition—as
anticipated. But our goal now is to see what we can do to sharpen the very
general result described above. We want to learn to distinquish the relevant
features of (506) from the less relevant, so that by discarding the latter we can
simplify our computational life.
Let us suppose that the source, though distributed, is “localized” in the
x) ≡ 0 for x exterior to a sphere of sufficiently large but finite
sense that ρ(x
309
radius a, and let us agree that our ultimate objective—what we are presently
getting in position to do—is to describe the electrostatic potential at points
external to that sphere (see Figure 153). Writing

x , x) = |x
R(x x − x| = x − x)·· (x
(x x − x)

= r2 − 2rr cos ϑ + r2

309
This weak assumption serves merely to exclude “infinite line charges” and
similar (unphysical) abstractions.
Multipole representation of a static source 423
√ √
with r ≡ x · x and r ≡ x · x , we note that the dimensionless ratios x/r,
y/r, z/r are in every instance less than unity. It becomes therefore natural to
contemplate expanding 1/R(x x , x) in powers of those ratios. To that end . . . we
recall that according to Taylor’s theorem

∞

∂
f (x + x) = ex ∂x f (x) = 1 n (n)
n! x f (x)
n=0

In the multivariate case we expect therefore to have

x ∂ + y ∂y
∂
+ z ∂z∂
f (x + x, y + y, z + z) = e ∂x f (x, y, z)
  ∂
∂ ∂
= 1 + x ∂x + y ∂y + z ∂z
 ∂2 ∂2 ∂2
+ 12 x2 ∂x 2 + 2xy ∂x∂y + 2xz ∂x∂z

∂2 ∂2 2 ∂2
+ y 2 ∂y 2 + 2yz ∂y∂z + z ∂z 2 + · · · f (x, y, z)

x , x) gives
which when applied in particular to the x -dependence of 1/R(x

1 1 1 
= + 3 · xx + yy + zz
|x
x − x| r r
1 
+ 5 · 12 x2 (3x2 − r2 ) + 6xyxy + 6xzxz
r
+ y 2 (3y 2 − r2 ) + 6yzyz + z 2 (3z 2 − r2 ) + · · ·

In a fairly natural (and quite useful) condensed notation we have

  
x x
= r−1 + r−3  y · y 
z z
  2  
x 3x − r2 3xy 3xz x
−5 1  
+r 2 y · 3yx 3y 2 − r2 3yz  y  + · · ·
z 3zx 3zy 3z 2 − r2 z

Feeding this expansion back into (506) we obtain

x) =
ϕ(x 1
4π r−1 q + r−3p · x + r−5 21 x · Qx x + ···


1
= 4π r−1 q + r−2 p · x̂
x + r−3 21 x̂
x · Q x̂
x + ··· (508)
424 Distributed charge systems

where

q≡ x) d3 x
ρ(x (508.0)
≡ so-called “monopole moment scalar” or total charge
 
 x
p≡  y  ρ(xx) d3 x (508.1)
z
≡ so-called “dipole moment vector ”
 2 
 3x − r2 3xy 3xz
Q≡  3yx 3y 2 − r2 3yz  ρ(xx) d3 x (508.2)
3zx 3zy 3z − r
2 2

≡ so-called “ quadrupole moment (tensor or) matrix ”

In higher order we lose the advantages of matrix notation . . . might appear in

3rd order to have to write something like
 
1 −4 1
r W abc x̂a b c
x̂ x̂ with W abc ≡ x) d3 x
Mabc (x, y, z) ρ(x
4π 3!
a,b,c
  
a+b+c=3 complicated cubic

but will soon be in position to proceed in a more orderly manner. As will

emerge, it is the lowest-order terms that are of highest practical importance, so
(508) is in fact quite useful as it stands: it will be useful also as a benchmark
against which to test more general formulæ as they become available. Several
comments are now in order:
1. The objects q, p, Q, . . . are called “scalar,”“vector,”“tensor,”. . . in recognition
of how they respond to rotations of the Cartesian frame: they are, in short,
tensorial with respect to the rotation group O(3), as one could demonstrate
without difficulty.
2. q is the 0th moment of the charge distribution ρ(x
x), p is assembled from
the 1st moments, Q is assembled from the 2nd moments, etc. Not surprisingly,
x)
if one possessed the moments of all orders then one could reconstruct the ρ(x
which generated those moments.310
3. Q is (like the energy/momentum tensor S: see again page 215) symmetric
and traceless. These properties are, moreover, preserved under coordinate
310
Usually, not always. The program would fail if, for example (see again
page 416), the distribution were Lorentzian

x) ∼ 1
ρ(x
x2 + y 2 + z 2 + a2

But such a distribution cannot be enclosed within a sphere of finite radius.

Multipole representation of a static source 425

rotation. From symmetric tracelessness it follows that Q contains (not 9, as one

would otherwise expect, but) only 5 adjustable constants (degrees of freedom).
Symmetry alone assures that Q can always be rotated to diagonal form
 
Q1 0 0
Q −−−−−−−−−−−−−−−−−−→  0 Q2 0 
properly chosen rotation
0 0 Q3

and tracelessness requires that the eigenvalues sum to zero: Q1 + Q2 + Q3 = 0.

Figure 154: Oblate spheroidal distribution, symmetric about the

z-axis. Spinning bodies (stars, planets, atomic nuclei) commonly
possess this shape, at least in leading approximation.

x) is symmetric about the z-axis (see the

If, as is quite commonly the case, ρ(x
figure) then Q acquires the structure
 
− 12 Q 0 0
 0 − 12 Q 0
0 0 Q

In such specialized contexts it is common (among nuclear physicists and others)

to speak of “the quadrupole moment,” the reference being to Q.
4. What is the origin of the monopole/dipole/. . . multipole terminology? The
answer has little/nothing to do with electrostatics per se, much to do with the
meaning of nth derivative. Look, for example, a 1 -dimensional model of the
situation in hand: suppose it to be the case that

ϕ(x) = ρ(x)F (x − x) dx
426 Distributed charge systems

1
where F (•) is some prescribed differentiable function (not necessarily the x− x
encountered in (506)) and where x remains “small” throughout the range of
integration. We expect then to have

∞
 
ϕ(x) = (−)n n!
1
· ρ(x)xn dx · F (n) (x)
n=0   
nth moment

where F (0) (x), F (1) (x), F (2) (x), F (3) (x), . . . acquire meaning from the following
scheme:

Figure 155: Representation of the mechanism by which iteration

of
    
δ ξ − (x + 21 ") − δ x − (x − 21 ")
F (1) (x) = lim F (ξ) dξ

↓0 "

gives rise to successive derivatives of F (x). Notice that 2n spikes

contribute to the construction of F (n) (x). This is the source of the
“di/quadu/octo. . . 2n -tuple pole” terminology.

In several dimensions one encounters only this new circumstance: one can
displace a sign-reversed monopole in several directions to create a dipole, can
displace a sign-reversed dipole in several directions to create a quadrupole, etc.
Electrostatic potential of a dipole 427

x)
5. We are led thus to the principle that an arbitrary localized distribution ρ(x
can be represented as the superposition of
• an appropriately selected monopole +
• an appropriately selected dipole +
• an appropriately selected quadrupole + etc:

= monopole + dipole + quadrupole + · · ·

6. Looking back again to (508) we notice that at sufficiently remote field points
one can drop all but the monopole term (ρ(x x) looks like a point charge). At
less remote points one can drop all terms subsequent to the dipole term. High
order multipole terms depend upon such high powers of 1/r that they are of
quantitative importance only in the near zone.
Equation (508) carries us a long way toward our goal, as stated on page 422.
But there remains a good deal of meat to be gnawed from the bone.

2. Electrostatic potential of a dipole. Consider the two-charge configuration (no

net charge) shown in Figure 156. The associated electrostatic potential can be
described


1 1
1
x) = 4π
ϕ(x q √ −√ (509.1)
r2 − 2ra cos ϑ + a2 r2 + 2ra cos ϑ + a2
  a 2 − 12   a 2 − 12

1
= 4π (q/r) 1 − 2 a r cos ϑ + r − 1+2a r cos ϑ + r

1 2qa cos ϑ 5 cos 2ϑ − 1  a 2
= 4π 2
1+ r (509.2)
r 4

63 cos 4ϑ − 28 cos 2ϑ + 29  a 4
+
64 r + ···

This describes, as a power series in a/r, the potential of a physical dipole.

Proceeding now to the double limit

a ↓ 0 and q ↑ ∞ in such a way that p ≡ 2aq remains constant

we obtain
↓
1 p cos ϑ 1 p · x̂
x 1 p·x
= 4π = 4π = 4π (510)
r2 r 2 r3
Notice that the dipole potential ϕ would simply vanish if q were held constant
during the compression process a ↓ 0. Equipotentials derived from (509) and
(510) are shown in Figure 157.
428 Distributed charge systems

+q

a ϑ

−q

Figure 156: Notation used in the text to describe the field of a

physical dipole •–•. A “mathematical dipole” results in the idealized
limit a ↓ 0, q ↑ ∞ with p ≡ 2aq held constant.

Figure 157: Central cross section of the equipotentials of a physical

dipole (on the left) and of an idealized dipole (on the right).
Electrostatic potential of an eccentric monopole 429

x ϑ

Figure 158: Notation used in the text to describe the field of an

“eccentric monopole,” i.e., of an isolated charge (or charge element)
that is arbitrarily positioned with respect to the coordinate origin.
The length of x is r, the length of x is r.

3. Electrostatic potential of an eccentric monopole. In what might at first sight

appear to be a step backward, but will soon be revealed to be a long step
forward, we look now to the potential of the primitive system shown above;
i.e., to the Coulomb potential of an eccentrically-positioned charge. This we
do by systematic elaboration of methods borrowed from the preceding section.
Immediately (which is to say: by the Law of Cosines)

x) = 1
√ 1
ϕ(x 4π q (511)
r2 − 2rr cos ϑ + r2

which—in preparation for implementation of our plan, which is to proceed by

power series expansion—we will write

  1
4π q r ·
1 1

 r  2 : adapted to the case r < r

 1 − 2 r cos ϑ + rr
=

 q ·
1 1 1

 r  2 : adapted to the case r > r
 4π r 1 − 2 r cos ϑ + rr

Thus do we acquire interest in the objects Pn (w) that arise as coefficients from
the series
∞
√ 1
= Pn (w) tn (512.1)
1 − 2wt + t2 n=0
430 Distributed charge systems

Mathematica supplies

P0 (w) = 1 


P1 (w) = w 





P2 (w) = 12 (3w2 − 1) 


P3 (w) = 12 (5w3 − 3w) (512.2)


P4 (w) = 1 4
− 30w + 3)
2 

8 (35w 



P5 (w) = 1 5
− 70w3 + 15w) 

8 (63w 
.. 
.
These are precisely the Legendre polynomials, the properties of which were first
described () by A. M. Legendre ( –) and are summarized in every
mathematical handbook.311 Graphs of some low-order Legendre polynomials
are shown in Figure 159.
Returning with this information to (511) we have
 ∞
   r n

 4π
1 1
qr ·

 r Pn (cos ϑ) in the far zone
n=0
x) =
ϕ(x ∞
(513)

   r n

 1 1
q ·
 4π r r Pn (cos ϑ) in the near zone
n=0

in which connection it becomes pertinent to notice that (ask Mathematica)


P0 (cos ϑ) = 1 



P1 (cos ϑ) = cos ϑ 





1
P2 (cos ϑ) = 4 (3 cos 2ϑ + 1) 

1
P3 (cos ϑ) = 8 (5 cos 3ϑ + 3 cos ϑ) (512.3)


1
P4 (cos ϑ) = 64 (35 cos 4ϑ + 20 cos 2ϑ + 9) 




P5 (cos ϑ) = 128 (63 cos 5ϑ + 35 cos 3ϑ + 30 cos ϑ) 
1


.. 

.
Looking specifically/explicitly to the far zone we have


ϕ(x 1
x) = 4π r−1 q + r−2 qrP1 (cos ϑ) + r−3 qr2 P2 (cos ϑ) + · · · (514)

which must comprise the multipole expansion—correct to all orders—of an

eccentrically placed monopole. How does this result compare with what (508)
has to say in such a specialized situation? Setting ρ(x x −x
x) = qδ(x x) and working
311
See, for example, W. Magnus & F. Oberhettinger, Formulas & Theorems
for the Functions of Mathematical Physics (), pages 50–59; J. Spanier &
K. B. Oldham, An Atlas of Functions (), Chapter 21; M. Abramowitz &
Irene Stegun, Handbook of Mathematical Functions (), Chapter 22. For
discussion of how the principal properties of the Legendre polynomials are
established see pages 471– 475 in classical electrodynamics ().
Electrostatic potential of an eccentric monopole 431

0.5

-1 1
-0.5

-1
1

0.5

-1 1
-0.5

-1

Figure 159: Graphs of Legendre polynomials of low odd order

(above) and low even order (below). Order can in each case be
determined by counting the number of zero-crossings. The Pn (w)
are orthogonal in the sense
 +1
2
Pm (w)Pn (w) dw = 2m+1 δmn
−1

and provide a natural basis within the space of functions defined on

the interval − 1, +1 .

from (508), we find that


q≡ x − x) d3 x = q
qδ(x

= qP0 (cos ϑ) : monopole terms agree trivially

 
 x
p≡  y  qδ(x
x − x ) d3 x
z
 
x
= q  y  so p · x̂
x = q r cos ϑ by definition of ϑ
z
= q rP1 (cos ϑ) : dipole terms agree
432 Distributed charge systems

and finally that

 2 
 3x − r2 3xy 3xz
Q≡  3yx 3y 2 − r2 3yz  qδ(x
x − x ) d3 x
3zx 3zy 3z − r
2 2
 2 
3x − r 2
3xy 3xz
= q  3yx 3y 2 − r2 3yz 
3zx 3zy 3z − r2
2

⇓


1
2
x
x̂ · Q x
x̂ = q 3
2
x
(x · x
x̂ )2
− 1 2
2 r
= qr2 21 (cos2 ϑ − 1)
= qr2 P2 (cos ϑ)

So though (508) and (514) look quite different, they do in fact say exactly the
same thing. Which is gratifying, but . . .

Equation (514) says in its complicated way what we could say quite simply
if we were to reposition our coordinate system (place the origin at the solitary
charge), so is of relatively little interest in itself. It acquires profound interest,
however, when put to its intended use:

4. Representation of an arbitrary potential by superimposed spherical harmonics.

x) d3 x of our distributed
The idea is to apply (514) to each constituent element ρ(x
charge. To implement the idea we introduce spherical coordinates in the usual
way
   
sin θ cos φ sin θ cos φ
x = r  sin θ sin φ  , x = r  sin θ sin φ 
cos θ cos θ

where evidently θ signifies colatitude (North and South poles are coordinated
θ = 0 and θ = π, respectively). Then

x · x̂
cos ϑ = x̂ x = cos θ cos θ + sin θ sin θ cos(φ − φ)

and
d3 x = r2 sin θ drdθdφ

so (514) supplies

∞ 
  r n  
r Pn cos θ cos θ + sin θ sin θ cos(φ − φ)
1 –1
x) =
ϕ(x 4π r
n=0
· ρ(r, θ, φ)r2 sin θ drdθdφ (515)
Entry of spherical harmonics into potential theory 433

Thumbing through the mathematical handbooks, we discover the wonderful

identity312
 
Pn cos θ cos θ + sin θ sin θ cos(φ − φ) (516.1)
 (n − m)!
n
= Pn (cos θ)Pn (cos θ) + 2 Pnm (cos θ)Pnm (cos θ) cos m(φ − φ)
m=0
(n + m)!

Here
1  d m
Pnm (w) ≡ (−)m (1 − w2 ) 2 m dw Pn (w) : m = 0, 1, 2, . . . , n
 d n
Pn (w) ≡ (−)n 2n1n! dw (1 − w2 )n

defines the so-called associated Legendre functions, the first few of which are
displayed below:313

P0 (w) ≡ P00 (w) = 1 =1

P1 (w) ≡ P10 (w) = w = cos θ


P11 (w) = − 1 − w2 = − sin θ

P2 (w) ≡ P20 (w) = 12 (3w2 − 1) = 14 (3 cos 2θ + 1)


P21 (w) = −3w 1 − w2 = − 32 sin 2θ
P22 (w) = −3(w2 − 1) = − 32 (cos 2θ − 1)

P3 (w) ≡ P30 (w) = 12 (5w3 − 3w) = 18 (5 cos 3θ + 3 cos θ)


P31 (w) = − 32 (5w2 − 1) 1 − w2 = − 38 (5 sin 3θ + sin θ)
P32 (w) = 15w(1 − w2 ) = − 15
4 (cos 3θ − cos θ)

P33 (w) = −15(1 − w2 ) 1 − w2 = − 15
4 (sin 3θ − 3 sin θ)

I have written these out to demonstrate that, while Pnm (w) is a polynomial
only if m is even, the associated Legendre functions are in all cases simple
312
Magnus & Oberhettinger,311 page 55; P. Morse & H. Feshbach, Methods of
Theoretical Physics (), page 1274. Identities of the frequently-encountered
design 
f (x + y) = gn (x)gn (y)
n

are called “addition formulæ.”

313
Use Mathematica to reproduce/extend the list. The commands are

LegendreP[n,m,w] and LegendreP[n,m,Cos[θ]]//TrigReduce

434 Distributed charge systems

combinations of elementary functions—nothing to become nervous about. If

we now write
eim(φ−φ) + e−im(φ−φ)
cos m(φ − φ) =
2
and accept the convention314 that

Pnm (w) and Pn−m (w) are two names for the same thing

then (516.1) becomes

 
Pn cos θ cos θ + sin θ sin θ cos(φ − φ)

m=+n
= Cnm · Pnm (cos θ)e−imφ · Pnm (cos θ)e+imφ (516.2)
m=−n
(n − |m|)!
Cnm ≡
(n + |m|)!

in which the (θ, φ) -variables and (θ, φ) -variables have been fully disentangled,
placed in nearly identical “piles.” Further simplifications become possible when
one reflects upon the orthogonality properties of eimφ and Pnm (w). Familiarly
 2π
e−imφ e+imφ = 2π δmm
0

Less familiarly—but as the handbooks inform us, and as (even in the absence
of explicit proof) we are readily convinced by a little Mathematica -assisted
experimentation—
 +1
Pnm (w)Pnm (w) = 2 m
2n+1 Cn δnn : 0
m
lesser of n and n
−1

So we construct

2n+1 (n+|m|)! m
Ym
n (w, φ) ≡(−)m
4π (n−|m|)! Pn (w)e
imφ
: m = 0, ±1, ±2, . . . , ±n
↑—a convention, fairly standard to the literature, and honored by Mathematica

which are orthonormal in the sense

 2π +1
∗ m
[Ym
n (w, φ)] Yn (w, φ) dwdφ = δ
mm
δnn
0 −1

Or—more suitably for the matter at hand—

Ynm (θ, φ) ≡ Ym
n (cos θ, φ)

314
Beware! The designers of Mathematica adopted at this point an alternative
convention.
Entry of spherical harmonics into potential theory 435

which are precisely the celebrated spherical harmonics , orthonormal on the

surface of the sphere
 2π π
[Ynm (θ, φ)]∗ Ynm (θ, φ) sin θdθdφ = δ mm δnn
0 0

just as the functions Em (φ) ≡ √12π eimφ were seen above to be orthonormal on
the surface of the circle. The functions Ynm (θ, φ) are relatively more complicated
than the functions Em (φ) not so much because they have an extra argument
as because the surface of a sphere is a topologically more complicated place
than the surface of a circle (or—more aptly—than the surface of a torus).
Mathematica, upon the command SphericalHarmonicY[n,m,θ,φ], produces
the following explicit list of low-order spherical harmonics:

Y00 (θ, φ) = 4π
1


Y1−1 (θ, φ) = + 8π3 −iφ
e sin θ

Y10 (θ, φ) = 4π3
cos θ

Y1+1 (θ, φ) = − 8π3 +iφ
e sin θ


Y2−2 (θ, φ) = + 32π
15 −2iφ
e sin2 θ

Y2−1 (θ, φ) = + 8π
15 −iφ
e cos θ sin θ

Y20 (θ, φ) = + 16π5
(3 cos2 θ − 1)

Y2+1 (θ, φ) = − 8π
15 +iφ
e cos θ sin θ

Y2+2 (θ, φ) = + 32π
15 +2iφ
e sin2 θ

There are 2n + 1 = 1, 3, 5, . . . of the things of order n = 0, 1, 2, . . .

By this point (516.2) has assumed the form
 
Pn cos θ cos θ + sin θ sin θ cos(φ − φ)

m=+n
4π m ∗ m
= 2n+1 [Yn (θ, φ)] Yn (θ, φ) (516.3)
m=−n

which when introduced into (515) gives

∞
 
m=+n
Ynm (θ, φ)
x) =
ϕ(x 1 –1
4π r
4π
2n+1 Qm
n (517)
rn
n=0 m=−n 
where Qnm ≡ [Ynm (θ, φ)]∗ ρ(r, θ, φ)rn+2 sin θ drdθdφ
436 Distributed charge systems

defines the multipole moments of the charge distribution :

Q00
Q−1
1 Q01 Q+1
1
Q−2
2 Q−1
2 Q02 Q+1
2 Q+2
2
..
.
Q−n −1 0 +1 +n
n . . . . . . . . . . . . . Qn Qn Qn . . . . . . . . . . . . . Qn

To remove any element of the mystery from the situation let us look to
some of the illustrative specifics:

Q00 = [Y00 (θ, φ)]∗ ρ(r, θ, φ)r2 sin θ drdθdφ
 
1
= 4π ρ(r, θ, φ)r2 sin θ drdθdφ

1
= 4π q (518)00

Q01 = [Y10 (θ, φ)]∗ ρ(r, θ, φ)r3 sin θ drdθdφ
 
= 4π 3
r cos θ · ρ(r, θ, φ)r2 sin θ drdθdφ
 
= 4π 3
z · ρ(x
x) d3 x

3
= 4π p3 (518)01

Q−1
1 = [Y1−1 (θ, φ)]∗ ρ(r, θ, φ)r3 sin θ drdθdφ
 
= + 8π 3
r(cos φ − i sin φ)∗ sin θ · ρ(r, θ, φ)r2 sin θ drdθdφ
 
= + 8π 3
(x + iy) · ρ(xx) d3 x

= + 8π 3
(p1 + ip2 ) (518)−1
1

Q+1
1 = − 8π 3
(p1 − ip2 ) (518)+1
1


Q02 = [Y20 (θ, φ)]∗ ρ(r, θ, φ)r4 sin θ drdθdφ
 
5
= 16π (3z 2 − r2 ) · ρ(x
x) d3 x

5
= 16π Q33 (518)02
Entry of spherical harmonics into potential theory 437

Q−1
2 = [Y2−1 (θ, φ)]∗ ρ(r, θ, φ) r4 sin θ drdθdφ
 
= + 8π 15
r2 (cos φ + i sin φ) cos θ sin θ · ρ(r, θ, φ)r2 sin θ drdθdφ
 
= + 8π 15
(x + iy)z · ρ(xx) d3 x

= + 8π 15 1
3 (Q13 + iQ23 ) (518)−1
2

Q+1
2 = − 8π 15 1
3 (Q13 − iQ23 ) (518)+1
2

Q−2
2 = [Y2−2 (θ, φ)]∗ ρ(r, θ, φ) r4 sin θ drdθdφ
 
= + 32π 15
r2 (cos 2φ + i sin 2φ) sin2 θ · ρ(r, θ, φ)r2 sin θ drdθdφ
  
= cos2 φ − sin2 φ + 2i cos φ sin φ
 
= + 32π 15
(x2 − y 2 + 2ixy) · ρ(xx) d3 x

= + 32π 15 1
3 (Q11 − Q22 + 2iQ12 ) (518)−2
2

Q+2 3 (Q11 − Q22 − 2iQ12 ) (518)+2
15 1
2 = + 32π 2

Here the notations pa and Qab have been taken from (508) on page 424. The
point is that same physical information is folded (if in a different way) into the
designs of Qm1 , Q2 , . . . as was folded into the designs of p, Q, . . . : equations
m

(517) and (508) are saying the same thing, but in different ways.
Were we to pursue the mathematical side of this subject we would want
to establish that & how the spherical harmonics Ynm (θ, φ) spring spontaneously
into being when one undertakes to

solve ∇2 ϕ = 0 in spherical coordinates by separation of variables

A little Mathematica-assisted experimentation315 may serve to convince the

reader—even in the absence of the formal demonstration—that

∇2 rp Ynm (θ, φ) = 0 if and only if p = n or p = −(n + 1)

315
Enter the commands

<<Calculus`VectorAnalysis`

and SetCoordinates[Spherical[r, θ, φ]]

and then test

Laplacian[rp SphericalHarmonicY[n,m,θ, φ]]

with various values of m, n and p.

438 Distributed charge systems

Solutions of the first type blow up as r ↑ ∞: at (517) we find ϕ(x

x) described as a
linear combination of solutions of the second type. Looking to the mathematics
of the situation from a somewhat different angle . . .

 r n m
ϕ(r, θ, ϕ) = Amn a Yn (θ, φ)
m,n

describes a solution of Laplace’s equation,316 and so also does

 n m
ψ(r, θ, ϕ) = a
r Amn a r Yn (θ, φ)
m,n

To say the same thing another way: if f (x, y, z) is a solution of Laplace’s

equation ∇2 f = 0 then so also is
 a2 a2 a2 
F (x, y, z) ≡ a
r f r2 x, r2 y, r2 z
Transformations of the form
x −−−−−−−−−−−−→ x = a
2

inversion r2 x
are called “inversions in the sphere of radius a” by geometers (they send interior
points to exterior points and visa versa, subject to the rule rr = a2 ), and are
self-inversive in the sense
x −−−−−−−−−−−−→ a r2
2

r 2 x = a2 x = x
inversion

Transformations of the form

x) −−−−−−−−−−−−→ f (x
f (x x) ≡ a a2
r f ( r2 x) (519)
Kelvin inversion

acquire their name from the fact that it was William Thompson (Lord Kelvin)
who first noticed () that they send “harmonic functions” (solutions of
Laplace’s equation) into harmonic functions: they are readily seen to be
self-inversive in the sense that
2
(Kelvin inversion) = identity transformation

Rotation of the charge distribution (equivalently: counter rotation of the

Cartesian frame) would clearly result in an altered set of coefficients Qmn that
refer to an altered set of spherical harmonics:

∞

m=+n m
1 –1 4π m Yn (θ, φ)
x) =
ϕ(x 4π r 2n+1 Qn
n=0 m=−n
rn
| rotation
↓
∞

m=+n
Ynm (θ, φ)
= 4π1 –1
r 4π
2n+1 Qm
n
n=0 m=−n
rn

316
Here and below: a is a constant “length” of arbitrary value, introduced for
a dimensional reason.
A geophysical application 439

Were we to pursue the theory of spherical harmonics we would certainly want

to explore the details of the now-fairly-evident fact that the harmonics of given
order n are rotationally induced to fold among themselves
 +n     +n 
Yn (θ, φ) Yn (θ, φ)
 ..    .. 
    
 0 .    0 . 
 Yn (θ, φ)  =  (2n + 1) × (2n + 1) matrix   
     Yn (θ, φ) 
 ..     .. 
 .    . 
−n −n
Yn (θ, φ) Yn (θ, φ)

in a why that provides a (2n+1)-dimensional representation of the 3-dimensional

rotation group O(3). When those details are approached algebraically (instead
off function-theoretically) it is found to make sense to speak also of cases

n = 12 , 32 , 52 , . . .

that give rise to even-dimensional matrix representations of O(3), and that those
have indispensible applications to the quantum theory of fractional spin. While
electrostatics served historically to inspire the initial development of the theory
of spherical harmonics, and does exploit some of the more superficial elements
of that theory, it is the quantum theory of angular momentum (equivalently: the
representation theory of O(3)) that first motivated people to explore (in order
to exploit) the riches hidden in the deeper nooks and crannies of the theory of
spherical harmonics. And it is because the theory is most naturally developed
in connection with its quantum mechanical applications317 that I am content
not to pursue it further here.318

5. A geophysical application. Though initially formulated in the language

of electrostatics, our results pertain also—obviously and quite usefully—to
gravitostatics . . . for reasons having to do with the structural similarity of the
statements
e
= electrostatic potential of a point charge e
4πr
GM
− = gravitostatic potential of a point mass M
r

Evidently the gravitational potential exterior to a sphere319 containing a blob

x) of matter—the earth is the “blob” of greatest interest to geophysicists—can
ρ(x
317
See, for example, David Griffiths, Introduction to Quantum Mechanics
(), Chapter 4 or J. Powell & B. Crasemann, Quantum Mechanics (),
Chapter 7.
318
In  I had not so much self-control: the missing details are sketched on
pages 486–510 of classical electrodynamics.
319
A mental sphere, of radius a, commonly identified with the maximal radius
of the geosphere (∼ 6.378 × 103 km).
440 Distributed charge systems

be described

x) = −G ρ(x 1
V (x x) d3 x
|x
x − x|

∞

m=+n
Ynm (θ, φ)
= −Gr–1 4π
2n+1 Qm n
rn
n=0 m=−n 
Qnm ≡ [Ynm (θ, φ)]∗ ρ(r, θ, φ)rn+2 sin θ drdθdφ
 
= −GM r1 1 + r1 P · x̂ x · Gx̂
x + r12 12 x̂ x + ···

where

M≡ x) d3 x = monopole moment = total mass
ρ(x

dipole moment vector
P ≡ 1
M
x) d3 x =
ρ(x
M
= center of mass coordinates

quadrupole moment matrix
G≡ 1
M 3xi xj − r2 δ ij ρ(x
x) d3 x =
M
Note that the dipole term drops away if one places the origin at the center of
mass.320 Dominant interest shifts therefore to the quadrupole term, which
“MacCullagh’s formula”
 
x) = −GM r1 1 − A − C
V (x (3 sin2
ψ − 1) + · · ·
2M r2
↑—signifies latitude

serves to relate to the geometrical parameters (A and C) that describe the

idealized oblate sphereoidal figure of the gravitating body (see again Figure 154).
Higher moments provide information about
• irregularities in the figure of the body
• inhomogeneities of the mass distribution.
Notice that (see again the formula that serves at the top of the page to define
the coefficients Qm n ) the higher moments depend most strongly upon details
near the surface of the body, and are of quantitative significance only in the
near zone: far away the body “looks like a monopole”:
↓
= −GM r1 : ra
For the earth the Qmn have been measured through at least n = 8, and in
the post-Sputnik era satellites have been used to fill in an “island” of higher
320
That would be a natural thing to do, but a conventional thing to do
(something one might elect not to do) . . . and should not be confused with the
physical fact that—because Nature provides no “negative mass”—gravitational
dipoles do not exist.
A geophysical application 441

Figure 160: Polar orbit of a satellite in polar orbit. Resolving the

spherical harmonics into their real/imaginary parts

Yn±m (θ, φ) = Cnm (θ, φ) ± iSnm (θ, φ)

we observe that Sn0 (θ, φ) and Cn0 (θ, φ) are φ-independent: they vanish
on circles parallel to the equator, thus partitioning the surface of the
sphere into “zones,” so are called “zonal harmonics.” At the other
extreme, the nodes of

Cnn (θ, φ) ∼ cos nφ sinn θ and Snn (θ, φ) ∼ sin nφ sinn θ

partition the sphere into sectors (bounded by great arcs of constant

longitude); such functions are called “sectoral harmonics,” while
spherical harmonics with 0 < m < n are called “tesseral harmonics.”
Some sectors have been painted on the earth, and rotate with the
earth (because they are taken here to refer to a property of the earth).

(m, n)-values—this by the pretty method that I now sketch. The period T of a
satellite in circular orbit can, in leading approximation, be described

a3  r  32
T = 2π
GM a
which in the case of the earth becomes
 r  32
= 84.5 a minutes
442 Distributed charge systems

The satellite will be in resonance with the sectoral harmonics Ynn of the earth’s
gravitational field if T = Tn , where Tn is the time it takes for the rotating earth
to replace one of the sectors of Ynn by the next. The sidereal day is 1436.07
minutes long, so

 89.75 minutes : n = 16


1436.07 minutes  95.74 minutes : n = 15
Tn = = 102.58 minutes : n = 14
n 


 110.47 minutes : n = 13
119.67 minutes : n = 12
3
and to achieve synchrony in those cases (solve 84.5x 2 = Tn for x) we must set
 16
 1.0410a : resonance with Y16 mode


 1.0868a : resonance with 15
Y15 mode
orbital radius = 1.1380a 14
: resonance with Y14 mode



 1.1956a : resonance with 13
Y13 mode
12
1.2611a : resonance with Y12 mode

If n
16 the satellite burns up in the atmosphere (or its orbit becomes

subterranean!), while if n  12 then r becomes so large that the (1/rn ) -factor
makes the effects of resonance unobservably small. The case n = 15 seems to
be nearly optimal, and indeed: scientists active in the field321 have been able
by this means to estimate the values of Q15 15 15 15
15 , Q17 , Q19 and Q21 . Since high
moments probe progressively more superficial properties of ρ(x x), one might
hope from such orbital data to extract information about the earth’s crust and
crust-mantle interface. The technique extends in principle to planetary bodies
other than the earth. And microphysical analogs do come to mind: an atom
with nuclear charge Ze has orbital radii given typically by (see again page 392)

2
R=
mZe2
which gets smaller when m is increased . One therefore expects that the
properties of µ-mesonic atoms might provide information about the surface
properties of complex nuclei.

6. Harmonic polynomials & Maxwell’s theory of poles. While the theory of

spherical harmonics has much to do with the representation of rotations in
3 -space, it has—contrary to the impression conveyed by some of the preceding
material—only incidentally to do with spherical coordinates. Important aspects
of the theory are, in fact, brought most simply/naturally into view by the
adoption of a Cartesian perspective . . . as I undertake now to demonstrate:
321
See R. D. Eberst, “Earth satellites and the gravitational potential” and
D. G. King-Hele & H. Heller, “Equations for the 15th -order harmonics in the
geopotential,” Nature Physical Science 235, 130 (1972). Also A. E. Roy, Orbital
Motion §10.4 () and H. F. R. Schöyer & K. F. Walker, Rocket Propulsion
and Space Flight Dynamics §18.6 ().
Harmonic polynomials & Maxwell’s theory of poles 443

x) ≡ a · x and notice
Introduce the (rotationally invariant!) monomial T (x
that, by quick calculation,

∇2 T n = n(n − 1)T n−2 a · a

Dismissing as trivial the cases n = 0 and n = 1, we conclude that the nth powers
x) will be harmonic iff a is null. But a · a = 0 entails that a be complex:
of T (x
a = b + icc with b2 − c2 = 0 and b · c = 0. If a · a = 0 is formulated

 
a3 = −(a21 + a22 ) = i (a1 + ia2 )(a1 − ia2 )

then it becomes fairly natural to introduce complex parameters


u≡ a1 + ia2

v ≡ a1 − ia2

in terms of which we can write

2

a1 = 1
2 (u + v2 ) 

a2 = 1
− v2 )
2i (u
2 (520)


a3 = i uv

which provide a (u, v)-parameterized description of the set of all null 3-vectors a.
In this notation
 2 n
2n (u + v )x + i (u − v )y + 2iuvz
2 2 2
T n (x
x) = 1 1
 n
2n u (x − iy) + 2iuvz + v (x + iy)
1 2 2
=
!
polynomial of degree n in variables (x, y, z)
=
polynomial of degree 2n in parameters (u, v)

To emphasize the latter point of view we write

m=+n
= 1
2n un−m v n+m Hnm (x
x)
m=−n

This, since harmonic for all values of u and v, entails that the polynomials
Hnm (x
x) are individually harmonic:

∇2 Hnm = 0
444 Distributed charge systems

Arguing from
 2

T ·Tn = 1
2n+1 u (x − iy) + 2iuvz + v 2 (x + iy) un−m v n+m Hnm
m


= 1
2n+1 u (n+1)−(m−1) (n+1)+(m−1)
v (x − iy)Hnm
m
+ u(n+1)−m v (n+1)+m (2iz)Hnm

+ u(n+1)−(m+1) v (n+1)+(m+1) (x + iy)Hm
m

= T (n+1)

1
= 2n+1 u(n+1)−m v (n+1)+m Hn+1
m

we obtain a relation
m
Hn+1 = (x − iy)Hnm+1 + 2izHnm + (x + iy)Hnm−1

from which—sprouting from the “seed”

!
1 : m=0
x) ≡
H0m (x
0 : m = ±1, ±2, . . .

—the harmonic polynomials Hnm (x

x) can be computed recursively: thus

H00 = 1

H1−1 = x − iy
H10 = 2iz
H1+1 = x + iy

H2−2 = (x − iy)2
H2−1 = 4i(x − iy)z
H20 = 2x2 + 2y 2 − 4z 2 = 2(r2 − 3z 2 )
H2+1 = 4i(x + iy)z
H2+2 = (x + iy)2
..
.

The harmonic polynomials are regular at the origin but blow up at ∞.

Kelvin inversion (519) permits us, however, to construct from them a population
of (non-polynomial) functions

x) ≡ r1 Hnm ( r12 x)
Jnm (x

which are assuredly also harmonic and, though singular at the origin, are
regular at ∞. Reading from the preceding list are led thus to the Kelvin
Harmonic polynomials & Maxwell’s theory of poles 445

transform of that list:

J00 = r−1 = + r1

J1−1 = r−3 · (x − iy) = −1(∂x − i∂y ) r1

J10 = r−3 · 2iz = −2(i∂z ) r1
J1+1 = r−3 · (x + iy) = −1(∂x + i∂y ) r1

J2−2 = r−5 · (x − iy)2 = + 13 · 1(∂x − i∂y )2 r1

J2−1 = r−5 · 4(x − iy)iz = + 13 · 4(∂x − i∂y )(i∂z ) r1
J20 = r−5 · 2(r2 − 3z 2 ) = + 13 · 6(i∂z )2 r1
J2+1 = r−5 · 4(x + iy)iz = + 13 · 4(∂x + i∂y )(i∂z ) r1
J2+2 = r−5 · (x + iy)2 = + 13 · 1(∂x + i∂y )2 r1
.. ..
. .

That the harmonic functions Jnm (xx) can be described by the highly patterned
formulæ on the right was discovered by Maxwell, who in the general case would
have us write

1  
Jn±m = (−)n 2n
(∂x ± i∂y )m (i∂z )n−m r1
1·3·5···(2n−1) n−m

where now m = 0, 1, 2, . . . , n.
We are by now not surprised to discover that if we at this point use

x ± iy = r sin θ · e±iφ and z = r cos θ

to pass from Cartesian to spherical coordinates, then the functions Jnm turn
out to differ only numerical factors from the functions r−(n+1) Ynm (θ, φ). The
detailed result can be expressed in several ways:

Yn±m (θ, φ) = (−)n (i)n+m 2n1n! 2n+1
4π (n−m)!(n+m)! · r
n+1 ±m
Jn (x x)
 1 n+1 ±m 
Yn (θ, φ) = (−)n 2n+1 4π (n−m)!(n+m)! (∂x ± i∂y ) (∂z )
1 m n−m 1
r r
  

From the latter we conclude that

≡ D±m
n

is a differential operator natural to the theory of spherical harmonics.

446 Distributed charge systems

Which brings us back again to very nearly our point of departure. We

established at (10.2) on page 12 that the function r1 = (x2 + y 2 + z 2 )− 2 is
1

harmonic except at the origin, where it blows up, but in a very interesting way:

∇2 r1 = −4πδ(x
x)

Application of D±m
n gives
↓
 
∇ 2 1 n+1
r Yn±m (θ, φ) = −4πD±m x)
n δ(x (521)

which shows that a similar remark pertains to the functions Ynm (θ, φ)/rn+1 ,
except that these possess singularities of higher order , the latter being described
by fancy derivatives of δ-functions. When, as at (517), we display ϕ(x x) as a
weighted superposition of the functions that appear on the left, we are in effect
x) is equivalent to an identically weighted superposition of the
claiming that ρ(x
singular functions (“distributions”) that appear on the right side of (521):
∞

m=+n
Ynm (θ, φ)
x) =
ϕ(x 1
2n+1 Qmn
rn+1
n=0 ↑ m=−n ↑
| |
—strength of D±m n δ(xx) singularity

—number of nth -order singularities

And we remarked already on page 426 the sense in which structured singularities
can be interpreted to refer to constellations of “poles.” We have arrived thus
at the essence of Maxwell’s “theory of poles.”
It is hard to let go of this beautiful subject. I allow myself the luxury of
one parting shot: It is an immediate implication of (520) that

a∗· a = 12 (u∗ u + v ∗ v)

The expression on the right is invariant under linear transformations

" # " # " #
u u u
−→ =U
v v v

provided U is unitary (inverse = conjugate transpose). Such transformations,

by (520), induce linear transformations

a −→ a = Ra
a

which, since norm-preserving, must describe 3-dimensional rotations. From this

germ of an idea one gains direct access to the rich subject matter to which I
allude at the end of §4.322
322
Some of the details are developed in my “Algebraic theory of spherical
harmonics” (Seminar Notes ). An excellent source is A. Erdélyi et al ,
Higher Transcendental Functions (), Volume 2, Chapter 11.
Harmonic polynomials & Maxwell’s theory of poles 447

The material described above—fruit of the genius mainly of Maxwell and

his friends, and of the generation that preceded them—takes Laplace’s equation

∇2 ϕ = 0

as its point of departure, but analogous methods are important in a variety of

other contexts. Look, for example, to the heat 1-dimensional equation

(∂x2 − ∂t )ϕ(x, t) = 0

2
It is clear that exz + tz describes a z-parameterized family of solutions. Taylor
expansion in z
2
exz + tz = 1 + xz + 12 (x2 + 2t)z 2
+ 16 (x3 + 6xt)z 3
+ 1
24 (x
4
+ 12x2 t + 12t2 )z 4 + · · ·
∞

≡ 1 n
vn (x, t) n! z
n=0

gives rise to a population of “heat polynomials,” analogous to the harmonic

polynomials encountered on page 444.323 And corresponding to the Kelvin
transformation (519) one has the (nearly inversive) Appell transformation ()

e−x /4t
2

ϕ(x, t) −−−−−−−−−−−−−−−−→ ψ(x, t) ≡ √ · ϕ( xt , − 1t )

Appell transformation 4πt

where the exponential factor is itself a solution—the so-called “fundamental

solution”—of the heat equation. We have seen that the Kelvin transformation
contributes importantly to the theory of harmonic functions. Just so the Appell
transformation: I have shown elsewhere that it is an object central to the theory
of the conformal group, and that in a quantum mechanical application it serves
as the bridge that links the standard formalism to the Feynman formalism.324

323
See D. V. Widder, The Heat Equation (), pages 8–14.
324
“Appell, Galilean & Conformal Transformations in Classical/Quantum
Free Particle Dynamics” (research notes ).