Research Article

Received 30 June 2011 Published online in Wiley Online Library

(wileyonlinelibrary.com) DOI: 10.1002/mma.1595

MOS subject classification: 35J05; 35J08; 35J15

On the Neumann function and the method of

images in spherical and ellipsoidal geometry
George Dassiosa and Johan C.-E. Stenb * †
Communicated by H. Ammari
The invention of an image system for a boundary value problem adds to a significant understanding of the structure of
the problem, both at the mathematical and at the physical level. In this paper, the interior and exterior Neumann functions
for the Laplacian in the cases of spherical and ellipsoidal domains are represented in terms of images. Besides isolated
images, the presence of the normal derivative in the Neumann condition demands an additional continuous distribution
of images, which in the spherical cases, can be restricted to a one-dimensional manifold, whereas for the ellipsoid, both a
one-dimensional and a two-dimensional distribution of images is needed. Copyright © 2012 John Wiley & Sons, Ltd.

Keywords: Neumann boundary condition; Neumann function; image system; ellipsoidal geometry; ellipsoidal harmonics

1. Introduction
The method of images is a well-known technique for solving boundary value problems when the fundamental solution of the related
differential operator is known. In fact, the method consists in finding the positions and strengths of a set of fictitious sources in the
complement of the fundamental domain, in such a way that the total fields generated from the real and the fictitious sources will auto-
matically satisfy the prescribed boundary conditions [1, 2]. In the case of a spherical boundary, the solution for the Dirichlet problem for
the Laplacian in the presence of a point singularity is associated with the well-known Green’s function, while the image consists of an
isolated singularity related to the real singularity by the Kelvin transformation [3]. This problem has a history of almost two centuries.
Any deviation from the spherical geometry causes a lot of complications. In fact, the image system for the Green’s function in the case
of the ellipsoid was found only recently [4]. Even the spherical case presents a lot of difficulties if the boundary has more than one con-
nected part [5]. Common to these cases is that, in addition to the isolated point singularities, continuous line and surface distributions
must be employed as well.
The situation becomes much harder if in the complications, caused by a non-symmetric geometry, we add the difficulties coming
from boundary conditions that involve differential operators, such as the Neumann condition. This is the case we investigate in the
present work. The fact that the derivative is defined as the limiting value of the difference between two values of the function leads to
the need of using a continuous distribution of images representing the approaching values of the function.
The sphere exhibits a rotational symmetry with respect to the axis passing through the center of the sphere and the point where the
singularity is located. This symmetry implies that the whole set of images, discrete and continuous, has to lie on the axis of rotation,
and we have found such sets both for the interior and the exterior spherical problems with the Neumann condition prescribed on
the boundary.
On the other hand, the ellipsoid has only a finite set of symmetries and, therefore, no such simplicity in the geometry of the images
can be expected. Nevertheless, the architecture of an image system for the ellipsoid has to be designed with the characteristics of
the ellipsoid itself, which are nothing else but the defining ellipsoidal coordinate system. Making use of this idea, we obtained an
image system which involves point images, line images over portions of the coordinate curves that are defined as the intersection of
the hyperboloid surfaces of one and two sheets, which are the analogue of the rays for the sphere, and finally, because of the lack
of rotational symmetry, a surface distribution of images over an ellipsoid confocal to the boundary lying in the complement of the
fundamental domain.
For the line images of the interior problems, where the images are distributed over an unbounded set, the existence of the related
improper integrals are secured by choosing appropriately the corresponding densities.

a Departement of Chemical Engineering, University of Patras, and ICE-HT/FORTH, Greece

b Technical Research Centre of Finland, P.O.Box 1000, FIN-02044 VTT, Finland
*Correspondence to: Johan C.-E. Sten, Technical Research Centre of Finland, P.O.Box 1000, FIN-02044 VTT, Finland.
† E-mail: [email protected]
482

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

The paper is organized as follows. Section 2 contains the interior and the exterior Neumann problem for the sphere using the
well-known theory of spherical harmonics and Kelvin inversion. The much less diffused theory of ellipsoidal harmonics is exposed
in Section 3 at the most elementary level. The book by Hobson [6] and Whittaker and Watson [7] are standard references for ellip-
soidal harmonics. Sections 4 and 5 involve the analysis of the exterior and interior Neumann problems, respectively. The conclusions
of this work are summarized in the final Section 5. The existing literature of the theory of images is very extensive. The image principle
was originally described by William Thomson (Kelvin) in 1845 [8]. The oldest example of the use of a continuous image distribution in
connection to a sphere is found in the book by Carl Neumann [9]. The review [10] and the last two chapters of the book [11] provide
some indication of the work that is available in this area. Also [12, 13] and [14] involve some interesting material on images for spherical
boundaries. For the use of the ellipsoidal harmonics in the solution of boundary value problems, we refer to [15, 16] and [17] among
many others.

2. The Neumann function for the sphere

Given any smooth closed surface S in R3 , with i its interior and e its exterior. The interior Neumann function [2] is defined as the
solution of the boundary value problem for the potential Ui as
Ui .r/ D ı.r  r0 /, r 2 i , (2.1)
@ i 1
U .r/ D , r 2 S, (2.2)
@n jSj
@
where  is the Laplacian, ı.r/ is the Dirac measure, and @n is the outward normal derivative on the surface S. jSj stands for the total
surface area of S and r0 is a fixed point in the open domain i .
The exterior Neumann function [2] is defined as the solution of the boundary value problem for the potential Ue as
Ue .r/ D ı.r  r0 /, r 2 e , (2.3)
@ e 1
U .r/ D  , r 2 S, (2.4)
@n jSj
1
Ue .r/ D O. 2 /, r D jrj ! 1, (2.5)
r
@
where r0 now belongs to the open domain e , and @n continues to represent outward normal differentiation on S. The asymptotic
e
behavior of U at infinity is a direct consequence of Equations (2.3), (2.4) and the divergence theorem.
The Neumann function is the analogue of the Green’s function, where the Dirichlet boundary condition is replaced by a boundary
condition of the Neumann type. In fact, the Green’s function is the solution of Equation (2.1) with Ui equal to zero on the boundary.
However, it is not possible to demand from the corresponding Neumann function that the normal derivative vanishes on the boundary,
because by the Gauss theorem, Equation (2.1) implies that
I
@Ui
ds D 1. (2.6)
S @n
Nevertheless, if we demand that the normal derivative is constant on S, then we arrive at the Neumann condition (2.2).

2.1. The exterior problem

For the case of the sphere, the exterior Neumann function is an exterior harmonic function with a monopole singularity at the point r0 ,
outside the sphere, which satisfies the boundary condition
@ e 1
U .r/ D  , jrj D a, (2.7)
@r 4a2
where a is the radius of the sphere.
Because of its symmetry, the sphere problem can be simplified by using as polar axis for the spherical coordinate system the line
that passes through the point singularity r0 . Once the solution of this problem has been obtained, we can find the solution of the
generic case through a rotation of the system and the use of the addition theorem for spherical harmonics. Nevertheless, to preserve
the analogy with the ellipsoidal case, where no such possibility exists, we will work with the spherical case in its generic form as well.
A particular solution of Equation (2.3) is provided by the fundamental solution, which for r < r0 has the expansion
1 X
X n
1 1 rn m?
 D nC1
Yn .Or0 /Ym
n .Or/, (2.8)
4jr  r0 j mDn
2n C 1 r 0
nD0

where Ymn .Or/ denotes the normalized complex form of the surface spherical harmonics [18]. Hence, the general solution of
Equation (2.3) is
1 Xn
" #
X 1 Am rn
n
Ue .r/ D nC1
 nC1
Ym? m
n .Or0 /Yn .Or/, (2.9)
mDn
2n C 1 r r 0
nD0
483

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

and the boundary condition (2.7) gives

A00 D 1 (2.10)
and
n a2nC1
Am
n D , n  1. (2.11)
n C 1 r0nC1
Then, the solution (2.9) is written as
 
n  a 2nC1 m?
1 X
X n
e 1 1 1 rn
U .r/ D   1C Yn .Or0 /Ym
n .Or/, (2.12)
4r 4r0 mDn
2n C 1 r0nC1 nC1 r
nD1

which holds for a < r < r0 , or

1
Ue .r/ D  C Ue,ref .r/, (2.13)
4jr  r0 j
where
1 n  2 nC1
e,ref 1 1X X n a
U .r/ D  Ym? m
n .Or0 /Yn .Or/ (2.14)
4r a mDn
.2n C 1/.n C 1/ rr 0
nD1

is the reflected potential, which holds for every r > a.

Our goal at present is to invent an image system in the complementary domain r < a, which produces the potential Ue,ref for r > a.
These images will then replace the actual boundary condition. Obviously, the first term in Equation (2.14) is represented as the poten-
tial generated by a monopole at the origin with strength 1. We also assume a monopole with strength Q0 at the point r00 rO0 D r00 with
r00 < a, and a continuous distribution of monopoles with constant strength equal to q0 along the line segment connecting the origin
with the point r00 , as illustrated in Figure 1.
This system will provide the potential
X1 X n
1 1 r00n m?
Ue,im .r/ D  Q0 Y .Or0 /Ym
n .Or/
4r 2n C 1 rnC1 n
nD0 mDn
X1 X n  0 nC1
1 r0
q0 Ym? m
n .Or0 /Yn .Or/, (2.15)
mDn
.2n C 1/.n C 1/ r
nD0

which we want to be equal to the reflected potential Ue,im .r/ given in Equation (2.14). By employing the orthogonality of the surface
spherical harmonics we see that the image potential Ue,im becomes identical to the reflected potential Ue,ref if we choose
a
Q0 D , (2.16)
r0
1
q0 D  (2.17)
a
and
a2
r00 D . (2.18)
r0
Therefore, the image-point r00 has to be the Kelvin image of the point singularity r0 . The image system for the Neumann function in the
exterior of a sphere then consists of a uniform distribution of monopoles with strength 1=a on the line segment that extends between
the origin and the Kelvin image r00 of the given singularity at r0 , plus a monopole at the origin with strength 1, plus a monopole at
the Kelvin image r00 with strength a=r0 .

r´0
r0 Q´ q ´ -1
r

Figure 1. Geometry of the exterior Neumann boundary problem for the sphere and the pertinent image system.
484

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

2.2. The interior problem

Let us now consider the corresponding interior problem of Equations (2.1) and (2.2), where r0 < a and we are looking for an exterior
image system that will represent the solution for r < a. For r 2 .r0 , a/ the solution is written as
1 X
X n  
1 r0n
Ui .r/ D Am
n r n
 nC1
Ym? m
n .Or0 /Yn .Or/ (2.19)
mDn
2n C 1 r
nD0

and the Neumann condition (2.7) demands that A00 is arbitrary and

n C 1 r0n
Am
n D , n  1. (2.20)
n a2nC1

Using the additional condition that Ui .0/ D 0 we obtain A00 D 1=4r0 and arrive at

n C 1  rr0 n m?
1 n
1 1 1X X
Ui .r/ D  C  Yn .Or0 /Ym
n .Or/. (2.21)
4jr  r0 j 4r0 a mDn
n.2n C 1/ a2
nD1

Assume now an image-system which consists of a monopole of strength a=r0 at the point r00 D r00 r0 , with r00 > a, and a continuous
distribution of monopoles at  D r0 with strength f . /, which extends from the point r00 to infinity along the ray in the direction rO0 .
This image-system will give rise to the potential
1 n
a X X 1 rn m?
Ui,im .r/ D  0nC1
Yn .Or0 /Ym
n .Or/
r0 mDn
2n C 1 r0
nD0
1 X
n C1
Z
X 1 f . /
 rn Ym? m
n .Or0 /Yn .Or/ d . (2.22)
mDn
.2n C 1/  nC1
nD0
r00

We observe here that, in contrast to the exterior problem, where the continuous image was extended over the bounded interval .0, r00 /,
the corresponding interval of images is the unbounded interval .r00 , C1/, which creates problems with the leading term n D 0 of the
last series on the right-hand side of expression (2.22). That means that we cannot distribute monopoles of constant strength, as it is the
case with the exterior problem, but we have to choose the density distribution f . / in a way that secures the existence of all integrals
in Equation (2.22). Because the problem is created by the term n D 0, we can choose the strength function to be
 
1 j  rj
f . / D 1 . (2.23)
a 

Then the second series on the right-hand side of Equation (2.22) becomes
C1
Z Z 
C1 
1 f ./ 1 1 1
 d D   d
4 j  rj 4a  j  rj
r00 r00

1 n C1
Z
1X X 1 d
D rn Ym?
n .O
r /Y
0 n
m
.O
r / nC1
a mDn
2n C 1 
nD1
r00
1 n
1X X 1 r
D . 0 /n Ym? m
n .Or0 /Yn .Or/. (2.24)
a mDn
n.2n C 1/ r 0
nD1

Setting r00 D a2 =r0 in Equation (2.22), we obtain the following image representation:
a Z  
1 r0 1 j  rj 1
Ui,im .r/ D   1 d . (2.25)
4 j a22 r0  rj 4a  j  rj
r0 1
a2
r0

Hence, the interior Neumann function in terms of images is written as

Z  
1 1 1 a 1 j rO0  rj 1
Ui .r/ D  C C   1 d . (2.26)
4jr  r0 j 4r0 4a 4r0 j a22 r0  rj 4a  j rO0  rj
r0 a2
C1
r0
485

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

Thus, the image system for the interior spherical Neumann function consists of a monopole with strength a=r0 at the Kelvin image-point
.a=r0 /2 r0 of the singularity at r0 and a continuous distribution of monopoles, with strength given in Equation (2.23), which extends from
the Kelvin image-point .a=r0 /2 r0 radially to infinity, as shown in Figure 2.
1 1
The constant 4 . r0 C a1 / comes from the condition

Ui .0/ D 0, (2.27)

which is the analogue of the condition

lim Ue .r/ D 0 (2.28)

r!1

for the exterior problem.

3. Harmonic functions in the ellipsoidal geometry

The ellipsoid is the sphere of the anisotropic space. It appears naturally in physical problems whenever some quantity exhibits
directional differentiations. The ellipsoidal coordinates of Lamé are denoted by ., , /, where  corresponds to the radial coordi-
nate and (, ) correspond to the angular coordinates of the spherical system. An ellipsoidal coordinate system refers to a fundamental
ellipsoid [6]

x12 x22 x32

C C D 1, (3.1)
a21 a22 a23

where the three semiaxes 0 < a3 < a2 < a1 < C1 specify the fixed focal ellipse, whose semi-axes are defined by .0, ˙h1 , 0/, .˙h2 , 0, 0/,
and the semi-focal point by .˙h3 , 0, 0/, where

h21 D a22  a23 , (3.2)

h22 D a21  a23 , (3.3)
h23 D a21  a22 . (3.4)

Because

h21  h22 C h23 D 0, (3.5)

any ellipsoidal system defines a 2-parameter coordinate system, which are usually taken to be the semi-axes h3 < h2 .
Then the points of the first Cartesian octant are identified by the relations

x1 D , (3.6)
h2 h3
q q q
2  h23 2  h23 h23   2
x2 D , (3.7)
h1 h3
q q q
  h2 h22  2 h22   2
2 2
x3 D , (3.8)
h1 h2
where, in the first octant, 0 <  < h3 <  < h2 <  < C1. Corresponding relations for the other seven Cartesian octants are obtained
by considering the signatures .x1 , x2 , x3 / D ., C, C/, .C, , C/, .C, C, /, ., , C/, ., C, /, .C, , /, ., , /.
Laplace’s equation in ellipsoidal coordinates is separable in terms of the coordinate variables and produces three identical equations,
known as the Lamé equation, governing the behavior of the harmonic function in the intervals of variation of the variables ., , / [6,7].

r´0 r0
f Q´
r

Figure 2. Geometry of the interior Neumann boundary problem for the sphere and the pertinent image system.
486

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

The Lamé equation has the interior solutions

Enm .x/, n D 0, 1, 2, : : : , and m D 1, 2, : : : , 2n C 1, (3.9)
which are known as Lamé functions of the first kind, and the exterior solutions
Fnm .x/ D .2n C 1/Enm .x/Im
n .x/, (3.10)
C1
Z
dt
Im
n .x/ D q q , n D 0, 1, 2, : : : , and m D 1, 2, : : : , 2n C 1, (3.11)
x ŒEnm .t/ 2 t2  h23 t2  h22

which are known as Lamé functions of the second kind.

Corresponding to the surface spherical harmonics Ym
n . , '/ we now have the surface ellipsoidal harmonics

Sm m m
n ., / D En ./En ./ (3.12)
and we define the interior ellipsoidal harmonics
Em m m
n ., , / D En ./Sn ., / (3.13)
and the exterior ellipsoidal harmonics
Fnm ., , / D Fnm ./Sm
n ., /. (3.14)
The relative coordinate surfaces are expressed as
x12 x22 x32
C C D 1, (3.15)
u2 u2  h23 u2  h22
where u D  > h2 describes a confocal family of ellipsoids, for h3 < u D  < h2 describes a confocal family of hyperboloids of one
sheet, and for 0 < u D  < h3 describes a confocal family of hyperboloids of two sheets. For u D h2 , Equation (3.15) identifies the focal
ellipse
x12 x22
C D 1, x3 D 0. (3.16)
h22 h21
For u D a1 , Equation (3.15) recovers the reference ellipsoid (3.1) and as u ! C1, equation (3.15) describes a sphere of infinite radius.
The ellipsoidal coordinate system is orthogonal and has the metric
.ds/2 D h2 .d/2 C h2 .d/2 C h2 .d/2 , (3.17)
where the metric coefficients assume the expressions
.2  2 /.2   2 /
h2 ., , / D , (3.18)
.2  h23 /.2  h22 /
.2  2 /.2   2 /
h2 ., , / D , (3.19)
.2  h23 /.h22  2 /
.2   2 /.2   2 /
h2 ., , / D . (3.20)
.h23   2 /.h22   2 /
Then, the surface element on the ellipsoid  D 0 is given by
dS0 ., / D h .0 , , /h .0 , , /dd, (3.21)
while for the volume element, we obtain
dV., , / D h ., , /h ., , /h ., , /ddd. (3.22)
The local ellipsoidal system is given by
3
 X xi
O D xO ,
2  a2 C a2 i
(3.23)
h  1 i
iD1
3
 X xi
O D
 xO ,
2  a2 C a2 i
(3.24)
h  1 i
iD1
3
 X xi
O D xO ,
2  a2 C a2 i
(3.25)
h  1 i
487

iD1

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

where the hat on top of a vector signifies that the vector has unit length. In the local ellipsoidal framework, the position vector is
written as
  
rD O C OC
 O (3.26)
h h h

and the gradient as

O @ O @ O @
rD C C . (3.27)
h @ h @ h @

Hence, the outward normal derivative at any point (, ) on the ellipsoid  D 0 is expressed as

@ 1 @
D O  r D . (3.28)
@n h .0 , , / @

The surface ellipsoidal harmonics Sm

n are orthogonal over any ellipsoid  D 0 with respect to the weighted function

1
`0 ., / D q q . (3.29)
02  2 02   2

The normalization constants

I
m
n ./ D ŒSm 2
n ., / ` ., /ds., / (3.30)
S

evaluated on the reference ellipsoid  D a1 , assume the values [15]

1
0 D 4 (3.31)
i 4 h21 h22 h23
1 D , i D 1, 2, 3 (3.32)
3 h2i
8
1
2 D  .ƒ  ƒ0 /.ƒ  a21 /.ƒ  a22 /.ƒ  a23 / (3.33)
5
8
2
2 D .ƒ  ƒ0 /.ƒ0  a21 /.ƒ0  a22 /.ƒ0  a23 / (3.34)
5
6i 14 2 2 2 2
D h h h h , i D 1, 2, 3, (3.35)
2
5 1 2 3 i
where ƒ and ƒ0 are the two roots of the equation
3
X 1
D 0. (3.36)
iD1
ƒ  a21

For n D 3, the constants 3i , i D 1, 2, : : : , 7 are given in [15], whereas for n  4, the exact values of nm in closed form are unknown.
In the present paper, we are going to need the exact form of the harmonics only for n D 0, 1, 2, and for completeness we provide
these functions in the sequel, both in ellipsoidal and in Cartesian form.

E10 ., , / D 1. (3.37)

1 1 1 xi
Ei1 ., , / D j2  a21 C a2i j 2 j2  a21 C a2i j 2 j 2  a21 C a2i j 2 D h1 h2 h3 , i D 1, 2, 3. (3.38)
hi

E12 ., , / D .2  a21 C ƒ/.2  a21 C ƒ/. 2  a21 C ƒ/

" 3 #
X x2
2 2 2 i
D .ƒ  a1 /.ƒ  a2 /.ƒ  a3 / C1 . (3.39)
iD1
ƒ  a2i

E22 ., , / D .2  a21 C ƒ0 /.2  a21 C ƒ0 /. 2  a21 C ƒ0 /

" 3 #
X x2
0 2 0 2 0 2 i
D .ƒ  a1 /.ƒ  a2 /.ƒ  a3 / C1 . (3.40)
ƒ0  a2i
488

iD1

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN
q q q q q q
 2  h23 2  h23 h23   2 2  h22 h22  2 h22   2
E6i
2 ., , / D q q q
hi j2  a21 C a2i j j2  a21 C a2i j j 2  a21 C a2i j
x1 x2 x3
D h1 h2 h3 , i D 1, 2, 3. (3.41)
xi
As a final item from the theory of ellipsoidal harmonics we provide the expansion of the fundamental solution for the Laplace operator
in ellipsoidal harmonics [16], which reads
1 2nC1
X X 4
1 1 m
D E .r0 /Fnm .r/, (3.42)
jr  r0 j 2n C 1 nm n
nD0 mD1

where r D ., , /, r0 D .0 , 0 , 0 / and 0 < .

4. The exterior Neumann function for the ellipsoid

Consider the ellipsoid (3.1), which we will refer to as S, and choose an ellipsoidal system with S as its reference ellipsoid. Then S is deter-
mined by  D a1 . A unit source is located at the point r0 D .0 , 0 , 0 / exterior to the ellipsoid  D a1 , that is, 0 > a1 . Then, the exterior
Neumann function is the solution of the following problem for the potential V e :

V e ., , / D ı.r  r0 /,  > a1 , (4.1)

@V e ., , / 1
D  ,  D a1 , (4.2)
@n jSj
1
V e ./ D O. 2 /,  ! 1, (4.3)


where jSj denotes the area of the ellipsoid S.

A particular solution of Equation (4.1) is given by Equation (3.42) and, therefore, the general solution of Equation (4.1) is
1 2nC1
X X 1 1 m
V e ., , / D  E ., , /Fnm .0 , 0 , 0 /
2n C 1 nm n
nD0 mD1
1 2nC1
X X 1 1 m m
C B F ., , /Em
n .0 , 0 , 0 /, (4.4)
2n C 1 nm n n
nD0 mD1

where Bm n are constants that have to be determined in such a way as to satisfy condition (4.2). The first series on the right-hand side of
Equation (4.4) converges in the region  2 .a1 , 0 /, where the boundary condition can be applied. Using the surface normal operator
(3.28) at  D a1 , we obtain
1 2nC1
X X
@V e ˇˇ 1 1  m0
Da1 D a2 a3 `a1 ., /  E .a1 /Fnm .0 , 0 , 0 /  Bm m0 m m
n Fn .a1 /En .0 , 0 , 0 / Sn ., /, (4.5)
@n 2n C 1 nm n
nD0 mD1

which, because of orthogonality and the boundary condition (4.2), implies that
Z Z
dS B1
 D a2 a3 01 I010 .a1 /`a1 ., /ds
S jSj 0 S
Z
B10 1
D a2 a3 . /`a ., /ds
1
0 S a2 a3 1
D B10 (4.6)

and finally B10 D 1.

For n  1, Equation (4.5) and orthogonality implies that

Enm0 .a1 /Fnm .0 , 0 , 0 / E m0 .a1 /Im

n .0 /
Bm
n D D nm m .a / 0 . (4.7)
Fnm0 .a1 /Em . ,
n 0 0 0  ,  / ŒEn .a /I
1 n 1

Therefore, the Neumann function is given by

1 2nC1
1 I1 ./ X X 1 1 Enm0 .a1 / m
V e ., , / D  C 0 C F .0 , 0 , 0 /Fnm ., , /, (4.8)
4jr  r0 j 4 2n C 1 nm Fnm0 .a1 / n
nD1 mD1
489

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

where
Enm0 .a1 / 1 Enm .a1 /Enm0 .a1 /a2 a3
m0 D . (4.9)
Fn .a1 / 2n C 1 En .a1 /Enm0 .a1 /a2 a3  1
m

Our next goal is to determine an image system for the reflected part of the solution that is

1 2nC1
I01 ./ X X 1 1 Enm0 .a1 / m
V e,ref ., , / D C F .0 , 0 , 0 /Fnm ., , /. (4.10)
4 2n C 1 nm Fnm0 .a1 / n
nD1 mD1

Because of the lack of symmetry, it may seem peculiar that a simple system of images, like in the spherical one, should exist for the
ellipsoid at all. However, it is easily observed that the potential I01 ./=4 is generated by a monopole distribution over the focal ellipse
Sh2 of surface density [17]

`h2 ., /
f ., / D  . (4.11)
4
Indeed, such a distribution gives rise to the potential
Z
1 f .0 ,  0 /
 ds.0 ,  0 /
4 Sh2 jr  r0 j
1 2nC1 Z
1 X X 4 1 m 0 0 0 0 0 0
D m F n ., , / Em
n .h2 ,  ,  /`h2 . ,  /ds. ,  /
.4/2 2n C 1 n Sh2
nD0 mD1
I01 ./
D . (4.12)
4
The source strength, or (in Electrostatics) the total amount of ‘charge’ over the focal ellipse is given by
Z  
`h .0 ,  0 /
 2 ds.0 ,  0 / D 1. (4.13)
Sh2 4

This total ‘charge’ of strength 1 corresponds to the point ‘charge’ of strength 1 that has to be put at the origin in the case of the
sphere.
Next, we conjecture that the remaining expansion, which begins with the dipole terms n D 1, can be interpreted as being partly
because of a system of point and line singularities, that is, a combination of a point source of strength Q at r0 D .0 , 0 ,  0 /, and a
one-dimensional distribution of sources density q.% /, extending from h2 to 0 < a1 along the coordinate curve passing through r0 , as
depicted in Figure 3.
Such a system will give rise to the following potential
0
Z
Q 1 q./
V e,im ., , / D   dl./. (4.14)
4jr0  rj 4 j  rj
h2

The length element dl along the line-segment of integration

˚
 D .t, 0 ,  0 /, t 2 .h2 , 0 /

r0 0 0 0
Sa
1
d
Q r´ ´ ´ ´
t ´ ´
q( ) Sh2

Figure 3. Geometry of the exterior Neumann boundary problem for the ellipsoid and the pertinent image system.
490

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

is given by
p p
t2  02 t2   02
dl./ D h .t, 0 ,  0 /dt D q q dt (4.15)
t2  h23 t2  h22

and we can simplify the calculation by assigning

1
q./ D (4.16)
h .t, 0 ,  0 /

so that

q./dl./ D dt. (4.17)

Expanding the potential V e,im we obtain

2 0
3
1 2nC1
X X Z
1 1 6 m 0 7 0 0 m
V e,im ., , / D  4QEn . / C Enm .t/dt5 Sm
n . ,  /Fn ., , /
2n C 1 nm
nD0 mD1 h2
1  0
D  1
Q C .  h2 / I01 ./
0
2 0
3
3 Z
1 X 1 6 m 0 7 0 0 m
 m 4QE1 . / C E1m .t/dt5 Sm
1 . ,  /F1 ., , /
3 1
mD1 h2
e,im
C Q ., , /, (4.18)

where
2 0
3
1 2nC1
X X Z
1 1 6 m 0 7 0 0 m
Qe,im ., , / D  4QEn . / C Enm .t/dt5 Sm
n . ,  /Fn ., , / (4.19)
2n C 1 nm
nD2 mD1 h2

denotes the contribution of all multipoles of order higher than the dipoles.
Because the monopole contribution is already represented by the focal distribution we should have

Q D h2   0 . (4.20)

Then for the dipole terms, using orthogonality between the expansions (4.10) and (4.18), we obtain
2 0
3
Z
6 0 m 0 m 7 m 0 0 E1m0 .a1 / m
4.  h2 /E1 . /  E1 .t/dt5 S1 . ,  / D m0 F .0 , 0 , 0 /, m D 1, 2, 3. (4.21)
F1 .a1 / 1
h2

Let
0
Z
0 0
gm . / D .  h2 /E1m .0 /  E1m .t/dt, m D 1, 2, 3. (4.22)
h2

Then,
1 02
g1 .0 / D .  h22 /, (4.23)
2 0 1
 0 q   1 2 2
 1 B h2 C h1 h2  h3 C
g2 .0 / D  h2 02  h23 C h2 h1 h22  h23  h23 ln @ q A, (4.24)
2 2 2 0 C 02  h2 3
0 1
q
0 1 B h2 C
g3 .0 / D .  h2 / 02  h22  h22 ln @ q A. (4.25)
2 2 0 C 02  h2 2
491

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

Furthermore, if we set
E1m0 .a1 / m a1 a2 a3 Em m
1 .0 , 0 , 0 /I1 .0 /
Dm
1 D m0 F1 .0 , 0 , 0 / D m , m D 1, 2, 3. (4.26)
F1 .a1 / a1 a2 a3 I1 .a1 /  1

then the system (4.21) is written as

g1 .0 /0  0 D D11 , (4.27)

q q
g2 .0 / 02  h23 h23   02 D D21 , (4.28)
q q
g3 .0 / h22  02 h22   02 D D31 , (4.29)

which has to be solved with respect to the coordinates (0 , 0 ,  0 ) of the image-point.
Because
02  02 .02  h23 /.h23   02 / .h22  02 /.h22   02 /
C C D 1, (4.30)
h22 h23 h21 h23 h21 h22

it follows that 0 has to solve the equation

!2 !2 !2
D11 h1 D21 h2 D31 h3
C C D h21 h22 h23 . (4.31)
g1 .0 / g2 .0 / g3 .0 /

With 0 known, Equations (4.28) and (4.29) give

!2 !2
02 02 1 D21 1 D31
 C D h22 C h23  2  2 , (4.32)
h1 g2 .0 / h1 g3 .0 /

which, combined with Equation (4.27),

!2
D11
02  02 D (4.33)
g1 .0 /

provides the values of 02 and  02 as the roots of a quadratic equation.

The remaining part of the multipole expansion will be represented by a surface distribution of monopoles over a confocal ellipsoid
in the interior of S. Let us consider the ellipsoid that passes through the point image at (0 , 0 ,  0 ). If we denote by d.0 ,  0 /, the density
of sources over the image ellipsoid S0 , then we want to identify the coefficients Gm
n of the expansion

1 2nC1
X X
d., / D `0 ., / Gm m
n Sn ., / (4.34)
nD2 mD1

in such a way that

I
0
d., /Em
n . , , /ds., /
S0
2 0
3
Z
6 7 Enm0 .a1 / m
C 4.h2  0 /Enm .0 / C Enm .t/dt5 Sm 0 0
n . ,  / C m0 F .0 , 0 , 0 / D 0 (4.35)
Fn .a1 / n
h2

for every n D 2, 3, : : : and m D 1, 2, : : : , 2n C 1.

The expansion (4.34) is chosen to start from the quadrupole term n D 2 to avoid any contribution from the monopole and dipole
terms, n D 0 and n D 1, which are already been taken care of by the point and line images. Indeed, by orthogonality, we see that
I
d., /Em
n ., , /ds., / D 0 (4.36)
S0

for n D 0 and n D 1.
Inserting Equation (4.34) in Equation (4.35) and using orthogonality once more we obtain
2 0
3
Z m0
1 6 0 m 0 0 0 m 0 0 E .a1 / m 7
Gm
n D  m m 0 4.h2   /En . ,  ,  / C Sn . ,  / Enm .t/dt C nm0 F .0 , 0 , 0 /5 (4.37)
n En . / Fn .a1 / n
h2
492

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

for every n D 2, 3, : : : and m D 1, 2, : : : , 2n C 1. The values of Gm

n given in Equation (4.37) provide the surface image density d given by
Equation (4.34).
Hence, an appropriate image system for the exterior Neumann function, for the ellipsoid, is seen to consist of:
(I) a surface monopole distribution over the focal ellipse with density f given in Equation (4.11),
(II) a point monopole of strength h2  0 at the point .0 , 0 ,  0 / inside the ellipsoid  D a1 ,
(III) a line monopole distribution with density q given by Equation (4.16) over the coordinate segment (, )= constant from the point
.h2 , 0 ,  0 / upon the focal ellipse to the point (0 , 0 ,  0 ), where the values 0 , 0 and  0 are solutions of the algebraic systems
(4.31)–(4.33), and
(IV) a surface distribution of density (4.34), (4.37) over the ellipsoid that is confocal to  D a1 and passes through the point .0 , 0 ,  0 /.

5. The interior Neumann function for the ellipsoid

In this section, we consider the interior Neumann problem for the ellipsoid. The interior Neumann function V i has to solve the following
boundary value problem:

V i ., , / D ı.r  r0 /, h2   < a1 (5.1)

@V i ., , / 1
D ,  D a1 , (5.2)
@n jSj

where

r0 D .0 , 0 , 0 /, h2  0 < a1 . (5.3)

In the neighborhood of the boundary, the solution has the form

1 2nC1
X X 1 1 m 
Vi D  E .0 , 0 , 0 / Fnm ./ C Am m m
n En ./ Sn ., / (5.4)
2n C 1 nm n
nD0 mD1

and the boundary condition (5.2) demands that

1 2nC1
X X 1 1 m  1
 a2 a3 `a1 ., / m En .0 , 0 , 0 / Fnm0 .a1 / C Am m0 m
n En .a1 / Sn ., / D . (5.5)
2n C 1 n jSj
nD0 mD1

Integrating Equation (5.5) over the ellipsoid  D a1 and using orthogonality we obtain
I I
1 1
 a2 a3 1 F010 .a1 / `a1 ., /ds., / D ds., / (5.6)
0 S jSj S

and
Fnm0 .a1 /
Am
n D , n  1. (5.7)
Enm0 .a1 /

Because
1
F010 .a1 / D  , (5.8)
a2 a3

it follows that Equation (5.6) holds as an identity. Therefore, A10 is the arbitrary constant of the solution V i .
Hence, the interior Neumann function is given by
1 2nC1
X X
1 A1 1 1 Fnm0 .a1 / m
V i ., , / D   0 C E .0 , 0 , 0 /Em
n ., , /, (5.9)
4jr  r0 j 4 2n C 1 nm Enm0 .a1 / n
nD1 mD1

where the value of the arbitrary constant A10 may be fixed by demanding the additional condition

lim V i .r/ D 0. (5.10)

r!0

The origin in ellipsoidal coordinates corresponds to the point (h2 , h3 , 0), i.e., all three coordinates (, , ) take their lowest values.
Because

Em m m m
n ., , / D En ./En ./En ./, (5.11)
493

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

it follows that at the origin, the only class of ellipsoidal harmonics that survive in this expansion are those that are constructed from
Lamé functions of class K and of even degree [6]. Consequently,
1 nC1
X X  m0 
A10 I1 .0 / 1 1 F2n .a1 / m
 D 0  E
m E m0 .a / 2n .0 , 0 , 0 /  F m
2n .0 , 0 , 0 / Em
2n .h2 , h3 , 0/, (5.12)
4 4 4n C 1 2n 2n 1
.K/nD1 mD1

1
P
where the notation indicates that only Lamé functions of class K are considered. There are in all n C 1 functions of class K and
.K/nD1
degree 2n.
Finally, we turn into an image system that represents the reflected part of the solution
1 2nC1
X X
A10 1 1 Fnm0 .a1 / m
V i,ref ., , / D  C E .0 , 0 , 0 /Em
n ., , /. (5.13)
4 2n C 1 nm Enm0 .a1 / n
nD1 mD1

Let us consider an image system consisting of a point source of strength Q at the exterior point r0 D .0 , 0 ,  0 /, a line distribution of
monopoles with density q.t/=h .t, 0 ,  0 /, which extends from the point r0 to infinity along the coordinate curve (0 ,  0 )= constant, and
a surface distribution of monopoles with density
1 2nC1
X X
d.0 , , / D `0 ., / Jnm Sm
n ., / (5.14)
nD2 mD1

on the confocal ellipsoid S0 that passes through the point r0 . The image is graphically represented in Figure 4.
This system of images generates the following potential in the interior of the ellipsoid  D a1
C1
Z I
i,im Q 1 q.t/ 1 d.0 , , /
V ., , / D   dt  ds., /, (5.15)
4jr0  rj 4 j  rj 4 S0 js  rj
0

where s D .0 , , /.

By duly expanding Equation (5.15) in ellipsoidal harmonics and equating the resulting expansion with the reflected potential given
in Equation (5.13), we obtain
2 3
1 2nC1 C1
Z
X X 1 1 6 m 0 7 0 0 m
4QFn . / C q.t/Fnm .t/dt5 Sm
n . ,  /En ., , /
2n C 1 nm
nD0 mD1 0
1 2nC1
X X 1
C Jm F m .0 /Em
n ., , /
2n C 1 n n
nD2 mD1
1 2nC1
A10 X X 1 1 Fnm0 .a1 / m
D  E .0 , 0 , 0 /Em
n ., , /, (5.16)
4 2n C 1 nm Enm0 .a1 / n
nD1 mD1

where the surface integrals have been calculated via orthogonality. The monopole term of Equation (5.16) gives
C1
Z
QI01 .0 / C q.t/I01 .t/dt D A10 (5.17)
0

d S 1
t ´ ´
Sa1
q( ) r´ ´ ´ ´

r0 0 0 0
Q

Figure 4. Geometry of the interior Neumann boundary problem for the ellipsoid and the pertinent image system.
494

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
G. DASSIOS AND J. C.-E. STEN

and if we choose the density on the line to be

1
q.t/ D 1  , (5.18)
tI01 .t/

then we obtain
2 3
C1
Z
1 6 1 1 7
QD 4 A0  .I01 .t/  /dt5 , (5.19)
I01 .0 / t
0

where the improper integral in Equation (5.19) exists owing to the particular choice (5.18) of the density function q.
Next, from the dipole terms in Equation (5.16) we obtain
2 3
C1
Z
6 m 0 1 7 0 0 F1m0 .a1 / m
4QF1 . / C .1  1 /F1m .t/dt5 Sm 1 . ,  / D  m0 E .0 , 0 , 0 /, m D 1, 2, 3, (5.20)
tI0 .t/ E1 .a1 / 1
0

where Q is given by Equation (5.19). The three equations in Equation (5.20) form a system for the three unknowns (0 , 0 ,  0 ), which is
similar to the system (4.21) and, therefore, it can be solved in an identical manner. That will specify the point r0 .
Finally, for the terms n  2, orthogonality implies the following values for the coefficients of the surface source density (5.14)
2 3
0 C1
Z
m m
1 F .a1 / En .0 , 0 , 0 / m 0 0
1 6 m 0 1 7 Sn . ,  /
Jnm D  m nm0 m  m 4 QF n . / C .1  /Fn
m
.t/dt 5 (5.21)
n En .a1 / Fn .0 / n tI01 .t/ Fnm .0 /
0

for every n D 2, 3, : : : and m D 1, 2, : : : , 2n C 1. That specifies completely the density d.0 , , / in Equation (5.14).
Collecting everything together, we see that the image system for the interior Neumann function for the ellipsoid consists of:
(I) a point source of strength Q, given by Equation (5.19), at the point (0 , 0 ,  0 ) specified as the solution of the algebraic system
formed by Equation (5.20),
(II) a line source distribution of monopoles with density q.t/=h .t, 0 ,  0 / with q.t/ given by Equation (5.18), which extends from the
point (0 , 0 ,  0 ) all the way to infinity, along the coordinate line (0 ,  0 )= constant, and
(III) a surface distribution with density d.0 , , / given by Equations (5.14), (5.21) over the confocal ellipsoid S0 passing through
the point r0 . We finally note that this last ellipsoid could be replaced by any other confocal ellipsoid lying outside S with an
appropriately chosen density.

6. Conclusions
In the present work, we have constructed image systems for the interior and exterior Neumann function for the Laplace equation in
spherical and ellipsoidal geometry. Because of the normal derivative that enters the Neumann boundary condition, it is not possible to
find an image system consisting of isolated singularities. Nevertheless, a combination of point images as well as one-dimensional and
two-dimensional continuous distributions can provide appropriate image systems for Neumann functions. Thanks to the high degree
of symmetry of the spherical system, point and line images are enough to represent the Neumann functions for the sphere. On the
other hand, for the case of the ellipsoid, besides point and line images, we need a surface distribution as well over an ellipsoid which
has to be confocal to the given ellipsoidal boundary. This is a consequence of the truly three dimensional character of the ellipsoidal
geometry as it compares with the one dimensional character of the sphere.

References
1. Lebedev NN, Skal’skaya IP, Uflyand YS. Problems in Mathematical Physics. Pergamon Press: Oxford, 1966.
2. Garabedian PR. Partial Differential Equations. John Wiley and Sons: New York, 1964.
3. Dassios G. The Kelvin Transformation in Potential Theory and Stokes Flow. IMA Journal of Applied Mathematics 2009; 74:427–438.
4. Dassios G, Sten JC-E. The Image System and Green’s Function for the Ellipsoid. In Imaging Microstructures. Mathematical and Computational
Challenges, Vol. 494, Ammari H, Kang H (eds), Contemporary Mathematics AMS, 2009; 185–195.
5. Sten JC-E, Dassios G. Image distribution and surface potential of a dipole in a one-shell conducting sphere. IMA Journal of Applied Mathematics 2010;
75:720–731.
6. Hobson EW. The Theory of Spherical and Ellipsoidal Harmonics. Cambridge University Press: Cambridge, 1931.
7. Whittaker ET, Watson GN. A Course of Modern Analysis, 4th edn. Cambridge University Press: Cambridge, 1962.
8. Thomson W. Extrait d’une lettre de M. William Thomson. Journal de Mathématiques Pures et Appliquées 1845; 10:364–367.
9. Neumann C. Hydrodynamische Untersuchungen. Teubner: Leipzig, 1883.
10. Lindell IV. Application of the image concept in electromagnetics. In URSI Review of Radio Science 1990-1992, Stone WR (ed.). Oxford University Press:
Oxford, 1993; 107–126.
495

11. Lindell IV. Methods for Electromagnetic Field Analysis. IEEE Press: New York, 1995.

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496
 G. DASSIOS AND J. C.-E. STEN

12. Choy TC, Alexopoulos A, Thorpe MF. Dielectric Function for a Material Containing Hyperspherical Inclusions to O(c2 ) I. Multipole Expansions.
Proceedings of the Royal Society of London A 1998; 454:1973–1992.
13. Choy TC, Alexopoulos A, Thorpe MF. Dielectric function for a material containing hyperspherical inclusions to O(c2 ) II. Method of images. Proceedings
of the Royal Society of London A 1998; 454:1993–2013.
14. Nayar BM. Neumann function for the sphere. I. Indian Journal of Pure and Applied Mathematics 1981; 12:1266–1282.
15. Dassios G, Hadziloizi D, Kariotou F. The octapolic ellipsoidal term in magnetoencephalography. Journal of Mathematical Physics 2009; 50:1–18.
013508.
16. Dassios G. The magnetic potential for the ellipsoidal MEG problem. Journal of Computational Mathematics 2007; 25:145–156.
17. Miloh T. The ultimate image singularities for external ellipsoidal harmonics. SIAM Journal of Applied Mathematics 1974; 26(2):334–344.
18. Colton D, Kress R. Inverse Acoustic and Electromagnetic Scattering Theory, 2nd edn. Springer: Heidelberg, 1998.
496

Copyright © 2012 John Wiley & Sons, Ltd. Math. Meth. Appl. Sci. 2012, 35 482–496