Computer Physics

Communications

Computer Physics Communications 68 (1991) 118

North-Holland

An introduction to the method of moments ~

E.H. Newman and K. Kingsley
The Ohio State University, Department of Electrical Engineering, ElectroScience Laboratory, 1320 Kinnear Rd., Columbus,
OH 43212 USA
Received 31 August 1990; in revised form 20 December 1990

This paper will present an introduction to the theory and application of the method of moments (MM) to problems of
electromagnetic radiation and scattering. The MM procedure for solving a linear operator equation, by transforming it into
a matrix equation, is reviewed. The integral equation and MM solution for an arbitrary perfectly conducting body is
presented, and then illustrated by the simple example of scattering by a 2D perfectly conducting strip. Numerical results are
used to illustrate the accuracy, convergence, and typical computer CPU times. The MM solution for radiation or scattering
by a rectangular dielectric cylinder is also presented. Finally, the internal resonance problem is discussed.

1. Introduction
This paper will present an introduction to a
numerical technique known as the moment
method or method of moments (MM), especially
as it applies to problems in electromagnetic radiation and scattering [1,2]. The MM is a numerical
technique for solving a linear operator equation
by transforming it into a system of simultaneous
linear algebraic equations, i.e. a matrix equation.
Over the last 30 years the MM has been extensively applied to virtually every area of electromagnetics including radiation and/or scattering
by perfectly conducting and material bodies, thin
wire antennas, aperture penetration, printed circuit structures, etc [3]. This paper will describe
some of the basic features of the MM, rather
than present an extensive bibliography of MM
research and applications,
In electromagnetics a moment method solution usually refers to a problem in which the
MM is used to solve a linear integral equation for
a current distribution representing a body. This is
*

This work was sponsored by the Joint Services Electronics

Program under Contract N00014-78-C-0049 with the The
Ohio State University Research Foundation.

0010-4655/91/$03.50 1991

basically a two-step procedure: The first step is to

obtain the integral equation. Basically, the surface or volume equivalence theorems are used to
replace the body by free space and by equivalent
currents [4]. The use of these equivalence theorems is crucial in that it allows the integral equation and MM solution to be formulated entirely
in terms of the free space fields of currents. The
integral equation for the equivalent currents is
obtained either as a statement of the boundary
conditions of the problem, or as a statement ol
the equivalence theorems used. The second ste
is to solve the integral equation by the MM. Jr
brief, the unknown current is expanded in tenn~
of an appropriate set of basis functions. If A
terms are retained in the expansion for the cur
rent, then N weighted averages of the integra
equation are enforced. This MM procedur
transforms the integral equation into an order J~
matrix equation for the N coefficients in th
expansion for the current. Once the current i~
known, most parameters of engineering interest
such as input impedance or radiated fields, cai
be found in a straightforward manner, since the
involve only the free space fields of the currents
The main advantage of MM solutions is hig}
accuracy. The MM is a direct numerical solutio]

Elsevier Science Publishers B.V. All rights reserved

E.H. Newman, K Kingsley

An introduction to the method of moments

of the (usually) exact integral equation. All phenomena of the problem are inherent in the integral equation, and are thus automatically included in the MM solution. For example, MM
solutions in electromagnetics automatically indude surface waves, creeping waves, multiple
diffractions, shadowing effects, etc. A second advantage of the MM is that it is capable of dealing
with very complex geometries. In fact, several
user-oriented computer codes have been written
which can treat geometries as simple as a dipole
or as complex as an airplane [510].
The main limitation of the MM is a result of
the fact that N, the number of terms which must
be retained in the expansion for the current in
order to obtain reasonable accuracy, is proportional to the electrical size of the body. The CPU
time to set up and store the MM matrix equation
is proportional to N2. The CPU time to solve the
MM matrix equation is proportional to N3 [11].
Thus, as the frequency is increased, the required
computer power also increases, and at some point
it becomes so large that the MM solution is
impractical. However, it should be emphasized
that this is a machine or hardware limitation, and
that as computers become more powerful, it will
be practical to apply the MM to larger problems.
Consider the example where the unknown is
the current on the surface of a 3D body. Accounting for both polarizations of the vector current,
typically N will be on the order of 100 unknowns
per square wavelength of surface area. Thus,
treating a body of 10A2 would require dealing
with an order N = 1000 matrix equation. Setting
up and solving the MM matrix equation typically
will require about an hour of CPU time. Storage
of the MM matrix equation will involve more
than N2 = 1 million complex numbers. Note that
doubling the frequency will increase the surface
area in A2 and thus N by a factor of 4. This will
increase the required storage by a factor of 16,
the CPU time to set up the MM matrix equation
by a factor of 16, and the CPU time to solve the
matrix equation by a factor of 64. Thus, the MM
is often referred to as a low frequency technique,
applicable when the body is not electrically large.
Section 2 of this paper describes the MM
procedure, and discusses the important question

of convergence. Section 3 describes the MM solution for an arbitrary perfectly conducting body,
and then illustrates the method by the simple
example of scattering by a perfectly conducting
cylinder. Numerical data will illustrate the accuracy, convergence, and CPU times for the MM
solution. Section 4 presents the MM solution for
radiation or scattering by a dielectric body. Finally, section 5 will briefly describe the internal
resonance problem.
In this paper all electromagnetic fields and
currents are considered to be time harmonic,
with the e3~~t
time dependence suppressed. Also,
all CPU times are for a VAX 8550 computer
which is about six times faster than a VAX
11/780.

2. The method of moments

2.1. The MM procedure

This section will describe the mathematical
procedure known as the method of moments
(MM), that is the procedure for solving a linear
operator equation by transforming it into a systern of simultaneous linear algebraic equations,
commonly referred to as a matrix equation. The
description given here for the MM largely follows
that originally presented by Harrington [1,21.
An inhomogeneous linear operator equation
can be written as
L ( f)

g,

(1)

where L is a known linear operator, g is a known

source or excitation function, and f is the unknown response function. In electromagnetics f
is typically an equivalent current which produces
the radiation or scattering from the body, L(f) is
typically the electromagnetic field of the current
f, and g is a known incident field. The linear
operator equation is typically an expression of
either the boundary conditions of the problem or
of the equivalence theorems [41used to define
the current f.

E.H. Newman, K Kingsley

An introduction to the method of moments

The first step in the MM solution of eq. (1) is

to expand the unknown function f as

results in
N

~ a,,KLf,,,

f~fN

(2)

Wm)

=Kg, Wm),

1, 2,...,N.

n=1

(5)

n~1

where fN is an N term expansion or approximation off, the f,, are a series of N known linearly
independent expansion or basis functions in the
domain of L, and the a~ are a series of N
unknown constants to be determined by the MM.
Substituting the fN expansion of eq. (2) into the
original operator (1), and using the linearity of L
yields
N

LfN= >.aL(f)g.

(3)

Equation (5) can be written as

(LfN

Wm)=

0,

1, 2,..., N.

(6)

Thus, the MM determines the a~so that the

difference between U N and g is orthogonal to
the N weighting functions.
Equation (5) can be recognized as N simultaneous linear algebraic equations in the N unknowns a~,n = 1, 2,..., N. It can be more cornpactly written in matrix form as
[L]A=G,
(7)

n=1

For finite N, f in eq. (2) will generally not be

a complete expansion for f~
and thus in practice
it is not possible to choose the a~to exactly
satisfy eqs. (2) or (3). Instead, the a~are chosen
so that N weighted averages of eq. (3) are satisfied. Define a series of N linearly independent
weighting or testing functions in the range of L,
denoted Wm, m = 1, 2,.. N. If the weighting
functions are chosen identical to the expansion
functions, i.e. Wm fm then the MM solution is
referred to as Galerkins method. The bracket
notation Ku, Wm) will be used to denote the inner
product between the functions u and Wm~ Typically, the inner product is taken as

where [LI is an N X N coefficient matrix, G is

the length N right-hand-side vector, and A is the
length N solution vector which contains the a~
from eq. (2). Typical elements of [LI and G are
given by
Lmn = KLf~,Wm), m, n = 1, 2,..., N,
Gm = Kg, Wm), m = 1, 2,..., N.

(8)
(9)

.,

Ku,

Wm)

f uw,~dr,
m

(4)

where the integral is over the region (line, surface, or volume) of Wm and the * implies complex
conjugate. In Harringtons original description of
the MM the weighting function was not conjugated. Although the choice is arbitrary, here we
include the complex conjugate so that we can
directly use the results of recent work concerning
the convergence of Galerkins method.
Taking the inner product of both sides of eq.
(3) with each Wm and using the linearity of U

Using standard matrix algebra, eq. (7) can now be

solved for the solution vector A, which when
substituted into eq. (2) provides an approximation
to f. In electromagnetics, Galerkins method results in a symmetric [LI matrix. Note that the
matrix [LI is independent of the excitation function g. Thus, an advantage of MM solutions is
that a relatively small effort is required to obtain
the solution for a second or subsequent excitation, since on the first solution one must set up
and LU decompose [LI.
2.2. Convergence of the MM
One of the most important questions concerning the MM is that of convergence. That is, as
N
co, does fN
f? Very little can be said
concerning the convergence of the general MM
in which weighting functions can be chosen differently from the expansion functions, and thus

El-I. Newman, K Kingsley

An introduction to the method of moments

the remarks in this section will refer to Galerkins

method in which Wm fm~
The convergence of Galerkins method is dependent upon the properties of the operator U
[12,13]. If L is positive, i.e. if KLf, f) > 0 for all
f * 0, then it can be shown that f converges in
energy to f [131, i.e.
tim ~KL(f~f),

(f~f))

0.

(10)

Note that convergence in energy does not imply

that f converges to f.
In electromagnetics the U operators are not
positive, and thus nothing can be said concerning
the convergence of Galerkins method or the
MM. However, experience has shown that with a
reasonable choice of expansion and weighting
functions the MM does converge, and in fact is
often used as a reference solution and referred to
as an exact method. Unfortunately, the choice of
reasonable expansion and weighting functions
remains somewhat of an art. Basically, one wishes
to choose expansion functions which:
1. incorporate as many of the known properties of the unknown function f as is possible. For
example if f is continuous, then it is desirable to
choose expansion functions which are continuous.
If f is zero at the boundaries, then it is desirable
to choose expansion functions which are zero at
the boundaries. If f is not zero at the boundaries, then using expansion functions which are
zero at the boundaries would be considered an
unreasonable choice, and the MM would not
be expected to converge.
2. permit the inner products in eqs. (8) and (9)
to be evaluated with reasonable ease.
In practice the choice of basis functions is a
compromise between the above two criteria,
Choosing very simple basis functions often reduces the difficulty in developing the MM code,
however, the code may be slowly converging and
require a relatively large N to achieve accurate
results. On the other hand, one can choose very
sophisticated basis functions which yield accurate
results with a relatively small N. However, this
can increase both the man hours required to

develop the code and also the CPU time to

evaluate the inner products.
2.3. The method of least squares
This section will show that the original operator equation can be modified so that the linear
operator is positive, and thus convergence in energy is guaranteed. The method employs the adjoint operator, L~,defined by
KLu v) Ku LaD)
11
=

Operation by U~on both sides of eq. (1) yields

the new operator equation
L~U(f)=L~(g).
VU is a positive operator since
KLaLU, u)

if

KLu, Lu)

(12)

II Lu 112> 0,

~* 0

(13)

and thus the Galerkin Method solution of eq.

(12) will converge in energy. The elements in the
matrix equation (7) are given by
Lmn

Gm

KLf~,LfTh), m, n 1, 2,..
Kg, Lfm), m 1, 2,..., N.
=

.,

N,

(14)
(15)

Equations (14) and (15) can be recognized as the

method of least squares which can be derived by
direct minimization of II Lf
g I [14]. Comparison with eqs. (8) and (9) shows that the method of
least squares is equivalent to a MM solution of
the original operator (1), but with weighting functions chosen as Wm Lfm. The main disadvantage
of the method of least squares is that the matrix
elements are typically more difficult and time
consuming to evaluate.
Figure 1 shows a comparison of the mean
square error, I
II 2, versus N in the solution of
a simple differential equation by Galerkins
method and by the method of least squares [151.
Note that the error in the least squares solution
monotonically decreases with increasing N, while
that for Galerkins method does not. Also, the
error for the least squares solution is always
lower than that for Galerkins method, however,
the difference is small. In electromagnetics the

E.H. Newman, K Kingsley

I

An introduction to the method of moments

-

ci O~lxI<1/2

dx2~~i
d

f(i)=f(1)=O

~.

f,, =( I x 2n~
~ n=i ,2,....N

I-

0
I..
L

ca

Galerkins Method
Mnlmum Mean

the surface S. Here (js0, e0) are the constitutive

parameters of free space, and is the unit outward normal to the closed surface S. When radiating in free space, (Ji, M) produce the known
incident fields denoted (E, H). If the impressed
sources are very close to the conducting body,
then the geometry of fig. 2a is referred to as an
antenna or radiation problem. If the impressed

a,

o~

\~quoreError

C
0
--

.-~

8..

4
N

,~

No. Terms

Fig. 1. A comparison of the mean square error in a Galerkin

method and a least square error solution of a differential
equation,

method of least squares is employed far less than

the MM, probably because the advantage in the
increased rate of convergence does not justify the
increased computational effort.

currents
are
planeare
waves,
far removed,
and fig. then
2a is the
a scattering
incident fields
problem.
As illustrated in fig. 2b, the first step in obtaining the integral equation is to use Schelkunoffs
surface equivalence principle [4,16,171 to replace
the perfectly conducting body by free space and
the equivalent electric surface current
J=FIxH onS.
(16)
The free space fields of the equivalent current J
are referred to as the scattered fields, and are
denoted (Es, Hs). In the equivalent problem of
fig. 2b, the total fields are the superposition of
the free space fields of (J, M) and J, i.e.
E = E + Es,
H=H+Hs.

(17)
(18)

3. Perfectly conducting bodies

It is important to emphasize that in the equivalent problem of fig. 2b, all currents radiate in free

This section will outline the integral equation

and MM solution to the problem of radiation or
scattering by a perfectly conducting body. First,
the solution for a general perfectly conducting
body will be presented, and then the method will
be illustrated by the simple problem of plane
wave scattering by a perfectly conducting cylinder.

space. As a result, the integral equation and MM

solution for J is formulated entirely in terms of
the free space fields of currents.
The electric field integral equation (EFIE) is a
statement of the boundary condition that the
total electric field tangential to S must vanish,
_..jz

X Es

XE

on S.

(19)
n

3.1. The integral equation

This section will obtain the electric and magnetic field integral equations for the current distribution on a perfectly conducting surface. The
general problem is illustrated in fig. 2a where the
assumed known impressed currents (J, M) radiate the unknown total fields (E, H) in a
medium which is free space except for the presence of a perfectly conducting body enclosed by

s._

(~)

(~,H)
~

(0.0) ~
S...

(/.Lo,Co)
(a)

(b)

Fig. 2. (a) The impressed currents (J, M) radiate in the

presence of a perfectly conducting body. (b) The body is
replaced by the equivalent electric surface current J.

E.H. Newman, K Kingsley

An introduction to the method of moments

Equation (19) is referred to as an integral equation for J, since the scattered field can be written
as

Es

(ft. G ds,

(20)

where G is the free space electric dyadic Greens

function [18].
The magnetic field integral equation (MFIE) is
a statement of the surface equivalence theorem
(16). Inserting eq. (18) into (16) yields

ii X H Ofl 5~
(21)
where S is a surface an infinitesimal distance
exterior to ~
In the above derivations of the EFIE and the
MFIE, the use of Schelkunoffs surface equivalence principle requires that S be a closed surface. For a closed surface, such as a sphere or a
closed box, J is the current flowing on the extenor of S, and there is no current on the interior.
However, for an open surface, such as a zero
thickness plate or a box with one side removed,
there will in general be different surface currents
flowing on either side. By considering a zero
thickness plate as the limiting case of a closed
box as the thickness goes to zero, it can be shown
that when the EFIE is applied to open surfaces,
J will be the vector sum of the current on the two
sides [19]. Since it is the vector sum current which
radiates the scattered fields, the EFIE is applicable to closed as well as open surfaces. By contrast, when the MFIE is applied to open surfaces
the resulting J is the vector difference between
the top and bottom currents. The vector difference current has no use except in the special case
where the current on one side is zero, i.e. for
closed surfaces. Thus, the MFIE is only applicable to closed surfaces.
X

JJS

~J

3.2. The MM solution

This section will describe the MM solution of
the basic EFIE of eq. (19). Comparing eq. (19) to
(1) it can be seen that:
the excitation function g is the incident electric
field tangential to 5;

the unknown response

is the surface current

J flowing on S;
the linear operator L is minus the tangential to
S component of the free space electric field of
an electric current.
The first step in the MM solution is to expand
the unknown current in terms of some basis functions. Thus we expand J as

N
j ~jN

~ J,,j,,,

(22)

where the .1,, are a sequence of N known linearly

independent vector expansion functions, and the
I~are a sequence of N unknown complex coefficients (n = 1, 2,
N). Note that the J~must in
general account for both components of the vector current J on S. Denoting U(J~)= E,, as the
free space field of .1,,, eq. (19) becomes
. . . ,

~ IRE,,

XE

(on 5).

(23)

Now define Wm (m = 1, 2,.

N) as a sequence
of N linearly independent weighting functions on
5, and with vector direction tangential to S. Taking the vector inner product of both sides of eq.
(23) with the Wm and integrating over S yields
. . ,

i~ff~

Wm

ds

= ffEi

w ds,

m=1,2,...,N.

(24)

Equation (24) is a system of N simultaneous

linear equations, which can be more compactly
written in matrix form as

[z]I

v,

(25)

where, in analogy with Ohms law, [Z] is the

order N impedance matrix, V is the length N
voltage vector, and I is the length N current or
solution vector which contains the I,, in eq. (22).
Typical elements of [Z] and V are given by
Zmn

Vm

hfm~

Wm

fJi~

Wm

ds,

ds, m

m, n

1, 2,.. .,N,
(26)

1, 2,..., N,

(27)

E.H. Newman, K Kingsley

An introduction to the method of moments

where the integration is over that portion of S

where Wm is non-zero. Note that the matrix equation (25) requires only the evaluation of the free
space fields of known currents. If the weighting
functions are considered as surface currents with
units A/m, then the elements of [ZI and V have
units VA, and the I,, are dimensionless. Equation
(25) can now be solved for I, which when substituted into eq. (22) provides an approximation to
J. Once J is known most parameters of interest,
such as radiated or scattered fields, can be easily
computed since they are simply the free space
fields of ~~1
3.3. TM scattering by a perfectly conducting strip
This section will present the MM solution of
the EFIE, eq. (19), for the relatively simple 2D
problem of TM scattering by a perfectly conducting strip. In particular, it is desired to obtain
simple exPressions for the elements in the MM
matrix equation that can be coded with a minimum of effort, and evaluated with a minimum of
CPU time. Figure 3 shows a TM (to z) polarized
plane wave incident upon a perfectly conducting
strip of width D. The incident electric field is
=

(28)

2ejkc0js~~~j),

where b, is the angle of incidence and k = 2 IT/A

is the free space wavenumber. For the TM polarization, the electric fields and currents are purely
2 directed. Thus, the vector notation will be
dropped and the 2 component is understood. For
example, the vector EFIE, eq. (19), reduces to
the scalar EFIE
=

E on S.

(29)

Following eq. (22), the current on the strip is

expanded as
N
j

jN

(30)

=~

where as described in section 3.1, J is the sum of

the surface current on the top and bottom surfaces of the strip. In order to define the MM
basis functions, the strip of width D is split into
N smaller strips or segments of width d = D/N,
and with center point Xn~ The expansion functions are chosen as the simple piecewise constant
or pulse functions
1
J,, = ~A/m

on segment n,

~d x

+ ~d,

Xn

= 0,
otherwise.
(31)
Note that the ~n have been normalized so that
they have 1 A total current. For the pulse expansion to be accurate, d must be chosen small
enough that the current is essentially constant in
each strip. Typically this requires strips of width
0.05A d 0.25A, with the accuracy increasing as
d/A decreases.

In order to find Zmn, we will first find E~.The

free space electric field of a unit amplitude line
source is
EL

2A H~)(k I

I),

(32)

where the characteristic impedance of free space,

~/j.~

0/~0
377
and I p p I is the distance
from the line source to the field point. Using
=

~,

primed and unprimed coordinates to denote the

source and field point, respectively,
______________________________

Using eq.= (32)

~
and superposition,
2pp cos(4 the~).free space
(33)

Conducting Strip

field of
1hi2131

IppI

TM Scattering by a Perfectly

Pulse Basis

PC. Strip

~n

is

E~(x,y)=

2A ~
n

Fig. 3. Geometry for the MM solution of a TM plane wave

incident upon a perfectly conducting strip of width D.

(34)

E.H. Newman, K Kingsley

An introduction to the method of moments

where the integral is over the width of segment n,

i.e. x,, ~d <x x,, + ~d. Employing Galerkins
method, with weighting functions Wm ~m the
elements in the MM matrix equation, eqs. (26)
and (27), reduce to

ality, fig. 4 shows the expansion mode J,, centered

about the y axis. Evaluation of the self
impedance, Z,,,,, requires finding the electric field
of .1,, directly on J,,. The electric field of J,, at a
field point
d/2 x d/2, y = 0) on J,, is

Zmn=_fEnJmdX
m

En(X)

ITfl

2A

J~H~2~(kIxxI)dx

d/2

d/2

ITll

ITll I
=

xl) dx] dx, (35)

2Ad2~[~H0

Vm=JEJmdX=_

2Ad
~d/2

H~2~(k(xx))dxl.

(37)

ix

ej~c~0~~1dx

[If-d/2 H~2~(k(xx))dx

dL

2 sin(kd cos 4~/2)

(36)

eikc~0s*,

kd dOS

where Xm is the center point of segment m, and

m, n = 1, 2,..., N. Since this is a Galerkin method
solution, the impedance matrix is symmetric, i.e.
Zmn = Znm.
As is typical for MM solutions, the computation of the N elements of the voltage vector, V, is
straightforward and fast (small CPU time). By
contrast, the computation of the N2 elements of
the impedance matrix is more complex, and can
require a great deal of CPU time. The evaluation
of the Zmn involves a double integral which must
either be done numerically or by some approximate method. In addition, for self impedance
(m = n) and adjacent mutual impedance (I m
n I = 1) terms, the integrand is singular and can
not be evaluated by straightforward numerical
integration. Below, we will present simple expressions for the Zmn which can be evaluated without
the need for numerical integration,
The self impedance terms are typically the
most important and most difficult terms to evaluate in the MM [ZI matrix. Without loss of gener-

Note that the integrands in eq. (37) are singular

when x =x, and thus they can not be evaluated
by straightforward numerical integration.
Since d ~zA, one method for treating the singularity is to replace the Hankel function by its
small argument approximation, and then perform
the integrals analytically. Using the approximation [201
H~2~(u) (1

jO.0738)

j2
in u,

IuI

IT

4Z

1,
(38)

in eq. (37) and integrating yields

En(X)

ITfl 1/

2Ad

II.~
(c +j
IT)I

j2
d

IT

((x + ~d) ln k(x

+ (~d x)

ln k ~d

+ ~-d)

)]~

(39)

where the complex constant C = 1 + jO.0738. Note

that E~(x)in eq. (39) is well behaved and contains no singularities. Finally, inserting (39) into
(35) and integrating yields
=

fd/2 En(X)Jn
d/2

dx

3\
j2
1
IlC+j~lnkdI.
2A
IT /
IT
j
IT7J1/

_______________

d/2

d/2

I i
3d/2

Fig. 4. Geometry for the computation of the self and touching

impedances.

[~

(40)

Figure 5 shows the self impedance, Z,,,, = Rn,, +

jX,,,,, versus d/A computed by eq. (40) and by the

E.H. Newman, K Kingsley

An introduction to the method of moments

the voltage vector. Using standard matrix algebra,

eq. (25) can then be solved for the current vector,
I, which when substituted into eq. (22) yields jN

TM Strip Self Impedance

Present Method

_____

Xflfl

on the strip. The next section will discuss the

1/
-

computation
which
is an approximation
of the far zonetoscattered
the true fields
current
andJ

-.

echo width of the strip.

_________________________
~
~

3.3.1. Far zone fields

In scattering problems, normally the parameter ofwidth.
interestThe
is the
far zone scattered
fields or
echo
computation
of the scattered

d/A
Fig. 5. A comparison of the self impedance, Z,,,, = R,,,, +jX,,,,
computed by eq. (40) and by the more exact method of Wang
121].

fields is straightforward since by definition the

scattered fields are the free space fields of J.
Replacing J by its N term approximation jN

R~

more exact methods of Wang [21].Note that eq.

(40) is reasonably accurate, even for strips as
wide as kd = +IT.
Referring to fig. 4, adjacent modes have a

common endpoint, and thus their mutual

impedance will also involve a singular integrand.
Using the same small argument approximation

for the Hankel function, the mutual impedance

between adjacent modes is
Zn ,n + 1

f3d/2E,,( x) J~

dx

E5

~ mEn,

(43)

P1

where the coefficients I,, (n

= 1, 2, . .
N) are
evaluated by the MM solution described in section 3.3, and E~is the free space electric field of
the basis function, J~.
This section will derive simple expressions for
the scattered fields in the far zone, i.e. in the
limit as kp oo In general, E~is given by eq.
(34). In the far zone, the Hankel function can be
replaced by its large argument approximation [20]
.,

d/2

ITfl

j2
(2ln2kdln kd)
IT

H~2~(u ~V

(41)
.

If modes m and n do not touch m n I > 1)

then Zmn can be evaluated using eq. (35) and
numerical integration. However, if modes m and
n are not too close, then Zmn can be approximated by
(I

ITU

>>

(44)

In the far zone p >>p, and eq. (33) for the

distance between source point x on the strip and
the field point (p, 4) reduces to
__________________

I~p I

2pp cos(4~ 4)

~p x cos 4i.

(45)

Using eq. (45) to approximate the exponential

phase term in eq. (44), and 1/I p p I i/p for
the amplitude, the far zone approximation for the

Zmn

2~k(Ixm x,, I),

~~H
1~

I xm

X,~I >> d.

(42)

A simple MM matrix equation (25) can be set

up using eqs. (40), (41) and (42) for the computation of self, touching mutual, and non-touching
mutual impedances, respectively, and eq. (36) for

Hankel function becomes

H~2~(k
I pp
k~I >> 1

I)

ejTT/4 ~

and p

p.

(46)

10

E.H. Newman, K Kingsley

An introduction to the method of moments

Inserting eq. (46) into (34) and integrating yields

I

E~(p,4)
=

e_Jkp

dx

fei~c0~

___________

~o.

,,

o~

e~4ije~j
2V~ ~

Exact Elgenfunction

2dA~/~k

MM

II

(47)

Strip

Note that in the far zone E~is proportional to V

1~,
of eq. (27), and in fact eq. (47) could have been
derived from reciprocity rather than by direct
integration. Since the p dependence of E,, (p, qS)
in the far zone is simply

~._

-~

2ID

3b

~Exact

E~(p,4) is usually written as

55

Eigenfunction

(48)

where EflF(4) is the far zone electric field of J,.

~.

__________________________

Comparing eqs. (47) and (48) gives

=

4~ 5~ ~

MM

E~(p,~)= e

EflF(4)

________________________

~1~Je~~E,,(p,
~

4~

2VX
e3TT/

l,(4).

~
The total far zone scattered electric field is

W has dimensions of meters, and is often expressed in terms of dB m which is obtaine4by

e~~4~
N

Fig. 6. A convergence curve for the current in the center of

the strip.

taking 10 log
10 W.

E~(~

n~1 ~

EflF( ~

2V~

n~1 ~

3.3.2. Numerical results

(50)

This section will present numerical results

based upon the above MM solution for TM scat-

For 2D problems, the power density of the far

tering by a perfectly conducting strip. This data

zone scattered field is usually expressed in terms

of the echo width. The echo width, W, is defined
as a width, when multiplied by the power density

will include strip current distribution, echo width,

and CPU times, and will be designed to illustrate
the accuracy and convergence of the MM solu-

of the incident wave, that would yield sufficient

power that could produce by isotropic radiation

tion. As illustrated in the insert in fig. 6, all data

in this section will be for a 1 A wide strip illumi-

the same radiation intensity as that in a given

direction from the scattering object. For the unit
amplitude incident plane wave being considered
here, the echo width is related to the far zone
field by

nated by a TM wave incident from the angle

4~=
with respect to the x axis.
The MM solution basically determines JN(x)
an N term approximation to the current induced
on the strip by the incident field. Figure 6 shows
the magnitude and phase of the current in the
center of the strip, J(x = 0), versus N = the

W( 4)

2IT~E~(4)

I 2.

(51)

450

E.H. Newman, K Kingsley

1.

An introduction to the method of moments

I

I,1.11,1.I.1.1.

f300

11
I

I
I

Mhz

~~1~

:
E

Strip J( x)

~.

~
____

Ii

Exact Eigenfunction

S. ~

o~o d.

-.5-&4-&3-&2-&I

Strip

____

______

MM
N=20
_Exact

1~

O~2 O~3O~405

Eigen.

x/X

-~

Fig. 7. A comparison of the N = 20 MM and eigenfunction

solution for the current on the strip.

I.
I

I
20

40

60

80

number of pulse basis functions used in eq. (30)

to expand the strip current. The dashed line

shows the exact eigenfunction solution for the
strip current [22,231.For small N, the MM solu-

100

120

140

160

180

~O(Deg)
Fig. 9. A comparison of the N = 20 MM and eigenfunction
solution for the bistatic scattering from the strip.

tion is erratic, however, as N increases it con-

verges almost exactly to the eigenfunction solution. For N = 20, which, corresponds to a segment

size of 0.05A, fig. 7 shows a comparison of the

magnitude of strip current computed by the MM

and the eigenfunction solution. The agreement is

very good near the center of the strip, however, it
does worsen near the edges where the true current has a 1/ V~singularity (s = distance to the
strip edge).

In most cases one is more interested in the

echo width than the strip current. Figure 8 shows

the bistatic echo width at 4, = 1350 versus N.
Note that it converges uniformly to the eigenfunction solution shown as the dashed line. Figure 9 shows a comparison of the eigenfunction
and N = 20 mode MM solution for the bistatic
scattering pattern of the strip. The agreement is
less than a few tenths of dB for all angles.

I,

MM CPU Vs. N
Q)

=300 Mhz
~

Sn

~i.

ES

a-

Next ES

Strip

I-

~o ~

x
Io

O~O

[z]

D5

x
do

~o do

~o

No. of MM Modes

Fig. 8. A convergence curve for the bistatic echo width of the

strip.

05

l~0

05

Log

2,0

2.5

Fig. 10. The CPU time on a VAX 8550 forvarious parts of the
MM solution.

12

E.H. Newman, K Kingsley

An introduction to the method of moments

The final set of data will illustrate the CPU

time for the strip MM solution. As a function of
N, fig. 10 shows the CPU time to:
1. compute the N x N impedance matrix, [ZI;
2. make the first computation of the scattered
field, Es, which involves (1) computing the
voltage vector, V, using eq. (36), (2) LU decomposing the [Z] matrix, (3) back substituting
to find the solution vector
and (4) finding
the far zone scattered fields using eq. (50);
3. make a second or subsequent computation of
Es which requires the above 4 steps except
step (2), the LU decomposition of [z I
For large N, the CPU time is dominated by
the first computation of the scattered field. Of
the four steps listed in item 2 above, the LU
decomposition of the [Z I matrix requires by far
the largest CPU time. For large N the first ES~~
curve is nearly a straight line with slope
3.0,3
indicating
thatnext
the largest
LU decomposition
an Nto
process. The
CPU time isis that
compute the impedance matrix [ZI. Since there
are N2 elements in [ZI this is an N2 process.
For large N, the slope of this curve is about 1.94.
It should be noted that the expressions used here
to evaluate the [Z] matrix are extremely simple
and fast to evaluate. In a more typical situation
the computation of the elements of [ZI require
one or more integrals which must be evaluated
numerically. In this case, the CPU time would be
dominated by the computation of the [Z I matrix
for small N, and the LU decomposition for large
N. The smallest CPU time is that for a second or

m,

subsequent computation of the scattered field,

since the [Z] matrix has already been LU decomposed. However, the total CPU time to
5 compute
must be
aevaluated
pattern at
can
be
significant,
since
E
many angles. For example, computation of a backscatter pattern at 10 steps for
N = 300, will require 3 s to compute the [ZI
matrix, 33 s for the first computation of E5, and
360 X 0.63 s = 227 s for the remaining 360 angles.

(.~)

(~)
(~

0,e0)

(a)

~ (~)
(~~~)/

(b)

Fig.
11. (a) The
(J, M)
thematerial
fields (E,body
H) in
the presence
of sources
a material
body;radiate
(b) the
is
replaced by free space and the equivalent current (J, M).

scattering by a dielectric and/or ferrite material

body. The method will be illustrated by the relatively simple problem of TM scattering by a rectangular dielectric cylinder.
4.1. The volume integral equation
This section
will obtain
volume
integral
equation
for scattering
by a the
material
body.
The
general problem is illustrated in fig. ha where
the assumed known impressed currents (Ji, M)
radiate the known incident fields (E, W) in free
space, and the unknown total fields (E, H) in a
medium which is free space except for the presence of a possibly lossy and inhomogeneous dielectric/ ferrite material body with constitutive
parameters (p., ) in the volume V.
As illustrated in fig. lib, the volume equivalence theorem is used to replace the material
body by free space and by the equivalent electric
and magnetic volume polarization currents [41
J=jw(e
0)E

M = jw(p.

mV,

(52)

p.0)H in V.
(53)
Note that J and M exist only in the volume V
where e ~ and p. # p.,~,respectively. In the
equivalent problem of fig. lib the total fields at

any point in space (interior or exterior to V) are

the superposition of the free space fields of the
impressed currents and the equivalent currents,
i.e.

E=E+E+EM,

(54)

H=H1+Hj+HM,

(55)

4. Material bodies
This section will outline the volume integral

equation and MM solution to the problem of

where (E, H) and (EM, HM) are the free space

fields of J and M, respectively.

E.H. Newman, K Kingsley

An introduction to the method of moments

The volume integral equation is obtained by

enforcing the volume equivalence theorems in V.

Solving eqs. (52) and (53) for E and H, and
substituting the results into eqs. (54) and (55)
yields
I
j0(e e0)

__
M

jw(p. p.o)

in V,

(56)

ME

mV.

-HM=H

(57)

TM Radiation or Scattering
by a Dielectric Cylinder

0,y0) ~
\PI___________

tI.~J
I
I

jw(e

~)

jw(

E~+

E~r+ jw(e

(58)

E,,

__________

e0) =Ez

for (ix J~,J~)in V. Figure 12 shows a rectangular dielectric volume which has been segmented
into a number of smaller rectangular cells for the
purpose of defining the MM expansion functions.
Each cell contains three MM basis functions,
corresponding to 2, .9, and I components of J.
Thus, M cells will result in N = 3M unknowns.
Employing the MM with three orthogonal vector

c.N

E
_______

_______

Equations (56) and (57) are a pair of coupled

vector integral equations for (I, M). For nonmagnetic material (p. = p.
0 and thus M = 0), eq.
(56) is equivalent to the three scalar equations

13

_______

Fig. 13. A rectangular dielectric cylinder

is segmented into N

rectangular cells corresponding to the MM expansion modes.

weighting
eachequation.
cell will reduce eq.
(58)to an functions
order 3M in
matrix

4.2. TM dielectric cylinder

This section will outline the MM solution for
TM radiation or scattering by a rectangular dielectric cylinder. Figure 13 shows a rectangular
cylinder of height H, width W, and permittivity e.
A lossy dielectric is described by its complex
permittivity
E=ErEO(l

j tan 6)

~r~0J

(59)

where Er is the relative permittivity, tan 3 is the

loss tangent, and r is the conductivity. For inhomogeneous dielectrics e will be a function of
position. The dielectric is shown excited by either
a TM polarized plane wave with incident electric

field given by eq. (28), or by a unit amplitude

(I = 2 A) line source located at (x0, y0) and with
incident electric field

E=I 2:~IlHO(2)(kPl),

Pulse Basis Functions in a

R.ctanguior Dl.iectric Volum.

Fig. 12. A rectangular dielectric body is segmented into smaller

rectangular cells corresponding to the MM expansion func-

tions.

(60)

where the local Pi is the distance from the line

source to the field point.
For a TM to z polarized incident electric field
all electric fields and electric currents will be 2
directed. Thus, we can now drop the vector notation, and the 2 component is understood. The

14

E.H. Newman, K Kingsley

An introduction to the method of moments

general vector integral eq. (56) reduces to the

scalar equation

wm=6(xxm) 6(yym),

E +

in R,

~ ~

(62)

n l

where the .1,, are a sequence of N known linearly

independent expansion functions, and the I,, are
a sequence of N unknown complex coefficients
(n = 1, 2,..., N). To define the expansion funclions, the rectangular cylinder is segmented into
N smaller rectangular cells. If we denote R~ as
the cross-section region of cell n, then the piecewise constant or pulse expansion functions are
defined by

J~=
0

(66)
where (xm, Ym) is the center of cell m. Multiplying both sides of eq. (65) by the Wm (m =
1, 2,
N) and integrating over R will reduce
eq. (65) into an N x N system of simultaneous
linear equations which can be compactly written
in matrix form as

[Z+~Z]m=v.

jN

m=1,2,...,N,

(61)

for the 2 component of ~f. Since this is a 2D

problem, eq. (61) applies in the cross-sectional
region R of the cylinder.
The first step in the MM solution of eq. (61) is
to expand the unknown volume polarization current as
j

The weighting functions will be Dirac delta

functions located at the center of the N cells,

2 in R,,,
A/rn
otherwise,

(67)

Using the sampling property of the Delta function, the elements of the MM matrix equation are

simply
Zmn

~En(Xm,

(68)

Ym),

1
=

=
=

jA(0(E,,
0, m ~ n,

(69)
for general excitation,

E( Xm, y,,,)
e jk(Xm cos b

(70)

1 y,,,sin 4I)

for
wave excitation,
2A plane
H~2)(k~/(x
2
+
~Xm)
0

(Yo

ym)2) (71)

(63)

where A is the cross-section area of the cells. ~

was the case for the strip modes defined in eq.
(31), the dielectric expansion modes are normalized to have unit total current.
The free space electric field of ~ at the field
point (x, y) is

for line source excitation,

(72)
for m, n = 1, 2,.. N. Note that the only place
that the permittivity enters the MM solution is in
the evaluation of the diagonal [z~Z] matrix. Inhomogeneous dielectrics are treated by simply using
e,,, the value of
at the center of cell n, in
evaluating the z~Z,,,,.Also note that the difference between plane wave and line source excita.,

1T77

E,,(x,

y)

2A

~(

fjJ,,H~2)(kp) ds,

(64)

where p =
x x ~2 + ~
~r 2 is the distance
from the source point to the field point and the
double integral is over R,,, the region of cell n.
Using this notation, eq. (61) becomes
N

m,,E,,

+ ~
~

m
j(e

e
0)

in R.

(65)

tion is a simple change in the voltage vector, V.

Choosing the weighting functions as Dirac
delta functions is equivalent to enforcing the integral equation at N points at the center of the N
cells, and is therefore referred to as a point
matching MM solution. The advantage of the
point matching solution is that it simplifies the
evaluation of the MM matrix equation. For example, evaluating the Zmn for point matching requires a double integral to find E,,(x~, Ym)~By

E.H. Newman, K Kingsley

An introduction to the method of moments

contrast, in a Galerkin MM solution

15

(Wm = ~m) a
quadruple integration would be required to find
the Zmn. The disadvantage of the point matching
solution is that, as compared to Galerkins
method, it tends to be more slowly converging,
The double integral required to find E~or
Zmn must be done either numerically or by some
approximate technique. For off diagonal terms
(m n), numerical integration is reasonably fast
and accurate. In fact, when the separation between the cells is much larger than the cell size,
J,, can be replaced by a unit amplitude line source

4.3. Numerical results

at the center of cell n, and the off diagonal terms

are shown for the dielectric cylinder segmented

canbe approximatedby

into N=6x3=18, 10x5=50, 20x10=200,

and 30 x 15 = 450 cells. The CPU times for these
runs were approximately 1.63, 5.27, 55.79, and
319.15 5, respectively, on a VAX 8550. The MM
solution is well converged for the 20 x 10 segmentation, which corresponds to a segment size
of 0.05A or 0.1 wavelength in the dielectric.
One of the useful features of the volume current formulation is that it is trivial to treat inhomogeneous dielectrics and to compute the total
electric field in the dielectric. Inhomogeneous
dielectrics are treated by simply using e,, = the
value of e at the center of cell n in evaluating the

ITll

Zmn

(73)

2~(kpmn),

~Hi~

where Pmn = 1/(Xm x~)2+ (Ym ~y)2

is the
distance between the centers of cells m and n.
However, for the self impedance terms (m = n)
the integrand is singular and straightforward numerical integration fails. Providing that the cells
are nearly square, Richmond presented a very
simple method for evaluating the self impedance
integrations [241. Basically, Richmond used the
approximation that the self impedance of a nearly
square cell is the same as that of a circular cell of
the same cross_section area. For a circular cell of
radius a = %IA/IT the point matching self
impedance is
Z,,,,

E~(p= 0)

ITfl
=

-ffJ,,H~2(kp)

This section will present numerical results

based upon the point matching MM solution for
scattering by a rectangular dielectric cylinder
given in section 4.2. For simplicity, self impedance
terms are evaluated using eq. (76) and all other

mutual impedances using eq. (73).

Figure 14 illustrates the convergence of the
backscatter echo width, W, at f = 300 Mhz for a
1 x 0.5 m dielectric cylinder with relative permittivity Er = 4 and loss tangent tan 6 = 0.1. Curves

ds

1f21rfH(2)(kp)p

dp d4,.

(74)

TM Dielectric Cylinder Convergence

20x10
-.
10x5
-.N=6x3
30x15

~.
I

0-,

2AA
0

In eq. (74),2IT,
the and
d4, integration
simply results
the dp integration
caninbea
factor
of using the identity [251
integrated

f=300 Mhz

,si.
I

I
I

o~
.c
7

_________

H
3m

~H~(kp).

(75)

2~(
ka) j _~j.
2 1
j?~~
aH~

____

U,

In this case, eq. (74) becomes

E?r4

41__fanOO.1

fpHo(2)(kp) dp

2b
(76)

3~1

4~

Fig. 14. Convergence of the MM

solution for TM scattering by
O(Deg)
a rectangular dielectric cylinder.

16

E.H. Newman, K Kingsley / An introduction to the method of moments

AZ~~tennsineq.(69).Usingeqs.(52)and(62),
the total electric field in the center of cell n is
simply

IIIIIIIIIII

11111111111111111111111.

x)

Er(

Yl

...~ N8O

I~

.0075m1

E(x~,y~)= jw(e~e0)A

0.6m

f1 Ghz

(77)

,~

1I

where
I, is the
cross-section
areacoefficient
of the cells.of J,~and A is the
The insert in fig. 15 shows an f= 1 Ghz TM
plane wave with edge on incidence (4,, = 0) to a

,u

thin dielectric slab of width 0.6 m and thickness

.~

electricslab
slabwith
with Er = 16, and for an inhomogeneous
Er =

15

~~(0.3

I x I)~

(78)

in which Er tapers linearly from 16 in the center

of the slab to 1 at the edges. When the incident
waves hits the edge of the homogeneous Er = 16
slab, a surface wave is produced by the abrupt
change in permittivity seen by the incident field.
The surface wave propagates toward the other
edge of the slab and is then largely reflected. The
result is the oscillatory or standing wave pattern

Er( x)1 6

IIf

I~

0.0075 m. For the MM solution, the slab is segmented into a single row of N = 80 cells. The
magnitude of the total internal fields along the
slab centerline is shown for a homogeneous di-

7
E

0.3lxl)

~ X

20

I~I

60

0.3
4o

I I~I

idot~o i
(Deg)

Fig. 16. The bistatic echo width for a homogeneous (dashed

line) and inhomogeneous (solid line) dielectric slab.

shown for the internal fields. By contrast, since

the edges of the inhomogeneous slab have Er = 1,
very little surface wave is produced, and the
internal fields display almost no oscillatory behavior. The bistatic echo widths for the homogeneous and inhomogeneous slabs are shown in fig.

16.
I

III~~I~

I I~~IIII~~
I

~~I~~III

~IIi~IIII

~Er(x)16
-

I I
II

0,.

/
o

/
Er(

/
I,

x)1 +~(o.3HxI)

,.

-~

Er( x)~,YfN=80
.0075m1

~
l~- O.6m
flGhz

&~o

o~1

o~2

o.~

x (m)
Fig. 15. The total internal fields in a homogeneous (dashed
line) and inhomogeneous (solid line) dielectric slab.

5. The internal resonance problem

Every technique, no matter how powerful and
general, has its limitations and problems. This
section will describe the internal resonance problem which occurs in MM solutions for currents on
closed surfaces.
In solving eq. (1) for f it is implicit that we
desire the particular or forced response, i.e. the
response due to the excitation g. However, if
there is a solution to the homogeneous equation
L(f)=0, then the total response is f=f~+cf
0
where f~is the desired particular solution, f0 is
the homogeneous solution, and c is an arbitrary
constant. If the operator equation has a homogeneous solution, then ideally, the MM matrix eq.

E.H. Newman, K Kingsley

An introduction to the method of moments

(7) will also have a homogeneous solution. The

rank of the matrix [LI will be less than N, and
[LI will be a singular matrix with a zero determinant and no inverse. However, since the matrix
[LI is a finite-dimensional approximation to the
operator L, in practice, [LI will be nearly singular
with a small but non-zero determinant and a very
large condition number. In this case, the numerical solution of eq. (7) will be extremely sensitive
to roundoff error, and obtaining an accurate solulion will be difficult if not impossible. Thus, if the
original operator equation has a homogeneous
solution, the MM solution is expected to fail.
In electromagnetics, homogeneous solutions
arise in the analysis of closed surfaces. For exampIe, consider the problem of scattering by a closed
perfectly conducting box. The current on the box
is the solution of the EFIE, eq. (19), which basically enforces the boundary condition that the
total electric field tangential to the surface of the
box must be zero. However, it is well known that
a cavity mode can exist in the interior of a closed
perfectly conducting surface [261. A cavity mode
is a solution of Maxwells source free equations,
which satisfy all boundary conditions. Thus, a
cavity mode is a solution of the homogeneous
EFIE. If the operating frequency is close to the
internal resonance frequency of one of the cavity
modes of the box, the result will be an ill conditioned MM matrix equation and a failure of the
MM solution. The failure can usually be recognized by the erratic behavior of the MM solution
over a narrow frequency range [27].
Mautz and Harrington found that the internal
resonance problem is a result of the fact that, at
frequencies corresponding to internal cavity resonances, the exact solutions to the EFIE and the
MFIE are not unique, and thus their MM solutions fail [281. However, they showed that the
combined-field integral equation (CFIE), which is

obtained via a linear combination of the EFIE

and the MFIE, does have a unique solution at all
frequencies. As a result, the MM solution of the
CFIE does not suffer from the internal resonance
problem. Note that the internal resonance problem is more a problem with the non-uniqueness
of the operator equation, than it is problem with
the MM solution [291.

17

6. Summary
This paper has presented an introduction to
the MM and its application to problems of electromagnetic radiation and scattering. The MM is
a numerical technique which is used to solve the
linear integral equations which arise in electromagnetics by transforming them into a system of
simultaneous linear algebraic equations, i.e. a
matrix equation. The method is illustrated by
considering both perfectly conducting and dielectric bodies. In particular simple MM solutions
were presented for TM scattering by a perfectly
conducting strip and a rectangular dielectric
cylinder. Although there is no mathematical guarantee of convergence, as is typical, the solutions
presented did convergence.

References
Ii]

R.F. Harrington, Field Computations by Moment Methods (Macmillan, New York, 1968).
12] R.F. Harrington, Matrix methods for field problems,
Proc. IEEE 55 (1967) 136.
13] R.C. Hansen, Moment Methods in Antennas and Scattering (Artech, Boston, 1990).
[4] CA. Balanis, Advanced Engineering Electromagnetics
(Wiley, New York, 1989) ch. 7.
[5] E.H. Newman, A users manual for the electromagnetic
surface patch code, Ohio State University, Dept. of Electrical Engineering, ElectroScience Lab Report 716199-11,
prepared under Grant NSG 1498 with the National
Aeronautics and Space Administration, Hampton, VA
(August 1988).
[6] E.H. Newman, Polygonal plate modeling, Electromagnetics 10 (1990) 65.
17] E.H. Newman, P. Alexandroupoulos and E.K. Walton,
Polygonal plate modeling of realistic structures, IEEE
Trans. Antennas Propag. 32 (1984) 742.
[8] G.J. Burke and A.J. Poggio, Numerical electromagnetic
code (NEC) method of moments, Naval Ocean Systems
Center, San Diego, CA, Tech. Doe. 16 (July 1977).
[9] S.M. Rao, D.R. Wilton and A.W. Glisson, Electromagnetic scattering by surfaces of arbitrary shape, IEEE
Trans. Antennas Propagat. 30 (1982) 410.
[10] W.A. Johnson and D.R. Wilton, Modeling scattering
from and radiation by arbitrary shaped objects with the
electric field integral equation triangular surface patch
code, Electromagnetics 10 (1990) 41.
111] E.K. Miller, A selective survey of computational electromagnetics, IEEE Trans. Antennas Propag. 36(1988)1281.

18

E.H. Newman, K Kingsley

An introduction to the method of moments

[12] T.K. Sarkar, A note on the variational method

(RaleighRitz), Galerkins method, and the method of
least squares, Radio Sci. 18 (1983) 1207.
113] D.G. Dudley, Error minimization and convergence in
numerical methods, Electromagnetics 5 (1985) 89.
[14] I. Statgold, Boundary Value Problems of Mathematical
Physics, Vol. II (MacMillan, London, 1968).
[15] H. Shamansky and C. Yang, Test functions for the moment method which yield the minimum mean square
error, Radio Sci. 26 (1991) 133.
[16] R.F. Harrington, Time-Harmonic Electromagnetic Fields
(McGraw-Hill, New York, 1961) section 37.
117] S.A. Schelkunoff, On diffraction and radiation of electromagnetic waves, Phys. Rev. 56 (1939).
[18] CT. Tai, Dyadic Greens Functions in Electromagnetics
(Intext, Scranton, PA, 1971).
[19] E.H. Newman and MR. Schrote, On the current distribution for open surfaces, IEEE Trans. Antennas Propag.
31(1983) 515.
[20] M. Abramowitz and IA. Stegun, Handbook of Mathematical Functions, Applied Mathematics Series 55 (National Bureau of Standards, Washington, DC, 1970) ch. 9.
[21] N.N. Wang, Reaction formulation for radiation and scattering from plates, corner reflectors, and dielectric-coated
cylinders, Ph.D. dissertation, The Ohio State University,
Dept. of Electrical Engineering (1974).

[22] J.S. Asvestas and RE. Kleinman, Electromagnetic and

Acoustic Scattering by Simple Shapes, J.J. Bowman,
T.B.A. Senior and P.L.E. Uslenghi, eds. (Hemisphere,
New York, 1987) ch. 4.
[23] A.K. Dominek, A computer code to generate the strip
eigenfunction data, The Ohio State University ElectroScience Lab., private communication.
[24] J.H. Richmond, Scattering by a dielectric cylinder of
arbitrary cross section shape, IEEE Trans. Antennas
Propag. 13 (1965) 334.
[25] IS. Gradshteyn and I.M. Ryzhik, Tables of Integrals,
Series and Products (Academic, New York, 1980).
[26] J. Van Bladel, Electromagnetic Fields (Hemisphere, New
York, 1985) ch. 10.
[27] C. Klein and R. Mittra, Stability of matrix equations
arising in electromagnetics, IEEE Trans. Antennas
Propag., 21(1973) 902.
[28] J.R. Mautz and R.F. Harrington, H-field, E-field, and
combined-field solutions for conducting bodies of revolution, Arch. Elek. Ubertragung 32 (1978) 159.
[29] E. Arvas and JR. Mautz, On the non-uniqueness of the
surface EFIE applied to multiple conducting and/or
dielectric bodies, Arch. Elek. Ubertragung 42 (1988) 364.