1722

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 36. NO 12, DECEMBER 1988

Generalized Formulations for Electromagnetic

Scattering from Perfectly Conducting and
Homogeneous Material Bodies-Theory
and Numerical Solution

Abstract-Generalized E-field formulation for three-dimensional scattering from perfectly conducting bodies and generalized coupled operator
equations for three-dimensional scattering from material bodies are
introduced. The suggested approach is to use a fictitious electric current
flowing on a mathematical surface enclosed inside the body to simulate
the scattered field and, in the material case, to use in addition a fictitious
electric current flowing on a mathematical surface enclosing the body to
simulate the field inside the body. Application of the respective boundary
conditions leads to operator equations to be solved for the unknown
fictitious currents which facilitate the fields in the various regions through
the magnetic vector potential integral. The existence and uniqueness of
the solution are discussed. These alternative operator equations are
solvable via the method of moments. In particular, impulsive expansion
functions for the currents in conjunction with a point-matching testing
procedure can be used without degrading the capability of the numerical
solution to yield accurately near-zone and surface fields. The numerical
solution is simple to execute, in most cases rapidly converging, and is
general in that bodies of smooth but otherwise arbitrary surface, both
lossless and lossy, can be handled effectively. Boundary condition checks
to see the degree to which the required boundary conditions are satisfied
a t any set of points on the body surface are easily made for validating the
solution. Finally, results are given and compared with available analytic
solutions, which demonstrate the very good accuracy of the moment
procedure.

I. INTRODUCTION

HREE-DIMENSIONAL problems of electromagnetic

scattering by perfectly conducting and material bodies
have been a subject of intense investigation and research to the
electromagnetic community for many years. The study of
electromagnetic scattering is not solely of academic interest,
but of practical importance as well in many application areas.
These efforts have led to a development of a large number of
analysis tools and modeling techniques for quantitative evaluation of electromagnetic scattering by various objects. Among
these methods, surface integral equation formulations are
probably the most suitable ones for numerical solutions. The
general procedure is to reduce the three-dimensional problem
to two dimensions by casting the problem in terms of unknown
functions defined on the surface of the body rather than in
terms of unknown volume functions. In considering scattering
from a conducting body (Fig. l), the problem is formulated in
Manuscript received September 17, 1986; revised September 25, 1987.
The authors are with the Department of Electrical Engineering, TechnionIsrael Institute of Technology, Technion City, Haifa, 32000 Israel.
IEEE Log Number 8823639.

J'

Mm

1I

perfectly conducting
closed surface S

Fig. 1. General problem of scattering by perfectly conducting body

terms of the yet to be determined surface current J, induced on

the conducting body surface S. This can be done in two
alternative ways discussed both by Poggio and Miller in [ 11.
One formulation, known as the E-field integral equation, is
derived by setting the component tangential to S of the sum of
the incident electric field and the electric field due to J,, both
calculated with the conducting body absent, equal to zero on S.
The other formulation, known as the H-field integral equation,
is derived by setting the component tangential to S of the sum
of the incident magnetic field and the magnetic field due to J,,
both calculated with the conducting body absent, equal to zero
just inside S. In considering scattering from a homogeneous
material body (Fig. 2 ) , the problem can be formulated in terms
of yet to be determined equivalent electric and magnetic
currents J,, M, over the body surface S. Application of
boundary conditions leads to a set of four integral equations to
be satisfied. Linear combinations of these four equations leads
to a coupled pair of integral equations to be solved. One choice
of combination constants gives the formulation described by
Poggio and Miller [ l j . Another choice of combination
constants gives the formulation obtained by Muller [ 2 ] . If
either the E-field or the H-field integral equation for the
conducting body case were solved exactly, we would have the
true solution. Similarly, if the coupled pair of integral
equations for the material body case were solved exactly, we
would have the true solution. To obtain approximate solutions,
these equations are reduced to matrix equations via the method
of moments [3j. The solution of the matrix equations is then

0018-926X/88/1200-1722$01.00 0 1988 IEEE

LEVIATAN et al. : GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

Fig. 2.

General problem of scattering by homogeneous material body

carried out in the computer by inversion or elimination, and

sometimes by iterative techniques. Once the unknown surface
current J, in the conducting body case and the unknown
equivalent surface currents J,, M, in the material body case are
determined, the analysis of these scattering problems is
completed as the fields and the field-related parameters may
then be calculated in a straightforward manner.
In this paper, we introduce alternative formulations for
scattering from perfectly conducting and homogeneous material bodies with smooth surfaces. The novel idea is to
formulate the problem in terms of unknown functions that for a
conducting body are all defined on a mathematical surface
enclosed in the body, and for material body are partially
defined on a mathematical surface enclosed in the body and
partially on a mathematical surface enclosing the body.
Specifically, in considering scattering from a perfectly conducting body, we simulate an equivalence for the region
exterior to the body by means of a fictitious electric current J,,
flowing on a smooth mathematical surface SI enclosed in S.
This current is assumed to radiate in free space. The operator
equation for J,, is then formally derived by setting the
component tangential to S of the sum of the incident electric
field and the electric field due to J,,, both calculated with the
body absent, equal to zero on S . Similarly, in considering
scattering from a material body, we simulate an equivalence
for the region exterior to S by means of current J,, as we do in
the conducting-body case and, in addition, simulate an
equivalence for the region interior to the body by means of a
fictitious equivalent current J,, flowing on a smooth mathematical surface S, enclosing S . This current is assumed to
radiate in an unbounded space filled with the medium
composing the object. The operator equations for J,, and J,,
are then formally derived by an enforcement of the boundary
condition, namely, the continuity of the tangential components
of the electric and magnetic fields across S , which leads to a
set of two operator equations to be satisfied by the fields in the
two simulated equivalent situations. It should be pointed out
that it is certainly not claimed that exact solutions to the
suggested operator equations are guaranteed for any selection
of S, and S,. The existence of an exact solution is intimately
related to the analytic continuability of the scattered fields
toward the interior region and of the internal field toward the

1723

exterior region. Exact solutions, if they exist, are actually

equivalent currents, which produce the true fields in the
respective regions. The existence question will be addressed
further in Section 111.
To solve the proposed operator equations, we can apply a
method of moments numerical solution. Being numerical, our
solution will never be exact whether a mathematically admissable solution exists or not. Our objective is thus to match the
boundary condition to some desired computational accuracy
and this can be effected for certain choices of Si and S, even if
the existence of the solution cannot be guaranteed from a
strictly mathematical point of view. In particular, the choice of
impulsive sources as expansion functions for the unknown
currents is well-suited. Good results can be obtained using an
expansion of impulsive currents that lie a distance away from
the surface because the fields these currents generate on the
surface constitute a basis of smooth field functions. Being
smooth field functions, they are suitable for representing
smooth quantities on the boundary and are likely to render the
final solution accurate, not only in the far zone but in the near
zone and on the surface as well. The notable advantage of the
displaced implusive currents is that they not only yield smooth
field functions on the surface but also enable us to determine
the fields anywhere analytically. The quality that the fields are
known anywhere analytically is appealing as one can save
laborious surface current integrations when calculating the
fields at the various stages of the solution. Note that there are
quite a few field calculations involved. First, when constructing the generalized impedance matrix. Second, when testing
the solution by checking the degree to which the required
boundary conditions are satisfied over a denser set of points on
the boundary. Third, when computing field-related quantities
of interest after the solution has been established. Furthermore, since we are actually using a basis of smooth field
functions for representing fields on the boundary surface, a
simple point-matching procedure can be conveniently adopted
for testing. Notice that the attractive combination of an
impulsive current expansion and a point-matching testing
procedure which renders the solution trivial cannot be successfully applied to the standard surface formulations. First,
impulsive currents on the surface would inherently yield poor
surface and near-zone field approximations. Second, even if
one is interested merely in far-field quantities, one should
refrain from testing procedures, such as point matching, which
give emphasis to surface quantities but rather resort to a testing
procedure that will average out the inaccuracies.
As already stated, our objective is to match the boundary
conditions over the surface to some desired accuracy. In our
solution, however, we force the boundary conditions to be
obeyed only at a finite number of points on the boundary
surface. Surely, the field between the match points might
happen to be quite different from what is required by the
boundary conditions. Therefore, the convergence of the solution must be validated from a check on how well the boundary
conditions are matched between the points. A question that
naturally arises is the relationship between the error in the
boundary condition match and the errors in the exterior
scattered field and in the interior field. We are not aware of a

I724

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 36, NO. 12, DECEMBER 1988

mathematical relationship, but studies have shown that as the

boundary condition error becomes smaller, so do the errors in
the scattered field and the internal field. Clearly the boundary
condition check can indicate faulty results. However, even if
the boundary condition shows a sound convergence behavior,
one should test the computed quantity under consideration by
increasing the number of sources and match points and
comparing the results. If the results are sufficiently close, the
solution may be taken as satisfactory. In fact, the suggested
formulations have already been successfully applied to twodimensional scattering problems involving metallic posts in
rectangular waveguides [4], [ 5 ] , dielectric posts in rectangular
waveguides [6], and metallic and dielectric cylinders in free
space [7]. Furthermore, for the metallic case this approach
bears some relationship to a spherical-wave expansion technique for scattering from smooth perfectly conducting bodies
presented recently by Ludwig [8]. In Ludwig's technique
several spherical wave expansions with different origins are
used simultaneously to represent the scattered field. Ludwig's
paper also contains a list of references to other earlier work
using the concept of multiple expansions.
The paper is organized in the following manner. The
generalized formulations for the metallic and material cases
are presented in the next section. The existence and uniqueness
of the solution are discussed in Sections I11 and IV. The
method of moments solution is described in Section V.
Numerical examples are given in Section VI. Finally, a few
concluding remarks are outlined.
11. GENERALIZED
FORMULATIONS

A . Perfectly Conducting Body

The problem to be considered in this subsection is that of a
perfectly conducting body of smooth surface S situated in free
space ( p o , eo) and excited by impressed sources J', Mi.
Harmonic ejw' time dependence is assumed and suppressed.
The geometry of the problem is illustrated in Fig. 1. The total
field on and exterior to S is the sum of the incident field (E'"',
HI"') due to the impressed sources and the field (E",HS)
scattered from the conducting body. In general, we seek the
scattered field and field-related parameters of interest such as
the current induced on the conducting surface.
To solve this problem, we introduce the following approach, which in a sense constitutes a generalized E-field
formulation. In the proposed method, the perfectly conducting
object is replaced by free space with yet unknown fictitious
surface current distribution JSiflowing on a smooth mathematical surface SI enclosed in S as shown in Fig. 3. This current,
radiating in free space, is assumed to simulate the electromagnetic field scattered by the conducting body on and
external to S . We denote the total field on and exterior to S in
the simulated equivalence for the region exterior to S shown in
Fig. 3 by (Einc + E(J,;), HI"' + H(Jsi)), where (E(JSi),
H(JSi))is the field due to Jsi, calculated in free space. This
total field is simulating the total field (Elnc + E", Hi"' + H")
present in the region exterior to S in the original situation
shown in Fig. 1 .
Again, we should remark that from a strictly mathematical

unbounded homogeneous spoce

( F a. C O )

(E_'nr'~(J,,l,
_?'

C'"'*cU >11,

MI

/In

mothemoticol closed
suifoce S

\
mothernotical closed
surface S ,

Fig. 3.

Simulated equivalence for region exterior to S.

point of view we cannot, in general, guarantee the existence of

a current distribution J,, for an arbitrarily selected SI which
H")on and exterior to
exactly produces the scattered field (E",
S . The questions of the choice of SI for which the existence of
J,, can be rigorously justified, in general, and the choice of
implusive sources suitable for a numerical solution, in
particular, will be dealt with later. For present purposes, we
will suppose that for the considered choice of inner surface S, a
current distribution J,, exists which exactly produces the
HS)on and external to S . Note that in this
scattered field (Es,
event, the current J,, is in fact an equivalent current.
Hence for such inner surface SI, an operator equation for J,,
can be derived by setting the component tangential to S of the
total electric field in the simulated equivalence for the exterior
region equal to zero on S . This leads to

fi x E(J,,) = - fix E'"' on S

(1)

where fi is a unit vector outward normal to S . Clearly, if (1) is

satisfied, then by uniqueness [9, sec. 3.31, the electromagnetic
field (E(J,,), H(J,,)) in the region exterior to Swill be exactly
equal to the true scattered field there. Equation (1) consitutes a
generalized E-field operator equation for the problem of Fig. 1
in which E'"' is known and J,,, for a given SI, is the unknown
to be determined. Once J,, is found, the analysis of the
scattering problem is completed as fields and field-related
quantities can be readily evaluated. Note that the standard Efield formulation [ l ] is a special case of the generalized E-field
formulation (1) with S, equal to S . Of course, the standard
formulation does not suffer the difficulties associated with the
existence proofs when SI # S . Another specialization of this
approach is the familiar thin-wire approximation in which the
surface current is represented by a filament of current on the
wire axis while the field is evaluated on the wire surface. This
completes the exposition of the suggested generalized E-field
formulation for three-dimensional scattering from perfectly
conducting bodies.

B. Material Body
The problem to be considered in this subsection is that of a
homogeneous material body of smooth surface S situated in

1725

LEVIATAN et al. : GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

free space (po,eo) and excited by impressed sources (Ji, Mi).

Harmonic eJwttime dependence is assumed and suppressed.
The geometry of the problem is shown in Fig. 2. The
permeability of the material composing the body is p and its
permittivity is E . Both p and E are considered complex to
account for dissipation. The total field on and exterior to S is
Hi"') and the field (ES, Hs)
the sum of the incident field (Elnc,
scattered from the material body. The field inside the material
body is denoted by (E, H). We primarily seek the scattered
field in the region exterior to the obstacle as well as the field
inside the obstacle.
The proposed method for solving the considered problem is
to set up two simulated equivalent situations, one for the
region exterior to S and the other for the region interior to S ,
by means of two electric surface current distributions. In the
simulated equivalence for the exterior region, we employ the
model used in the metallic case and shown in Fig. 3 . The
scatterer is thus replaced by free space with yet unknown
fictitious surface current distribution J,; flowing on a smooth
mathematical surface Si enclosed in S . We denote the total
field on and exterior to S in the simulated equivalence for the
region exterior to S shown in Fig. 3 by (Einc+ E(J,,), Hinc+
H(J.$J),where (E(J,;), H(J,;)) is the field due to J,;, calculated
in free space. This total field is simulating the total field (Einc
+ E", HI"' + H S )present in the region exterior to S in the
original situation shown in Fig. 2 . Similarly, in the simulated
equivalence for the interior region, shown in Fig. 4, the
impressed sources are removed, the exterior region is filled
with homogeneous material indentical to that composing the
object, and yet unknown fictitious surface current J,, is
distributed on a smooth mathematical surface S, enclosing S .
This current, radiating in an unbounded homogeneous medium
of constitutive parameters p and e , is assumed to simulate the
electromagnetic field inside the material body. We denote the
interior to S in the simulated equivalence for the region
interior to S shown in Fig. 4 by (E(J,,), H(J,,)). This field is
simulating the field (E, H) present inside the body in the
original situation shown in Fig. 2.
The question as to the existence of current J,; and J,, for
arbitrarily selected surfaces S; and S, which exactly produce
the scattered field (E,, H") on and exterior to S and the
field (E, H) inside S , respectively, will be discussed later.
Like in the metallic case, it is impossible to guarantee the
existence of such currents, in general, from a strictly
mathematical point of view. Again, we will assume for present
purposes that for the considered choices of inner and outer
surfaces S; and So there exist current distributions J,; and J,,
which produce the true fields in the respective regions. Note
that in this event the currents JSiand J,, are in fact equivalent
currents.
Hence, for such inner and outer surfaces SI and So, the two
simulated equivalent situations can be pieced together by
enforcing the simulated fields in the two regions to obey the
continuity conditions for the tangential components across the
material boundary S. This leads to the operator equations:

ii x [E(JSi)- E(J,,)]

ii x E'"' on S

~ X [ H ( J , ~ ) - H ( J , , ) ] =-iixHInCon S

unbounded homogeneous space

(P , * )

mathematical closed
surface S o

mathematical closed
surface S

Fig. 4. Simulated equivalence for region interior to S .

where ii is a unit vector outward normal to S. Clearly, if ( 2 )

and ( 3 ) are satisfied, then by uniqueness [9, sec. 3-31, the
electromagnetic fields (E(J,,), H(J,,)) in the region exterior to
S and (E(J,,), H(J,,)) in the region interior to S will be
exactly equal to the true scattered and total fields in these
respective regions. Equations (2) and (3) thus constitute a
generalized set of coupled operator equation for the problem
of Fig. 2 in which (Elnc,HInc)is known and J,, and J,,, for
given SI and So, respectively, are the unknowns to be
determined. Once these currents are found, the analysis of the
problem is completed as field and field related quantities can
be readily calculated.
111. EXISTENCE

In this section, we first examine the strictly mathematical

requirements that guarantee the existence of current distributions J,, and J,, on arbitrarily selected S, and So, which
produce the true fields in their respective regions.
As discussed by Millar [lo], the exterior scattered field in
both the metallic and material cases can be continued
analytically into the region interior to S provided that in this
process no singularity of the exterior scattered field is crossed.
Note that the scattered field must have singularities within (or
on) S for otherwise it would vanish identically [ 1 1 1 . The
problem of locating singularities of the exterior scattered field
has been studied by Millar [ 121. The location of the singularities depends on the form of the scatterer surface and the
smoothness of the incident field on this surface. Therefore,
each problem would demand detailed consideration on its own
merits. Similarly, the interior field in the material case can be
continued analytically into the region exterior to S provided
that in this process no singularity of this field is crossed. The
internal field must also have singularities outside (or on) S for
otherwise it would vanish indentically. Again, the location of
these singularities depends on the form of the scatterer surface
(2) and the smoothness of the incident field on this surface and
each problem would demand detailed consideration on its own
( 3 ) merits. Notice that the continuation of the scattered field

1726

IEEE TRANSACTIONS ON ANlENNAS AND PROPAGATION, VOL. 36, NO. 12. DECEMBER 1988

unbounded homogeneous space

toward the region interior to S is effected in free space with the
(Po
scatterer absent. Similarly, the continuation of the internal
1
field toward the region exterior to S is effected in homogeneous unbounded space filled with the same material composing
the scatterer. Also, each analytic continuation, if exists, is, of
mathematical closed
surface S
course, unique.
As an example, consider the simple two-dimensional
scattering problem of a current filament situated at distance p
from the axis of a perfectly conducting cylinder of circular
cross section of radius a (a < p ) . The solution to this
problem is analytically derivable and can be found in [9, sec.
5-91. Applying large order asymptotic expansions for Bessel
and Hankel functions, it can be readily shown that the series
[9, eq. (5-120)] representing the scattered field on and exterior
mathematical closed
to the cylinder p 2 a is also uniformly convergent in the
surface S ,
interior annular region a 2 / p < p < a. This series diverges,
however, in the circular region p Ia 2 / p . It thus follows that Fig. 5 . Analytic continuation of scattered field towards region interior to S.
the scattered field can be analytically continued backwards all
unbounded homogeneous space
the way from the cylinder surface p = a at least down to p =
( P * <1
mathematical closed
a 2 / p .As another example, consider scattering of TM plane
surface S o
incident wave by a perfectly conducting elliptic cylinder with
semiaxes a, b(a > 6) [lo]. For this case, the expansion in
terms of Mathieu functions converges uniformly on and within
the cylinder as far as the interfocal segment. Thus excluding
the interfocal segment, the analytic continuation of the exterior
scattered field is valid everywhere throughout the interior
region.
We now assume, without further comment, that the scattered field (ES, Hs) can be analytically continued to some
extent into the region inside S and, in the material case, that
the internal field (E, H) can be analytically continued to some
extent into the region exterior to S. Notice again that the
continuation of the scattered field toward the region interior to
S is effected in free space (with the scatterer absent) and that of
mathematical closed
surface S
the internal field is effected in homogeneous unbounded space
filled with the same material composing the scatterer. We Fig. 6 . Analytic continuation of internal field toward region exterior to S .
denote the former by (E Sp, Hp) and the latter by (EP, HP) as
shown, respectively in Figs. 5 and 6. Now given a smooth similar manner, given a smooth mathematical surface So
mathematical surface SI inside S lying within the region outside S lying within the region through which the analytic
through which the continuation of the scattered field (ESPP, continuation (Ep, Hp) of the internal field is valid and
Hsp) is valid and enclosing the region containing the singulari- enclosing S, we can use the equivalence principle of set up an
ties, we can use the equivalence principle [9, sec. 3-51 to set up equivalent situation. A pertaining illustration is shown in Fig.
an equivalent situation. A pertaining illustration is shown in 6. Let the original field (E, H) exist internal to S, the field
Fig. 5. Let the original scattered field (ES, HS) exist on and (Ep, Hp) between S and So and the source-free field, denoted
external to S, the analytic continuation of the scattered field by (EO, HO), having a tangential electric field over So equal to
(ESPp,
Hp)exist between S and SI, and let the source-free field, that of EP and satisfying the radiation condition at infinity,
denoted by (ES,H),having a tangential electric field over SI exist exterior to So. T o support these fields there must be a
equal to that of ESP, exist internal to SI. To support these surface equivalent current J,, over So to account for the
fields, there must be a surface equivalent current J,, over S, to discontinuity in the tangential component of the magnetic field
account for the discontinuity in the magnetic field across SI. across So. This current is
This current is
J,,=fi,x(HP-H) on So
(5)
J,, = fi, X (HSP- HS1 on SI
(4)
where ii, is a unit vector inward normal to So. From the
where fi, is a unit vector outward normal to SI. From the uniqueness theorem again we know that the field exterior to So
uniqueness theorem [9, sec. 3-31, we know that the field produced by J, radiating in an unbounded space filled with the
interior to SI produced by J,, radiating in free space will be same material composing the scatterer will be (E, HO),
(E,H), between SI and S the field will be (ESP, H), and between So and S the field will be (EP, HP), and interior to S
farther out, exterior to S, the field will be (ES, H). In a the field will be (E, H).
*CO)

,??I

(ECE,C.

1727

LEVIATAN et al. : GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

The existence of the solution to (1) in the metallic body case

and to (2) and (3) in the material body case are thus evident
whenever the surfaces S, and So both lie within regions
through which the analytic continuation of the respective fields
is valid. Furthermore, the surface S, must enclose all the
singularities of the scattered field within S while the surface So
should enclose the scatterer.

IV. UNIQUENESS
In this section, we examine the uniqueness of the solution of
(1) for a given smooth surface S, lying within the region inside
S through which the analytic continuation of the scattered field
is valid and enclosing the region containing the singularities.
Next we examine the uniqueness of the solution to the coupled
equations (2) and (3) for given surfaces S, defined as above and
So lying within the region outside S through which the analytic
continuation of the internal field is valid and enclosing S .
The solution to a linear operator equation is not unique if the
corresponding homogeneous solution has a nonzero solution.
For the operator equation ( l ) , the solution will therefore be
unique if

where P,, is the complex power flow of the electromagnetic

field due to J,, out of S and P,, is the complex power flow of
the electromagnetic field due to J,, into S. The real part of (1 1)
is

(12)

Re (P,,)= -Re (P,,),

but both Re (PSJand Re (Pro)are either greater or equal to

zero. Hence it follows from (12) that
Re (Psi)= Re (PSo)= 0.

(13)

Since there are no external resonances, (13) implies that

fi x E(JSi)= fi x H(J,,) = 0

on S.

(14)

Now, substituting (14) into (9) and (lo), we have

fixE(J,,)=fixH(J,,)=O

on S .

(15)

Furthermore, since J,; is allowed to reside only on S, internal

to S and J,, is allowed to reside only on So enclosing S , it
follows that the solution to (2) and (3) will not be unique if

f i j ~ E ( J , j ) = i i j ~ H ( J , j ) = O just outside Si
fix E(J,,) = 0

on S

has only the trivial solution J,, = 0. Since there are no external
resonances (6) implies that
n x H(J,,) = 0

on S .

(7)

Furthermore, since J,, is allowed to reside only on S, internal

to S , it follows that the solution to (1) will not be unique if

fi, x E(J,,) = a, x H(J,,) = 0

just outside S,

(16)

(6)

(8)

has a nonzero solution. However, (8) is known to have a

nonzero solution whenever the frequency is such that S,, when
covered by a perfect electric conductor, forms a resonant
cavity. Hence the uniquenss of the solution here bears
reservations similar to those of the E-field equation [13].
Specifically, the resonant mode current produces no field
external to S, and therefore the field (E(J,,), H(J,,)) external to
S theoretically should be unique. However, when numerically
computing the solution, the matrix representing the continuous
operator becomes ill-conditioned at and in the vicinity of the
modal resonances and then the solution for (E(J,,), H(J,,))
external to S might degenerate. It should be remarked
furthermore that while the resonances of the E-field equations
occur at frequencies corresponding to resonances of the given
scatterer physical geometry, those of (1) occur at frequencies
corresponding to resonances of the selected cavity formed by a
hollow conductor of the same shape as S,.
For the operator equations (2) and (3), the solution will be
unique if

fi, x E(J,,)

= fi,

x H(J,,)

=0

just inside So

(17)

have a nonzero solution. However, following the argumentation stated just after (S), we know that (16) might have a
nonzero solution. Equation (17), on the contrary, implies that
J,, = 0 as there are no resonances external to So. Hence
although theoretically the solution for the scattered and
internal field should be unique, the numerical solution may
exhibit difficulties in the vicinity of the modal resonance of S,
and then the solutions for (E(JSi),H(JSi))external to S and,
consequently, of (E(J,,), H(J,,)) internal to S might degenerate.
V. METHOD
OF MOMENTS
SOLUTION

A . Perfectly Conducting Body

We here retrace the generalized formulation presented in

Section I1 for the problem of Fig. 1 and set up a simulated
equivalence for the region exterior to S using impulsive
electric current elements. To aid in future notation, we refer to
the region exterior to S as region I . In the simulated
equivalence for region I, shown in Fig. 3, the current J,i is
now represented by two sets of N 1fictitious implusive electric
N,
current elements, {II;,,} and {U;,,},n = 1, 2, * *
situated on S, enclosed in S . For conciseness, we denote the
electromagnetic field (E (JSi),H (J,i)) for this particular choice
of impulsive JSi by (ES1, HS1). The field (ES1, HS1) thus
simulates the true scattered field (ES,H)present in region I in
the original situation (Fig. 1). The nth current element Ilf, is
situated at r: on the surface Si, oriented in the i?,,
direction and
is
of,
yet
to
be
determined,
moment
Zlf,,.
The
nth current
fi x (E(J,,) - E(J,,)) = 0
on S
(9)
element Ilin is also situated at r; but is oriented in the i:,,
ax(H(J,,)-H(J,,))=O
on S
(10) direction and is of yet to be determined moment Zli,,. Here,
iy,, and i:,,
are orthogonal unit vectors tangent to S; at the
have only the trivial solution J,, = J,, = 0. From (9) and (lo), source point r: on Si.
we can readily find that
If sets of impulsive current elements {II;,,} and {U;,,}on S,
S,=
-T
.O
(1 1) were found which exactly satisfy the boundary condition (l),
a ,

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 36. NO. 12. DECEMBER 1988

1728

then (E"', H") would equal to the true field (ES, H") in region
I. To obtain an approximate solution, the boundary condition
(1) is imposed either strictly at Ns = "field points r;, m =
1, 2, . . . , NS,on S or, alternatively, in the least-square error
sense, at NS > N' field points r;, m = 1, 2, . . . , N", on S.
The result is
[ z , l i , = V,
(18)

where the dagger denotes the transpose complex conjugate of

the matrix. The matrix ([Z,] [Z,])- [Z,] + is called the
pseudoinverse of [Z,] . This completes the solution of matrix
equation (18). Once the unknown current is derived, either
from (22) or (23), one can straightforwardly evaluate an
approximate scattered field (ES/,
HS') and other field-related
quantities of interest.

where

B. Material Body

In (18), [Z,] is a 2N" by 2N' generalized impedance matrix,

is a 2NI-element generalized unknown current column
vector, and PA is a 2Ns-element generalized voltage source
column vector. In (19), [Z;,,] is an N" by NI matrix whose
(m, n) element is the
component of the electric field at
observation point r; on S due to current element ll;, of unit
moment (Zl;, = 1). Similarly, [ZL,,] is an N" by NI matrix
whose (m, n) element is the is, component of the electric field
at observation point r; on S due to current element Ili, of unit
moment (Zl;, = 1). The matrix [Z;,,] is an Ns by NI matrix
whose (m, n) element is the i;, component of the electric field
at observation point rk on S due to a current element Il;, of
unit moment (Zl;,,= 1). Similarly [ZL,,] is an N " by NI matrix
whose (m, n) element is the I
;,component of the electric field
at observation point r; on S due to a current element ll;, of
unit moment (Zli,, = 1). Here, I:, and i;, are orthogonal unit
vectors tangential to S at observation point r; on S. In (20),
is an "-element column vector whose nth element is Zl;,.
Similarly, 7; is an "-element
column vector whose nth
element is Zl;,. Finally, in (21),
is an Ns-element column
vector whose mth element is the negative of the is component
'J
of E'"' at observation point r; on S . Similarly, Ve2is an Nselement column vector whose mth element is the negative of
the is, component of E'"' at observation point r; on S.
Having formulated the matrix equation (18), the unknown
current vector can be found in a simple manner. If the
boundary condition is forced at N" = NI selected points on S ,
then [Z,] will usually be one-to-one and onto. In this case,
[Z,] is invertible and the exact solution to (18) is

?I,

vel

i, = [ Z,]

FA

We here retrace the generalized formulation presented in

Section I1 for the problem of Fig. 2, and set up two simulated
equivalences, one for the region exterior to S (region I) and the
other for the region interior to S using impulsive current
elements. To aid in future notation, we refer to the region
interior to S as region 11. In the simulated equivalence for
region I, we employ exactly the same model used in the
metallic case, which is described in the preceding subsection.
In the simulated equivalence for region 11, shown in Fig. 4, the
current J,, is now represented by two sets of N" fictitious
implusive current elements, {IIY,} and {II:',,}, n = 1, 2, * . .,
N", situated on So enclosing S . For conciseness, we denote
the electromagnetic field (E(JSo),H (JSo))for this particular
choice of impulsive J,, by (E", H"). The field (E", H") thus
simulates the true field (E, H) present in region I1 in the
original situation (Fig. 2). The nth current elements U?,, is
situated at
on the surface So, oriented in the
direction
and is of yet to be determined moment I/{',,. The nth current
element 117, is also situated at rsp but is oriented in the
direction and is of yet to be determined moment I/:,. Here, ty,
and iynare orthogonal unit vector tangent to So at the source
point rp on So.
If sets of impulsive current elements {U:,,} and {Ili,} on S,
and {Ily,} and {II?,} on So were found which exactly satisfy
the boundary conditions (2) and ( 3 ) , then (E"', H") would be
equal to the true scattered field (ES, HS) in the region I and
(E", H") would be equal to the true field (E, H) in region 11.
To obtain an approximate solution, the boundary conditions
(2) and (3) are imposed either strictly at N " = (NI + N")/2
field points r;, m = 1, 2, . . ., NS,on S , or, alternatively, in
the least-square error sense at Ns > (N' + N")/2 field points
r;,m = 1 , 2 , - . . , N S o n S . T h e r e s u l t i s

rr

i;"n

[ z ~ I & =V B
where

(22)

where [Z,]-' is the inverse of [Z,]. If, on the other hand, the
boundary condition is forced at N' > NI selected points on S ,
then [Z,] will usually be one-to-one but not onto. In this case,
unless ?A is in the range of [Z,], there is no exact solution to
is not in the range of [Z,], we pursue the smallest
(18). If
least-square error solution to (1 8), that is, the solution which
minimizes the standard norm of the vector [Z,]
- PA.This
solution of (18) is known to be

vA

TA

V,

= ([zA1+[ZAl)-l[zAl+

(23)

(24)

1729

LEVIATAN er al.. GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

VI. NUMERICAL
RESULTS

In (24), [Z,] is a 4 N " by 2(N' + N") generalized impedance

matrix,
is a 2(N' + N")-element generalized unknown
current column vector, and
is a 4Ns-element generalized
voltage source column vector. In (25), the matrices [Z&], p ,
q = 1, 2, are precisely the matrices introduced in (19) and
defined in detail thereafter. Specifically, each [Z&,q]denotes
an Ns by NI matrix whose (m, n ) element is the iirn
component of the electric field at observation point rk on S
due to a current element I l b of unit moment (Zl;, = 1).
Similarly, each [ 2 & ] , p , q = 1, 2, denoted an Ns by N"
matrix whose (m, n) element is the negative of the i:,,
component of the electric field at observation point r;, on S
due to a current element II;, of unit moment (Zlf, = 1).
Further, each [ZLp,],p , q = 1, 2, denotes an N Sby N' matrix
whose (m, n) element is the ii,, component of the magnetic
field at observation point r;, on S due to a current element ll;,
of unit moment (Z/q,l = I). Similary, [Z','pq],p , q = 1, 2,
denotes an N " by NI' matrix whose (m, n) element is the
negative of the i;, component of the magnetic field at
observation point r; on S due to a current element llf,, of unit
moment (I/;, = 1). In (26), the vectors
q = 1, 2 , are
precisely the vectors introduced in (20) and defined in detail
thereafter. Specifically, each T;denotes an "-element column
vector whose nth element is I/:,. Similarly, each F:, q = 1,
2, denotes an NI'-element column vector whose nth element is
I/;,,. Finally, in (27), the vectors Vep,p = 1, 2, are precisely
the vectors introduced in (21) and defined in detail thereafter.
Specifically, each
denotes the Ns-element column vector
whose mth element is the negative of the i;,,, component of
E'"' at observation point rk on S . Similarly, each Fhp,p = 1,
2, denotes an Ns-element column vector whose mth element is
component HI"' at observation point rk
the negative of the iirn
on S .
Having formulated the matrix equation (lo), the unknown
current vector can be found in a simple manner. If the
boundary condition is imposed at Ns = 1/2(N' + NI') points
on S , then the solution will be, in analogy to (22),

%,

vep

If, on the other hand, the boundary condition is forced at N " >
1/2(N1 + N") points on S , then the solution will be, in
analogy to (23),

This completes the solution of matrix equation (24). Once the

unknown current is derived, either from (28) or (29), one can
readily proceed to evaluate an approximate scattered field
(E"', H"') in the exterior region, an approximate field (E",
HI') in the interior region, and, of course, any other fieldrelated quantity of interest.

A versatile computer program has been developed using the

formulation of the preceding section. To check the accuracy of
the suggested method, we consider a conducting sphere, a
conducting cylindrical rod with rounded ends whose axis of
symmetry coincides with the z axis, and a dielectric sphere
illuminated by an incident plane wave of unit magnitude.

E'"' = U, exp (-jk,z)

(30)

1
HI"'= uy - exp ( -jk,z)

(31)

70

propagating in the z direction. For the spherical cases, the

exact solution can be found in [9, sec. 6-91. For the conducting
cylindrical rod with rounded ends case, a numerical solution is
available in [ 141. Some computational results obtained with
the program are given in this section and compared with the
available solutions. To limit the data displayed, respresentative results will be shown, without loss of generality, in the
principal xz plane.
The location of the sources may affect the rate of convergence. Based on previous studies [4]-[7], one would expect
the numerical results to converge faster to a sufficiently
accurate value when the sources are situated on surfaces
concentric with S and of figure similar to S. For the spheres,
the mathematical surfaces S, and So are thus taken to be
spherical surfaces of radii r' and rI1, respectively, concentric
with S. The current elements are evenly spaced along the
latitudinal and longitudinal lines of the respective spherical
surfaces. The match points are also evenly spaced on S . It was
found that for a sphere of radius rs selections of rI between
0.2rs and 0.W and of rl'greater than 1.5rs have a comparable
rate of convergence. In contrast, the rate of convergence
deteriorates when the inner sources approach either the sphere
center or the sphere surface, and when the outer sources
approach the sphere surface. For the numerical examples, the
inner source points are located on a spherical surface of radius
rl = 0.2rS and the outer source points are located on a
spherical surface of radius rl' = 2.0rs. In line with the above
criteria, for a conducting cylindrical rod with rounded ends of
diameter d' and length ,'I the mathematical surface S, is taken
to be a finite hemisphere-capped cylinder of diameter 0.2d'
and length 1'- 0.8d' (i.e., the distances between the centers of
hemispherical caps of Siand of S are equal). Furthermore,
with regards to the option of imposing the boundary condition
in the least square error sense, it was empirically found that,
although in some cases one can achieve the same accuracy
using fewer sources, thereby gaining the advantage of inverting smaller matrices, in general this option is redundant. In the
metallic case, we thus take an equal number of inner source
points N' and match points Ns. In the material case, we take
an equal number of inner source points NI, outer source points
N", and match points Ns. This common number is denoted for
convenience by N.Note that in the material case the value of
N represents twice as many sources as in the metallic case. It
should be remarked that the sources do not have to be split
equally between the inner and outer regions. Other combina-

....... N = 4

N.16

___-

****

....... N = 16

- N.25
____

N.36
exact

N :3 6

e
0

45

90

135

135

90

45

180

(degms)

I80

e ( d e veps)

Fig. 7. Plots of boundary condition error AE, versus 8 in xz plane for

metallic sphere of radius rs = 0.2X, for various numbers of sources and
match points N .

Fig. 8. Plots of surface current Je on a metallic sphere of radius rs = 0.2X

versus 8 in xz plane, for various numbers of sources and match points N.
...... N.8
__-.
N.16

tions can be used and may even yield a more rapidly

converging solution. Clearly, in any event, one should test the
solution by increasing the number of sources and match points
and verify the fulfillment of the boundary conditions between
the match points. Furthermore, for any desired quantity of
interest, one should examine the numerical convergence by
comparing the results for an increasing number of sources and
match points. If the computed results are sufficiently close, the
solution can be taken as satisfactory.

- N.36

**** exoci

o m i

I
I

A . Perfectly Conducting Sphere

Results for the problem of plane wave scattering by a
perfectly conducting sphere are shown in Figs. 7-10. The
conducting sphere in Figs. 7-9 is of radius rs = 0.2X, where X
is the wavelength in free space. In Fig. 10, a larger conducting
sphere of radius rs = 1.OX is examined.
First, we study the convergence of the boundary condition
error AE,, defined by

0 154

0 00

'
0

AE, =

IEincI

45

90

135

180

9 (degrees)

(fix(E"'+ ElnC)/on S
(32)

This quantity reveals how well the boundary condition is

satisfied between the match points. Plots of AI?,, as a function
of the polar angle 8 in the xz plane for various values of the
parameter N are presented in Fig. 7. Cases considered are N
= 16, 25, and 36. The boundary condition error, which by (1)
is zero at the match points, increases smoothly and reaches a
maximum between the points. As the number of sources and
match points increases, the maximum of AEbc on the surface
falls sharply. Note that even for N as small as 36 the maximum
is smaller than 0.35 percent. This nature of convergence has
been observed in other cases involving spheres of other radii

Fig. 9. Plots of scattering cross section U versus 8 in xz plane for various

numbers of sources and match points N for case of metallic sphere of radius
rs = 0.2h

and when the sources were distributed differently. Thus, by

forcing (1) to be obeyed at a sufficiently dense set of points on
S , (1) can be satisfied within a low error between the match
points as well, and consequently fields and field-related
quantities of interest can be approximated to a satisfactory
degree of accuracy.
Fig. 8 shows plots of the 8 component of the surface induced
current J given by
J = fi x (H"'+ HInc)
(33)

1731

LEVIATAN et al. : GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

0 DO

160

..___.
...N.65
......
.

__-_
__

****

120

_ _ _ - - N.96
-N=192

N.25
N =49

N.100

* **

0 75

Andreasen

exoct

0 60

"x
\ 80
b

"4

\o

40

45

0 JC

0 1:
0

1:15

DO

45

180

0 (dryrce.)
Fig. 10. Plots of scattering cross section U versus 6 in xz plane for various
numbers of sources and match points Nfor case of metallic sphere of radius
rs = 1.OX.

versus 8 in the xz plane for various values of N . Here, the

interval from 0" to 90" on 8 is in the "shadow" region of S
while 90" to 180" interval is in the "lit" portion of S . Note
that for N = 36 the results are in excellent agreement with the
exact eigenvalue solution [9, sec. 6-91. It should be remarked
that the accuracy obtained here is equivalent to that obtained
by Rao et al. [15] using 96 triangular patches to model the
sphere.
Next, we compute the scattering cross section defined by
U=

lim 4ar2
7-

IE"I
~

(Einc
I

(34)

where r is the radial distance from the origin. Plots of U versus

8 in the xz plane for various values of N are shown in Fig. 9
and compared with the exact solution. Again, very good
agreement with the exact solution is seen for N = 36.
Finally, we compute the scattering cross section (34) for the
larger sphere. Plots of U versus 8 in the xz plane for various
values of N are depicted in Fig. 10 and compared with the
exact solution. Of course, we expect that more sources will
now be required to render the solution accurate. Here the
results converge to the exact solution for N not larger than
100.

B. Perfectly Conducting Cylindrical Rod with

Hemispherical Caps at the Ends
Results for the problem of plane wave scattering by a
perfectly conducting cylindrical rod with hemispherical caps at
the ends are shown in Fig. 11. The rod is of diameter d' =
0.4X and length 'I = 1.5h. Plots of U as a function of 8 in the
xz plane for various values of N are dipicted in Fig. 11 and
compared with Andreasen's numerical result [ 141. Observe

0 O(

00

45

135

180

(dcyrerr)

Fig. 1 1 . Plots of scattering cross section U versus 0 in xz plane for various

numbers of sources and match points for case of perfectly conducting
cylindrical rod with rounded ends of diameter d' = 0.41 and length I' =
1.51.

that for N = 192 our result is in excellent agreement with

Andreasen's result. Note also that even for N as small as 96,
our result is quite close to that of Andreasen.

C. Dielectric Sphere
Results for the problem of plane wave scattering by a
dielectric sphere are shown in Figs. 12-14. The sphericai
scatterer considered is of radius rs = 0.2X. The sphere is of
permeability p = po and permittivity E = 3eO.
In a manner analogous to that presented in the metallic case,
we first carry out a study of the convergence of the boundary
condition errors AE, and AHk defined by

Ifi x (E5'+E'"'-E")I on S
IE'"'I

(35)

lfi x (H"+ HI"'- HI1)( on S

(H'"')

(36)

AEb, =

AHbc =

Plots of U, and AHk as a function of 8 in the xz plane for

various values of parameter N are presented in Fig. 12. Cases
considered are N = 25, 36, and 49. The boundary condition
errors AE, and AHbc, which by (2) and (3) are zero at the
match points, increase smoothly and reach a maximum
between the points. As the number of sources and match points
increases, the maxima of AE, and AH, on the surface fall
sharply. Note that even for N as small as 49, the maxima of
AEk and AH, are smaller than 0.7 percent.
To give some additional information on the convergence as
the number of expansion functions and match points is

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION. VOL 36, NO. 12, DECEMBER 1988

1732

........ N = 25

4
........

R!

N: 25

____

__

N=36

N.36
N = 49

N = 49
;
,

'

- I

>:

*. -3 I1

w 4

X 4

0
0

45

I35

90

45

90

135

180

8 (degrr6i)

I80

O(degire\)

(b)

(b)
Fig. 12. (a) Plots of boundary condition error AEh versus 8 in xz plane for
dielectric sphere of radius rs = 0.2X and permittivity t = 3to, for various
numbers of sources and match points N.(b) Plots of boundary condition
error AHh versus 0 in xz plane for dielectric sphere of radius r' = 0.2X and
permittivity E = 3t0, for various numbers of sources and match points N.

increased, we investigate the convergence of the approximate

scattered field to the exact solution on the surface of the
sphere. For this purpose, we define the scattered field errors
A E and A H as follows:

(ESr-EZxactI
on S
AE=

IEinCI
I

(37)

Fig. 13. (a) Plots of scattered field error A E on boundary versus 8 in xz

plane for dielectric sphere of radius rs = 0.21 and permittivity t = 3to, for
various numbers of sources and match points N.(b) Plots of scattered field
error A H on boundary versus 0 in xz plane for dielectric sphere of radius rs
= 0.2X and permittivity t = 3to, for various numbers of sources and r.atch
points N.

(HSr-H&actI
on S
AH=

IHinCI

(38)

where (E",xact,
H",x,c,)denote the exact values of the scattered
field obtained using the result of [9, sec. 6-91. Plots of A E and
A H as a function of 19 in the xz plane are depicted in Fig. 13.
Note the convergence of the fields as the number N increases.

LEVIATAN et al. : GENERALIZED FORMULATIONS FOR ELECTROMAGNETIC SCATTERING

0 16

0 08

I
ooo!

, ,

45

90

135

180

8 (degiee3)
Fig. 14. Plots of scattering cross section U versus e in xz plane for various
numbers of sources and match points N for case of dielectric sphere of
radius rs = 0.2X and permittivity E = 3to.

A similar convergence to the exact solution is also found for

the scattered field outside the sphere and for the field inside the
sphere. These plots will not be shown here.
Finally, the scattering cross section given by (34) is
computed for the dielectric sphere case. Plots of U versus 0 in
the xz plane for various values of N are shown in Fig. 14 and
compared with the exact solution. Very good agreement with
the exact solution is seen for N = 36.
VII. CONCLUSION
Generalized formulations for three-dimensional problems of
scattering by perfectly conducting and homogeneous material
bodies have been proposed. The innovative approach is to use
the field of a fictitious electric current flowing on a mathematical surface enclosed within the body to simulate the exterior
scattered field. In the material body case, the field of an
additional fictitious electric current flowing on a mathematical
surface enclosing the body is used to simulate the internal
field. Application of the boundary conditions at the body
surface leads to alternative operator equations to be solved for
the unknown currents which facilitate the fields in the various
regions via the magnetic vector potential integral.
Attention has been paid to the questions of existence and
uniqueness of the solution. It is found that from a strictly
mathematical point of view, one cannot, in general, guarantee
the existence of current distributions, for arbitrary selected
surfaces inside and outside the body, that will produce the true
fields in the respective regions. The existence of an exact
solution is intimately related to the analytic continuability of
the scattered field towards the interior region and the
internal field, in the material body case, towards the exterior
region. Exact solutions, if they exist, are actually equivalent
currents which produce the true fields in the respective
regions.

1733

The operator equations are solvable by the method of

moments. In particular, an impulsive expansion for representing the unknown current can be used. The noticeable
advantage of the impulsive expansion is that it enables us to
determine the fields anywhere in space analytically, thereby
rendering the quite a few field calculations involved in the
various stages of the solution trivial. At the same time, this
expansion yields accurate results not only in the far-zone but in
the near-zone and on the surface as well because the displaced
impulsive sources generate smooth fields on the surface
suitable for representing smooth quantities on the surface.
Finally, since we are using, though indirectly, a smooth
expansion for the fields on the surface, a point-matching
procedure can be selected for testing.
The proposed method has already been applied successfully
to two-dimensional waveguide [4]-[6] and free-space [7]
scattering problems, where the various fields have been
approximated using filamentary currents situated on suitably
selected surfaces. The numerical procedure was found to be
simple to apply, of wide range of applicability, and rapidly
converging. An application of this method to three-dimensional scattering problems where the various fields are
approximated using impulsive current elements has been
presented in this paper. The numerical solution for threedimensional problems is also simple to execute, rapidly
converging, and general in that bodies of smooth but otherwise
arbitrary surface both lossless and lossy can be handled
effectively. It should be clear that it is almost impossible to
state a rule of thumb as to the choice of source location and
number. In any case, one should test the solution by examining
both the degree to which the boundary conditions are satisfied
over a denser set of points on the boundary and the numerical
convergence of the considered quantity as the number of
sources and match points is increased. Some choices of source
location may speed up the convergence, but even for choices
less than optimal the solution will usually converge to the
appropriate limiting value without seriously taxing the computing system. In any case, it is immediately known if the
results are inaccurate by using the boundary condition check.
As presented here, the formulation deals exclusively with a
single scatterer. The extension of this formulation to encompass the multiscatterer case, say Mones, is straightforward. In
this case, the exterior scattered field is simulated by the field
of M sets of sources, each situated inside its corresponding
body, while the field inside each of the material bodies is
simulated, as before, by the field of an appropriate set situated
outside the body. Boundary conditions are then simultaneously
applied at selected points on the M surfaces.
The suggested technique is mainly applicable to scatterers
with smooth surface. A lack of smoothness can cause the
method to fail since the fields generated on the surface by the
impulsive current elements that lie a distance away from the
surface are smooth, and clearly a sum of such smooth fields is
not best suited for representing fields near edges where the
fields are singular. This deficiency can in principle be
overcome by incorporating, in addition to the impulsive
current elements, surface currents capable of representing the
correct edge singularity in subdomains near the edges.

1734

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 36, NO. 12, DECEMBER 1988

However, it might even be possible to use impulsive Sources

near the edges to
the edge behavior to accuracy that
may be sufficient for engineering needs. These approaches are
currently under investigation by us.

R. F. Harrington, Time-Harmonic Electromagnetic Fields. New

York: McGraw-Hill, 1961.
[IO] R. F. Millar, Rayleigh hypothesis in scattering problem, Electron.
Lett., vol. 5 , pp. 416-418, Aug. 1969.
[ I I] R. Courant and-D. Hilbert, Methods of Mathematical Physics, vol.
2. New York: Interscience, 1962, pp. 317-318.
[ 121 R. F. Millar, The Rayleigh hypothesis and singularities of solutions to
the Helmholtz equation, Bull. Radio Elec. Eng. Div. Nut. Res.
COumnC. Can.. Vol. 20., DD. 23-27. Aor. 1970.
J. R. Mautz and R. F. Harrington, H-field, and E-field, and
combined-fields solutions for conducting bodies of revolution, Arch.
Elek. Ubertragung., vol. 32, pp. 157-164, Apr. 1978.
M. G. Andreasen, Scattering from bodies of revolution, IEEE
Trans. Antennas Propagat., vol. AP-13, pp. 303-310, Mar. 1965.
S . M. Rao, D. R. Wilton, and A. W. Glisson, Electromagnetic
scattering by surfaces of arbitrary shape, IEEE Trans. Antennas
Propagat., vol. AP-30, pp. 409-418, May 1982.
[9]

..

REFERENCES
111 A. J. Poggio and E. K . Miller, Integral equation solutions of threedimensional scattering problems, in Computer Techniques for
Electromagnetics, R. Mittra, Ed. Oxford, England: Pergamon,
1973, ch. 4.
I21 C. Muller, Foundations of the Mathematical Theory of Electromagnetic Waves. New York: Springer Verlag, 1969.
131 R. F. Harrington, Field Computation by Moment Methods. New
York: Macmillan, 1968.
[4] Y. Leviatan, P. G. Li, A. T . Adams, and J . Perini, Single-post
inductive obstacle in rectangular waveguide, IEEE Trans. Microwave Theory Tech., vol. MTT-31, pp. 806-811, Oct. 1983.

..

1984.
[6] Y. Leviatan and G. S. Sheaffer, Analysis of inductive dielectric posts
in rectangular waveguide, IEEE Trans. Microwave Theory Tech.,
vol. MTT-35, pp. 48-59, Jan. 1987.
171 Y. Leviatan and A. Boag, Analysis of electromagnetic scattering from
dielectric cylinders using a multifilament current model, IEEE
Trans. Antennas Propagat., vol. AP-35, pp. 1119-1 127, Oct. 1987.
[8] A. C. Ludwig, A comparison of spherical wave boundary value
matching versus integral equation scattering solutions for a perfectly
conducting body, IEEE Trans. Antennas Propagat., vol. AP-34,
pp. 857-865, July 1986.

Amir Boag, for a photograph and biography please see page 1127 of the
October 1987 issue of this TRANSACTIONS,

Alona b a g , for a photograph and biography please see Page 1607 of the
November 1988 issue of this TRANSACTIONS.