IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO.

6, JUNE 2001

961

Localized Iterative Generalized Multipole Technique

for Large Two-Dimensional Scattering Problems
Fernando Obelleiro, Member, IEEE, Luis Landesa, Member, IEEE, Jos L. Rodrguez, Marcos R. Pino,
Ruth V. Sabariego, and Yehuda Leviatan, Fellow, IEEE

AbstractIn this work, we propose a novel and efficient solution

for the generalized multipole technique (GMT): the localized iterative generalized multipole technique (LIGMT). In LIGMT, an analytic constraint is imposed on the power radiated by the set of multipole sources sharing the same origin, rendering it minimum over
a given angular sector. In this way, the power radiated by each set
of multipoles is confined to a different section of the scatterer surface. It follows that each set of multipole coefficients can be solved
step by step via an iterative process, which circumvents the need
to solve the large and full matrix equation. This implies a significant reduction of the computational and storage cost, enhancing
the scope of application of the GMT method to larger problems.
Index TermsElectromagnetic scattering, equivalent sources,
integral equations.

I. INTRODUCTION

N RECENT years, fictitious source methods, such as the

generalized multipole technique (GMT) [1], [2] and the current model method [3][5] have been found to be very efficient
alternatives to the standard Method of Moments (MoM) [6] for
analyzing time-harmonic electromagnetic scattering problems.
In these methods, the scattered field is represented by the summation of fields radiated by a set of adequately placed fictitious
sources. If the location, number, and amplitudes of the fictitious
sources are appropriately chosen, the boundary conditions are
satisfied, in some approximate sense, on the scatterer surface,
and the scattered field can be obtained without any integration
over the surface currents. A review of recent advances of GMT
can be found in [7] and [8].
These fictitious source methods reduce the number of unknowns with respect to the conventional MoM solution, but at
the expense of being very dependent on the location of the fictitious sources. Nevertheless, the size of the scattering objects is
still restricted due to the computational cost and storage requirements of formulations of this kind. Recently, efforts have been
made to enable the application of these methods to larger bodies.
In tune with the impedance matrix location (IML) method proManuscript received January 22, 1999; revised August 31, 2000. This
work was supported by the Spanish Comisin Interministerial de Ciencia
y Tecnologa (CICYT), Project TIC99-0355 and by INTAS-International
Association for the promotion of co-operation with scientists from the New
Independent States of the former Soviet Union, Project INTAS-96-2139. T
F. Obelleiro J. L. Rodrguez, M. R. Pino, and R. V. Sabariego are with the
Depto. Tecnoloxas das Comunicacins, E.T.S.E. Telecommunicacion, Universidade de Vigo, 36200 Vigo, Spain (e-mail: obi@tsc.uvigo.es).
L. Landesa is with the Depto. de Informtica, Escuela Politcnica, Universidad de Extremadura, 10071 Cceres, Spain.
Y. Leviatan is with the Department of Electrical Engineering, TechnionIsrael, Institute of Technology, Haifa 32000, Israel.
Publisher Item Identifier S 0018-926X(01)05243-7.

posed by Canning [9][11], directional sources and testing procedures have been employed to obtain generalized impedance
matrices with a banded structure. In [12], the directivity arises
when the equivalent sources are arranged in such a manner as to
produce a field focus on the bounding surface. In [13], [14], the
directivity is achieved by positioning point sources in complex
space. A similar solution has been applied in the complex multipole beam approach (CMBA) [15], [16], where the scattered
fields are produced by a set of multipole sources located in complex space. Finally, in [17], [18] the directivity is achieved by
choosing arrays of fictitious sources and arrays of testing points
with directional radiation and receiving patterns.
In this paper, a new method, the localized iterative generalized multipole technique (LIGMT), is presented. LIGMT allows
for a significant reduction of the computational and storage cost,
thereby enhancing its scope of application to very large scattering problems. This new method is based on the GMT but
presents some substantial differences. In LIGMT, not only are
boundary conditions imposed, but we also require that the power
radiated by each group of multipole sources sharing the same
origin will be confined only to part of the surface. In this manner,
a substantial reduction in the coupling between different origins is readily achieved. The LIGMT method is completely different from the block-iterative solutions [19], [20] previously
suggested for GMT (specially suitable for problems involving
multiobject scatters). In these previous works, the GMT matrix
was simply partitioned and rearranged in order to obtain diagonally dominant block matrices following criteria based on physical considerations rather than on purely numerical reasoning.
This enables one to solve independently for the complex amplitudes of the multipole sources at each origin and solve the
whole problem by a simple iterative process avoiding the need
to invert the whole matrix. Note, however, that while a significant reduction in the computational cost with respect to conventional GMT can be attained, this approach may often result in a
less accurate satisfaction of the boundary conditions (BC). Nevertheless, the BC error levels obtained by LIGMT (below 1%)
have been found to be sufficient for far-field scattering problems. Some numerical results are presented to illustrate the convergence and accuracy of the proposed method. For the examples given, the computational cost is reduced by up to 2% of
the direct GMT solution, and even further reduction may be attained for scatterers of larger size. Regarding the storage cost, it
can be reduced by a factor of (number of multipole origins)
and shows a linear relation with the size of the scatterer.
The paper is organized as follows. Section II contains a short
description of the problem; the conventional GMT method is

0018926X/01$10.00 2001 IEEE

962

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 6, JUNE 2001

sources are placed within and are assumed to be radiating in

free space.
The complex amplitudes of the multipole sources are calculated by imposing the -field boundary condition over . We
have
at

(1)

where is a unit vector outward normal to

is the scattered field given by
field, and

is the incident

(2)
is the field due to the th multipole source, and
in which
is its yet unknown complex amplitude. Here, refers the
and designates the multipole order (demultipole origin
can be found in [21]). The coeffitailed expressions for
are obtained via generalized point matching [1] by
cients
imposing (1) at a set of uniformly distributed matching points
on as
(3)
This can be expressed in a matrix form as
(4)
where
Fig. 1. (a) General problem of scattering by a perfectly conducting body. (b)
Simulated equivalence for region V .

briefly outlined in Section III. Section IV is devoted to a description of the LIGMT algorithm, Section V contains some results illustrating the capabilities of the method, and finally, the
discussion and conclusions are presented in Section VI.

full impedance matrix;

GMT coefficients vector;
excitation vector due to the incident wave.
Obtaining accurate results requires a number of matching points
,
greater than the number of unknowns [1], i.e.,
and accordingly (4) is solved for in a least-squares error sense
via
(5)

II. PROBLEM DESCRIPTION

Consider a perfectly electric conducting infinite cylinder
of arbitrary cross section, illuminated by an electromagnetic
source. Let denote the perfectly conducting boundary surthe region interior to , and
the unbounded free
face,
exterior to , as shown in Fig. 1(a). The electric
space
are considered to be the sum of
and magnetic fields in
the incident fields (radiated by the source without obstacles)
and the scattered fields that are the contribution of the scatter.
We are concerned with obtaining the scattered field in
and other related quantities; both TE and TM polarizations
time dependence is assumed and
will be considered. An
suppressed in the rest of the document.
III. CONVENTIONAL GENERALIZED MULTIPOLE TECHNIQUE
(GMT)
To solve this problem, we formulate an equivalent problem
for the region . In this equivalent problem, shown in Fig. 1(b),
the scattered field due to the conducting body is simulated by
the fields radiated by sets of multipole sources. These multipole

Once has been found, it is straightforward to calculate the

scattered field and other related quantities of interest.
IV. LOCALIZED ITERATIVE GENERALIZED MULTIPOLE
TECHNIQUE (LIGMT)
As an alternative to the direct solution of the problem along
the lines mentioned in (5), a new solution is proposed in this
paper, in which the coefficients of the expansions located in
each origin will be obtained separately throughout an iterative
process.
A. Description of the Method
First, we divide the surface region
. Thus, we have

into

subregions,

(6)
Each of these subregions is associated with a different multipole
origin. Specifically, is the part of which is closer to than

OBELLEIRO et al.: LOCALIZED ITERATIVE GENERALIZED MULTIPOLE TECHNIQUE FOR LARGE 2-D SCATTERING PROBLEMS

963

In an ideal situation in which the multipoles do not radiate

outside the region that is under their immediate influence, there
is no coupling between the different origins, and therefore the
solution can be obtained by decomposing the original problem
into a set of
subproblems each corresponding to a different
origin as stated in (10) and (11). However, in practice, even
considering a very strict power constraint, the multipoles will
still have a residual radiation outside region , which implies
that they are actually coupled. Hence, the solution cannot be
achieved in a direct way but rather through an iterative process.
B. Iterative Process

Fig. 2. Associated subregions corresponding to the multipole origins.

to any of the other multipole origins (see Fig. 2). We refer to

as the the subregion of which is mainly under the influence
of the fields radiated by the multipoles centered at .
Next, we require that the total power radiated by the
multipoles located in the -th origin and passing through the
be as small as possible. This can be effected
area of
by requiring that the power radiated by the multipoles located
be minimal outside the angular sector
at the -th origin
subtended by
(see Fig. 2). The expression for this power is
derived in Appendix A. We have

Each stage of the iterative process consists of solving (11) for

each of the origins separately. Once the sequence of origins
has been completed, the process is repeated in a new iteration, in
which the coefficients previously obtained are used to provide a
modified excitation. The process is repeated until the algorithm
has converged to an accurate solution.
Specifically, the multipole coefficients corresponding to the
th iteration
are obtained via the following
minimization procedure:
for

(12)

(7)
where
a constant;
vector containing the multipole coefficients corresponding to the th origin;
a
far-field power-coupling matrix
whose elements are given by (33) (see Appendix A).
Equation (7) can be further transformed throughout a
as follows:
Cholesky decomposition
(8)
Finally, restricting the power radiated outside
effected by requiring that

is readily

In (12), the excitation vector is obtained using the sum of

the incident field and the fields radiated by multipole sources
that are located at the other multipole origins, taking their coefth iteraficients to be those calculated in the preceding
tion. That is
(13)
and by setting
The process begins with
.
The convergence of the above iterative process can be efficiently improved by actualizing the excitation vectors with those
coefficients of the multipole origins that have already been calculated within the same iteration:

(9)
is a threshold power level.
where
At this point, we can combine the power constraint stated
in (9) with the solution of the multipole coefficients corresponding to the th origin .
which describe the interDefining the matrix by blocks
and
action between the multipoles located at the th origin
subregion , we pursue the following minimization problem:
(10)
This minimization problem can be readily effected by solving
(11)
is a parameter
via a least-squares procedure [22]. Here,
which weights the relevance of the power constraint versus the
boundary condition fulfillment.

(14)
It is worth mentioning that the power constraint imposed in
LIGMT also implies a very useful secondary effect, namely, an
implicit regularization process similar to the one presented in
[23], [24]. This regularization improves the convergence of the
iterative process (reducing the condition number associated with
the matrix formulation of the problem) and also reduces the dependence of the method on the sources location.
C. Location of Multipole Expansions
In this subsection, we describe a set of simple rules that can
be followed to easily locate the multipole sources to apply the
LIGMT method:
1) The distance from each origin to the perfect electric con(see
ductor (PEC) surface must be approximately
Fig. 3).

964

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 6, JUNE 2001

Note that a different number of multipole expansions

has been considered for the LIGMT method (LIGMT may
),
require more multipole expansions than GMT,
while and have been considered to be the same as in
the GMT solution.
2) Matrix-vector multiplications described in (14) and the
matrix-vector product required to obtain the final solution
(17)
accounts for the number of iterations.
where
Thus, the total cost of the iterative LIGMT solution is
(18)

Fig. 3. Location of multipole origins.

2) The distance between two successive origins approaches

(as shown in Fig. 3).
to
3) A density of 10 matching points per wavelength are considered over the scatterer surface.
4) Twenty-one multipole expansions per multipole origin
has been found to be an optimal choice for the
LIGMT method. This averages 10 unknowns per wavelength which is a typical parameter in conventional MoM
formulations.
It must be pointed out that these rules have been derived empirically from our experience with LIGMT. These rules do not
necessarily follow those of the conventional GMT method and
may result in a somewhat larger number of unknowns, but they
are easy to implement in a general-purpose computer code.
D. Computational Cost
Let us now consider the computational cost of the LIGMT
method, comparing it with the conventional GMT method. To
this end, we take into account the number of floating point operations (FLOPs) involved in the different steps of the algorithm.
The definition of the FLOP count and a brief discussion about
the number of FLOPs involved in elementary matrix operations
can be found in [25].
The direct GMT solution leads to the least-squares problem
stated in (5). The solution of the least-squares problem is usually obtained throughout a QR decomposition in order to avoid
the numerical problems related to the normal equations method.
This implies the following computational cost:
(15)
is the number of matching points,
is the
where
is the number of
number of multipole origins, and
multipole expansions considered for each origin.
On the other hand, the iterative solution proposed in this paper
has the following computational cost:
1) Precalculation of the pseudo inverse matrices corresponding to the least-square problems stated in (12). This
allows one to avoid repeating this calculation in each iteration
(16)

In order to compare (15) and (18), recall the source location

rules given in the previous section. These rules provide the fol, and the perimeter of
lowing approximate relation between
and
. Furthe scatterer ( in wavelengths):
thermore, the number of multipole expansions in each origin is
for the LIGMT method and let us consider
chosen to be
for the conventional GMT (which averages less
only
than five unknowns per linear wavelength). Upon these considand
could be defined
erations, the ratio between
and the number of
as a function of the scatterer perimeter
. Fig. 4 plots this ratio versus the number of iteriterations
ations for different values of the perimeter .
Regarding the storage cost, it must be pointed out that the
LIGMT method can work storing only matrices of dimensions
, at the expense of recalculating all the out-of-diagonal
blocks for each iteration, although this improvement implies an
overload in the computational cost which has not been included
matrix must be considered
above. On the other hand, an
for the conventional GMT. So, the storage requirements of the
LIGMT method can be reduced approximately by a factor of
with respect to GMT.
V. NUMERICAL RESULTS
In this section, some numerical examples are shown to illustrate the convergence and accuracy of the method described
above. Numerical results have been presented only for the TE
polarization, although the same behavior has been found for the
TM case. In all the simulations, the multipole sources used in the
LIGMT method have been distributed according to the location
above rules. We have avoided taking advantage of the particular geometry of the circular cylinder (i.e., when the multipole
sources are located in the center, they become orthogonal). We
have located the multipoles all over the boundary of the scatterer, following the rules suggested for a general case.
The first example is a perfectly conducting circular cylinder
illuminated by a TE-polarized plane wave
of radius
. Fig. 5 shows the convergence of the LIGMT
algorithm, plotting the normalized residual error in the boundary
defined as
condition
(19)

OBELLEIRO et al.: LOCALIZED ITERATIVE GENERALIZED MULTIPOLE TECHNIQUE FOR LARGE 2-D SCATTERING PROBLEMS

Fig. 4.

965

LIGMT/GMT computational cost ratio versus the number of iterations for different perimeter lengths.

TABLE I
SUMMARY OF PARAMETERS.

Fig. 5. LIGMT residual error in the boundary condition versus the number of
iterations for different values of the parameter (which weights the relevance
of the power constraint versus the boundary condition fulfillment). Cylinder of
20.
radius r

versus the number of iterations for several values of the parameter . In this example, 64 multipole origins and 1256 matching
points were used as summarized in Table I. Although a good
convergence is obtained for all the values of , it can be seen
that the faster convergence corresponds with smaller values of
. On the other hand, a too small parameter presents an undesired oscillating behavior, thus a compromise value must be
would be a good choice.
chosen for ; in this example
It is worth mentioning that the error levels achieved by the
LIGMT in this case (0.1% order) are clearly worse than the
ones that could be obtained with the conventional GMT method
(0.01% order for this case). Nevertheless, this is what can be expected because, as previously mentioned, LIGMT has to satisfy
a set of additional constraints that enables a reduction in both

computational cost and storage requirements, and this naturally

comes at the expense of some loss of accuracy.
Fig. 6 shows the LIGMT behavior for several circular cylinders of different radii when a fixed value of the parameter
is used. It can be seen that the convergence of the method
is similar or even better as the size of the scatterer increases.
This makes the LIGMT method especially suitable for large
scattering problems, which was our original purpose. This behavior can be understood as a result of the localized effect of
each multipolar origin that only interacts with the near matching
points, avoiding its dependence on far matching points and consequently on the scatterer size.
In order to prove the ability of the LIGMT method to deal
with more complex geometries, the following examples have
been considered: a square cylinder (Fig. 7), an L-shape cylinder
(Fig. 8), and a U-shape cylinder (Fig. 9) are analyzed plotting
the error defined in (19) versus the number of iterations. Once
again, it can be seen that the LIGMT method presents a convergent behavior, providing residual errors (1% order for the square
and L-shape cylinder and 0.1% order for the U-shape cylinder)
that guarantee accurate results when dealing with far-field scattering problems. It must be pointed out that the residual error

966

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 6, JUNE 2001

Fig. 6. LIGMT residual error in the boundary condition versus the number
of iterations for circular cylinders of different sizes and a fixed value of the
3.
parameter

Fig. 7. LIGMT residual error in the boundary condition versus the number of
iterations for a square cylinder (d = 40) and = 3.

Fig. 9. LIGMT residual error in the boundary condition versus the number of
iterations for a U-shape cylinder (r = 10) and = 3.

in the square and the L-shape cylinders (mainly concentrated

at the corners) is not only due to the iterative algorithm, but
it also corresponds to the basic multipole expansions we have
used that are not able to account for an accurate description
of the fields near nonsmooth regions like corners. The inclusion of special expansions in the GMT, as for example MoM
basis or specific GMT models [26][31], could reduce these
error levels. Anyway, these new expansions could easily be included in the LIGMT algorithm, although this is outside the
scope of this paper. On the other hand, the residual error obtained for the U-shape cylinder is larger than the one provided
by the GMT, which in this case could reduce the residual error
by up to 0.01%.
Finally, we show a scattering result obtained for a large geometry. Fig. 10 plots the magnitude of the surface electric current
, comparing the
on a PEC circular cylinder of radius
given by an analytical solution [32] to
reference solution
the current obtained with the LIGMT method as
(20)
where
the incident magnetic field;
the scattered magnetic field;
a unit vector outward the surface .
The same geometry is considered in Fig. 11, where the mono, defined as follows, is shown:
static RCS
(21)

Fig. 8. LIGMT residual error in the boundary condition versus the number of
iterations for an L-shape cylinder (d = 40) and = 3.

Both Figs. 10 and 11 show that no difference can be observed

between the reference and the LIGMT solutions, demonstrating
the accuracy of the LIGMT method for large scattering problems.
The parameters involved and the computational cost reduction achieved in all the previous examples are summarized in
Table I.

OBELLEIRO et al.: LOCALIZED ITERATIVE GENERALIZED MULTIPOLE TECHNIQUE FOR LARGE 2-D SCATTERING PROBLEMS

967

operations, which are associated with the practical implementation of the the two methods, that have not been accounted for in
(15) and (18). Finally, for the sake of completeness, it must be
pointed out that the above CPU times have been obtained on a
Pentium III 450-MHz processor.
With respect to the computational efficiency of the method,
it is worth mentioning that the speed of convergence was found
to be constant (or even faster) as the size of the scatterer increases. This is illustrated in Table I, where it can be observed
(see the first four rows) that both the boundary condition error
and the computational cost are reduced as the size of the scatterer increases. It can be seen that the computational efficiency
increases as the size of the scatterer increases.
VI. DISCUSSION AND CONCLUSION
Fig. 10. Current magnitude on a PEC circular cylinder of radius r
parameter = 3.

= 40 and

Fig. 11. RCS of a PEC circular cylinder of radius r = 40 and parameter
= 3.

The selection of 20 iterations for Table I is justified because

at this point the algorithm has reached the 1% error level in the
boundary condition, which already guarantees accurate results
for the scattered fields. The use of more iterations would not be
justified in most cases because it would not imply any significant improvement of our goals.
The computational cost reduction included in the last column
of Table I has been obtained using (15) and (18) presented in
Section IV-D. In order to give an idea of the CPU time required
to obtain the above-mentioned numerical results, let us include
the following information about the CPU time needed in the
computation of the multipole coefficients corresponding to the
. The GMT needs 359 s (QR decircular cylinder of radius
composition with pivoting) while each iteration of LIGMT takes
only 0.52 s. With these CPU times, the computational cost of 20
iterations of LIGMT is about 2.92% that of the GMT. This is in
good agreement with the theoretical estimation of 2.27% presented in Table I. The small deviation is due to some additional

A new solution has been proposed to deal with large electromagnetic scattering problems: the located iterative generalized
multipole technique (LIGMT). The main goal of this method,
based on the GMT but with significant differences, is to provide
a significant reduction of the computational and storage costs,
enhancing its scope of application to large scattering problems.
The LIGMT method does not aim to be as accurate as conventional GMT. In the LIGMT method, not only the boundary
conditions are imposed, but also a set of additional constraints
whose aim is to produce a relatively local pattern, resulting in
a substantial reduction in the coupling between different origins. The fulfillment of these conditions enables the fast iterative process to be applicable, but also implies a worse fulfillment
of the boundary condition. Otherwise, the additional conditions
imposed in the LIGMT method introduce some kind of regularization which improves the convergence of the method and
reduces its dependence on the sources location.
The LIGMT method yields a rapidly converging solution
which is almost independent of the size of the scatterer. However, the solution does depend on the weighting parameter .
This parameter weights the relevance of the power constraint
versus the boundary condition fulfillment. A small value of
implies that the power restriction is less important, so the
radiation of each multipole origin is not concentrated over its
respective region, and so the iterative process would present a
poor convergence. On the contrary, large values of implies
a more strict power restriction, so multipoles do not radiate
outside the region which is under their immediate influence,
although the solution achieved in this case may provide a poor
solution in the boundary condition.
Consequently, the value of must be very carefully chosen.
Small values between 1 and 5 have been found to provide good
results.
Although LIGMT outperforms the conventional GMT from
a computational efficiency point of view, the computational
operations (which is the
cost of each iteration is still
standard matrix-vector multiplication cost in a problem with
unknowns). This cost could be easily reduced by combining
LIGMT with modern fast solvers (a recent review of which can
be found in [33]), but such a study is beyond the scope of this
paper. It should only be remarked that LIGMT is particularly
suitable for such fast solvers because the directional nature of

968

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 6, JUNE 2001

where is a column vector containing the

multipole
corresponding to the th origin, and is a row
coefficients
elements
vector of
where

(27)

Using the approximation of Hankel functions for large arguments

(28)
vector
Fig. 12.

becomes

Coordinate system centered at a given multipole origin  .

(29)

the multipole-source LIGMT inherently ensures the desired

feature of negligible interaction between the various groups
of sources and almost all except but a few far-field matching
points.
Some numerical results have been shown to illustrate the capabilities of the proposed method. For the examples considered,
the computational cost is reduced by up to 2% with respect to direct GMT solution, providing an error in the boundary condition
approaching 1% (which does not disturb the scattered fields).
Nevertheless, it should be noted that further advantages can be
gained as the size of the object increases.

where
(30)
In order to obtain the outgoing power through an angular
and
(see Fig. 12), the
sector defined between
must be integrated as follows:
(31)
which can be expressed as

APPENDIX A
MULTIPOLE POWER CONSTRAINTS
A. TM Polarization
The electric field at an observation point due to the multipole expansion located at a given th multipole origin
(22)
If we use a coordinate system centered at the origin , the
(Fig. 12). The
coordinate transforms into
previous expression becomes

(32)
Here, is a square matrix of dimensions
whose elements are given by
(33)
or, more explicitly, by

(23)
and the average Poynting vector

can be obtained as
(34)

(24)
where

that is,

(35)
B. TE Polarization
(25)

In this case, we consider the magnetic field in an observation

point due to the multipole expansion located in a given multipole origin .

(26)

(36)

which can be expressed in a matrix form as follows:

OBELLEIRO et al.: LOCALIZED ITERATIVE GENERALIZED MULTIPOLE TECHNIQUE FOR LARGE 2-D SCATTERING PROBLEMS

The average Poynting vector

can be obtained as
(37)

which can be expressed in a matrix form as follows:

(38)
is a vector containing the
multipole cowhere
, with
efficients located in the -th multipole origin
, and is a row vector whose elements are the same
defined for the TM case.
must be inIn order to obtain the outgoing power, the
and
tegrated over a given angular sector defined between
. We have
(39)
which can be expressed as

(40)
where the matrix

is the same as that obtained for the TM case.

REFERENCES

[1] C. Hafner, The Generalized Multipole Technique for Computational

Electromagnetics. London, U.K.: Artech House, 1990.
[2] A. Ludwig, A new technique for numerical electromagnetics, IEEE
Antennas Propagat. Newsletter, vol. 3, pp. 4041, Feb. 1989.
[3] Y. Leviatan and A. Boag, Analysis of electromagnetic scattering form
dielectric cylinders using a multifilament current model, IEEE Trans.
Antennas Propagat., vol. AP-35, pp. 11191126, Oct. 1987.
[4] Y. Leviatan, A. Boag, and A. Boag, Analysis of TE scattering from dielectric cylinders using a multifilament magnetic current model, IEEE
Trans. Antennas Propagat., vol. 36, pp. 10261031, July 1988.
[5]
, Analysis of electromagnetic scattering using a current model
method, Computer Physics Comunications, vol. 68, pp. 331345, 1991.
[6] R. F. Harrington, Field Computation by Moment Method. New York,
NY: IEEE Press, 1993.
[7] T. Wriedt, Ed., Generalized Multipole Techniques for Electromagnetic
and Light Scattering, The Netherlands: Elsevier Science , 1999.
[8] C. Hafner, Post-Modern ElectromagneticsUsing Intelligent Maxwell
Solvers. Chichester, U.K.: Wiley, 1999.
[9] F. X. Canning, The impedance matrix-localization (IML) method for
moment-method calculations, IEEE Antennas Propagat. Mag., pp.
1830, Oct. 1990.
[10]
, Improved impedance matrix localization method, IEEE Trans.
Antennas Propagat., vol. 41, pp. 659667, May 1993.
[11] F. X. Canning and K. Rogovin, Fast direct solution of standard momentmethod matrices, IEEE Antennas Propagat. Mag., vol. 40, pp. 1526,
June 1998.
[12] R. J. Pogorkelski, Improved computational efficiency via near-field localization, IEEE Trans. Antennas Propagat., vol. 41, pp. 10811087,
Aug. 1993.
[13] E. Erez and Y. Leviatan, Analysis of scattering form structures containing a variety of length-scales using a source-model technique, J.
Acoust. Soc. Amer., vol. 93, pp. 30273031, June 1993.
[14]
, Electromagnetic scattering analysis using a model of dipoles located in complex space, IEEE Trans. Antennas Propagat., vol. 42, pp.
16201624, Dec. 1994.

969

[15] A. Boag and R. Mittra, Complex multipole beam approach to electromagnetic scattering problems, IEEE Trans. Antennas Propagat., vol.
42, pp. 366372, Mar. 1994.
, Complex multipole-beam approach to three-dimensional
[16]
electromagnetic scattering problems, J. Opt. Soc. Amer., vol. 11, pp.
15051512, Apr. 1994.
[17] Y. Leviatan, Z. Baharav, and E. Heyman, Analysis of electromagnetic
scattering using arrays of ficticious sources, IEEE Trans. Antennas
Propagat., vol. 43, pp. 10911098, Oct. 1995.
[18] Z. Baharav and Y. Leviatan, Scattering analysis using ficticious
wavelet array sources, J. Electromagnetic Waves Applicat., vol. 10,
pp. 16831697, 1996.
[19] N. Kuster and L. H. Bomholt, A block iterative technique to expand
MMPs applicability to EM problems of higuer complexity, IEEE
Trans. Microwave Theory Tech., vol. 42, pp. 875883, May 1994.
[20] F. Obelleiro, J. L. Rodrguez, L. Landesa, and A. G. Pino, An iterative
algorithm to extend the applicability of the hybrid GMT-MoM method
to composite scattering problems, Micr. Opt. Technol. Lett., vol. 16, pp.
267271, Dec. 1997.
[21] L. H. Bomholt, A Computer Code for Electromagnetic Scattering
Based on the GMT, Ph.D. Thesis, Swiss Federal Institute of Technology, Zurich, 1990.
[22] P. C. Hansen, Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects of Linear Inversion. Philadelphia, PA: SIAM, 1998.
[23] M. R. Pino, L. Landesa, J. L. Rodrguez, and F. Obelleiro, A regularized
solution for the generalized multipole technique, presented at the IEEE
APS Trans. Ant. Prop. Intl. Symposium Dig., vol. 2, Atlanta, Georgia,
June 1998.
[24] L. Landesa, F. Obelleiro, J. L. Rodrguez, and M. R. Pino, Stable solution of the GMT-MoM method by Tikhonov regularization, in Progress
in Electromagnetics Research, PIER 20, J. A. Kong, Ed. Cambridge,
MA, USA: EMW Publishing, 1998, ch. 3.
[25] G. H. Golub and C. F. V. Loan, Matrix Computations. Laurel Park,
MD: The Johns Hopkins University Press, 1989.
[26] S. Eisler and Y. Leviatan, Analysis of electromagnetic scattering from
metallic and penetrable cylinders with edges using a multifilament
current model, in IEE Proceedings, Pt. H, vol. 136, Dec. 1989, pp.
431438.
[27] C. Hafner, J. Waldvogel, J. Mosig, J. Zheng, and Y. Brand, On the combination of MMP with MoM, Appl. Comp. Elec. Soc. J., vol. 9, no. 3,
pp. 1827, 1994.
[28] A. Boag, E. Michelssen, and R. Mittra, Hybrid multipole-beam approach to electromagnetic scattering problems, Appl. Comp. Elec. Soc.
J., vol. 9, no. 3, pp. 717, 1994.
[29] J. L. Rodrguez, F. Obelleiro, and A. G. Pino, A hybrid multipolar-expansion-moment-method approach for electromagnetic scattering problems, Micr. Opt. Tech. Lett., vol. 11, pp. 304308, Apr. 1996.
[30] F. Obelleiro, J. L. Rodrguez, and A. G. Pino, An automatic location
algorithm of MoM basis in the hybrid GMT-MoM method, Micr. Opt.
Technol. Lett., vol. 13, pp. 327329, Dec. 1996.
[31] U. Jakobus, H.-O. Ruo, and F. M. Landstorfer, Analysis of electromagnetic scattering problems by an iterative combination of MoM with
GMT using MPI for ghe communication, Micr. Opt. Tech. Lett., vol.
19, pp. 14, Sept. 1998.
[32] C. A. Balanis, Advanced Engineering Electromagnetics. New York:
Wiley, 1988.
[33] W. C. Chew, E. M. J. M. Jin, C. C. Lu, and J. M. Song, Fast solution
methods in electromagnetics, IEEE Trans. Antennas Propagat., vol. 45,
pp. 533543, Mar. 1997.

Fernando Obelleiro (M95) was born in Forcarei,

Galiza, Spain, in April 1968. He received the M.S.
and Ph.D. degrees in telecommunication engineering
from Universidad de Vigo, Spain, in 1991 and 1994
respectively.
He is currently an Assistant Professor with the
Department de Tecnoloxas das Comunicacins at
the Universidad de Vigo, where he teaches courses in
electromagnetic fields and radar. His current research
interests are applied mathematics and numerical
techniques for computational electromagnetics.

970

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 6, JUNE 2001

Luis Landesa (M98) was born in Vigo, Spain, in

November 1971. He received the M.Sc. and Ph.D. degrees in electrical engineering from the University of
Vigo, Spain, in 1995 and 1998, respectively.
During 1999, he was an Assistant Researcher
with the Spanish Council for the Scientific Research
(CSIC). He is currently an Assistant Professor
with the Computer Department at the University of
Extremadura, where he teaches courses in antennas
and signal processing. His current research interests
are array antennas in complex environments,
computational electromagnetics, and applied mathematics.

Jos L. Rodrguez was born in Monforte de Lemos,

Lugo, Spain, in 1967. He received the M.S. and Ph.D.
degrees in telecommunication engineering from the
University of Vigo, Vigo, Spain, in 1992 and 1997,
respectively.
During the fall of 1993, he was a Visiting
Researcher at the ElectroScience Laboratory, The
Ohio State University, Columbus. He is currently
an Associate Professor of the Departamento de
Tecnoloxas das Comunicacins at the Universidad
de Vigo, where he is teaching several courses on
radar and remote sensing. His research areas are radar cross-section prediction and measurement, ISAR, and numerical techniques for computational
electromagnetics.

Marcos R. Pino was born in Vigo, Spain, in 1972.

He received the M.Sc. and Ph.D. degrees in electrical
engineering from the University of Vigo, Vigo, Spain,
in 1997 and 2000, respectively.
During 1998, he was a Visiting Scholar at the
ElectroScience Laboratory, The Ohio State University, Columbus. He is currently teaching courses
on electromagnetic fields and remote sensing at the
Departamento de Tecnoloxas das Comunicacins at
the University of Vigo. His research areas are radar
cross section, rough surface scattering, and applied
mathematics for computational electromagnetics.

Ruth V. Sabariego was born in Vigo, Spain, in 1974.

She received the M.Sc. degree in telecommunication
engineering from the University of Vigo, Spain, in
1998. She is currently working toward the Ph.D.
degree at the Department of Electrical Engineering
(ELAP), Institut Montefiore, University of Lige,
Belgium.
Her research interests include numerical techniques for computational electromagnetics, applied
mathematics, radiation hazards, and bio-electromagnetics.

Yehuda
Leviatan
(S81M81SM88F98)
received the B.Sc. and M.Sc. degrees in electrical
engineering from the Technion-Israel Institute of
Technology, Haifa, Israel, in 1977 and 1979, respectively, and the Ph.D. degree in electrical engineering
from Syracuse University, Syracuse, NY, in 1982.
He spent the 19821983 academic year as an
Assistant Professor at Syracuse University and
subsequently joined the Department of Electrical
Engineering at the Technion, where at present he is a
Professor. During his tenure at the Technion, he held
short-term visiting positions at Cornell University, the Swiss Federal Institute
of Technology (ETH) in Zurich, the Catholic University of America, the University dAix-Marseille III, the University of Illinois at Urbana-Champaign, the
University of Washington, Bell Laboratories, and the University of Michigan.
During 1989-1991, while on sabbatical leave from the Technion, he was with
the Department of Electrical Engineering and Computer Science at The George
Washington University as a Distinguished Visiting Professor. His research
interest is primarily in computational electromagnetics. He has published more
than 75 journal papers and presented many others at international symposia.
Dr. Leviatan is a member of Commissions B of the International Union of
Radio Science.