You are on page 1of 7

Engineering Analysis with Boundary Elements 27 (2003) 325–331

www.elsevier.com/locate/enganabound

A volume/surface potential formulation of the method of moments


applied to electromagnetic scattering
Philippe De Doncker
Université Libre de Bruxelles, Elecgen CP 165/51, Avenue Roosevelt 50, 1050 Bruxelles, Belgium
Received 15 February 2002; revised 25 March 2002; accepted 2 April 2002

Abstract
A new integral method to compute the electromagnetic scattering by general inhomogeneous dielectric bodies is presented. The method is
based on a volume/surface equivalence principle and it uses the electromagnetic potentials as primary unknowns. The main advantage of this
method comes from the continuity of the potentials across interfaces separating different media so that very easy to handle nodal basis
functions can be used to solve the integral equations with the method of moments.
q 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Method of moments; Electromagnetic scattering; Potential theory

1. Introduction the electric field was supposed to be constant (pulse


approximation) and the collocation method at the center of
The method of moments (MoM) [1] is one of the most the cells was used to discretize the integrals (point matching
popular methods to compute the scattering of electromag- procedure). This discretization technique raised numerous
netic waves by dielectric, magnetic and conducting bodies. comments and in 1984 Massoudi et al. [4] showed that the
In this method, the scatterer is taken into account thanks to electric field discontinuities inside homogeneous regions
equivalent sources and integral formulations are derived for induce spurious charge densities on the boundaries of the
the electric and/or magnetic fields (electric field integral cells which drastically reduce the accuracy of the method.
equation methods, EFIE or magnetic field integral equation This kind of problem is not restricted to the pulse
methods, MFIE). Notwithstanding their success, these approximation, but it is common to all the nodal
formulations are very difficult to handle, namely due to interpolation techniques which cannot take into account
the fields discontinuities existing across material interfaces. the discontinuities of the electric field in the EFIE or of the
To obtain accurate results, these discontinuities must be magnetic field in the MFIE. Several attempts were still
taken into account in the discretization process and high undertaken to develop nodal methods for the MoM [5 – 8],
order methods are very difficult to define and to implement, but, till now, only vectorial basis functions have been found
decreasing the attractiveness of the integral formulations to be accurate for integral formulations in electromagnetics.
compared to the differential formulations like the finite- The most widespread vectorial interpolation method has
difference time-domain method (FDTD). been defined by Schaubert et al. It is based on linear
The discretization problems linked to the MoM are not vectorial basis functions in combination with tetrahedral
new. Considering for instance the scattering by dielectric cells. The definition of high order vectorial basis functions is
bodies, the EFIE formulation is most often used. Its first real a tedious task which is still under intense developments
implementations date back to 1974 and they concerned [10 –13], giving rise to very elegant methods, which are
wave propagation inside the human body [2] on the one unfortunately difficult to implement. Moreover, the integral
hand, and geophysics [3] on the other hand. In these works, formulations based on the fields also suffer from another
the scatterer was divided into cubical cells inside which difficulty: the kernel singularity. Even if these two problems
do not seem to be linked, it will be shown in Section 2 that in
E-mail address: pdedonck@ulb.ac.be (P. De Doncker). fact they share the same origin and that it is possible to solve
0955-7997/03/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 5 - 7 9 9 7 ( 0 2 ) 0 0 1 2 0 - 0
326 P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331

both problems by adopting a new point of view: using the it is possible to show that Eqs. (4) and (5) together with the
electromagnetic potentials in place of the fields as continuity properties of the fields imply the continuity of the
the primary unknowns in the integral formulations. In the scalar potential and of each component of the vector
remaining of the text, a time dependence ejvt is assumed. potential at any point of the equivalent problem [15]. The
electric field discontinuities across material interfaces in the
initial problem are modeled by the discontinuities of 7V
2. The equivalence principle for the potentials across superficial layers of electric charges in the equivalent
problem. Moreover, with this choice of gauge, a few
Let D be a dielectric scatterer bounded by a closed manipulations show that the potentials satisfy the wave
surface S, immersed in a homogeneous medium ð10 ; m0 Þ and equations
characterized by its relative permittivity 1r ðx; y; zÞ continu- 1
ous inside D and a permeability m ¼ m0 as shown in ~ þ k02 A
72 A ~ ¼ 2m0 J;
~ 72 V þ k02 V ¼ 2 r ð6Þ
10
Fig. 1(a). The scatterer is supposed to be linear and
isotropic. Inside D, the electric and magnetic fields satisfy so that they can be computed by convolution of the sources
the source-free Maxwell’s equations with the scalar free-space Green’s function:
~
~ ¼ 2jvm0 H; ~ ¼ jv10 1r E;
~ ð 1 ð
7£E 7£H ~ rÞ ¼ m0 JGð~
Að~ ~ r; ~r0 Þdv0 ; Vð~rÞ ¼ rGð~r; ~r0 Þdv0 ð7Þ
ð1Þ 10
~ ¼ 0;
7·10 1r E ~ ¼0
7·m0 H
where the integrations are performed over the whole space
Applying the volume equivalence principle [14], let us (outside D, J~ and r represent the sources producing the
define the equivalent electric current and charge densities, incident fields impinging the scatterer) and where Gð~r; ~r0 Þ is
respectively, by given by
 
1
J~ ¼ jv10 ð1r 2 1ÞE;
~ ~
0
r ¼ 10 1r E·7 ð2Þ 0 e2jk0 l~r2~r l
1r Gð~r; ~r Þ ¼ ð8Þ
4pl~r 2 ~r0 l
so that re-ordering the terms of Eq. (1), the electric and Integrations (7) can be split into two parts. First, if they are
magnetic fields alternatively satisfy in D the Maxwell’s restricted to the volume enclosing the sources producing the
equations corresponding to a homogeneous medium ð10 ; m0 Þ incident fields, they give the so-called incident potentials A~i
containing the equivalent sources (Fig. 1(b)): and Vi : Next, if they are restricted to the volume of D, they
~
~ ¼ 2jvm0 H; ~ ¼ jv10 E
~ þ J;
~ give the so-called scattered potentials A~ s and Vs produced by
7£E 7£H
ð3Þ the equivalent sources (Eq. (2)).
~ ¼ r;
7·10 E ~ ¼0
7·m0 H
In the homogeneous space of this equivalent problem, the
~ and an 3. The classical formulation
fields derive from a magnetic vector potential A
electric scalar potential V according to
In the classical approach, the integral formulations are
~ 2 7V;
~ ¼ 2jvA
E ~ ¼ 1 7£A
H ~ ð4Þ
derived in three steps as shown in Fig. 2. Considering for
m0 instance the EFIE formulation, the starting point is always
the identity:
To univocally define the vector potential, a gauge condition
must be imposed on its divergence. If the Lorentz gauge is ~ rÞ ¼ E
Eð~ ~ s ð~rÞ þ E
~ i ð~rÞ ð9Þ
pffiffiffiffiffiffi
chosen ðk0 ¼ v 10 m0 Þ
where E~ is the electric field, E
~ i is the incident electric field
2jv ~ ~
and Es is the scattered field produced by the equivalent
7·A ¼ 2k0 V ð5Þ
k0 sources modeling the scatterer. However, since the fields are

Fig. 1. (a) Initial problem; (b) equivalent problem.


P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331 327

Starting from Eq. (11), the equation system for the


potentials can now be explicitly derived. According to the
first of Eq. (7), the scattered vector potential is given by
ð
~ s ð~rÞ ¼ m0
A ~ r; ~r0 Þdv0
JGð~ ð12Þ
D

so that using Eqs. (2) and (4), the first of Eq. (11) gives an
integral equation for the vector potential:
ð
~ rÞ 2 k02
2jvAð~ ð1r 2 1Þð2jvA ~ 2 7VÞGð~r; ~r0 Þdv0
D

~ i ð~rÞ
¼ 2jvA ð13Þ
Fig. 2. Classical and potential integral formulations. To obtain an integral equation for V; it is important to note
that on S, due to the discontinuity of 1r in the initial problem,
the electric charges collapse into a layer of superficial
not naturally linked to the sources (real or equivalent), charges s giving rise to surface terms in the integrals for Vs
generally, in a first step, the incident and scattered so that the integral equation for V must be more carefully
electromagnetic potentials are computed according to Eq. derived. In a first step, let us apply the Green’s theorem for
(7) by convolution of the sources with Gð~r; ~r0 Þ (step 1 on Vð~rÞ in D:
Fig. 2). They are next differentiated to obtain the fields
V0 ð þ  ›G 
0 r 0 ›V
(step 2): Vð~rÞ ¼ Gð~r; ~r Þ dv 2 V 2G ds0 ð14Þ
4p D 10 ›n ›n
E ~ i;
~ i ¼ 27Vi 2 jvA E ~s
~ s ¼ 27Vs 2 jvA ð10Þ S

where r is the equivalent volumic charge density, V0 is


so that using Eqs. (2), (7), (9) and (10) gives an integral
~ the solid angle under which the point ~r sees the volume
equation for E:
D ðV0 ¼ 4p if ~r is inside D, V0 [ ½0; 4p if ~r lies on S
In this derivation, the integration – differentiation process
and V0 ¼ 0 if ~r is outside DÞ; n~ is the normal pointing in
has two important consequences: first, the differentiation of
the outward direction. The surface integral is performed
the scattered potentials leads to integral equations with
‘just inside’ D:
heavily singular kernels, and next and more important,
Next, the Green’s theorem can be applied to the space
although the potentials are continuous throughout the whole
surrounding D and after some transformations we obtain
space, their derivatives (the fields) are discontinuous at
interfaces separating different materials. The two difficulties V1 þ  ›G ›V

mentioned in the introduction have thus the same origin: Vð~rÞ ¼ Vi ð~rÞ þ V 2G ds0 ð15Þ
4p ›n ›n
the integration – differentiation process of the field S

formulations. where V1 is the solid angle under which the point ~r sees the
space surrounding D (with V1 ¼ 0 if ~r [ D) and where the
surface integral is performed ‘just outside’ D:
4. The potential formulation Summing up the two last equations and using the
boundary condition for the normal derivative of the scalar
To circumvent these problems, the solution we propose is potential on S; an integral equation for V valid ;~r [ D or S
to transfer the differentiation process at the end of the can be written
computation, keeping the vector and scalar potentials as the ð þ
primary unknowns in the integral formulation. The diagram r
Vð~rÞ ¼Vi ð~rÞ þ Gð~r; ~r0 Þ dv0 þ Gð~r; ~r0 Þð1r 2 1Þ
of the new method is also drawn in Fig. 2: the incident and D 10
S
scattered potentials are computed by convolution of the
sources with the low-singular scalar Green’s function (step ~ 2 7VÞ·~n ds0
ð2jvA ð16Þ
1). Next, by splitting the integrals of Eq. (7) into two parts,
the total potentials are equated to the sum of the incident and so that Eqs. (13) and (16) constitute a system of coupled
scattered potentials (step 2): integral equations for the potentials.
8 All the problems seem to be solved with the potential
~ rÞ ¼ A
< Að~ ~ i ð~rÞ þ A
~ s ð~rÞ formulation: the potentials are linked to the sources thanks
ð11Þ
: Vð~rÞ ¼ V ð~rÞ þ V ð~rÞ to the low-singular scalar Green’s function and the
i s
potentials are continuous across interfaces so that very
Finally, if necessary, at the end of the computation the fields easy to handle nodal basis functions can be used in
are derived by differentiation of the potentials (step 3). combination with the most general curvilinear cells.
328 P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331

Let us now consider the special case of a general mapping


dielectric scatterer divided into Nc discretization cells Dc 0 1
xðj1 ; j2 ; j3 Þ
inside which the permittivity is supposed to be constant, as B C
usual in the MoM. In this case, the scattered potentials of T~ : ðj1 ; j2 ; j3 Þ ! ~r ¼ B C
@ yðj1 ; j2 ; j3 Þ A ð20Þ
each cell must be summed up, and since there are no
zðj1 ; j2 ; j3 Þ
volumic equivalent charges inside the cells, system (13) –
(16) becomes: of the cube I : ½0; 1 £ ½0; 1 £ ½0; 1 in the ðj1 ; j2 ; j3 Þ
8 parameters space onto the curvilinear cube R in the physical
> XNc ð
>
> ~ 2 k02 ~ 2 7VÞG dv0 ¼ 2jvA
~i ðx; y; zÞ space so that each face of R corresponds to a surface
>
> 2jv A ð1r 2 1Þð2jvA
>
< D c
ji ¼ constant ði ¼ 1; 2; 3Þ in the parameters space (Fig. 3).
c¼1
Each face of a curvilinear cube R is eventually an
> XNc þ
>
> ~ 2 7VÞG·~n ds0 ¼ Vi interface between different dielectrics and although the
>
> V 2 ð1r 2 1Þð2jvA
>
: c¼1
potentials are continuous across such an interface, the
Sc normal component of the electric field can experiment
ð17Þ jumps, the tangential part being always continuous. This
discontinuity corresponds to a discontinuity in the normal
It is noteworthy that the kernels of these integral equations ~ themselves.
derivative of V, but not in the potentials V or A
exhibit a 1=R singularity very easy to handle numerically (to Since (if J is the Jacobian of the transformation T)~ [16]
compare to the 1=R2 or 1=R3 singularities of the classical !
formulations). A drawback of this formulation is coming 1 X 3
›~r ›~r ›V
7V ¼ £ ð21Þ
from the first order derivatives of the potentials, which are lJl i¼1 ›jj ›j k ›j i
difficult to compute when using curvilinear discretization
cells. In this last case, it is easier to remove the 7V terms with ð›~r=›jj Þ £ ð›~r=›jk Þ normal to the face ji ¼ constant; on
from the integrals, unfortunately increasing the kernel a face ji ¼ constant; in any curvilinear coordinates system,
singularity. Using the Green’s theorem, it can be shown ›V=›jj and ›V=›jk must be continuous to ensure the
that the first of Eq. (17) can then be written continuity of the tangential part of E; ~ but ›V=›ji has to be
eventually discontinuous to allow the discontinuity of the
0 ~ To reach a good accuracy, the
X
Nc ð normal part of E:
~ rÞ2k02
2jvAð~ @ ~ dv0
ð1r 21Þð2jvAÞG discretization of V must satisfy these conditions while V
c¼1 Dc
and each component of A ~ being continuous. It is possible to
1 ~ are interpolated
þ ð show that if V and each component of A
C ~ i ð~rÞ inside each cell thanks to the nodal Lagrange interpolation
2 ð1r 21ÞVG~n ds0 þ ð1r 21ÞV70 G dv0 A ¼ 2jvA
Dc functions of any order in the ðj1 ; j2 ; j3 Þ variables, all these
Sc
ð18Þ conditions are met, for any mapping T: ~ The continuity of the
tangential part of E ~ and the eventual discontinuities of its
and the second of Eq. (17): normal component can thus be simulated without vectorial
0 basis functions and the definition of high-order methods in
XNc þ þ any geometry is trivial.
B
Vð~rÞ2 @ ð1r 21Þð2jvAÞG·~ ~ n ds0 2 ð1r 21ÞV ›G ds0 Using Lagrange nodal basis functions, to each cell
c¼1
›n
Sc Sc corresponds a set of interpolation nodes as shown in Fig. 3
1 in the linear case. After expanding the potentials in terms of
V0c
2 ð1 21ÞV A ¼ Vi ð~rÞ ð19Þ these functions in Eqs. (17) or (18) and (19), the point
4p rc matching procedure at the interpolation nodes is used to

where V0c is the solid angle under which the point ~r sees the
cell Dc (with V0c ¼ 0 if ~r lies outside Dc ).
Eqs. (17) or (18) and (19) constitute the nodal volume-
surface formulations we propose to replace the fields
formulations.

5. Discretization of the potential formulations

To discretize the potential formulations, the scatterer is


divided into Nc curvilinear cubes. A curvilinear coordinates Fig. 3. Curvilinear discretization cell and Lagrange linear interpolation
system can be built in each cube by defining an univalent nodes in the ðj1 ; j2 ; j3 Þ and ðx; y; zÞ spaces.
P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331 329

~ i impinging a dielectric sphere of radius a.


Fig. 4. Plane wave E

obtain an algebraic system, so that four unknowns is


associated to each node (the value of scalar potential and of
the three components of the vector potential) in place of
three for instance in the EFIE formulation (the three
components of E). ~ However, in the potential formulation Fig. 6. Normalized magnitudes of Ez and 2jvAz along the y axis of a
homogeneous dielectric sphere ða ¼ 0:25l0 ; 1r ¼ 8 2 j4Þ: The solid line is
this drawback is balanced by the special structure of the the analytical result and the dotted lines are the numerical ones.
integral formulations. Thanks to the fact that the Green’s
function appearing in the first of Eq. (17) or in Eq. (18) is
scalar, there is no coupling between the different com- ði – jÞ are zero:
ponents of A~ so that the matrix of the discretized problem is
sparse. Indeed, the formulations Eqs. (17) or (18) and (19) 0 10 1 0 1
for the four unknowns Ax ; Ay ; Az and V can be split into four Ax A x 0 0 Ax V Ax Axi
scalar equations (the projections on x; y and z of Eq. (17) or B CB C B C
B 0 Ay V C B C B C
B Ay A y 0 CB Ay C B Ayi C
(18) and the second equation of Eq. (17) or (19), so that the B CB C ¼ B C
B CB C B C
MoM discretization matrix can be divided into 4 £ 4 sub- B 0 0 A z Az Az V C B Az C B Azi C
@ A@ A @ A
matrices noted Ai Aj ; Ai V; VAi and VV ði; j ¼ x; y; z):
VAx VAy VAz VV Ve Vei
0 10 1 0 1
Ax Ax Ax Ay Ax A z Ax V Ax Axi
B CB C B C Moreover, three sub-matrices are equal
BA A Ay V CB C B C
B y x Ay Ay Ay A z CB Ay C B Ayi C (Ax Ax ¼ Ay Ay ¼ Az Az ) so that only eight sub-matrices
B CB C ¼ B C ð22Þ
B CB C B C
B Az Ax Az A y Az A z Az V CB Az C B Azi C must be computed and stored. It is also noteworthy that
@ A@ A @ A
the charge densities appearing in the second of Eq. (17) or
VAx VAy VAz VV Ve Vei
in Eq. (19) depend on the permittivity jump ð1ri 2 1rj Þ
between the two cells i and j so that in homogeneous
In this notation, the first letter numbers the four equations regions, their contributions to V is zero and the
and the second one numbers the four unknowns. Since in corresponding terms in the sub-matrices VAx ; VAy ; VAz
Eq. (17) or (18) there is no coupling between the different and VV are zero: these sub-matrices are also most often
components of A; ~ the cross-coupling sub-matrices Ai Aj very sparse.

Fig. 5. Normalized magnitudes of Ez ; 2jvAz and k0 V along the z axis of a ~ 0 (dotted


Fig. 7. Normalized magnitudes of k0 V (solid line) and of 2jv7·A=k
homogeneous dielectric sphere ða ¼ 0:25l0 ; 1r ¼ 8 2 j4Þ: The solid line is line) along the z axis of a homogeneous dielectric sphere ða ¼ 0:16l0 ;
the analytical result and the dotted lines are the numerical ones. 1r ¼ 4Þ:
330 P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331

Fig. 8. Normalized magnitudes of (a) Ez and (b) Ey along the z axis of a layered dielectric sphere ða ¼ 0:13l0 ; the inner layer has a radius 0:065l0 and a
permittivity 1r ¼ 16; the outer layer has a permittivity 1r ¼ 9Þ: The solid lines are the analytical results and the dotted lines are the numerical ones.

6. Numerical results To check the method accuracy it was interesting to compute


if the potentials satisfy the Lorentz gauge (5) inside the
To validate the potential formulation, the system (18) dielectric. This gauge has been used to define the potential
and (19) was applied to the solution of the scattering of an z- formulation but it is not explicitly imposed on the solution
polarized plane wave propagating in the þy direction by a so that its fulfillment can be considered as a good estimator
homogeneous dielectric sphere (Fig. 4), one of the few cases of the global accuracy. Fig. 7 shows that the Lorentz gauge
whose analytical solution exists. In all our simulations, a is indeed satisfied, the discontinuous nature of 7·A ~ coming
linear interpolation of the potentials has been considered. from the linear interpolation used.
The incident potentials corresponding to the incident wave It was also interesting to verify that the discontinuity of
~ i ¼ E0 e2jk0 y 1~ z are given by:
E the electric field across interfaces was also well simulated.
8 Fig. 8 shows the results of our simulations in the case of a
< 2jvA ~i ¼ E~i
ð23Þ two-layers dielectric sphere. To validate our method, the
:V ¼ 0 grid has been chosen especially coarse ð8 £ 8 £ 8Þ; so that
i
the result is very satisfactory, the normal component of the
Figs. 5 and 6 show the perfect agreement between the electric field satisfying the theoretical jump while the
analytical and numerical solutions in the case of a very tangential component being continuous.
rough discretization: the spheres were drawn inside a 13 £
13 £ 13 grid. The different values drawn on these figures are
normalized by the incident electric field E0 : Along the y
diameter, V is zero so that it does not appear on Fig. 6, but
along the z diameter, it is possible to see that due to the
induced surface charge densities, it varies almost linearly.

Fig. 9. Normalized magnitudes of Ez along the y axis of a homogeneous


dielectric sphere ða ¼ 0:09l0 ; 1r ¼ 35 2 j13:65Þ: The solid line is the Fig. 10. Normalized radar cross-sections suu =l20 and sff =l20 (in dB) of a z-
analytical result, the piecewise constant line is the result of Ref. [9] and the directed finite length dielectric cylinder ðheight ¼ l0 ; radius ¼ 0:05l0 ;
dotted line is the potential formulation one. 1r ¼ 2:54Þ:
P. De Doncker / Engineering Analysis with Boundary Elements 27 (2003) 325–331 331

To compare our method with the classical ones existing References


in the literature, we chose to compare our results with those
obtained by Schaubert et al. in Ref. [9] which is the most [1] Harrington RF. Field computation by moment methods. New York:
widespread method as we have seen in the introduction. To The MacMillan Company; 1968.
carry out this comparison, we have computed all the results [2] Livesay DE, Chen K. Electromagnetic fields induced inside arbitrarily
shaped biological bodies. IEEE Trans Microwave Theor Techn 1974;
presented in Ref. [9]. In this last paper, the simulations were 22(December):1273–80.
done by using a tetrahedral modeling consisting of 512 cells. [3] Hohmann GW. Three-dimensional induced polarization and
To place ourselves in equivalent conditions, we considered electromagnetic modeling. Geophysics 1975;40(April):309– 24.
a cubical mesh of 552 cells. It is important to note that these [4] Massoudi H, Durney CH, Iskander MF. Limitations of the cubical
grids are too coarse to obtain very accurate results but they block model of man in calculating SAR distributions. IEEE Trans
Microwave Theor Tech 1984;32(August):746– 51.
are sufficient to compare the methods. In general, our results
[5] Catedra MF, Gago EM, Nuno L. A numerical scheme to obtain the
were at least as good as those presented in Ref. [9]. Fig. 9 RCS of three-dimensional bodies of resonant size using the conjugate
shows for instance a comparison in the case of the scattering gradient method and the fast Fourier transform. IEEE Trans Antennas
by a homogeous sphere. The potential formulation thus Propagat 1989;37(May):528–37.
appears as a good alternative to the fields formulations. [6] Graglia RD, Uslenghi PLE, Zich RS. Moment method with
Finally, Fig. 10 shows a comparison between the radar isoparametric elements for three-dimensional anisotropic scatterers.
Proc IEEE 1989;77(May):750–60.
cross-section of a finite length dielectric cylinder obtained [7] Graglia RD. The use of parametric elements in the moment method
with the FDTD method and our potential formulation. In solution of static and dynamic volume integral equations. IEEE Trans
this last method, the cylinder has been discretized into Antennas Propagat 1988;36(May):636–46.
curvilinear elements exactly matching the cylinder cross- [8] Su C. The three-dimensional algorithm of solving the electric field
section shape so that the computation has been carried out integral equation using face-centered node points, conjugate gradient
method and FFT. IEEE Trans Microwave Theor Techn 1993;
with very few cells. The agreement is between 1 or 2 dB.
41(March):510–5.
[9] Schaubert DH, Wilton DR, Glisson AW. A tetrahedral modeling
method for electromagnetic scattering by arbitrarily shaped inhomo-
7. Conclusion geneous dielectric bodies. IEEE Trans Antennas Propagat 1984;
32(January):77–85.
[10] Graglia RD, Wilton DR, Peterson AF. Higher order interpolatory
To circumvent the problems inherent to the use of the
vector bases for computational electromagnetics. IEEE Trans
fields in integral equations methods, a nodal formulation Antennas Propagat 1997;45(March):329– 42.
based on the electromagnetic potentials has been developed. [11] Wang J, Webb JP. Hierarchal vector boundary elements and p-
It has been shown that the use of the potentials in adaptation for 3D electromagnetic scattering. IEEE Trans Antennas
combination with Lagrange interpolation of any order in Propagat 1997;45(December):1869–79.
any curvilinear cell allows the simulation of the continuity [12] Graglia RD, Gheorma I. Higher order interpolatory vector bases on
pyramidal elements. IEEE Trans Antennas Propagat 1999;47(May):
of the tangential part of the electric field and the eventual 775–82.
discontinuity of its normal part without using vectorial basis [13] Martini E, Selleri S. Innovative class of curvilinear tetrahedral
functions. Since the discretization matrix of the new method elements. Electron Lett 2001;37(April):557–8.
is sparse, it does not demand more memory than the [14] Balanis CA. Advanced engineering electromagnetics. New York:
classical approaches for the same discretization mesh. Wiley; 1989.
[15] Stratton JA. Electromagnetic theory. New York: McGraw-Hill;
Finally, the formulation has been validated on near- and
1941.
far-field calculations and it has been verified that using [16] Thompson JF, Warsi ZUA, Wayne Mastin C. Numerical grid
linear nodal basis functions, the accuracy is of the same generation, fundations and applications. Amsterdam: Elsevier;
order as with the classical linear vectorial basis functions. 1985.

You might also like