IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 55, NO.

3, MARCH 2007 931

Non-Maxwellian Locally-Conformal PML Absorbers

for Finite Element Mesh Truncation
Ozlem Ozgun, Student Member, IEEE, and Mustafa Kuzuoglu, Member, IEEE

Abstract—We introduce the locally-conformal perfectly matched

layer (PML) approach, which is an easy and straightforward PML
implementation, to the problem of mesh truncation in the finite el-
ement method (FEM). This method is based on a locally-defined
complex coordinate transformation which has no explicit depen-
dence on the differential geometric characteristics of the PML-free
space interface. As a result, it is possible to handle challenging PML
geometries with interfaces having arbitrary curvature, especially
those with curvature discontinuities. In order to implement this
approach, we also introduce the concept of complex space FEM
using elements with complex nodal coordinates. After developing
the analytical background of this method, we present some nu-
merical results to demonstrate the performance of this method in
three-dimensional electromagnetic scattering problems.
Index Terms—Complex coordinate stretching, electromagnetic
scattering, finite element method (FEM), perfectly matched layer
(PML), radar cross-section. Fig. 1. Some challenging PML geometries, whose problematic regions are
shaded by the gray-colour.

I. INTRODUCTION
P ERFECTLY matched layers (PMLs) have been widely
used for mesh truncation in the numerical approximation
of electromagnetic radiation and/or scattering problems by the
finite element method (FEM) and the finite difference time
domain method (FDTD). The PML concept has been intro-
duced by Berenger [1] in the context of the FDTD method by
using a split-field formulation in cartesian coordinates. Another
formulation, yielding a PML action, has been introduced by
Chew and Weedon [2] via the concept of complex coordinate
stretching, which is essentially the analytic continuation of
the electromagnetic field variables to complex space. Both of
these approaches yield non-Maxwellian fields within the PML
domain. It has been demonstrated by Sacks et al. [3] that a
Maxwellian PML can be realized as an anisotropic layer in
cartesian coordinates with suitably defined permittivity and
permeability tensors. The anisotropic PML concept, originally
introduced in cartesian coordinates, has been extended to
cylindrical and spherical coordinates, and has been used in the
design of conformal PMLs using a local curvilinear coordinate
system [4]–[7].
All of the previous PML implementations in FEM sim-
ulations are based on the design of a Maxwellian PML as
an anisotropic medium having suitably defined constitu-
tive parameters. In this paper, however, we introduce the
“locally-conformal PML” method without using artificial
anisotropic materials. The locally-conformal PML, which
Manuscript received August 12, 2006; revised November 20, 2006.
O. Ozgun and M. Kuzuoglu are with the Department of Electrical and Elec-
tronics Engineering, Middle East Technical University, 06531 Ankara, Turkey.
Digital Object Identifier 10.1109/TAP.2007.891865
0018-926X/$25.00 © 2007 IEEE
is based on the analytic continuation of the frequency-do-
main Maxwell’s equations to complex space, is an easily
implementable, non-Maxwellian PML technique for mesh
truncation in FEM applications. The PML region is constructed
in complex space by keeping the same constitutive parame-
ters everywhere and replacing the real coordinates with their
complex counterparts calculated via the complex coordinate
transformation. In this formulation, Maxwell’s equations are
modified accordingly inside the complex space, and the weak
variational form of the wave equation is derived in terms of the
complex coordinates. Then, the weak variational form of the
wave equation is discretized using the complex elements (i.e.,
elements with complex nodal coordinates) which are obtained
by the complex coordinate stretching. The equations related to
the FEM formulation in the PML region depend directly on the
node coordinates, thus, the replacement of the node coordinates
with the complex coordinates is sufficient to achieve the design
of the locally-conformal PML.
The locally-conformal PML has some important practical
advantages in terms of the easy design of PMLs having chal-
lenging geometries. Although the anisotropic PML has been
used in the design of conformal PMLs using a local coordinate
system [5], [7], this approach has some difficulties in terms
of the computational and analytical effort, especially for the
implementation of the PML at the intersection regions or in
the case of abrupt changes in curvature. The implementation of
the anisotropic PML may be easy for some smooth geometries
(such as a spherical or cylindrical shell), but it may be very
difficult to design an anisotropic PML for geometries in Fig. 1,
due to the presence of some problematic regions shown in the
figure. The PML geometries in Fig. 1(a) and (c) may be inter-
preted as the partial cross sections of a conformal PML region
932 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 55, NO. 3, MARCH 2007

typical points in ( , , and ), for the sake of il-

lustrating the method.
In order to obtain the PML action, the region , which
is originally a subset of , is extended to , which is a
manifold in . During this procedure, locally defined coordi-
nate systems are totally avoided, and each point in is
mapped to in by using the parameter

(1)

and the unit vector

Fig. 2. Locally-conformal PML implementation.

(2)

designed to enclose a conesphere object in a scattering problem where and are the position vectors of the points and ,
(see Fig. 5 or Fig. 6). Similarly, the geometry in Fig. 1(b) is respectively. The point (i.e., , , or in Fig. 2)
a cross-sectional sketch of a PML domain conformal to an is located on , and is the solution of
Ogive object (see Fig. 8). The main advantage of our approach
is its flexibility to design a PML domain which encloses an
arbitrarily-shaped convex spatial domain, possibly with abrupt (3)
changes and even discontinuities in the curvature of the free
space-PML interface. The parametric representation of the Such a exists and is uniquely defined because is the
complex coordinate transformation is obtained by only a single boundary of the convex set . The unit vector in (2) is actually
parameter and a single unit vector, which are easily derived the unit vector which is normal to , except for the PML points
from the node coordinates in an existing FEM mesh using some inside the subdomain (i.e., problematic region) which is shown
computational search techniques. between the dashed lines in Fig. 2. That is, for all points inside
The organization of this paper is as follows: We first derive the problematic region (such as and ), the solution of the
the equations governing the parametric construction of the lo- minimization problem given by (3) yields the same value for ,
cally-conformal PML method in Section II. Then, in Section III, as shown in Fig. 2.
we formulate the FEM in complex PML space using tetrahedral Now, we define the complex coordinate transformation
edge elements for the solution of the three dimensional vector as follows:
wave equation. In Section IV, we present some numerical appli-
cations involving electromagnetic scattering problems. Finally,
we draw some conclusions in Section V. (4)

II. LOCALLY-CONFORMAL PML FORMULATION where is the position vector of the point in complex
The locally-conformal PML method is based on the complex space, and is a function satisfying the following two condi-
coordinate stretching concept introduced by Chew and Weedon tions: (i) , and (ii) is a monotonically increasing
[2]. The initial phase of the approach is the construction of the function of . A possible expression satisfying these properties
spatial region occupied by the PML such that the white space is given by
within the computational domain is minimized. If the problem
involves a single convex object (i.e., the spatial domain occu-
pied by the object is a convex subset of ), we can place the (5)
PML region at a suitable distance, which is typically a fraction
of the wavelength , from the surface of the object. For an ob- where is a positive parameter, is a positive integer (typi-
ject which is not convex or for a group of objects, we can choose cally 2 or 3) related to the decay rate of the field strength in-
a PML region which is conformal to the convex hull [i.e., the side , and is the position vector of the point (i.e.,
smallest convex set that encloses the object(s)]. We illustrate the , , or in Fig. 2), which is actually the inter-
locally-conformal PML method by considering the geometry in section of the line passing through and (i.e., the dotted line
Fig. 2, which represents a partial cross-section of an arbitrary in Fig. 2) and . We also note that is the local
PML region . In this figure, we do not show the ob- PML thickness for the corresponding PML point. The choice of
ject(s) in the domain , which actually contains the free-space exponents in (5) (i.e., in the numerator and in the de-
region and the region occupied by object(s), for the purpose of nominator) is for the purpose of possible simplifications in the
simplification. The PML region is enclosed within the derivative terms of the Jacobian tensor appearing in the complex
boundaries and . The whole computational domain coordinate transformation. As shown in Section III, the transfor-
is the union of the regions and . We choose four mation in (4) can be implemented easily in a FEM program by
OZGUN AND KUZUOGLU: NON-MAXWELLIAN LOCALLY-CONFORMAL PML ABSORBERS 933

replacing the real-valued node coordinates inside by their PML may also be extended to obtain an anisotropic PML, that
complex-valued counterparts calculated by (4). is, the resulting non-Maxwellian fields in the locally-conformal
Now, consider an outgoing wave in the neighborhood of an PML can be mapped to Maxwellian fields in an anisotropic PML
arbitrary point . This wave can be locally represented as a medium [12], [13]. However, this anisotropic PML implemen-
superposition of plane waves (i.e., local plane wave spectrum) tation is computationally much more complicated than the non-
and a typical representative is given below as (the suppressed Maxwellian locally-conformal PML implementation for FEM
time dependence is of the form ) applications.
The analysis of the locally-conformal PML approach leads us
(6) to the following important conclusions.
(i) The implementation of the method requires the deter-
where is the free-space wave number, is the unit vector mination of the unit vector and the parameter
representing the direction of incidence, and is the unit vector , which can be obtained via the minimization problem
denoting the polarization. It should be mentioned that is not given by (3), which always has a unique solution. It is
arbitrary, and it must satisfy the constraint clear that the differential geometric properties of the
to be in conformity with the assumption that the original field PML-free space interface have no direct effect on the
is outgoing. In order to achieve an efficient PML design, the application of the method. As can be seen from Fig. 2,
plane wave in (6) must be transmitted into without any the analytic continuation operation works equally well
reflection, and furthermore, it must be subject to a monotonic even in the presence of discontinuities in the curvature
decay within in such a way that the magnitude of the of the interface. This is due to the fact that the stretching
transmitted field must be negligible on . operation is locally-defined without using any particular
Then, let be the analytic continuation of to coordinate system.
(ii) The choice of the unit vector and the parameter
complex space given by the expression
, as well as the function , guarantees the decay of
outgoing waves away from the interface. It is clear from
(7) the transformation that this decay be-
havior holds at all PML points, even in problematic re-
It is evident that the explicit form of the expression in (7), gions close to the interface points where the surface cur-
which is , indicates vature is discontinuous.
that the field decays within the PML region away from the PML- (iii) Another important observation is the continuity of
free space interface. , interpreted as follows:
Thus, inside the complex space, the equations governing the Given any , there exists ( depends on ) such that
fields and become
(11)
(8.a) which implies that two closely-located points and in
(8.b) are mapped to also closely-located points and
in . This follows from the definition of the transforma-
where and are free-space permittivity and permeability tion (i.e., the evaluation of and ), and this result
values, respectively, is the analytic continuation of holds for all points in , even close to the interface
points with curvature jumps in problematic regions. It is
to complex space, and is the nabla operator in complex obvious that this result is critical in the implementation of
space and is given by the complex space FEM, explained in Section III.
(iv) It is also clear that we must be cautious while choosing
the value of in order to achieve an efficient PML de-
(9) sign. The value of must be specified properly in order
to satisfy the smooth decay of the transmitted wave in-
where is the Jacobian tensor defined as (in cartesian side . In addition, the magnitude of the transmitted
coordinates) field must attain a negligible value on . With re-
gard to the last requirement, one can claim that the value
of must be as large as possible. However, we also de-
(10) sire that the decay must be smooth, which means that the
decay rate must not exceed a certain value. In addition, a
larger value for yields deformations in the transformed
The fields and in the complex PML region node coordinates calculated by (4) due to very large imag-
are non-Maxwellian in terms of the real coordinate variables, inary parts. In other words, the shape of the mesh elements
but this fact does not imply any difficulty for the implemen- in complex space becomes poor in quality due to the im-
tation of the locally-conformal PML in a FEM application. In proper transformation of the coordinates. This may result
Section III, we show that the FEM formulation can easily be car- in ill-conditioning in the global matrix equation, yielding
ried out within the PML region which is a subset of the complex inaccurate analysis results. We, thus, need to set the op-
space. If a Maxwellian PML is desired, the locally-conformal timal value of . Practically, for a PML thickness between
934 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 55, NO. 3, MARCH 2007

Fig. 3. Mapping of tetrahedral elements to complex tetrahedral elements.

Fig. 4. Isoparametric mapping in FEM formulation.

and , the value of gives reliable results
for an efficient PML design.
From (15), we construct a 6 6 local matrix whose entries
are given by
III. COMPLEX SPACE FEM FORMULATION
In the locally-conformal PML method, Maxwell’s equations
are modified through the complex coordinate transformation,
given by (8). Using these equations, the vector wave equation (16)
in complex space is derived as
The integration in (16) is not carried out directly in terms of
the global coordinates, but the element is mapped to a master
(12) element in local coordinates using the “isoparametric mapping”
(see Fig. 4). In this mapping, both the global coordinates and
The weak form of this wave equation can be calculated using the unknown field are expressed in terms of the same shape
the method of weighted residuals, and is given in complex space functions.
as In each local element, the vector shape functions are ex-
pressed as [8]

(17)
(13)
where is the length of the edge, and are the nodes as-
where is a vector weight function in complex space. sociated with the edge, and and are the scalar shape
In FEM, we solve the weak form of the wave equation in (13) functions for the nodes and , respectively. The edge num-
by discretizing the computational domain using tetrahedral edge bers and the associated nodes and are defined in Fig. 4 by
elements. In the complex coordinate transformation, the tetra- the direction of arrows over each edge. That is, for the edge,
hedral elements are mapped to complex tetrahedral elements is the node at the beginning of the arrow, and is the node at
(i.e., elements with complex node coordinates), as illustrated in the end of the arrow. For instance, for the first edge, is 1 and
Fig. 3. is 2.
Within each element, the unknown field is approximated as In each local element, the scalar shape functions are defined
as [8]
(18.a)
(14) (18.b)
(18.c)
where is the unknown tangential field along the th edge, (18.d)
and is the vector shape function for the th edge.
Using the isoparametric mapping, the coordinate variable
When this expression is substituted into (13), and we use the variations are expressed in terms of the scalar shape functions
Rayleigh-Ritz approach where the weight functions are chosen and the global node coordinates (in cartesian coordinates) as
to be equal to the shape functions (i.e., ), the weak follows:
form of the wave equation becomes
(19.a)

(19.b)

(19.c)
(15)
OZGUN AND KUZUOGLU: NON-MAXWELLIAN LOCALLY-CONFORMAL PML ABSORBERS 935

The unknown field is also expressed as follows: Finally, the implementation algorithm of the locally-con-
formal PML approach in a FEM program is summarized as
follows.
1. Construct the PML region enclosed within the
(20)
boundaries and (see Fig. 2). Preserve the node
information (i.e., ID’s and coordinates) inside , and
Using the expressions in (19), the Jacobian matrix is calcu- on the boundaries and , separately.
lated as 2. Replace the node coordinates inside by their com-
plex counterparts calculated by the complex coordinate
transformation in (4).
3. For each element inside , construct the 6 6 local
matrix whose entries are obtained by (22).
4. Assemble the global matrix by combining the local ma-
trices calculated in step 3.
(21)
5. Solve the global matrix equation for the unknown fields.

The expression in (21) shows that the entries of the Jacobian
matrix are constant, and depend only on the node coordinates.
Then, the expression in (16) becomes
(22)
In order to evaluate this expression, we need to calculate both
and expressions. It should be ob-
vious from (17) and (18), these expressions depend only on , The RCS is calculated by
and terms. The components of these terms are entirely
determined by the inverse of the Jacobian matrix in (21), whose
entries are simply given in terms of global node coordinates, as
follows:
IV. NUMERICAL EXAMPLES
In this section, we present the results of some numerical ex-
periments to test the performance of the locally-conformal PML
method in three dimensional scattering problems. All simula-
tions are performed using our vector FEM software employing
tetrahedral edge elements, and are validated with the results in
[9]–[11]. The common parameters are: is (i.e., is 0.1 m),
is , and is 3. In all examples, the distance between the
PML and the scatterer is approximately , and the approxi-
mate PML thickness is also chosen as . In all the examples
given below, the performance of the PML is tested in terms of
radar cross section (RCS) calculations of objects with “compu-
tationally difficult” geometries that demonstrate the superiority
of the locally-conformal PML approach in handling these cases.
(24)

where and are the far-zone scattered field and the inci-
dent field, respectively. The far-zone scattered field is obtained
by the Huygens’ principle, which is utilized by employing a
closed surface enclosing the object(s) in the free-space region.
From the knowledge of the field values on , the equivalent cur-
rents are calculated by

(25.a)
(25.b)
where and are the total fields on , and is the normal
(23) unit vector directed outward on . Then, the far-zone scattered
field is calculated using these equivalent currents as follows
where is the th entry of the inverse Jacobian matrix.
This analysis shows that the expression in (22) is evalu-
ated in terms of the nodal coordinates. Consequently, the FEM
formulation can easily be implemented in the complex space (26)
by using the complex-valued node coordinates obtained via the
complex coordinate transformation. Although the evaluation of where and are the observation and source points respec-
the expression in terms of the nodal coordinates is given for tively, and is the free space impedance. Since only the far-
tetrahedral elements, it is obvious that similar derivations are field values are taken into account in the RCS calculations, it is
possible for other types of elements (for example, hexahedral). evident that the smoothing effect of the far-field calculation may
936
Fig. 5. Bistatic RCS profile of PEC conesphere: (a)  polarization, (b) 
polarization.
Fig. 6. Bistatic RCS profile of sharp PEC conesphere in  polarization.
reflect a reduction in the magnitude of errors present in the near
field terms. A detailed analysis of the near field error variations
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 55, NO. 3, MARCH 2007
Fig. 7. Monostatic RCS profile of PEC cone as a function of ka where a is the
base radius of the cone.
Fig. 8. Monostatic RCS profile of PEC Ogive as a function of the maximum
diameter in wavelengths.
as a function of PML parameters, is presently under investiga-
tion and will be presented elsewhere.
The first example is a PEC conesphere whose half-cone
angle is 45 degrees, base radius is and cone-height is
. The bistatic RCS profile of the conesphere is plotted in
Fig. 5, assuming an axial (nose-on) plane-wave incidence. The
mesh cross-section of the whole computational domain and the
mesh of the scatterer are also shown in Fig. 5. These meshes
are just coarse meshes given for the purpose of visualization.
The different gray-shades in the mesh cross-section correspond
to the scatterer, free-space and PML regions, respectively. A
similar approach is followed in the remaining examples to
present the results.
The second example is a sharp PEC conesphere whose half-
cone angle is 10 degrees, base radius is and cone-height
is . The bistatic RCS profile of the conesphere is plotted in
OZGUN AND KUZUOGLU: NON-MAXWELLIAN LOCALLY-CONFORMAL PML ABSORBERS
Fig. 6, assuming an axial (nose-on) plane-wave incidence. The
third example is a PEC finite cone with circular base, whose
half-cone angle is 20 degrees. The monostatic RCS profile of
the cone is plotted in Fig. 7 as a function of ka where is the
base radius of the cone. The fourth example is a PEC Ogive,
whose nose angle is 75 degrees. The monostatic RCS profile
of the Ogive is plotted in Fig. 8 as a function of the maximum
diameter of the Ogive in wavelengths. In the last two examples,
we again assume an axial (nose-on) plane-wave incidence.
The test examples demonstrate that the results obtained by
the locally-conformal PML are very close to the results in the
literature, and the locally-conformal PML is very reliable even
in handling challenging geometries.
V. CONCLUSION
In this paper, we have introduced the locally-conformal
PML method as a generalization of the complex coordinate
stretching concept. We have shown that the implementation of
this method in a FEM program is very simple and practical,
and it makes easier the design of conformal PMLs enclosing
arbitrarily-shaped convex sets, which contain especially some
intersection regions with different interface curvatures. We
have concluded that we do not need to employ any artificial
anisotropic material or any local coordinate frame to imple-
ment a PML region. After we interchange the real-valued node
coordinates inside the PML region with the complex-valued
node coordinates obtained by the complex coordinate transfor-
mation, we achieve a successful PML design without changing
the original FEM formulation. Finally, we have investigated
the accuracy of the method by means of some numerical com-
parisons with the literature data, and we have shown that the
locally-conformal PML provides very accurate results even in
difficult geometries.
REFERENCES
[1] J. P. Berenger, “A perfectly matched layer for the absorption of electro-
magnetic waves,” J. Comput. Phys., vol. 114, pp. 185–200, Oct. 1994.
[2] W. C. Chew and W. Weedon, “A 3D perfectly matched medium from
modified Maxwell’s equations with stretched coordinates,” Microw.
Opt. Technol. Lett., vol. 7, pp. 599–604, 1994.
[3] Z. S. Sacks, D. M. Kingsland, R. Lee, and J.-F. Lee, “A perfectly
matched anisotropic absorber for use as an absorbing boundary con-
dition,” IEEE Trans. Antennas Propag., vol. 43, pp. 1460–1463, Dec.
1995.
937
[4] M. Kuzuoglu and R. Mittra, “Investigation of nonplanar perfectly
matched absorbers for finite element mesh truncation,” IEEE Trans.
Antennas Propag., vol. 45, pp. 474–486, 1997.
[5] ——, “Mesh truncation by perfectly matched anisotropic absorbers in
the finite element method,” Microw. Opt. Technol. Lett., vol. 12, pp.
136–140, 1996.
[6] F. L. Teixeira and W. C. Chew, “Systematic derivation of anisotropic
PML absorbing media in cylindrical and spherical coordinates,” IEEE
Microw. Guided Wave Lett., vol. 7, pp. 371–373, Nov. 1997.
[7] ——, “Analytical derivation of a conformal perfectly matched absorber
for electromagnetic waves,” Microw. Opt. Technol. Lett., vol. 17, pp.
231–236, Mar. 1998.
[8] J. Jin, The Finite Element Method in Electromagnetics. New York:
Wiley, 2002.
[9] L. N. Medgyesi-Mitschang and D. S. Wang, “Hybrid solutions for scat-
tering from perfectly conducting bodies of revolution,” IEEE Trans.
Antennas Propag., vol. AP-31, pp. 570–583, 1983.
[10] J. B. Keller, “Backscattering from a finite cone—Comparison of theory
and experiment,” IEEE Trans. Antennas Propag., vol. 9, pp. 411–412,
1961.
[11] W. E. Blore, “The radar cross section of Ogives, double-backed cones,
double-rounded cones, and cone spheres,” IEEE Trans. Antennas
Propag., vol. 12, pp. 582–590, 1964.
[12] I. V. Lindell, Methods for Electromagnetic Field Analysis. New York:
Oxford Univ. Press, 1992.
[13] F. L. Teixeira and W. C. Chew, “General closed-form PML constitutive
tensors to match arbitrary bianisotropic and dispersive linear media,”
IEEE Microw. Guided Wave Lett., vol. 8, pp. 223–225, 1998.
Ozlem Ozgun (S’05) was born in Turkey on July 8,
1976. She received the B.S. and M.S. degrees in elec-
trical engineering from Bilkent University, Ankara,
Turkey, in 1998 and 2001, respectively. She is cur-
rently working toward the Ph.D. degree in electrical
engineering in the Middle East Technical University,
Ankara, Turkey.
From 2000 to 2004, she worked in the Scien-
tific and Technical Research Council of Turkey
(TUBITAK)—National Research Institute of Elec-
tronics & Cryptology (UEKAE), Ankara. From 2004
to 2005, she worked in the Aselsan Inc., Microwave & System Technologies
Division—Radar Electronic Warfare Department, Ankara. Her main research
interests include numerical electromagnetics, FEMs, electromagnetic propaga-
tion and scattering.
Mustafa Kuzuoglu (M’92) received the B.Sc., MSc.,
and Ph.D. degrees in electrical engineering from the
Middle East Technical University (METU), Ankara,
Turkey, in 1979, 1981, and 1986, respectively.
He is currently a Professor at METU. His research
interests include computational electromagnetics, in-
verse problems and radars.