IEEE MICROWAVE AND GUIDED WAVE LETTERS, VOL. 7, NO.

9, SEPTEMBER 1997 273

A Locally Conformal Finite-Difference

Time-Domain (FDTD) Algorithm for Modeling
Three-Dimensional Perfectly Conducting Objects
Supriyo Dey and Raj Mittra

Abstract— A novel conformal finite-difference time-domain

(CFDTD) technique for locally distorted contours that accurately
model curved metallic objects is presented in this paper. This
approach is easy to implement and is numerically stable. Several
examples are presented to demonstrate that the new method
yields results that are far more accurate than those generated
by the conventional staircasing approach. Example geometries
include cylindrical and spherical cavities, and a circular mi-
crostrip patch antenna. Accuracy of the scheme is demonstrated
by comparing the results derived from analytical and Method of
Moments (MoM) techniques.
Index Terms— Finite-difference time-domain (FDTD), locally
conformal grid, Maxwell solver.

I. INTRODUCTION

T HE staircasing approach to analyzing objects with curved

metallic surfaces using the Yee algorithm [1] not only
introduces errors [2] due to inaccurate approximation of the
geometry, but can also generate spurious solutions. Several
techniques have been proposed in the literature for overcoming
these difficulties. These include the globally curvilinear grid
technique [3], which requires a special structured type of mesh
that may be difficult to generate when modeling an arbitrary
Fig. 1. Cross-sectional view of the quarter of the spherical cavity.
shaped object, and the contour path finite-difference time-
domain (CPFDTD) scheme [4] that deforms the grid only
locally to accommodate the curvature of the surface. Although II. CONFORMAL FINITE-DIFFERENCE
simple and efficient, CPFDTD frequently leads to instabilities TIME-DOMAIN (CFDTD) ALGORITHM
because of noncausal and nonreciprocal nearest neighbor Consider the cross section of a spherical cavity resonator
approximation [5], [6]. Various modifications in the CPFDTD shown in Fig. 1, where, for convenience, we display only
algorithm have been proposed in the literature to obviate this one of its quadrants. The mesh for this structure is depicted
instability problem [7], [8]; however, the bookkeeping, mesh by the solid dark lines in the figure, which also shows an
generation, and programming are considerably more complex overlay of a uniform Cartesian mesh. The undistorted cells
in these schemes than they are in the conventional FDTD are treated in the usual way in the CFDTD algorithm, for
algorithm. both the - and -fields. For the distorted cells, no special
In a recent communication, the present authors have re- treatment is needed for the electric field, which is assumed
ported a simple yet accurate technique for the FDTD analysis to be constant along the edge of a cell that resides within
of curved two-dimensional (2-D) perfectly conducting bodies the cavity, and zero if it is located on the metallic surface
[9] using a locally conformal grid. In this paper, we extend or within the conductor. However, the treatment of the -
this technique to three-dimensional cases and illustrate its field is somewhat different. It is assumed to be located at
application by investigating a number of test geometries. the center of the corresponding undistorted cell (the Cartesian
cell obtained by removing the partial filling) for the purpose
of numerical calculations, irrespective of whether the location
of the center is inside or outside the computational domain.
Manuscript received March 19, 1997. It is also assumed to be constant over the region of the cell
The authors are with the Electromagnetic Communication Research Labo-
ratory, Pennsylvania State University, University Park, PA 16802 USA. that is inside the cavity. Thus, the updating procedure of the
Publisher Item Identifier S 1051-8207(97)06173-4. -field is changed little from the conventional FDTD scheme.
1051–8207/97$10.00  1997 IEEE

Authorized licensed use limited to: ULAKBIM UASL - Akdeniz Universitesi. Downloaded on December 29,2023 at 08:59:26 UTC from IEEE Xplore. Restrictions apply.
274 IEEE MICROWAVE AND GUIDED WAVE LETTERS, VOL. 7, NO. 9, SEPTEMBER 1997

We now offer a possible justification for the fact that we

always locate the -field at the center of the cell, regardless
of whether it is partially filled or not. We begin with the
observation that, conventionally, the updating of the magnetic
field in a regular cell is carried out by summing up the
contributions of the electric fields tangential to the boundaries
of the cell. Next, we consider the case where the cell is
partially filled with a lossy material. It may be verified that
the updating of the -field as well as its location remains
unchanged with the introduction of this modification, and this
is the basis of our argument for following the procedure for
time-stepping of the -field that we have presented above.
However, it is also obvious that the electric field in the region
interior to the conducting medium becomes negligible in the
limit of a perfectly conducting filling, and, hence, we need Fig. 2. Resonant frequencies of cylindrical cavities calculated using various
only retain the contributions of the -field on the portions of methods.
the cell contour that are outside of the conducting region.
The above procedure enable us to employ the regular FDTD 2) The ratio between the maximum length of the side of
equations to update the magnetic field by using the electric a cell and its area is less than 15 (for 50% of Courant
field values along the distorted contour that are appropriately limit) and 10 (for 70% of limit).
weighted with the lengths of the contours. The updating
Finally, if desired, the CPFDTD scheme can be used to
equation for the -field along the -direction reads
supplement the CFDTD algorithm to deal with the cells in
which the above conditions are violated.

III. NUMERICAL RESULTS

To demonstrate both the accuracy and stability of the pro-
(1) posed approach, we have calculated the resonant frequencies
of the dominant TE and TM modes of cylindrical cavities and
where and are the cell lengths along the and the two lowest order resonant frequencies of spherical cavities.
directions, respectively. The usefulness of the approach has been further demonstrated
Once the -fields have been computed, the -fields are by computing the resonant frequency of a circular microstrip
updated in the conventional manner by using the adjacent patch antenna. All of these results were obtained by using 8192
-field values. For example, the updated -field along the time steps. Additional runs were carried out with up to 16 000
-direction is derived from time steps and no instability was ever observed.

A. Cylindrical Cavity
The proposed algorithm was employed to calculate the
resonant frequencies of the dominant TE and TM modes
for circularly cylindrical cavities, with heights of 30 cm and
radii ranging from 18 to 26 cm. In each case, the spatial
discretization was chosen to be 3 cm. The results are shown
(2) in Fig. 2 and are compared with those obtained by using the
staircase approximation, as well as with the analytical results.
where, and are the step size along and directions. In the above figure, the legend CFDTD refers to the present
Since all the field values are updated without borrowing conformal FDTD method. The worst case and the average error
from any of the adjacent cells, as is done in CPFDTD, due to staircasing is 8.45% and 3.76%, respectively, while the
the associated stability problems are no longer present in corresponding figure for the present scheme is only 0.3% and
this scheme. However, the stability of this algorithm is still 0.14%.
governed by the nature of the mesh and the choice of the time
step. Numerical experiments have shown that a time step of 50 B. Spherical Cavity
to 70% of the Courant limit associated with the undistorted cell To demonstrate the full three-dimensional capability of the
is adequate to ensure the stability of the algorithm, provided algorithm, two lowest order resonant frequencies of the spher-
that the following conditions are met. ical cavity have been computed. Calculations were carried out
1) The area of the distorted cell (partially filled) is greater for spheres with radii ranging from 14 to 24 cm by using
than 1.5% (for 50% of Courant limit) and 2.5% (for a spatial discretization of 4 cm in size. It is evident that
70% of limit) of the area of the undistorted cell area. this choice of the discretization leads to a relatively coarse

Authorized licensed use limited to: ULAKBIM UASL - Akdeniz Universitesi. Downloaded on December 29,2023 at 08:59:26 UTC from IEEE Xplore. Restrictions apply.
IEEE MICROWAVE AND GUIDED WAVE LETTERS, VOL. 7, NO. 9, SEPTEMBER 1997 275

IV. CONCLUSION
In this paper, we have shown that a simple accurate stable
and locally conformal FDTD algorithm, previously developed
for the 2-D case, can be successfully extended to three-
dimensional geometries. The scheme has been found to be
numerically stable, because the problem associated with bor-
rowing from adjacent cell, as for instance in CPFDTD, is
not present here. Furthermore, the algorithm generates no
spurious solutions and yields results that are considerably more
accurate than those obtained by using the staircasing approach.
There are several attractive features of this algorithm: first, the
bookkeeping involved is relatively simple compared to those
required in other explicit nonorthogonal FDTD schemes. Sec-
ond, an automatic mesh generation procedure can be utilized
Fig. 3. Resonant frequencies of spherical cavities calculated using various in conjunction with this algorithm because the specifications
methods.
for the mesh are not very demanding. Finally, the allowable
time step for stable calculations can be much higher ( order
mesh and, hence, provides a stringent test for the algorithm. of magnitude) than dictated by the smallest-size distorted cell,
Fig. 3 shows the comparison between the resonant frequencies and this can result in a considerable saving of the computation
predicted by the present technique, the analytical results, and time.
those derived by using the conventional staircasing approach. REFERENCES
It can be seen that accurate results are obtained with the [1] K. S. Yee, “Numerical solution of initial boundary value problems in-
CFDTD method even with the relatively course mesh that has volving Maxwell’s equations in isotropic media,” IEEE Trans. Antennas
been employed. The worst and average staircasing errors are Propagat., vol. AP-14, pp. 302–307, May 1966.
[2] A. C. Cangellaris and D. B. Wright, “Analysis of the numerical error
12.96% and 4.8%, respectively, while the same for the present caused by the stair-stepped approximation of a conducting boundary
scheme are only 1.14% and 0.69%, respectively. in FDTD simulations of electromagnetic phenomena,” IEEE Trans.
Antennas Propagat., vol. 39, pp. 1518–1525, Oct. 1991.
[3] J. F. Lee, R. R. Palandech, and R. Mittra, “Modeling three-dimensional
C. Circular Microstrip Patch discontinuities in waveguides using nonorthogonal FDTD algorithm,”
IEEE Trans. Microwave Theory Tech., vol. 40, pp. 346–352, Feb. 1992.
[4] T. G. Jurgens and A. Taflove, “Three-dimensional contour FDTD
The last example considered was a probe-fed circular mi- modeling of scattering from single and multiple bodies,” IEEE Trans.
crostrip patch antenna with the following parameters: dielectric Antennas Propagat., vol. 41, pp. 1703–1708, 1993.
constant 4.0, thickness 4.0 mm, and radius 10 mm. The cell [5] N. K. Madsen, “Divergence preserving discrete surface integral method
for Maxwell’s curl equations using nonorthogonal unstructured grids,”
size along the thickness of the substrate was 0.8 mm while it Rep. UCRL-JC-109787, Lawrence Livermore National Laboratory,
was 2.5 mm along the other two directions. The fundamental 1992.
resonant frequency of the patch obtained by using the present [6] M. Okoniewski, J. Anderson, M. Mrozowski, and S. S. Stoutly, “Arbi-
trarily located metal surfaces in FDTD technique,” Progress in Electro-
method, the staircased FDTD and the MoM were, 3.75, 3.68, magnetics Research Symp., Seattle, WA, July 1995, p. 178.
and 3.80 GHz, respectively. The CFDTD result is 1.31% lower [7] C. J. Railton and I. J. Craddock, “Analysis of general 3-D PEC structure
than the one predicted by MoM, whereas the one computed using improved CPFDTD algorithm,” Electron. Lett., vol. 31, no. 20,
pp. 1753–1754, Sept. 1995.
by using the staircased FDTD algorithm is 3.16% lower. [8] J. Anderson, M. Okoniewski, and S. S. Stuchly, “Practical 3-D con-
The difference between the CFDTD and MoM results for the tour/staircase treatment of metal in FDTD,” IEEE Microwave Guided
resonant frequency can be attributed to the fact that the radius Wave Lett., vol. 6, pp. 146–148, Mar. 1996.
[9] S. Dey, R. Mittra, and S. Chebolu, “A technique for implementing the
of the feed probe modeled with the MoM is finite, whereas it FDTD algorithm on a nonorthogonal grid,” Microwave Opt. Technol.
is assumed to be infinitesimal in the FDTD simulation. Lett., vol. 14, no. 4, pp. 213–215, Mar. 1997.

Authorized licensed use limited to: ULAKBIM UASL - Akdeniz Universitesi. Downloaded on December 29,2023 at 08:59:26 UTC from IEEE Xplore. Restrictions apply.