354 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 45, NO.

3, MARCH 1997

The Finite-Difference Time-Domain (FDTD)

and the Finite-Volume Time-Domain (FVTD)
Methods in Solving Maxwell’s Equations
Kane S. Yee, Fellow, IEEE, and Jei S. Chen, Member, IEEE

Invited Paper

Abstract— The finite-difference time-domain (FDTD) and its physics and the numerical approximation can further be made
current generalizations have been demonstrated to be useful and transparent by using the integral forms of Maxwell’s equations.
powerful tools for the calculation of the radar cross section (RCS) The disadvantages of FDTD include the requirement of a
of complicated objects, the radiation of antennas in the presence
of other structures, and other applications. The mathematical large computer when the linear dimension of the scattering
techniques for conformal FDTD have matured; the primary object is large compared to the wavelength and the error
impediments to its implementation are the complex geometries introduced by the numerical dispersion of the algorithm. When
and material properties associated with the problem. Even under an object is large (compared to wavelength), the effect due to
these circumstances, FDTD is more flexible to implement because the interaction of distance points in the object is not accurate
it is based on first principles instead of a clever mathematical
trick. It is the purpose of this paper to give an account of some because of the error introduced by the numerical dispersion.
new results on conformal FDTD obtained by the authors and
their associates at Lockheed Martin Space Company since 1988.
The emphasis is on nonsmooth boundary condition simulation.
II. THE FDTD AND THE FVTD ALGORITHMS
This topic, to the best of our knowledge, has not been widely There are two integral forms of Maxwell’s equations that
investigated due to its complexity. are useful in the time-domain numerical solutions of electro-
Index Terms— FDTD methods. magnetic problems. These are

(1a)
I. INTRODUCTION

A. Literature on FDTD (1b)

and
T HE finite-difference time-domain (FDTD) method is now
widely used in various applications, and the literature on
it is extensive. We refer the readers to recent books [1], [2]
(2a)

and a literature survey [3]. Further discussion of work done

(2b)
in conformal FDTD can be obtained from the references cited
in [4].
where, in the above equations, and are arbitrary surfaces
and and are arbitrary volumes.
B. Advantages and Disadvantages of Using FDTD
The discretizations based on (1a) and 1(b) have been com-
The main advantages of FDTD-based techniques for solving monly known and will be referred to as FDTD, generalized,
electromagnetic problems are simplicity and the ability to and the discretizations based on (2a) and (2b) have been known
handle complex geometries (in principle). Derivation from and will be referred to as finite-volume time-domain (FVTD).
Maxwell’s equations is based on traditional mathematical Equations (1a) and (1b) are the mathematical description of
methods of approximating derivatives by finite differences Faraday’s law and Ampere’s law, respectively. These integral
and approximating line integrals, surface integrals, and vol- forms and the partial differential equations are all equivalent
ume integrals by summations. It involves basic arithmetic under smoothness assumptions. Our new proposed conformal
operations—addition, subtraction, multiplication, and division. method will be a combination of FDTD and FVTD.
FDTD can model material through parameter averaging or,
in case of a more sophisticated approach, choosing grids to A. The Leapfrog FDTD and FVTD ([5], [6])
conform to the geometries of the material boundaries. The
The numerical approach we present involves no more than
Manuscript received March 1, 1996; revised November 13, 1996. accurate evaluation of line, surface, and volume integrals. We
The authors are with the Lockheed Martin Palo Alto Research Laboratory,
Palo Alto, CA 94304 USA. shall use time leapfrog in updating the electric field and the
Publisher Item Identifier S 0018-926X(97)02285-0. magnetic field. We introduce the following terminology:
0018–926X/97$10.00  1997 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 YEE AND CHEN: FDTD AND FVTD METHODS IN SOLVING MAXWELL’S EQUATIONS 355

The line integral on the right-hand side of (1a) is the electric are the electric vector or ; we store the three rectangular
circulation around the curve ; the curve is an electric components of the electric field at these vertexes. A face
contour. The line integral on the right-hand side of (1b) is the (a set of ordered electric vertexes) of an electric element
magnetic circulation around the curve ; the curve is an electric face. The “center” of an electric element is a
is a magnetic contour. The surface integral on the right-hand magnetic vertex. Two magnetic vertexes are connected if the
side of (2a) is the electric vorticity over the surface . The two corresponding electric elements share a common face.
surface integral on the right-hand side of (2b) is the magnetic We thus generate a collection of magnetic elements; and
vorticity over the surface . the collection constitutes the magnetic grid. Located at the
The leapfrog time discretizations of the above equations are magnetic vertexes are the magnetic vector or . Updates
of and (we assume that and for
simplicity) are calculated using (4a) and (4b). The volume in
(4a) is an electric element, and the volume in (4b) is a magnetic
(3a) element. If a face of an electric element is not coplanar, we
make the electric face definite by adding the barycenter whose
rectangular coordinates are the average of those of the vertexes
of the face.
(3b)
B. The Overlapping-Grid FDTD/FVTD Hybrid
Our FDTD/FVTD hybrid uses a composite grid consisting
of a rectangular grid intersecting a body conformal grid. The
conformal grid consists of an electric grid and a magnetic
(4a)
grid. For smooth scatterers the electric grid is a few layers
of prisms erected along the normal of the scattering surface.
Our main goal is to make use of the triangular surface grid
usually generated for the MOM frequency-domain integral
(4b) equation methods. Therefore, our conformal electric grid will
consist of layers of triangular prisms. No more than five
In the above equations, the superscript indicates the time level layers are needed. The conformal electric grid intersects an
when the variables are evaluated. outer rectangular grid in which the FDTD algorithm is used.
The above identities are subject to the right-handed orien- Constructed from the conformal electric grid is the magnetic
tation of vector analysis. Many time-domain algorithms can grid. The height of the prisms is approximately the same as
be obtained by judicious choices of the curves and surfaces. , the spatial length of the FDTD grid. The lengths of
In fact, our updating algorithms are almost self evident from the sides of the surface triangles are comparable to . In
(1a)–(2b), and can be loosely stated as the interior electric vertexes or the magnetic vertexes (points
1) for FDTD: which are neither grid boundary points nor physical boundary
a) use the electric circulation around an electric contour points), the FVTD algorithm is used for updating. The interior
to update the component of the magnetic field along boundary of our rectangular grid is 0.5 (0.5 zone) to
the area vector of the surface enclosed by this (one zone) away from the scatterer. The outer boundary
contour; electric field on the FVTD electric grid is obtained through
b) use the magnetic circulation around a magnetic interpolation of the calculated rectangular FDTD electric field;
contour to update the component of the electric field the rectangular FDTD electric field at the inner boundary
along the area vector of the surface enclosed by this of the rectangular grid is obtained through interpolating the
contour. calculated FVTD field.
2) for FVTD:
C. The Double Interpolation
a) use the electric vorticity over a surface to update the
magnetic vector in the center of the volume enclosed At each time step, we carry out a double interpolation for the
by the surface; data from one grid to another grid. The two intersecting grids
b) use the magnetic vorticity over a surface to update must be such that the data needed for the interpolation are
the electric vector in the center of the volume calculated from the respective algorithms. Details of double
enclosed by the surface. interpolations are given in [5], [6].
The accuracy of the algorithm depends on careful selection
of the curves enclosing the surface and the surfaces enclosing D. Locally-Distorted-Grid FVTD and FDTD
the volume. In computation, we assume that we have a grid The FVTD algorithm, when applied to an electric grid
consisting of a collection of cubes, distorted cubes, prisms, consisting of cubes, is almost identical to rectangular FDTD. If
tetrahedrons, or any combination of these “elements.” This the outer boundary vertexes of the FVTD are the same as the
grid will be called an electric grid. The vertexes of the electric inner vertexes of the FDTD, it is quite simple to interpolate
grid are the electric vertexes. Located at the electric vertexes the FVTD field from the FDTD field and vice versa on the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 356 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 45, NO. 3, MARCH 1997

Fig. 1. The inner boundary of the FVTD grid.

interface between these two algorithms. Thus, when a locally

distorted electric grid is available, one could use the FVTD on
the distorted part of the grid and the FDTD on the undistorted Fig. 2. Illustrating the interpolation and extrapolation for the electric field
part of the grid. One can also use FVTD throughout. The only E at the point p.

advantage of FDTD in this case is its bookkeeping simplicity.

E. Totally Unstructured Grid

The time leapfrog and spatially staggered algorithms based
on the integral forms of the Maxwell’s equations do not
require a structured grid, although it is advisable to use a
structured grid when it is suitable for a particular problem.
An unstructured grid can be a union of electric elements
of various shapes with the corresponding magnetic elements.
Furthermore, with our newly discovered radiation boundary- Fig. 3. A bordering edge one through two.
condition approximation (Section IV), we do not require a
regular outer-computation boundary. (to the surface) component of the electric field is obtained
through extrapolation (from point two). Intuitively, this is more
III. BOUNDARY CONDITIONS accurate than the staircasing PEC boundary-condition approx-
The perfectly electrically conducting (PEC) boundary- imation and less accurate than the interpolation in a good
condition simulation and the impedance boundary condition conformal grid. More details of this method is given in [8].
(IBC) simulation for smooth surfaces with a conformal grid
are relatively easy to formulate and have been explained B. PEC Boundary Condition for Nonsmooth Scattering
in our previous papers [5]–[7]. Here, we shall give a new Surface with a Conformal Grid [9]
interpolation-and-extrapolation method based on a rectangular The boundary condition simulation for a smooth boundary
grid alone for a smooth PEC boundary, a more accurate yet requires a unique surface normal at each boundary point. In
general method to simulate PEC boundary with edges and general, it is not applicable for surfaces with corners or edges.
corners, and a method to simulate boundary condition across Here, we present a method that is similar to that in [10], yet
different media. The general method can also be used for different in that we do not average overall directions. Our
smooth boundaries. method will retain some directional dependence near singular
points on the surface. We introduce the following terminology
A. Smooth PEC Boundary Boundary Node: A boundary node (vertex) is a vertex at
In this approximation, the corners and edges of a surface the scattering surface. A bordering edge is an edge having
are rounded so that the surface is smooth. One then constructs one and only one end point at a boundary vertex. Let the edge
the union of FDTD cubes containing the closure of the be a bordering edge with point one at the boundary and
volume enclosed by the scattering surface. The outer-boundary point two at the interior of the computation volume (Fig. 3).
vertexes of this union of cubes will be interior boundary Let “ ” be the midpoint of the bordering edge and be the
vertexes of a rectangular computation grid for the FVTD unit vector of the area vector of the magnetic contour
algorithm (Fig. 1). Through each boundary vertex there passes surrounding the edge . FDTD will yield .
a surface normal; at each of the boundary nodes and at each 1) From the FVTD data, is known.
time step we fix the value of the electric vector. The foot of 2) Extrapolation will yield .
the normal at the scattering surface will give two tangential Repeat the above procedure for each of the bordering edges.
directions (Fig. 2). The tangential (to the surface) components Let be an element having a boundary point . This element
of the electric field at a boundary vertex are obtained through will have either one, two, or three bordering edges leading
interpolation (between points one and two), and the normal away from the point to the interior of the computation region

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 YEE AND CHEN: FDTD AND FVTD METHODS IN SOLVING MAXWELL’S EQUATIONS 357

Fig. 4. The edges leading away from a point p (a pie element).

Fig. 6. The path to update a tangential component of the electric vector E.

for the surface integral. The whole process is the same as if
we use the average of the two values for associated with
the elements and , respectively, in the calculation of the
electric vorticity over the face .
The above algorithm is quite easy to implement for a
Fig. 5. The edges leading away from a point p (a tetrahedon). prismatic type of grid.

(Figs. 3–5). The total number of bordering edges and boundary C. The Boundary-Condition Simulation
edges that belong to the element and that share the point as Between Two Different Media
a vertex is three. Let and be the midpoints of the three
In Fig. 6, regions one and two are lossy dielectrics. Across
edges and let and be the respective unit area vectors
the interface, both the tangential components of the electric
of the area vectors and enclosed by the respective
field and the magnetic field are continuous. The way FDTD
magnetic contours. We then have three pieces of information
positions the variables makes it natural and simple to simulate
obtained through FDTD. They are and
the boundary condition. In the FDTD/FVTD hybrid, we make
. These data can be used, either through extrapolation
use of FDTD to help us simulate the boundary condition
or through boundary conditions, to yield an approximation for
numerically.
and . Since the vectors
Region one has the material parameters and
and are linearly independent (the grid must be such that this
region two has the material parameters . These
condition holds), the rectangular components
two regions are separated by a surface (the interface). The
and of the vector can be determined. This vector
surface normals pointing into the respective regions are and
will be taken to be the value of the electric vector if the vertex
. Setting , the boundary conditions are
is approached from the interior of the element . We repeat
and . As a consequence
the process for each element having as a boundary vertex
and obtain the following.
Let be the number of elements sharing a boundary vertex and
; there will be values associated with the electric vector
at the point . These values will not be the same in general.
If is a boundary vertex of an open surface, these values Let be a tangential direction on the interface. We choose a
can not be all the same. magnetic contour so that the area vectors are and
How do we make use of these values of in calculations? . Then, the Ampere–Maxwell law yields
The purpose of having the electric vector at a node is
to supply the information for the calculation of the electric
vorticity over the faces of an electric element so that the
magnetic vector at the centroid can be updated. Each face is
shared by two elements. Let be a face having as one of it
vertexes. If is a boundary face, then the integral
is known. If it is not a boundary face, it will be shared by two (5)
elements and . We first calculate the integral Integrating along a curve on the interface yields
with the values of the electric vector at the vertexes associated
with the element . We then calculate the integral
with the values of the electric vector at the vertexes associated
with the element . We then take the average of the two values (6)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 358 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 45, NO. 3, MARCH 1997

(a) (b) (c)

Fig. 7. The tapered damping zones for a rectangular outer boundary. (a) Zone #1. (b) Zone #2. (c) Zone # . N
where , are areas in regions one and two , and as follows:
is the total area enclosed. In the approximation for the area
integral (5), we have assumed that is constant throughout
the enclosed area (flat top assumption). A more accurate
assumption (as numerically demonstrated later) would be that
is linear in and in along the surface normal
direction (roof-top assumption).

IV. THE RADIATION CONDITION SIMULATION [11]

In time-domain FDTD calculations in electromagnetic prob-
lems, one must truncate the domain of calculation. Since the
field variables at the outer-calculation boundary cannot be
updated by the numerical algorithm derived from the differ-
ential equations, an algorithm based on physical conditions (7a)
must be obtained. For the scattered electromagnetic field,
the so-called radiation boundary condition (RBC) states that
the electromagnetic energy propagates toward “infinity.” Until
recently, the mathematical implementation of the RBC has
been based on the differential equations describing outgoing
behavior. The splitting of the Maxwell’s equations in the
perfectly matched layer (PML) method by Berenger [15],
in 1994, is an innovative breakthrough, but it is limited to
rectangular grids. So far, splitting in nonrectangular coor-
dinates or in an irregular grid has not be discovered. For
years our experimental colleagues have approximately solved
the outgoing condition problems by aligning the experimental
chambers so that the incident electromagnetic wave does not (7b)
reflect back. The tapered damping described below is based
on this physical motivation. We observe that (7a) and (7b) are exactly the finite-difference
equations for a lossless medium except for the factors within
A. Motivation for an Alternative RBC Approximation
the first set of curly brackets. These two factors take on
In our implementation of the FDTD/FVTD method [5], [6], values between zero and one. In the following, we shall refer
we discovered that the existing RBC approximations (namely, to them as electric and magnetic tapered damping functions,
the Mur [12] or Liao [13]) are not readily amenable to respectively. It will be shown, heuristically, that these two
parallel computation. Furthermore, they are restricted mainly damping functions should be the same. Equations (7a) and
to a rectangular outer boundary. To alleviate this problem, (7b) enable us to interpret the numerical calculations as a two-
we searched for possible alternatives to the known RBC stage process. In the first step, the field variables are calculated
approximations. We have found that the following alternative as if there is no loss. The second step is to multiply the
is easy to explain, simple to implement, and above all is resulting field variables by the loss factor which is a function
suitable for parallel computation. of position only, if the material parameters are functions of
position only. We shall use the above interpretation of lossy
B. Heuristic Mathematical Justification for material to devise a damping function instead of the usual
Rectangular Outer Boundary RBC approximation. Fig. 7 shows the computation volume
Two of the FDTD finite-difference equations for an electri- and its outer computational boundary. A few zones around
cally lossy and magnetically lossy material can be expressed the outer boundary are also shown, as well as a tapered

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 YEE AND CHEN: FDTD AND FVTD METHODS IN SOLVING MAXWELL’S EQUATIONS 359

(a) (b)
Fig. 8. The tapered damping zones for a nonrectangular outer boundary. (a) Zone #1. (b) Zone #2.

function taking its value between zero and one. Inside the Let us consider our choice of the tapered damping function
smallest box, the value of this tapered function is one. Fig. 8 in more detail. Referring to Fig. 7(a) and (b), we let (
shows the outer boundary of another computation volume, is usually chosen to be ten in our simulation) be the number
indicating that there is no requirement that the boundary be of damping zones and let be the damping
a rectangular one. zone number starting from the outer computational boundary.
We define a tapered function
C. The Choice of the Tapered Damping Functions
for a Rectangular Outer Boundary
There is an infinite number of ways to choose the tapered
damping functions. We want to emphasize that we do not
choose the electric conductivity nor the magnetic conductivity. We then define the damping function
We choose the damping functions! The electric conductivity
1) , if ( ) does not belong to any
and the magnetic conductivity are implied when the damping
damping zone;
functions are chosen. For more physical insight, we recast the
2) for if
two damping factors as follows:
( ) belongs to the damping zone .
For the electric vector
We have also tried tapered functions such as
, , and exponential
functions. These make little difference in the results.

(8) D. Procedures of Calculation with

Rectangular Outer Boundary
It is straightforward to implement our tapered damping idea
as a substitute for the Mur or Liao RBC. Specifically, we do
For the magnetic vector
the following for the scattered field.
1) Set the scattered field variables located at the outer
computational boundary to zero and never update them.
2) Update the scattered magnetic field variables by the
FDTD algorithm for a lossless medium
(9)

The two damping factors would be the same if .

This is, however, just the impedance-matching condition. In
one-dimensional wave propagation, such a medium will not
cause any reflection. Therefore, both electric and magnetic 3) Multiply each magnetic field variable (obtained in step
damping should be used, and the two damping functions 2) by the value of the magnetic tapered damping function
ought to be the same. Therefore, we choose at that point
. Such a heuristic conclusion
has been demonstrated in our numerical experimentation in
the scattering by a square plate and in the scattering of
a sphere [11].

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 360 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 45, NO. 3, MARCH 1997

(a) (b)
Fig. 9. (a) RCS of a 1 m radius PEC sphere, calculated with a rectangular grid with interpolation and extrapolation (tapered damping RBC). (b) RCS of
a PEC ellipsoid, calculated with a rectangular grid with interpolation and extrapolation (tapered damping RBC).

(a) (b)
Fig. 10. RCS for EMCC target #1 (flat PEC triangle-circle plate), calculated with overlapping grids and Mur RBC. (a) V–V pol. (b) H–H pol.

4) Update the electric field variable by the FDTD algorithm boundary. This is necessary because it is sometimes more con-
for a lossless medium venient to use an unstructured irregular grid for computation.
When we use a single grid, the outer-computational boundary
is usually not rectangular [Figs. 14(a), (b), 15]. This can be
explained from the following development.
The time leapfrog equations obtained from a lossy medium
are ( is either the center of an area or the center of a volume)

(10a)
5) Multiply each electric variable by the value of the
electric tapered damping function at that
point
(10b)

E. Generalization of Tapered Damping for Nonrectangular

(Irregular) Outer Boundary in an Unstructured Grid
The foregoing tapered damping function for a rectangular (11a)
outer boundary can be generalized for an irregular outer

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 YEE AND CHEN: FDTD AND FVTD METHODS IN SOLVING MAXWELL’S EQUATIONS 361

(a) (b)
Fig. 11. RCS for EMCC target #1 (flat PEC triangle-circle plate), calculated with unstructured triangular prisms with tapered damping RBC. (a) V–V
pol. (b) H–H pol.

(a)
(a)

(b)

(b) Fig. 13. Bistatic RCS for a thinly coated PEC circular cylinde.r (a) TM-pol.
(b) TE-pol.
Fig. 12. RCS of the parabolic disk calculated with an unstructured triangular
prism grid with tapered damping RBC. (a) H–H. (b) V–V.
that the process of updating the variables can be separated
into two steps:
1) update the variables by means of the lossless algorithm;
2) multiply the results obtained in step 1) by a “damping
(11b) factor.”
The above mathematical manipulations show that if we can
identify layers from the outer boundary of our computation
The difference between a lossless case and a lossy case is in the region, we can apply the tapered damping functions for
loss factor (damping factor) on the left-hand side. This means FDTD or FVTD regardless of whether the outer boundary is

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 362 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 45, NO. 3, MARCH 1997

(a) (b)
Fig. 14. The bases of the triangular prisms for (a) the triangle-circle and (b) the parabolic disc.

rectangular or not. This generalization has been implemented quadrilaterals. We were not able to match the experimental
in our codes when the computation grid is irregular. data [14] very well for the H–H pol. when theta ranges
from 0–50 . This poor match to experimental data is due to
V. SOME COMMENTS ON STABILITY interpolation near the edge of the target where the field is
CONDITION AND COMPUTATION COST singular. Fig. 11(a) and (b) shows the RCS for the EMCC
target #1, calculated with an unstructured grid consisting of
We have not conducted any extensive analyses on the
triangular prisms with tapered damping RBC. The boundary-
numerical stability requirements of our computations. Instead,
condition simulation, here, is based on the discussion of
we rely on our naive (intuitive) approach to the limitation of
Section III-B. The grid used is constructed by translating the
the time step for our FDTD/FVTD grids. Roughly speaking,
“base” shown in Fig. 14(a), along the vertical direction. The
the size of our surface triangles should be comparable to
distance between two consecutive parallel surfaces is .
the FDTD spatial discretization length , which usually
Comparing the results shown in Figs. 10 and 11, we see that
is taken to be 1/15 of the wavelength. The time step is,
results with the boundary-condition simulation discussed in
therefore, taken to be . Rarely did we have
Section III-B is an improvement over that of the overlapping
to select to be smaller than to produce stable
grids. We used a coarser grid ( is 1/20 of a wavelength)
numerical calculations. At present the authors do not know
to produce the results in Fig. 11(a) and (b) than that used
how to establish a systematic analysis of the stability problem
to produce the results in Fig. 10(a) and (b) ( is 1/24 of
in an irregular grid. With overlapping grids we observe late-
a wavelength). Fig. 12 shows the RCS of the parabolic disc
time instability, but we eliminate it with an averaging process
calculated with an unstructured grid consisting of triangular
explained in [5]. We did not observe any late time instability
prisms. Again, the boundary condition is simulated with the
when we used only one irregular grid in our calculations.
technique discussed in Section III-B. The “base” of this grid is
The increase in computation cost is considerable. In the
the triangulated surface [Fig. 14(b)] containing the parabolic
examples we performed, the amount of time in computer for
disk. The base is translated up and down along the vertical
the overlapping grids is about five times that for the standard
direction to the number of zones desired (we used 15). The
FDTD. The cost for a single irregular grid is a little less (but
tapered damping RBC is used in these calculations. This
more) than three times.
unstructured grid is easy to generate and double interpolations
of the overlapping grids is eliminated. We have not, to
VI. SOME SAMPLE CALCULATIONS date, performed many three-dimensional calculations with
The techniques discussed in the previous sections have been thinly coated objects. But we have calculated the bistatic
applied to a variety of applications. RCS for a thinly coated PEC circular cylinder, based on the
Fig. 9(a) is the RCS of a 1-m radius PEC sphere, and technique discussed in Section III-C; the results are shown
Fig. 9(b) is the RCS of a PEC ellipsoid calculated with a in Fig. 13. Here, we use FDTD near the cylinder (region
rectangular grid, as discussed in Section III-A. The tapered one) and FVTD away (region two). The grid used is shown
damping RBC is used. For comparison, we also show the series in Fig. 15. Tapered damping RBC is used. The thickness
solution and the numerical results obtained with the overlap- of the coating is one half of a zone, and we subgrid in
ping FDTD/FVTD [with the Mur radiation boundary condition the normal direction into five zones. The rationale for this
(RBD)], as well as the result obtained with staircasing FDTD. is that the field changes much more drastically along the
It is seen, here, that this technique is an improvement over the radial direction of the cylinder. Across the dielectric boundary
staircasing FDTD, but inferior in accuracy to the conformal we use both the flat-top assumption as well as the roof-
overlapping grids FDTD/FVTD. Fig. 10(a) and (b) shows the top assumption. For high-dielectric constants, the roof-top
RCS for the electromagnetic code consortium (EMCC) target assumption is superior and the results compare well with the
#1 (flat PEC triangle-circle plate), calculated with overlapping exact series solution. In all these RCS calculations, the incident
grids and the Mur RBC. These are earlier results where we wave is a plane Gaussian pulse, and the frequency-domain
used the boundary-condition simulation, as discussed in [5], data have been obtained through a Fourier transform of the
[6]. Near the perimeter of the target we used one layer of time-domain data.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.
 YEE AND CHEN: FDTD AND FVTD METHODS IN SOLVING MAXWELL’S EQUATIONS 363

[6] , “Hybrid finite difference time domain and finite volume time
domain in solving Maxwell’s equations,” in ACES 1995 Conf. Proc.,
Monterey, CA, Mar. 1995, pp. 453–464.
[7] , “Impedance boundary condition (IBC) simulation in the
FDTD/FVTD hybrid,” IEEE Trans. Antennas Propagat., to be published.
[8] , “Conformal FVTD with a rectangular grid for PEC scattering
objects,” in 1996 ACES Conf. Proc., Monterey, CA, Mar. 1996, pp.
1049–1056.
[9] J. S. Chen and K. S. Yee, “Scattering of electromagnetic waves by open
surfaces modeled by the finite volume time domain (FVTD) or by the
FDTD/FVTD hybrid,” IEEE Trans. Antennas Propagat., to be published.
[10] K. N. Madsen “Divergence preserving discrete surface integral methods
for Maxwell’s curl equations using nonorthogonal unstructured grids,”
J. Computat. Phys., vol. 119, pp. 34–45, June 1995.
[11] J. S. Chen and K. S. Yee, “Tapered damping near the outer boundary
in the FDTD/FVTD hybrid,” IEEE Trans. Antennas Propagat., to be
published.
[12] G. Mur, “Absorbing boundary conditions for finite-difference approx-
imations of the time-domain electromagnetic field equations,” IEEE
Trans. Electromagn. Compat., vol. EMC-23, pp. 1073–1077, Nov. 1981.
[13] Z. P. Liao, H. L. Wong, B. P. Yang, and Y. F. Yuan, “A transmitting
boundary for transient wave analysis,” Scientia Sinica (Ser. A), vol. 27,
no. 10, pp. 1063–1076, 1984.
[14] T. G. H. Wang and A. Woo, “Radar cross section measurement data,
Fig. 15. The FDTD-grid for the circular cylinder with thin coating.
electromagnetic code consortium bench mark targets,” NWCTM 6985,
Naval Weapons Ctr., China Lake, CA.
[15] J. P. Berenger, “A perfectly matched layer for the absorption of
VII. CONCLUSION electromagnetic waves,” J. Computat. Phys., vol. 114, pp. 185–200,
June 1994.
In this paper, we have presented the FDTD and FVTD
algorithms. We have illustrated the various techniques with
some nontrivial calculation examples. The integral forms of the Kane S. Yee (M’89–SM’95–F’97) was born in
Maxwell’s equations have been demonstrated to be straight- China in 1934. He received the B.S. and M.S.
forward to understand and implement. Judicious use of these degrees in electrical engineering in 1957 and 1958,
respectively, and the Ph.D. degree in applied mathe-
two integral forms has made numerical simulation of initial- matics in 1963, all from the University of California
boundary value problems involving Maxwell’s equations sim- at Berkeley.
ple and intuitive. The improvement in accuracy and flexibility From 1966 to 1984, he was a Professor of electri-
cal engineering and mathematics at the University
with the integral forms is a far cry from the popular staircasing of Florida, Gainesville, and later at Kansas State
FDTD algorithm published in 1966. University, Manhattan, KS. Since 1966 he was a
The overlapping grids scheme was introduced because of Consultant for the Lawrence Livermore National
Laboratory, CA, where he worked on microwave vulnerability problems from
the ease to generate an irregular grid. If a single fairly 1984 to 1987. From 1987 to 1995 he worked for Lockheed Palo Alto Research
uniform irregular grid can be readily generated throughout Laboratory, CA, doing research on numerical electromagnetics. He retired
the computation volume, it is advisable to use it than the from Lockheed in January 1996. He is the author of the 1966 FDTD electro-
magnetic numerical algorithm. His research areas include electromagnetics,
overlapping grids. This is because of the late time instability in continuum mechanics and partial differential equations.
the overlapping grids and the extra computational burden with
the double interpolations in each time step when overlapping
grids are used.
Jei S. Chen (S’88–M’90) was born in Tainan,
Taiwan on October 10, 1952. He received the B.S.
REFERENCES degree in communication engineering and the M.S.
degree in electronic engineering, both from the
[1] K. S. Kunz and R. J. Luebbers, The Finite Difference Time Domain National Chiao Tung University, Hsinchu, Taiwan,
Method for Electromagnetics. Boca Raton, FL: CRC, 1993. in 1975 and 1979, respectively, and the Ph.D. degree
[2] A. Taflove, Computational Electromagnetics: The Finite-Difference in electrical engineering from the University of
Time-Domain Method. Norwood, MA: Artech House, 1995. Washington, Seattle, in 1990.
[3] K. L. Shlager and J. B. Schneider, “A selective survey of the finite- From 1979 to 1981, he worked as a Senior En-
difference time-domain literature,” IEEE Antennas Propagat. Mag., vol. gineer in the National Taiwan Telecommunication
37, pp. 39–56, Aug. 1995. Laboratory. From 1981 to 1983 he was an Instructor
[4] K. S. Yee, “Ten years of evolution of the FDTD-like conformal in the National Taiwan Institute of Technology. During 1984 to 1990 he was a
techniques,” in ACES 1995 Conf. Proc., Monterey, CA, Mar. 1995, pp. Research and Teaching Assistant in the Department of Electrical Engineering,
1241–1257. University of Washington. Since 1990 he joined the Lockheed Martin Missile
[5] K. S. Yee and J. S. Chen, “Conformal hybrid finite difference time do- and Space, Sunnyvale, CA. He is currently working on the Simulation-Based
main and finite volume time domain,“ IEEE. Trans. Antennas Propagat., Satellite Design Program. His research interests include numerical simulation,
vol. 42, pp. 1450–1455, Oct. 1994. computational electromagnetic, massively parallelization, and satellite design.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on January 30,2025 at 06:12:19 UTC from IEEE Xplore. Restrictions apply.