156 Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26

An Upwind Leapfrog Scheme for Computational

Electromagnetics: CL-FDTD
Yong Zhang, K. R. Shao, and X. W. Hu
Huazhong University of Science and Technology, China
J. D. Lavers
University of Toronto, Canada

Abstract
This paper extends upwind-leapfrog scheme, initially developed from computational aeroacoustics and elas-
todynamics, into computational electromagnetics, and develops a novel Characteristic Line FDTD (CL-FDTD)
method. In illustrating 2-D problems, this paper gives the implementation of the CL-FDTD method, and
points out its merits over the FDTD method chosen for comparison. The results demonstrate that the CL-
FDTD method is less numerical dispersion, no requirement to deal with outer boundary conditions and it can
precisely simulate accident signals that are difficult to treat with the conventional FDTD method.

Introduction
The upwind leapfrog method, originally developed from computational aeroacoustics and elastodynamics
[1][2], has a more compact stencil by compared with a classical leapfrog method such as the Finite-Difference
Time-Domain (FDTD) method. This approach preserves the time-reversibility of the leapfrog algorithm, which
results in no dissipation, and it permits more flexibility by the ability to adopt a characteristic based method.
Moreover, it leads to a more natural treatment of outer boundaries and material boundaries.
This paper introduces upwind-leapfrog scheme into computational electromagnetics and develops a novel
characteristic-line FDTD (CL-FDTD) method with combining the PDE features of electromagnetics. It also
gives the implementation of the CL-FDTD method for 2-D problems, and points out its merits over the FDTD
method chosen for comparison.

Complementation of CL-FDTD
Maxwell TM equations for 2-D linear homogeneous medium are

Dz Hy Hx
( ) + Dz = 0 (1)
t x y
1 Hx Dz 1
+c + Hx = 0 (2)
c t y c
1 Hy Dz 1
c + Hy = 0 (3)
c t x c

where c = 1/ , and are electric and magnetic conductivities respectively. The directions that waves
propagate with physical speed c are the characteristic lines. Therefore, the characteristic variables can be
defined as
1 1 1 1
P = D z Hy , Q = D z + Hy , R = D z + Hx , S = D z Hx (4)
c c c c
to represent the x and y propagating solutions. Then equations (1) to (3) are rewritten as

P P a b c (R S)
+c + P + Q+ =0 (5)
t x 2 2 2 y

Q Q a b c (R S)
c + Q+ P + =0 (6)
t x 2 2 2 y
Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26 157

R R a b c (P Q)
+c + R+ S+ =0 (7)
t y 2 2 2 x
S S a b c (P Q)
c + S+ R+ =0 (8)
t y 2 2 2 x
where a = / + / and b = / /. The stencil of the CL-FDTD method for 2-D wave is shown as Fig.1,
in which (n, i, j) is the point (i, j) at time leveln. Using the stencil shown as Fig.1, the difference equations for
(5) to (8) are

n+1 n n n1 n n n n n n
(Pi,j Pi,j )+(Pi1,j Pi1,j ) (Pi,j Pi1,j ) a n+1 b n c [(Ri,j Ri,j1 )(Si,j+1 Si,j )]
+c + Pi,j + Qi,j + =0 (9)
2t x 2 2 2 y

(Qn+1 n n n1
i,j Qi,j )+(Qi+1,j Qi+1,j ) (Qni+1,j Qni,j ) a n+1 b n c [(Ri,j
n n
Ri,j1 n
)(Si,j+1 n
Si,j )]
c + Qi,j + Pi,j + =0 (10)
2t x 2 2 2 y

n+1 n n n1 n n n n
(Ri,j Ri,j )+(Ri,j1 Ri,j1 ) (Ri,j Ri,j1 ) a n+1 b n c [(Pi+1,j Pi,j )(Qni,j Qni1,j )]
+c + Ri,j + Si,j + =0 (11)
2t y 2 2 2 x

n+1 n n n1 n n n n
(Si,j Si,j )+(Si,j+1 Si,j+1 ) (Si,j+1 Si,j ) a n+1 b n c [(Pi+1,j Pi,j )(Qni,j Qni1,j )]
c + Si,j + Ri,j + =0 (12)
2t y 2 2 2 x

Figure 1: Amplitude contours of the E-filed intensity

The iterative modes of equation (9) to (12) are

n+1 1  n1 n n

Pi,j = Pi1,j +(12x)(Pi,j Pi1,j )b tQni,j y (Ri,j
n n
Ri,j1 n
Si,j+1 n
+Si,j ) (13)
1+a t
1  n1 
Qn+1
i,j = Qi+1,j (12x)(Qni+1,j Qni,j )b tPi,j
n n
y (Ri,j n
Ri,j1 n
Si,j+1 n
+Si,j ) (14)
1+a t

n+1 1  n1 n n n n n

Ri,j = Ri,j1 +(12y )(Ri,j Ri,j1 )b tSi,j x (Pi+1,j Pi,j Qni,j +Qni1,j ) (15)
1+a t

n+1 1  n1 n n n n n

Si,j = Si,j+1 (12y )(Si,j+1 Si,j )b tRi,j x (Pi+1,j Pi,j Qni,j +Qni1,j ) (16)
1+a t
 158 Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26

where x = c t/ x, y = c t/ y are the x and y Courant numbers. When , there is condition

Pin+1 = Qn+1
i = Rin+1 = Sin+1 = 0. Field variables Dz , Hx , Hy have the relationship with characteristic P , Q,
R, S that

Dz = (P + Q + R + S)/4, Hy = c(P Q)/2, Hx = c(R S)/2 (17)

The procedure for the TE polarization is similar to that of the TM case, except the characteristic variables
are defined as

1 1 1 1
P = D y + Hz , Q = D y Hz , R = D x Hz , S = D x + Hz (18)
c c c c

Numerical Dispersion between CL-FDTD and FDTD

The stability conditions for 1-D and 2-D CL-FDTD are 1 and x , y 1/2 respectively by using
Fourier analysis, and the numerical dispersion between the FDTD method and the CL-FDTD method for 1-D
and 2-D cases are compared as well. Fig.2 shows the result above, in which Fig.2 (a) for normalized phase speed
at different wave-lattice ratio with the Courant number = 0.5 in 1-D case, and Fig.2 (b) for normalized phase
speed at different angel with the Courant number x = y = 0.5 and wave-lattice ratio / = 15. Based upon
the result, the CL-FDTD method gets more economical characteristics than that of the FDTD for the same
level of accuracy.

Figure 2: Numerical dispersion comparison between CL-FDTD and FDTD in 1D and 2D cases (a) 1D numerical
dispersion characteristic (b) 2D numerical dispersion characteristic

Comparison between CL-FDTD and FDTD for Simulating Signals

The FDTD method belongs to classical frog mode, it has no special choice in the direction that waves
propagate, and thus there comes numerical oscillation when it is used to simulate accident signals. However,
the CL-FDTD method has great selectivity in the direction wave propagates, and its numerical condition of
the lower reaches never effect that of the upper reaches, therefore, it can simulate accident signals with high
degree accuracy. Fig.3 shows the comparison between these two methods for simulating accident signals, in
which Fig.3 (a) is a comparison for simulating a single sine signal and Fig.3 (b) for simulating a single square
wave. It indicates from Fig.3 that the CL-FDTD method can simulate accident signals with high accuracy, but
to the FDTD method, it will comes severe oscillation. To simulate a square signal is so important because it
contains information in full frequency-domain, and has great significance to electromagnetics analysis.

Out Boundary Characteristics of CL-FDTD

Absorbing boundary conditions (ABC) are routinely used in electromagnetic simulations to minimize the
computer resources required for modeling an open system. Two classes of algorithms are usually in use, one of
them is one-way ABC, which is applied only to the wave leaving the computational domain [3][4], and another
is perfectly matched layer ABC [5][6], which absorbs waves regardless of their directions of propagation. The
latest one, with great flexibility and effectiveness, has more advantages over the former one. But no matter
which one is used, additional procedures and computational resources are required.
Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26 159

Figure 3: Comparison between CL-FDTD and FDTD on simulating accident signals (a) Comparison between
CL-FDTD and FDTD on simulating single-period sinusoidal signal (b) Comparison between CL-FDTD and
FDTD on simulating square wave signal

Figure 4: Contour lines of point source radiation simulated by CL-FDTD (a) TM case (b) TE case

However, the CL-FDTD method is a up-wind frog-leap method, it has great selectivity in the direction and
the directions of its characteristic lines are those waves propagating, its numerical condition of the lower reaches
never effect that of the upper reaches, thus it has natural ABC.
Wave propagates from a point source in two dimensional domain (1m1m) is illustrated here as an example.
The simulative parameters are = 0.1m, / = 10, x = y = 0.5 and time-steps for simulation is equal to
500. Fig.5 is the contour lines of point source radiation simulated by CL-FDTD, in which Fig.5 (a) for TM case
and Fig.5 (b) for TE case. It indicates from Fig.5 that the contour lines are regular concentric circles.

Conclusion
This paper extends upwind-leapfrog scheme, initially developed from computational aeroacoustics and elas-
todynamics, into computational electromagnetics, and develops a novel Characteristic Line FDTD (CL-FDTD)
method for 2-D problems. The CL-FDTD method demonstrates many distinct merits over the conventional
FDTD method.
First, the CL-FDTD method has less numerical dispersion than the FDTD method. Second, it has natural
absorbing boundary condition. Finally, it can precisely simulate accident signals that are difficult to deal with
using the FDTD method.

Acknowledgement
This work was supported by National Natural Science Foundation of China (NSFC) under Grant No.
90405004.
 160 Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26

REFERENCES
1. Nakazawa, S., The Application of a Two-Dimensional Upwind Leapfrog Scheme to Linear Elastodynamics,
Ph.D. Thesis, University of Michigan, Ann Arbor, MI, 2001.
2. Kim, C., Multimensional Upwind Leapfrog Scheme and Their Applications, Ph.D. Thesis, University of
Michigan, Ann Arbor, MI, 1997.
3. Mur, G., Absorbing Boundary Condition for the Finite-difference Approximation of the Time-domain
Electromagnetic Field Equations, IEEE Trans. Electromagn. Compat., Vol. 23, 377-382, 1981.
4. Engquist, B. and A. Majda, Absorbing Boundary Conditions for the Numerical Simulation of Waves,
Math. Comput., Vol. 31, 629-651, 1977.
5. Berenger, P., A Perfectly Matched Layer for the Absorption of Electromagneticwaves, J. Comput. Phys.,
Vol. 114, 185200, 1994.
6. Sacks, Z. S., D. M. Kingsland, R. Lee and J. F. Lee, A Perfectly Matched Anisotropic Absorber for Use as
An Absorbing Boundary Conditions, IEEE Trans. Anten. and Prop., Vol. 43, 1460-1463, 1996.