IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO.

3, MARCH 2014 1321

A Three-Dimensional Unconditionally Stable

Five-Step LOD-FDTD Method
Alok Kumar Saxena, Student Member, IEEE, and Kumar Vaibhav Srivastava, Senior Member, IEEE

Abstract—A three-dimensional unconditionally stable 2 in the LOD2-FDTD represents the number of sub-steps in
five-step locally one-dimensional finite-difference time-do- the two-step LOD-FDTD method and not its order of accuracy
main (LOD-FDTD) method is presented. Unlike the two-step as it has been used in [8] and [9]. In 2008, Ahmed et al. pro-
LOD-FDTD and three-step LOD-FDTD methods, the proposed
method has second order temporal accuracy. Hence, it gives less
posed a three-step LOD-FDTD (LOD3-FDTD) method [10] for
numerical dispersion than the two-step LOD-FDTD and three-step three-dimensional structures, which solves the Maxwell’s equa-
LOD-FDTD methods. It also gives less numerical dispersion than tions in three sub-steps. Like the LOD2-FDTD method, this
the alternating direction implicit finite-difference time-domain method is also only first-order actuate in the time domain. While
(ADI-FDTD) method. Moreover, for every propagation angle , the ADI-FDTD and LOD2-FDTD methods have one dispersion
it provides very small anisotropy error than the above-mentioned relation which is same for both the methods [11], the LOD3-
FDTD methods. Effects of the time step and the mesh size on
the performance of the proposed method are discussed in detail.
FDTD method has two dispersion relations [12]. Hence, the
In this paper, validation of the stability and the accuracy of the LOD3-FDTD method give different dispersion characteristics
proposed method is done with the help of simulation results. To as compared with the ADI-FDTD and LOD2-FDTD methods.
further show the advantage of the proposed method, performance However, all the three methods give nearly same accuracy in
of the proposed method with artificial coefficients (control param- their simulation results [1], [10]. Furthermore, to improve the
eters) is also discussed in this paper. accuracy of these FDTD methods, various techniques such as
Index Terms—Alternating direction implicit finite-difference control parameters with second-order space derivative approx-
time-domain (ADI-FDTD), anisotropy error, finite-difference imation [13]–[17], higher order approximation of space deriva-
time-domain (FDTD), locally one-dimensional finite-difference tives [18]–[20] and (2, 4) stencil [21], [22] are used.
time-domain (LOD-FDTD).
In this paper, a three-dimensional unconditionally stable five-
step LOD-FDTD (LOD5-FDTD) method is presented. Like the
I. INTRODUCTION LOD3-FDTD method, the proposed method has variation in ei-
ther -, - or -direction in each sub-step. The proposed method

T HE locally one-dimensional finite-difference time-do-

main (LOD-FDTD) method [1] is an unconditionally
stable method whose time step size is not restricted by the
is second-order accurate in the time domain as well as in the
space domain. It gives less numerical dispersion error and very
less anisotropy error [23] than the ADI-FDTD, LOD2-FDTD,
Courant–Friedrich–Lewy (CFL) stability condition [2]. Due and LOD3-FDTD methods. In this paper, unconditional stability
to less arithmetic operations, it takes less computational of the proposed method is verified with the help of simulation re-
time than the alternating direction implicit finite-difference sults. Moreover, simulation results show that the accuracy of the
time-domain (ADI-FDTD) method [3]–[5]. However, while proposed method is much better than those of the above-men-
the ADI-FDTD method is second-order accurate in the time tioned FDTD methods. Performance of the proposed method
domain, the LOD-FDTD method is only first-order accurate. is further investigated with control parameters [13], [14]. Our
It is worth mentioning that the LOD-FDTD method [1] is study shows that use of the control parameters improved the
similar to the split-step 1 scheme of the split step finite-dif- performance of the LOD5-FDTD method. Also, the improved
ference time-domain (SS-FDTD) method [6]. In 2007, Tan LOD5-FDTD scheme gives better performance than the similar
proposed a three-dimensional LOD-FDTD scheme [7] which control parameters schemes of the ADI-FDTD, LOD2-FDTD,
is second-order accurate in the time domain. However, this and LOD3-FDTD methods [14]–[16].
scheme has two extra steps for input and output processing.
The conventional LOD-FDTD method solves the Maxwell’s II. FORMULATION
equations in two sub-steps; hence, it can also be called the two-
step LOD-FDTD (LOD2-FDTD) method. In this paper, digit In a linear, lossless, nondispersive and isotropic medium
with permittivity and permeability , the time-dependent 3-D
Maxwell’s curl equations can be written in a matrix form as
Manuscript received August 07, 2013; revised November 16, 2013; accepted
November 29, 2013. Date of publication December 05, 2013; date of current (1)
version February 27, 2014.
The authors are with the Department of Electrical Engineering, In-
dian Institute of Technology Kanpur, Kanpur 208016, India (e-mail: where represents the field
[email protected]; [email protected]).
components and , , and are the sparse matrices [10]
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. whose elements are related to the spatial derivatives along -,
Digital Object Identifier 10.1109/TAP.2013.2293790 -, and -directions, respectively.

0018-926X © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1322 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014

By using Crank–Nicolson scheme in (1) (5d)

(2)

where is the time index. Equation (2) can be approximated as

follows:

(5e)

(3)
(5f)
For the LOD5-FDTD method, (3) can be updated in five sub-
steps as follows:
Sub-step 1: where and . Here is the time step
size and is the second-order finite-difference approxima-
(4a) tion of the first-order spatial derivative in the -direction and
can be given as
Sub-step 2:
(4b) (6)

where or , , and .
Sub-step 3:
Substituting from (5f) into (5b) and using (6), we will get
the updating equation for . Similarly, we can get the updating
(4c)
equation for . Since values of and are known by now,
by using (6) in (5e) and (5f), we can get the updating equations
Sub-step 4: for and , respectively.
Similarly, for the other four sub-steps, we can find out the
(4d) updating equations of electric and magnetic fields.

Sub-step 5: III. NUMERICAL STABILITY ANALYSIS

(4e) In this section, numerical stability of the proposed method

is discussed. Here, amplification matrix method [23] is used to
It can be observed that sub-steps 1 and 5 have variation only analyze the numerical stability, Let
in the -direction, sub-steps 2 and 4 have variation in the -di-
rection and sub-step 3 has variation in the -direction. With the (7)
help of procedure given in [6, Appendix], it can be easily veri-
fied that the proposed LOD5-FDTD method is second-order ac- where , , , and
curate in the time domain. denote space indices and wave numbers along the -, -,
For sub-step 1, (4a) can be given as and -directions, respectively. Here is magnitude of the wave
number and are the propagation angles.
(5a) For the sub-step 1, substituting (7) into the updating equations
of , , , , and will result in a matrix, which is
given by

(8a)

(5b) Similarly for the other four sub-steps, matrices will be given by

(8b)
(8c)
(8d)
(5c) (8e)
SAXENA AND SRIVASTAVA: 3-D UNCONDITIONALLY STABLE FIVE-STEP LOD-FDTD METHOD 1323

where where denotes the initial values of the field components.

Substituting (10) into (9) gives

(11)

where is a 6 6 identity matrix. For a nontrivial solution of

(11)

(12)

On solving (12), we get numerical dispersion relation of the

proposed method as

(13)

V. NUMERICAL RESULTS
In this section, numerical dispersion characteristics of the
proposed method is discussed using the numerical dispersion
relation (13). Without loss of generality, uniform mesh size
is considered in this paper. In
Section V-A, used notations are discussed. In Section V-B,
numerical dispersion characteristics of the proposed method
are discussed and compared with the other FDTD methods. In
Section V-C, effects of the time step and the mesh size on the
proposed method are discussed.

A. Used Notations
substituting (8a)–(8d) into (8e) gives
To compare the numerical dispersion characteristics of var-
(9) ious FDTD methods, following notations are used:
1) CFL Number (S): S is defined as the ratio of the time step
where . With the help of MAPLE software, size to the CFL limit :
the eigenvalues of amplification matrix are found to be
(14)

2) Cells Per Wavelength (CPW): CPW can be defined as the

number of cells per wavelength, i.e.,

(15)
where
where . Here is the speed of light in vacuum and
is the reference frequency.
3) Numerical Phase Velocity Error (NPVE): The numerical
phase velocity error (NPVE) is given by:

(16)
It can be observed that , hence
the proposed method is unconditionally stable.
where is the numerical phase velocity. Here
and is the frequency.
IV. NUMERICAL DISPERSION RELATION 4) Averaged Phase Velocity Error (APVE): The averaged
To find out the numerical dispersion relation of the proposed phase velocity error (APVE) over preselected angles
method, a monochromatic wave with angular frequency is can be given as
considered. Its field components can be represented as
(17)
(10)
1324 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014

Fig. 2. NPVE (%) against wave propagation angle for the LOD3-FDTD
Fig. 1. NPVE (%) against wave propagation angle for the ADI-FDTD (first dispersion relation) and LOD5-FDTD methods.
and LOD5-FDTD methods.

where and . In this paper,

(for ) are taken.
Since the LOD3-FDTD method has two dispersion relations,
the overall APVE over preselected angles for the first
and second dispersion relation can be given as

(18)

where and are the APVE of the first and second dis-
persion relation, respectively.
5) Anisotropy Error (AE): Anisotropy error (AE) [23] at any
propagation angle can be defined as

(19)
Fig. 3. NPVE (%) against wave propagation angle for the LOD3-FDTD
where is the maximum value of the normalized nu- (second dispersion relation) and LOD5-FDTD methods.
merical phase velocity at any propagation angle obtained by
the dispersion relation of any FDTD method while is Figs. 2 and 3 compare the NPVE (%) against wave propaga-
the minimum value. tion angle for the LOD5-FDTD method with those for
the first and second dispersion relations of the LOD3-FDTD
B. Numerical Accuracy method, respectively. One can see that at , both disper-
In this subsection, the proposed method is compared sion relations of the LOD3-FDTD method and the LOD5-FDTD
with the ADI-FDTD and LOD3-FDTD methods. Since the method give same numerical dispersion error. Fig. 2 shows that
ADI-FDTD and LOD2-FDTD methods have the same dis- the LOD5-FDTD method gives less numerical dispersion near
persion relation, both the methods give the same dispersion and around diagonal direction ( , )
characteristics. Hence, comparison of the proposed method than the first dispersion relation of the LOD3-FDTD method.
with the LOD2-FDTD method is not given here. To compare However, at other propagation angles it gives more numerical
the numerical dispersion characteristics of the various FDTD dispersion than the first dispersion relation of the LOD3-FDTD
methods, and are taken. method. Fig. 3 shows that at all propagation angles except
Fig. 1 shows the NPVE (%) against wave propagation angle , the LOD5-FDTD method gives less numerical dispersion
for the ADI-FDTD and LOD5-FDTD methods. It can be error than the second dispersion relation of the LOD3-FDTD
seen that at all propagation angles except , the LOD5- method. On comparing overall performance of the two methods,
FDTD gives less NPVE than the ADI-FDTD method. However, APVE value for the LOD3-FDTD method is 6.455%, whereas,
at , both methods give same NPVE value. One can also it is only 4.032% for the LOD5-FDTD method; which shows
see that the proposed method gives maximum error at that the LOD5-FDTD method gives less overall numerical dis-
and minimum error at . Also, at every , it gives very persion than the LOD3-FDTD method.
less anisotropy error. On comparing average dispersion error for Fig. 4 plots the AE (%) against wave propagation angle
all propagation angles, while for the ADI-FDTD method, APVE for the dispersion relations of various FDTD methods. From
value is 5.539%, it is only 4.032% for the LOD5-FDTD method. Fig. 4, it can be seen that at most of the propagation angles ,
SAXENA AND SRIVASTAVA: 3-D UNCONDITIONALLY STABLE FIVE-STEP LOD-FDTD METHOD 1325

TABLE I
APVE (%) FOR THE ADI-FDTD, LOD3-FDTD, AND LOD5-FDTD METHODS WITH DIFFERENT VALUES OF S AND CPW

TABLE II
AE (%) FOR THE ADI-FDTD AND LOD5-FDTD METHODS WITH
DIFFERENT VALUES OF S AND CPW

Fig. 4. AE (%) against wave propagation angle for the ADI-FDTD, LOD3-
FDTD (first and second dispersion relations) and LOD5-FDTD methods.

dispersion relations (DR) of the ADI-FDTD and LOD3-FDTD

methods give larger values of AE than that of the LOD5-FDTD
method. Also, for dispersion relations of the ADI-FDTD and
LOD3-FDTD methods, values of AE vary against ; whereas
for the LOD5-FDTD, values of AE are nearly constant at all
. Since at , all angles represent the same point,
AE values are zero for all the four dispersion relations. As de-
scribed in the Appendix, at , dispersion relation of
the ADI-FDTD method and both dispersion relations of the
LOD3-FDTD method turn into same equation, hence all the
three give same value of AE. Fig. 5. NPVE (%) against wave propagation angle for the LOD5-FDTD
method with different values of S, ( and ).
C. Performance With Different Time Steps and Mesh Sizes
In this subsection, to study the performance of the proposed LOD3-FDTD method are not compared here. One can see that
method with different time steps and mash sizes, values of for all the time steps and mesh sizes, values of AE are very small
APVE and AE for the proposed method are compared with for the proposed method than the ADI-FDTD method. It can
those of the ADI-FDTD and LOD3-FDTD methods. also be seen that for the ADI-FDTD method, as S increases or
APVE (%) for the various FDTD methods with different CPW decreases, AE increases. However, for the LOD5-FDTD
values of S and CPW are tabulated in Table I. It can be seen method, AE increases as CPW decreases but as S increases, first
that as S increases or CPW decreases, APVE increases for AE decreases and then it increases. This behavior of the pro-
all the three methods. However, for all the time steps and posed method with different time steps can be explained with
mesh sizes, the proposed method gives better performance the help of Fig. 5.
than the ADI-FDTD and LOD3-FDTD methods. Also, im- Fig. 5 shows the NPVE (%) against wave propagation angle
provement in the performance of the LOD5-FDTD method is for the LOD5-FDTD method at . It shows that
more for the larger time step or the coarser mesh size for the lower value of time step ( ), error is minimum at
than the smaller time step or the finer and maximum at , 90 . As S increases, the dif-
mesh size . ference between maximum and minimum dispersion error re-
Table II compares the AE (%) for the ADI-FDTD and LOD5- duces which reduces the value of AE. At a certain time step
FDTD methods at with different values of S and , AE is nearly zero. If S is increased further, AE in-
CPW. Since, the ADI-FDTD and LOD3-FDTD methods have creases again but for the higher value of time step , error
the same dispersion relation at , AE values for the is maximum at and minimum at , 90 .
1326 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014

TABLE IV
RELATIVE ERROR (%) OBTAINED BY THE ADI-FDTD AND LOD5-FDTD
METHODS WITH DIFFERENT VALUES OF S AND

Fig. 6. against time at the center of the cavity obtained by the LOD5-FDTD
method, ( and mm).

TABLE III
TIME TAKEN PER ITERATION BY VARIOUS FDTD METHODS

Fig. 7. Relative error in the calculation of resonant frequency (

mode) of the rectangular cavity obtained by various FDTD methods against S,
mm .

VI. NUMERICAL VALIDATION

obtained by the ADI-FDTD and LOD5-FDTD methods with
In this section, numerical validation of the stability and the different time steps (S) and mesh sizes . Here is defined
accuracy of the proposed method are presented. For the numer- as
ical stability validation, a homogeneous rectangular cavity of
dimension mm mm mm is simulated by the LOD5- (20)
FDTD method with a point source excitation [24]. For accuracy
validation, same cavity is simulated by the LOD5-FDTD and where is the analytical resonant frequency of the cavity
other FDTD methods with different time steps and mesh sizes. and is the resonant frequency obtained by different FDTD
methods. For the used cavity dimension, for the
A. Numerical Stability
mode is 19.437 GHz [24].
To validate the unconditional stability of the LOD5-FDTD Table IV illustrates that as the time step or the mesh size in-
method, the rectangular cavity is simulated with and creases, increases for both the methods. However, for all the
mm. time steps and mesh sizes, accuracy of the proposed method is
Fig. 6 plots the against time at the center of the cavity significantly better than that of the ADI-FDTD method. It can
obtained by the LOD5-FDTD method. Fig. 6 demonstrates that also be seen that the improvement in the performance of the pro-
for a bounded input, is bounded for large simulation time posed method is more for the larger time step or the
(with ) which validates the unconditional stability of the coarser mesh size mm than the smaller time step
LOD5-FDTD method. or the finer mesh size mm . Moreover,
with mm, though the ADI-FDTD method does not
B. Numerical Accuracy give any solution for , the proposed method gives a solu-
In this subsection, first, time taken by the proposed method is tion with even better accuracy than solution by the ADI-FDTD
compared with various FDTD methods and then its accuracy is method for .
compared. Fig. 7 presents the relative error in the calculation of
Time taken per iteration by various FDTD methods is com- resonant frequency ( mode) of the rectangular cavity
pared in Table III. From Table III, it can be observed that the obtained by various FDTD methods against S. One can see
proposed method takes nearly 33% more time than the ADI- that for every S, the relative errors given by the ADI-FDTD,
FDTD method and 66% more time than the LOD2-FDTD and LOD2-FDTD, and LOD3-FDTD methods are nearly same.
LOD3-FDTD methods. Also, the LOD5-FDTD method gives very less relative error
Table IV compares the relative error in the calculation than these FDTD methods. One can also see that as the time step
of resonant frequency ( mode) of the rectangular cavity increases, performance of the LOD5-FDTD method increases
SAXENA AND SRIVASTAVA: 3-D UNCONDITIONALLY STABLE FIVE-STEP LOD-FDTD METHOD 1327

in comparison with these FDTD methods. Furthermore, the

LOD3-FDTD method does not give any solution above
and the ADI-FDTD and LOD2-FDTD methods do not give
any solution above . However, proposed method gives
solution up to for which its accuracy is nearly same
as the accuracy given by the ADI-FDTD and LOD2-FDTD
methods for .
Table IV and Fig. 7 demonstrate that the accuracy of the pro-
posed method is significantly better than other mentioned FDTD
methods. Also, better accuracy is obtained for the larger time
steps or the coarser mesh sizes than the lower time steps or the
finer mesh sizes. Moreover, the simulation time can be reduced
by taking large values of the time steps for the proposed method.
These results validate the numerical results given in Section V.
Further, we have also compared the computational time taken
by the proposed method with the conventional FDTD method
Fig. 8. NPVE (%) against wave propagation angle for the LOD5-FDTD
[25]. Our study shows that for the same mesh size, the proposed and ILOD5-FDTD methods, ( and ).
method takes approximately 5.5 times more time per iteration
than the conventional FDTD method. One can see from Fig. 7 TABLE V
that for mm, the proposed method gives solution APVE (%) FOR THE LOD5-FDTD AND ILOD5-FDTD METHODS
WITH DIFFERENT VALUES OF S AND CPW
up to . Hence, with mm and , the
proposed method can be nearly times faster than
the conventional FDTD method. Furthermore, the ADI-FDTD
method takes approximately four times more time per iteration
than the conventional FDTD method. For mm, it
gives solution up to . Hence, with mm and
, the ADI-FDTD method can be nearly times
faster than the conventional FDTD method. This study shows
that the proposed method can be more fast than the ADI-FDTD
method with nearly same accuracy of the ADI-FDTD method.
B. Numerical Stability Analysis and Dispersion Relation
VII. IMPROVED FIVE-STEP LOD-FDTD METHOD For the numerical stability analysis and the dispersion re-
Use of artificial coefficients (control parameters) [13]–[17] on lation of the ILOD5-FDTD method, same procedures can be
any FDTD method is one of the main technique to improve the used which have been used for the LOD5-FDTD method in
accuracy of that method. In this section, controlled parameters Sections III and IV, respectively.
are used on the LOD5-FDTD method to improve its accuracy. It can be observed that if we change of the LOD5-FDTD
Formulation, numerical stability analysis and dispersion rela- method by , the Eigen values of the LOD5-FDTD method will
tion for the Improved LOD5-FDTD (ILOD5-FDTD) method become Eigen values of the ILOD5-FDTD method. Here, is
are discussed in brief. Numerical results of the ILOD5-FDTD given by
method are discussed with the help of analytical and simulation
results. Also, performance of the ILOD5-FDTD method is com- (22)
pared with the improved techniques of other FDTD methods.

A. Formulation where , and .

It can be seen that for real values of , and , magnitudes
For the ILOD5-FDTD method, the sparse matrices , ,
of , , , and are unity for the ILOD5-FDTD method.
and of (1)–(4) are replaced by , and , respec-
Hence, like the LOD5-FDTD method, the ILOD5-FDTD
tively. These modified sparse matrices can be given as
method is also unconditionally stable.
(21a) After replacing by , numerical dispersion relation of the
ILOD5-FDTD method can be given by (13).
(21b)
(21c) C. Numerical Results
where , , and are the controlled parameters. It can be To determine the values of the control parameters, axes-opti-
observed that for , the ILOD5-FDTD method mized scheme is used here. In this scheme, the numerical phase
will become the LOD5-FDTD method. velocity errors (NPVEs) are made to be zero along the -, - and
Updating equations for the electric and magnetic fields for the -axis. It is worth mentioning that same procedure is used in the
ILOD5-FDTD method can be formulated similarly as they have improved ADI-FDTD (IADI-FDTD), improved LOD2-FDTD
been formulated for the LOD5-FDTD method in the Section II. (ILOD2-FDTD) and improved LOD3-FDTD (ILOD3-FDTD)
1328 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014

TABLE VI
RELATIVE ERROR (%) OBTAINED BY THE LOD5-FDTD, IADI-FDTD AND ILOD5-FDTD METHODS WITH DIFFERENT VALUES OF S AND

methods [14]–[16]. On using axes-optimized scheme on the nu-

merical dispersion relation of the ILOD5-FDTD method, opti-
mized control parameters are found to be

for , , and

Fig. 8 plots the NPVE (%) against wave propagation angle

for the LOD5-FDTD and ILOD5-FDTD methods. It can
be seen that at every propagation angle, numerical dispersion
Fig. 9. Relative error in the calculation of resonant frequency (
error is very less for the ILOD5-FDTD method than that for mode) of the rectangular cavity obtained by various improved FDTD methods
the LOD-FDTD method. Also, along the -, - and -axis, the against S, mm .
ILOD5-FDTD method gives nearly zero NPVE.
Table V compares the APVE values for the LOD5-FDTD and other discussed FDTD methods. Also, as S increases, perfor-
ILOD5-FDTD methods with different values of S and CPW. mance of the ILOD5-FDTD method increases.
It illustrates that for all the values of S and CPW, the average
numerical dispersion is very less for the ILOD5-FDTD method
than that for the LOD5-FDTD method. Also, for the larger VIII. DISCUSSION AND CONCLUSION
values of S or the smaller values of CPW, the ILOD5-FDTD The three-dimensional five-step LOD-FDTD method has
gives more improvement than the smaller values of S or the been presented in this paper. The proposed method has
larger values of CPW. second-order temporal and spatial accuracy. Both the analyt-
Table VI compares the obtained by the ILOD5-FDTD ical and the simulation results have shown that the proposed
method for different S and with those of the LOD5-FDTD method is unconditionally stable. It gives better accuracy com-
and IADI-FDTD methods. From Table VI, it can be seen that for pared to the ADI-FDTD, two-step LOD-FDTD, and three-step
every time step and mesh size, the ILOD5-FDTD method gives LOD-FDTD methods which has been demonstrated by the
very less relative error than the LOD5-FDTD methods. Also, theoretical results and verified by the simulation results. More-
it gives significantly less error than the IADI-FDTD method. over, for every propagation angle , the proposed method has
It can also be seen that improvement is more for the coarser very less anisotropy error compared to the above mentioned
mesh size mm or the larger time step than FDTD methods. Also, for the coarser mesh sizes or the larger
time steps, the proposed method shows better performance than
the finer mesh size mm or the smaller time step
the finer mesh sizes or the smaller time steps. The improved
. Like the ADI-FDTD method, the IADI-FDTD method
five-step LOD-FDTD method which uses artificial coefficients
also does not give any solution for ( , mm).
(control parameters) to improve the accuracy of the five-step
We have also simulated the rectangular cavity by the ILOD2-
LOD-FDTD method has been also presented in this paper.
FDTD and ILOD3-FDTD methods. Fig. 9 shows the relative Improved five-step LOD-FDTD gives better performance than
error obtained by the various improved FDTD methods similar technique used on above mentioned FDTD methods
against S. It can be seen that both the ILOD2-FDTD and ILOD3- which further demonstrates the better performance of the
FDTD methods give nearly same relative errors as given by five-step LOD-FDTD method over the ADI-FDTD, two-step
the IADI-FDTD method. Like the LOD5-FDTD method, the LOD-FDTD, and three-step LOD-FDTD methods.
ILOD5-FDTD method also gives solution up to . How- The proposed method also gives better performance for the
ever, the IADI-FDTD and ILOD2-FDTD methods give solu- higher order approximation of the spatial differential operator
tion up to and the ILOD3-FDTD method gives solution and different boundary conditions (Mur’s and CPML). Due to
only up to . One can see that the proposed ILOD5-FDTD space limitation these advantages of the five-step LOD-FDTD
method gives better accuracy than the improved schemes of method will be discussed in our future publications.
SAXENA AND SRIVASTAVA: 3-D UNCONDITIONALLY STABLE FIVE-STEP LOD-FDTD METHOD 1329

APPENDIX [12] I. Ahmed, E. K. Chua, and E. P. Li, “Numerical dispersion analysis

DISPERSION RELATIONS FOR THREE-DIMENSIONAL of the unconditionally stable three-dimensional LOD-FDTD method,”
IEEE Trans. Antennas Propag., vol. 58, no. 12, pp. 3983–3989, Dec.
ADI-FDTD, LOD2-FDTD, AND LOD3-FDTD METHODS 2010.
[13] A. P. Zhao, “Improvement on the numerical dispersion of 2-D
Both the LOD2-FDTD and ADI-FDTD methods give same ADI-FDTD with artificial anisotropy,” IEEE Microw. Wireless
dispersion relation which can be written from [11], [26] as Compon. Lett., vol. 14, no. 6, pp. 292–294, Jun. 2004.
[14] I. Ahmed and Z. Chen, “Dispersion-error optimized ADI FDTD,” in
Proc. IEEE MTT-S Int. Microw. Symp. Dig., Jun. 2006, pp. 173–176.
(23) [15] E. Li, I. Ahmed, and R. Vahldieck, “Numerical dispersion anal-
ysis with an improved LOD-FDTD method,” IEEE Microw. Wirel.
where Compon. Lett., vol. 17, no. 5, pp. 319–321, May 2007.
[16] F. Liang, G. Wang, H. Lin, and B. Z. Wang, “Numerical dispersion
improved three-dimensional locally one-dimensional finite-difference
time-domain method,” IET Microw. Antennas Propag., vol. 5, no. 10,
pp. 1256–1263, Jul. 2011.
[17] J. Shibayama, R. Ando, J. Yamauchi, and H. Nakano, “A 3-D LOD-
Two dispersion relations of the LOD3-FDTD method can be FDTD method for the wideband analysis of optical devices,” J. Lightw.
written from [12], [20] as Technol., vol. 29, no. 11, pp. 1652–1658, Jun. 2011.
First dispersion relation: [18] W. Fu and E. L. Tan, “Stability and dispersion analysis for higher order
3-D ADI-FDTD method,” IEEE Trans. Antennas Propag., vol. 53, no.
11, pp. 3691–3696, Nov. 2005.
(24a) [19] Q. F. Liu, Z. Z. Chen, and W. Y. Yin, “An arbitrary-order LOD-FDTD
method and its stability and numerical dispersion,” IEEE Trans. An-
tennas Propag., vol. 57, no. 8, pp. 2409–2417, Aug. 2009.
Second dispersion relation: [20] A. K. Saxena and K. V. Srivastava, “Stability and dispersion analysis
of higher order unconditionally stable three-step locally one-dimen-
(24b) sional finite-difference time-domain method,” IET Microw. Antennas
Propag., vol. 7, no. 11, pp. 954–960, Sep. 2013.
[21] W. Fu and E. L. Tan, “A parameter optimized ADI-FDTD method
where based on the (2, 4) stencil,” IEEE Trans. Antennas Propag., vol. 54,
no. 6, pp. 1836–1842, Jun. 2006.
[22] Q. F. Liu, W. Y. Yin, Z. Z. Chen, and P. G. Liu, “An efficient method
to reduce the numerical dispersion in the LOD-FDTD method based on
the (2, 4) stencil,” IEEE Trans. Antennas Propag., vol. 58, no. 7, pp.
It can be observed that the ADI-FDTD, LOD2-FDTD, and 2384–2393, Jul. 2010.
LOD3-FDTD methods give same dispersion relation at [23] A. Taflove and S. C. Hagness, Computational Electrodynamics: The
and 90 (since ). Finite-Difference Time-Domain Method, 3rd ed. Norwood, MA,
USA: Artech House, 2005.
[24] Y. D. Kong and Q. X. Chu, “High-order split-step unconditionally-
stable FDTD methods and numerical analysis,” IEEE Trans. Antennas
REFERENCES Propag., vol. 59, no. 9, pp. 3280–3289, Sep. 2011.
[25] K. S. Yee, “Numerical solution of initial boundary value problems
[1] J. Shibayama, M. Muraki, J. Yamauchi, and H. Nakano, “Efficient involving Maxwell’s equations in isotropic media,” IEEE Trans. An-
implicit FDTD algorithm based on locally one-dimensional scheme,” tennas Propag., vol. AP-14, no. 3, pp. 302–307, May 1966.
Electron. Lett., vol. 41, no. 19, pp. 1046–1047, Sep. 2005. [26] F. Zheng and Z. Chen, “Numerical dispersion analysis of the uncondi-
[2] R. Courant, K. Friedrichs, and H. Lewy, “On the partial difference tionally stable 3-D ADI-FDTD method,” IEEE Trans. Microw. Theory
equations of mathematical physics,” IBM J., vol. 11, pp. 215–234, Mar. Tech., vol. 49, no. 5, pp. 1006–1009, May 2001.
1967.
[3] T. Namiki, “A new FDTD algorithm based on alternating-direction im-
plicit method,” IEEE Trans. Microw. Theory Tech., vol. 47, no. 10, pp. Alok Kumar Saxena (S’12) was born in Sitapur,
2003–2007, Oct. 1999. India, in June 1983. He received the B.Tech. degree
[4] T. Namiki, “3-D ADI-FDTD method-Unconditionally stable timedo- in electronics and communication engineering
main algorithm for solving full vector Maxwell’s equations,” IEEE from Madan Mohan Malviya Engineering College,
Trans. Microw. Theory Tech., vol. 48, no. 10, pp. 1743–1748, Oct. Gorakhpur, India, in 2009. He is currently working
2000. toward the Ph.D. degree in electrical engineering at
[5] F. Zheng, Z. Chen, and J. Zhang, “Toward the development of a the Indian Institute of Technology Kanpur, India.
three-dimensional unconditionally stable finite-difference time-do- His research interest is in computational electro-
main method,” IEEE Trans. Microw. Theory Tech., vol. 48, no. 9, pp. magnetics, particularly in FDTD.
1550–1558, Sep. 2000.
[6] J. Lee and B. Fornberg, “A split step approach for the 3-D Maxwell’s
equations,” J. Comput. Appl. Math., vol. 158, no. 2, pp. 485–505, Sep.
2003. Kumar Vaibhav Srivastava (M’09–SM’13) was
[7] E. L. Tan, “Unconditionally stable LOD-FDTD method for 3-D born in Allahabad, India, in May 1982. He received
Maxwell’s equations,” IEEE Microw. Wireless Compon. Lett., vol. 17, the B.Tech. degree in electronics engineering from
no. 2, pp. 85–87, Feb. 2007. Kamla Nehru Institute of Technology, Sultanpur,
[8] E. L. Tan, “Fundamental schemes for efficient unconditionally stable India, in 2002 and the M.Tech. and Ph.D. degrees
implicit finite-difference time-domain methods,” IEEE Trans. An-
in electrical engineering from the Indian Institute
tennas Propag., vol. 56, no. 1, pp. 170–177, Jan. 2008.
of Technology, Kanpur, India, in 2004 and 2007,
[9] T. H. Gan and E. L. Tan, “Unconditionally stable fundamental
LOD-FDTD method with second-order temporal accuracy and com- respectively.
plying divergence,” IEEE Trans. Antennas Propag., vol. 61, no. 5, pp. After the Ph.D. degree, he worked with the GE
2630–2638, May 2013. Global Research Centre Bangalore for one year
[10] I. Ahmed, E. K. Chua, E. P. Li, and Z. Chen, “Development of the three in 2008. At GE, he contributed significantly on
dimensional unconditionally stable LOD-FDTD method,” IEEE Trans. wireless power transfer, magneto-caloric refrigerator, perambulatory stroke
Antenna Propag., vol. 56, no. 11, pp. 3596–3600, Nov. 2008. detection, and subsea communication. He joined as an Assistant Professor the
[11] S. C. Yang, Z. Chen, Y. Yu, and W. Y. Yin, “An unconditionally stable Department of Electrical Engineering, Indian Institute of Technology, Kanpur,
one-step arbitrary-order leapfrog ADI-FDTD method and its numer- India, in 2009. His research interests include meta-materials, microwave filters
ical properties,” IEEE Trans. Antennas Propag., vol. 60, no. 4, pp. and antennas, dielectric resonators, and finite-difference time-domain (FDTD)
1995–2003, Apr. 2012. techniques.