NOLTA, IEICE

Invited Paper

Introduction to the FDTD method and its

application to implicit locally
one-dimensional calculations

Jun Shibayama A1
1
Faculty of Science and Engineering, Hosei University
3-7-2 Kajino-cho, Koganei, Tokyo 184-8584, Japan

Received July 31, 2023; Published January 1, 2024

Abstract: This paper introduces the finite-difference time-domain (FDTD) method widely
used for the analysis of electromagnetic problems. The original FDTD method is based an
explicit formulation in time, allowing simple arithmetic operations without calculating simul-
taneous equations. However, at the cost of such a simple calculation, the FDTD method has
a limitation on the choice of the time step size, known as the Courant-Friedrichs-Lewy (CFL)
condition. To remove the CFL condition, the locally one-dimensional (LOD) scheme has been
applied to the implicit FDTD formulation. Here, we review the formulation of the FDTD
method and describe its application to implicit calculations, particularly with the use of the
LOD scheme.
Key Words: FDTD method, cellular dynamic system, locally one-dimensional (LOD) scheme,
unconditionally stable time-domain technique, frequency-dependent formulation, surface plas-
mon polariton, surface plasmon resonance

1. Introduction
The finite-difference time-domain (FDTD) method has widely been used in the numerical analysis of
various electromagnetic problems dealing with waves. The FDTD method was proposed by Yee in
1966 [1], in which Maxwell’s equations are discretized in time and space with a simple finite-difference
formula [2]. In particular, the electric and magnetic fields in space are assigned to cubic cells, which
is well known as Yee’s mesh. This leads to quite simple discretization of Maxwell’s equations without
using advanced mathematical knowledge. The resultant finite-difference equations are explicitly solved
with simple arithmetic operations that do not require solving simultaneous equations. Owing to this
fact, the FDTD algorithm is also suited for parallel computing.
It should be noted, however, that at the cost of the explicit calculation, the FDTD algorithm
imposes a limitation on the choice of the time step seize, known as the Courant-Friedrichs-Lewy
(CFL) condition. The time step size is determined by the size of the spatial sampling meshes; the
smaller the spatial sampling meshes are, the smaller the time step seize becomes. This is one of
the main drawbacks of the FDTD method. When the space is discretized with very small sampling

This work is licensed under a Creative Commons Attribution Non Commercial, No Derivatives 4.0 License.
Nonlinear Theory and Its Applications, IEICE, vol. 15, no. 1, pp. 2–16 ⃝IEICE
c 2024 DOI: 10.1587/nolta.15.2
meshes relative to the wavelength of the problem, the time step size also becomes unnecessarily small,
giving rise to time-consuming calculations.
To remove the CFL condition, we can resort to implicit schemes applied to the discretization in
time. The Crank-Nicolson (CN) scheme is the most popular implicit one for unconditionally stable
calculations, in which the time step size can be chosen much larger than that determined with the CFL
condition. The CN scheme has been applied to the implicit FDTD formulation [3], the calculation of
which is very time-consuming since it requires solving a large system of simultaneous equations. To
efficiently deal with the implicit calculation, we often employ an operator splitting technique, such as
the alternating-direction implicit (ADI) and locally one-dimensional (LOD) schemes.
The ADI scheme was utilized almost simultaneously by two groups in Japan [4] and Canada [5]
in 1999 to formulate an implicit FDTD method. On the other hand, we applied the LOD scheme
to the implicit formulation in 2005 [6], and the group in the U.S. and Brazil [7] also published a
conference paper on the LOD application. The use of the ADI- and LOD-FDTD methods leads to
the fact that the problem reduces to solving a tridiagonal system of linear equations, which can be
efficiently solved with the Thomas algorithm. Note that the advantage of the LOD-FDTD method
over the ADI counterpart is that it reduces the number of arithmetic operation, while maintaining
almost the same calculation accuracy [8].
In this paper, we review the formulation of the traditional explicit FDTD method and provide
its application to implicit calculations using the LOD scheme. The efficient implementation tech-
nique called a fundamental scheme [9] is described for the LOD-FDTD method. Applications of the
LOD-FDTD method to 3-D problems, frequency-dependent media, cylindrical structures, periodic
structures, and other important topics will also be presented.

2. Formulation of the FDTD method

We start with the formulation of the traditional explicit FDTD method. Maxwell’s equations for
linear, isotropic, and lossless media are given as

∂H
∇ × E = −µ (1)
∂t
∂E
∇×H =ε (2)
∂t
where E and H are the electric and magnetic vectors, respectively, µ is the permeability, and ε is
permittivity. From Eqs. (1) and (2), we obtain

∂Hx ∂Ey ∂Ez

µ = − (3)
∂t ∂z ∂y
∂Hy ∂Ez ∂Ex
µ = − (4)
∂t ∂x ∂z
∂Hz ∂Ex ∂Ey
µ = − (5)
∂t ∂y ∂x
∂Ex ∂Hz ∂Hy
ε = − (6)
∂t ∂y ∂z
∂Ey ∂Hx ∂Hz
ε = − (7)
∂t ∂z ∂x
∂Ez ∂Hy ∂Hx
ε = − (8)
∂t ∂x ∂y

for the Cartesian coordinate system. Now, Eqs. (3)–(8) are discretized using the arrangement of the
electromagnetic fields shown in Fig. 1, in which (a) and (b) illustrate the electromagnetic fields in
space and time, respectively. From Yee’s mesh in Fig. 1(a), it is seen that the electric and magnetic
fields are spatially shifted by half a cell. ∆x, ∆y and ∆z are the side lengths of the mesh in the x, y
and z directions, respectively. The space point is expressed as x = i∆x, y = j∆y and z = k∆z, in
which i, j and k are integers. From Fig. 1(b), the electric and magnetic fields are shifted by half a

3
 Fig. 1. Arrangement of electromagnetic fields.

cell in time. The time point is expressed as t = n∆t, in which n is integer. As a result, the electric
and magnetic fields are shifted by half a cell in both time and space for the FDTD formulation. The
entire computational domain is discretized using Yee’s mesh.
Here, we derive the finite-difference equation of Eq. (3) as an example. When differentiating the
left-hand side of Eq. (3), the center of the time derivative is taken to be n. Therefore, we differentiate
Hx using the values at n + 1/2 and n − 1/2 as
n+1/2 n−1/2
n
∂Hx,i,j+1/2,k+1/2 Hx,i,j+1/2,k+1/2 − Hx,i,j+1/2,k−1/2
≃ (9)
∂t ∆t
The position of Hx in space is at i, j + 1/2, k + 1/2 in Fig. 1(a), which is denoted by the subscript of
Hx .
Then, we differentiate the right-hand side of Eq. (3). Since the center of the time derivative is at n,
we use Yee’s mesh shown in Fig. 1(a) at this time. For the first term in the right-hand side of Eq. (3),
the center of the space derivative is at i, j + 1/2, k + 1/2 of Hx in Eq. (9). Therefore, across this Hx ,
we differentiate Ey using the field at i, j + 1/2, k + 1 and i, j + 1/2, k as
n
∂Ey,i,j+1/2,k+1/2 n
Ey,i,j+1/2,k+1 − Ey,i,j+1/2,k
n
≃ (10)
∂z ∆z
Similarly, the second term in the right-hand side of Eq. (3) is differentiated as
n
∂Ez,i,j+1/2,k+1/2 n
Ez,i,j+1,k+1/2 − Ey,i,j,k+1/2
n
≃ (11)
∂y ∆y

Substituting Eqs. (9)–(11) into Eq. (3) and rearranging it, we derive the following finite-difference
equation to be solved:
n+1/2 n−1/2
Hx,i,j+1/2,k+1/2 = Hx,i,j+1/2,k−1/2
∆t ( n )
+ Ey,i,j+1/2,k+1 − Ey,i,j+1/2,k
n
µ∆z
∆t ( n )
− Ez,i,j+1,k+1/2 − Ey,i,j,k+1/2
n
(12)
µ∆y

The other two components Hy and Hz are similarly derived. We can obtain the H n+1/2 field from the
given initial values E n and H n−1/2 . With respect to the electric field, the finite-difference equations
are derived from Eqs. (6)–(8) and its solution is calculated with the initial field E n and the calculated

4
field H n+1/2 . As can be seen, the FDTD method is a simple algorithm, in which the fields at n and
n − 1/2 are given as the initial values. The entire domain is updated every time step ∆t, and the
calculation proceeds until the desired time.
Now, we consider the relationship between the physical meaning of Maxwell’s equations and the
assignment of electromagnetic fields in Yee’s mesh. Equation (1) shows that the rotation of E field
results in time variation of the H field. In Yee’s mesh, the E fields are assigned to the four sides of
a single plane, and the rotation of these fields yields the H field at the center of that plane. This is
true for Eq. (2). Consequently, the FDTD method is one that can numerically calculate Maxwell’s
equations in a very natural way.
The above formulation is based on an explicit scheme; we obtain the solutions explicitly without
solving any simultaneous equations. However, at the cost of these simple calculations, the upper limit
of the time step ∆t is restricted by the well-known CFL condition as
1
v∆t ≤ √( )2 ( )2 ( )2 (13)
1 1 1
∆x + ∆y + ∆z

where v is the maximum speed of the EM wave in the calculation region (in free space, v is regarded as
the speed of light c). As seen, the upper limit of ∆t is determined with the spatial sampling meshes.
Note, for the problems with small spatial sampling meshes, that the time step becomes unnecessarily
small resulting in time-consuming calculations.

3. Application of the LOD scheme

To remove the above-mentioned CFL condition, we can resort to an implicit scheme, such as the LOD
scheme. In 2005, we applied the LOD scheme to the two-dimensional (2-D) FDTD method for the
first time [6]. For the 2-D problem, the LOD-FDTD method solves only four equations, while the
ADI-FDTD method must solve six equations. In this case, the number of arithmetic operations in
the right-hand of the resulting equations is reduced to 26 for the LOD, while it is 36 for the ADI,
leading to efficient implicit calculations. The unconditional stability of the LOD-FDTD method was
also verified analytically. Here, we review the formulation of the three-dimensional (3-D) LOD-FDTD
method [10–12], in comparison with the ADI counterpart.
Maxwell’s equation is expressed in the matrix form as

∂ϕ
= [A]ϕ (14)
∂t
where ϕ = [Ex , Ey , Ez , Hx , Hy , Hz ]T and
 
0 0 0 0 − ε∂z
∂ ∂
ε∂y
 ∂ 
 0 0 0 ∂
0 − ε∂x 
 ε∂z 
 0 0 0 − ε∂y
∂ ∂
0 
[A] = 

ε∂x 

 0 ∂
µ∂z − µ∂y
∂
0 0 0 
 ∂ 
− µ∂z
∂
0 µ∂x 0 0 0 
∂
µ∂y − µ∂x
∂
0 0 0 0

in which T is the transpose. Here, we apply the CN scheme to Eq. (14) as [3]

ϕn+1 − ϕn ϕn+1 + ϕn
= [A] (15)
∆t 2
Note, for the implicit FDTD method, that the half time step offset is not usually used for stability.
Equation (15) is rewritten as follows:
( ) ( )
∆t ∆t
[I] − [A] ϕn+1 = [I] + [A] ϕn (16)
2 2

5
Equation (16) involves solving simultaneous equations, but it has a large banded matrix with a
time-consuming procedure. Instead of directly solving Eq. (16), we introduce an operator spitting
technique. The matrix [A] is split as
   
0 0 0 0 0 ∂
ε∂y 0 0 0 0 − ε∂z
∂
0
 ∂   ∂ 
− ε∂x
 0 0 0 0 0   0 0 0 0 0 
 ε∂z   
 0 0 0 0 ∂
0   0 0 0 − ∂
0 0 

[A] =  ε∂x   ε∂y 
0 ∂
0 0 0 0 + 0 0 − ∂
0 0 0 
 µ∂z   µ∂y 
 0 ∂   − ∂ 
 0 µ∂x 0 0 0   µ∂z 0 0 0 0 0 
∂
µ∂y 0 0 0 0 0 0 − µ∂x
∂
0 0 0 0
= [A] + [B] (17)

Equation (16) can be rewritten with Eq. (17) as

( ) ( )
∆t ∆t
[I] − ([A] + [B]) ϕ n+1
= [I] + ([A] + [B]) ϕn
2 2
We introduce operator splitting as follows:
( )( ) ( )( )
∆t ∆t ∆t ∆t
[I] − [A] [I] − [B] ϕn+1 = [I] + [A] [I] + [B] ϕn
2 2 2 2
which is rewritten as
( ∆t
)( ∆t
)
[I] + 2 [A] [I] + 2 [B]
ϕ n+1
=( )( ) ϕn (18)
[I] − ∆t
2 [A] [I] − ∆t
2 [B]

The ADI scheme can be used to solve Eq. (18) as [4, 5]

∆t
[I] + 2 [A] n
ϕn+1/2 = ϕ (19)
[I] − ∆t
2 [B]

for the first step and

∆t
[I] + 2 [B] n+1/2
ϕn+1 = ϕ (20)
[I] − ∆t
2 [A]

for the second step, where ϕn+1/2 is the intermediate field. Equations (19) and (20) result in solving a
tridiagonal system of linear equations, the solution of which is efficiently carried out with the Thomas
algorithm. Notice that each of (19) and (20) has both [A] and [B], which is why this scheme is called
the “alternating-direction” implicit one.
Instead of using the ADI scheme, we have applied the LOD scheme to the implicit FDTD method [6].
With the LOD scheme, Eq. (18) is calculated as
∆t
[I] + 2 [A] n
ϕn+1/2 = ϕ (21)
[I] − ∆t
2 [A]

for the first step and

∆t
[I] + 2 [B] n+1/2
ϕn+1 = ϕ (22)
[I] − ∆t
2 [B]

for the second step. Note that the only difference between the LOD and ADI schemes is that each
step has either [A] or [B].
From Eq. (21), we obtain the following six equations for the first step:
n+1/2
∆t ∂Hz ∆t ∂Hzn
Exn+1/2 + = Exn + (23a)
2ε ∂y 2ε ∂y

6
 n+1/2
∆t ∂Hx ∆t ∂Hxn
Eyn+1/2 + = Eyn + (23b)
2ε ∂z 2ε ∂z
n+1/2
∆t ∂H y ∆t ∂Hyn
Ezn+1/2 + = Ezn + (23c)
2ε ∂x 2ε ∂x
n+1/2
∆t ∂Ey ∆t ∂Eyn
Hxn+1/2 + = Hxn + (23d)
2µ ∂z 2µ ∂z
n+1/2
∆t ∂Ez ∆t ∂Ezn
Hyn+1/2 + = Hyn + (23e)
2µ ∂x 2µ ∂x
n+1/2
∆t ∂Ex ∆t ∂Exn
Hzn+1/2 + = Hzn + (23f)
2µ ∂y 2µ ∂y
n+1/2
To eliminate Hz in Eq. (23a), we substitute Eq. (23f) to Eq. (23a), leading to
n+1/2
∆t2 ∂ 2 Ex ∆t ∂Hzn ∆t2 ∂ 2 Exn
Exn+1/2 − = Exn + − (24)
4εµ ∂y 2 ε ∂y 4εµ ∂y 2

Similarly, substituting Eq. (23d) to Eq. (23b) and Eq. (23e) to Eq. (23c) results in
n+1/2
∆t2 ∂ 2 Ey ∆t ∂Hxn ∆t2 ∂ 2 Eyn
Eyn+1/2 − = Eyn + − (25)
4εµ ∂z 2 ε ∂z 4εµ ∂z 2
n+1/2
∆t2 ∂ 2 Ez ∆t ∂Hyn ∆t2 ∂ 2 Ezn
Ezn+1/2 − = Ezn + − (26)
4εµ ∂x2 ε ∂x 4εµ ∂x2

As can be seen from Eqs. (24) to (26), there exist derivatives only in the one direction in each equation.
This is why it is called the “locally one-dimensional” scheme. Applying the finite-difference scheme
to Eqs. (24) to (26) yields the resultant finite-difference equations to be solved with three unknown
terms in each equation, known as a tridiagonal matrix. First, we solve these tridiagonal matrices from
Eq. (24) to (26) with an efficient procedure such as the Thomas algorithm. After obtaining the E field
at n + 1/2, we explicitly calculate Eq. (23d) to (23f) to obtain H field at n + 1/2. The components
for the second step are similarly calculated. The advantage of the LOD-FDTD method is that the
number of arithmetic operators is reduced, in comparison with the ADI counterpart. That is, the
number of multiplications/divisions and additions/subtractions in the right-hand sides is 72 for the
LOD and that is 102 for the ADI, leading to efficient implementations [10].
Note that the above formulation completely eliminates the CFL condition. A large time step beyond
that determined from Eq. (13) can be used, as long as the computation accuracy is reserved. The
size of the time step in the implicit FDTD method (∆timp ) is sometimes defined as the CFLN (CFL
number), which means how large it is in comparison to the maximum time step (∆tCFL ) determined
from Eq. (13), i.e., CFLN=∆timp /∆tCFL .
The envelope LOD-FDTD has also been developed [8], in which only the slowly-varying envelope
function is treated. This envelope formulation can greatly relax the constraint on the time step for
a band-limited simulation. The numerical dispersion analysis was performed [13] and the envelope
formulation was carried out in the frequency-dependent one [14] given in the next section.

4. Extension to frequency-dependent formulation

We often encounter the problems of analyzing frequency-dependent (dispersive) media, e.g., metal
materials at optical frequencies, human bodies at microwave and terahertz frequencies, water at
terahertz frequencies and so on. In particular, quite small spatial meshes are required to analyze
metal at optical frequencies. For example, the skin depth of Au is about several tens of nano-meters
at optical frequencies. In order to accurately analyze electromagnetic waves thinly penetrating into
metals, we need to use a sampling mesh of a few nano-meters. This gives rise to an extremely
small time step from Eq. (13), which would result in a very long computation time if the traditional
explicit frequency-dependent FDTD method were used. To efficiently analyze dispersive media, we

7
often resort to the frequency-dependent LOD-FDTD method, since the restriction of the time step is
completely removed.
It is necessary to devise a way to analyze the above-mentioned dispersive media, since the FDTD
method is used in time domain. There are three major ways to treat dispersive media, i.e., the
recursive convolution (RC) [15–18], auxiliary differential equation (ADE) [19, 20], and Z-transform
techniques [21]. All these techniques have been implemented and their performance has been in-
vestigated in the LOD-FDTD method [22–26]. Here, we mention the formulations of the RC-based
techniques.
As an example, we consider the Drude-Lorentz model often used for treating metals such as Au
and Ag as [27]

2
ωD ∑
P
∆εGp ωp2
εr (ω) = ε∞ + + (27)
jω (νD + jω) p=1 jωνp + ωp2 − ω 2

where ε∞ is the dielectric constant of the material at infinite frequency, ω is the angular frequency, ωD
and ωp are the electron plasma frequencies, νD and νp are the effective electron collision frequencies,
Gp is the pole strength, and ∆ε is the weighting coefficient. Eliminating the last term in the right-hand
side of Eq. (27) results in the Drude model.
The D-E relation with Eq. (27) is written in the frequency-domain as

D(ω) = ε0 εr (ω)E(ω) (28)

where ε0 is the permittivity of free space. To analyze media expressed by Eq. (28) in time domain,
we have to take inverse Fourier transform of this equation, leading to the convolution form as
∫ ∞
D(t) = ε0 εr (τ )E(t − τ )dτ (29)
−∞

Hence, the D-E relation in the time domain is expressed as follows:

∫ n∆t
D(t) = ε0 ε∞ E(t) + ε0 E(n∆t − τ )χ(τ )dτ
0
≡ ε0 ε∞ E(t) + ε0 P (n∆t) (30)

where P is the linear polarization and χ(τ ) is susceptibility. The RC-related techniques are methods
for efficiently handling this convolutional computation on P .
For the RC technique [15], P is approximated as

∑
n−1
P (n∆t) = E n−m χm (31)
m=0

where
∫ (m+1)∆t
χm = χ(τ )dτ (32)
m∆t

in which the electric field is assumed constant over ∆t, resulting in first-order accuracy in time.
For the PLRC technique [16], P is approximated as

∑
n−1
P (n∆t) = {E n−m χm + (E n−m−1 − E n−m )ξ m } (33)
m=0

where
∫ (m+1)∆t
1
ξm = (τ − m∆τ )χ(τ )dτ
∆t m∆t

8
 Fig. 2. Relation among several RC techniques.

in which the electric field has piecewise linear functional dependence over ∆t, leading to second-order
accuracy. Note that the term regarding ξ m appears compared with Eq. (31), requiring an additional
calculation of the convolution integral.
As shown, the RC technique is first-order accurate with one convolution calculation, while the
PLRC technique is second-order accurate but needs two convolution calculations. In contrast, for the
trapezoidal RC (TRC) technique [17, 18], P is approximated using an average of the electric fields
over two consecutive time steps in the following form:

∑
n−1
E n−m + E n−m−1 m
P (n∆t) = χ . (34)
m=0
2

Obviously, the TRC technique requires only a single convolution integral regarding χm as in the RC
technique, and is second-order accurate as well. This is why we often employ the TRC technique for
various dispersion model.
Figure 2 shows the relationship among RC, PLRC and TRC techniques. The convolution calculation
requires the area between E n and E n+1 . As shown in Fig. 2, the RC technique uses only E n ,
underestimating the area. The PLRC technique provides more accurate results with the piecewise
linear approximation between E n and E n+1 . In contrast, the TRC technique averages E n and E n+1 .
Clearly, the area calculated from the TRC technique is identical to that from the PLRC technique.
Here, we present the basic equations of the frequency-dependent 3-D LOD-FDTD method based
on the TRC technique for the Drude-Lorentz model of Eq. (27). As a result of a careful derivation of
the convolution integral in accordance with the two-step 3-D LOD procedure, we obtain the following
equations for the first step (the normalized expression of field components is used) [28]:
( n+1/2
)
ε∞ − χ0 /2 n 1 c∆t ∂Hz ∂Hzn
Exn+1/2 = 0
Ex + 0
ϕnx + + (35a)
ε∞ + χ /2 ε∞ + χ /2 2 (ε∞ + χ0 /2) ∂y ∂y
( n+1/2
)
ε∞ − χ0 /2 n 1 c∆t ∂Hx ∂Hxn
Eyn+1/2 = E + ϕn + + (35b)
ε∞ + χ0 /2 y ε∞ + χ0 /2 y 2 (ε∞ + χ0 /2) ∂z ∂z
( n+1/2
)
ε∞ − χ0 /2 n 1 c∆t ∂Hy ∂Hyn
Ezn+1/2 = 0
Ez + 0
ϕnz + + (35c)
ε∞ + χ /2 ε∞ + χ /2 2 (ε∞ + χ0 /2) ∂x ∂x
( n+1/2
)
c∆t ∂Ey ∂Eyn
Hxn+1/2 = Hxn + + (35d)
2 ∂z ∂z
( n+1/2
)
c∆t ∂Ez ∂Ezn
Hyn+1/2 n
= Hy + + (35e)
2 ∂x ∂x
( n+1/2
)
c∆t ∂Ex ∂Exn
Hzn+1/2 n
= Hz + + (35f)
2 ∂y ∂y

9
 ∑ [ 0] ∑P [ n]
where χ0 = χ0D + P n n
p=1 Re χp and ϕδ = ϕD +
n n
p=1 Re ϕp , in which δ is x, y or z. ϕD and ϕp are
expressed as follows:

Eδn + Eδn−1
ϕnD = ∆χ0D + e−νD ∆t ϕn−1D
2
E n + Eδn−1
ϕnp = δ ∆χ0p + eγ∆t ϕn−1p .
2
√
where γ = −α + jβ, α = νp /2, and β = ωp2 − α2 . The parameters used in the above formulation
are given as
2 { }
ωD 1 −νD ∆t
0
χD = ∆t − (1 − e )
νD νD
2∆εGp ωp2
χ0p = − (1 − eγ∆t )
γ(γ − γ ∗ )
ω2
∆χ0D = − 2D (1 − e−νD ∆t )2
νD
∆χ0p = χ0p (1 − eγ∆t )

The calculation procedure is as follows: Eqs. (35d), (35e) and (35f) are respectively substituted into
Eqs. (35b), (35c) and (35a) and the resultant tridiagonal equations are implicitly solved, and then
Eqs. (35d), (35e) and (35f) are solved explicitly. The procedure for the second step is similarly carried
out. With these formulations, we have analyzed a 3-D plasmonic waveguide with Ag claddings. With
CFLN=10 (the time step is ten times as large as that of Eq. (13)), the computation time is reduced
to 60% of that of the traditional explicit FDTD method. It is noteworthy that we do not encounter
the so-called late-time instability problem, showing its unconditional stability.
The LOD-FDTD method based on the TRC technique has also been introduced into other dispersion
models, such as the Drude-critical points [29], multi-pole Debye [30], and Cole-Cole [31] models. For
the Drude-critical points model [29], we have analyzed a plasmonic waveguide with Al claddings. With
CFLN=10, the computation time is reduced to 25% compared to the explicit FDTD method. The
frequency-dependent LOD-FDTD method is quite effective in efficiently analyzing dispersive media.

5. Fundamental scheme
Tan of Nanyang Technological University has developed an efficient calculation technique for the
implicit FDTD method, called a fundamental scheme [9]. Note that the fundamental scheme offers
convenient matrix-operator free forms in the right-hand side of the resultant equations, but provides
numerical results completely equivalent to the conventional formulation. Although the fundamental
scheme can be applied to the ADI-FDTD method, we here give the application to the LOD-FDTD
method.
The LOD-FDTD equations of Eqs. (19) and (20) are written as follows:
( ) ( )
∆t ∆t
[I] − [A] ϕ n+1/2
= [I] + [A] ϕn (36)
2 2
( ) ( )
∆t ∆t
[I] − [B] ϕn+1 = [I] + [B] ϕn+1/2 (37)
2 2

Equation (36) can be rewritten as

( ) ( )
∆t ∆t
[I] − [A] ϕ n+1/2
= [I] + [A] ϕn
2 2
( )
∆t
= 2ϕ − [I] −
n
[A] ϕn+1/2 (38)
2

This is manipulated readily to give

10
 ( )( )
∆t
[I] − [A] ϕn+1/2 + ϕn = 2ϕn
2
where the vector terms in the bracket can be denoted by an auxiliary variable

v n+1/2 = ϕn+1/2 + ϕn

Similar manipulation can be applied to Eq. (37). As a result, we derive the following simplified
equations for the LOD-FDTD method:
( )
∆t
[I] − [A] v n+1/2 = ϕn (39)
2
ϕn+1/2 = v n+1/2 − ϕn (40)

for the first step and

( )
∆t
[I] − [B] v n+1 = ϕn+1/2 (41)
2
ϕn+1 = v n+1 − ϕn+1/2 (42)

for the second step. As can be seen, the above equations do not have any spatial derivatives in the
right-hand sides, that is they are matrix-operator-free forms, which leads to an efficient implementa-
tion of the LOD-FDTD method. The number of multiplications/divisions and additions/subtractions
in the right-hand sides of the fundamental LOD is reduced to 42, while for the original LOD it is 72.
The fundamental scheme can be applied to the frequency-dependent formulations. We have intro-
duced the fundamental scheme into the frequency-dependent LOD-FDTD method for various disper-
sion models [30, 32]. For the Debye model [30], we have analyzed the response of THz radiation to the
healthy skin and the basal cell carcinoma. The computation time is reduced to 19% with CFLN=20,
maintaining acceptable accuracy.
Incidentally, the fundamental scheme can be applied to any calculation technique that repeatedly
performs an implicit calculation. For example, we have applied the scheme to the beam-propagation
method (BPM) for the analysis of optical waveguides [33–35]. For the full-vectorial BPM [34], the
number of the terms regarding mixed derivatives is reduced from 8 to 2 with the aid of the fundamental
scheme, while maintaining the equivalent computation as in the conventional full-vectorial one.

6. Other applications of the LOD-FDTD method

Discussions given above is based on the FDTD method in the Cartesian coordinate system shown
in Fig. 1(a). Needless to say, the FDTD method can be formulated in another coordinate system.
The LOD-FDTD method has been formulated in the cylindrical coordinate system together with
the fundamental scheme [36], in which a metal disc-type surface wave splitter [37] was efficiently
analyzed and designed. The LOD-FDTD method in cylindrical coordinates has also been extended to
a frequency-dependent version for plasmonics. The plasmonic grating was analyzed with CFLN=50,
leading to a reduced computation time of only 9% of the explicit cylindrical FDTD method [38].
It should be noted, in cylindrical coordinates, that the fields are regarded as periodic in the cir-
cumferential (φ) direction. This means that the fields at a specific plane in the φ direction should be
connected using the periodic boundary condition. For the implicit FDTD method, this connection
of the fields gives rise to the problem with a cyclic matrix, which cannot be directly solved with
the Thomas algorithm. To efficiently handle the cyclic matrix, we utilize the Sherman-Morrison for-
mula [39], in which the cyclic matrix is transformed into the ones consisting of a tridiagonal matrix.
As a result, the equations for the cylindrical LOD-FDTD method can also be efficiently calculated
with the Thomas algorithm.
In contrast to the use of the cylindrical FDTD method, the body-of-revolution (BOR) one is quite
effective for the analysis of circularly symmetric structures, since only the 2-D plane is analyzed even
for the 3-D structure. The LOD scheme has been applied to the BOR-FDTD method [40] together

11
with the fundamental scheme [41], in which a fiber grating is efficiently analyzed with dispersion
control parameters [42]. The BOR-LOD-FDTD method has also been employed to analyze frequency-
dependent media for a plasmonic rod waveguide and grating [43].
For periodic structures such as meta materials, the FDTD method with the periodic boundary
condition has widely been utilized. The LOD scheme has been introduced into the periodic FDTD
method at normal incidence [44], in which the Sherman-Morrison formula is also used to handle a
cyclic matrix resulting from the periodic boundary. In addition, at oblique incidence, the split-field
technique is applied to the LOD-FDTD method [45]. The dispersion control parameters are also
introduced to reduce the numerical dispersion error. The computational time is reduced to 16% of
the explicit counterpart for the analysis of a broadband mirror consisting of a subwavelength grating.
The traditional explicit FDTD method is suitable for parallel computations due to its simple calcu-
lations without solving simultaneous equations. For the implicit calculations, however, it is generally
difficult to carry out parallel computations due to the need for solving simultaneous equations. Nev-
ertheless, some attempts have been made for parallel computations using the LOD scheme. Tan [46]
optimized the LOD-FDTD algorithm for graphics processor units. Hemmi et al., [47] described an
efficient message passing interface-parallel implementation of the LOD-FDTD method for Debye-
dispersive media. Zygiridis et al., Tan et al., and Li et al., also investigated the parallel LOD-FDTD
method for efficient calculations [48–50]. We have implemented an error-controllable scheme for the
LOD-FDTD method with the help of parallel computations, wherein the fast inverse Laplace trans-
form is employed to generate the electromagnetic field in arbitrary time domain from that in complex
frequency domain [51].
The structure to be analyzed is not necessarily a polygonal shape that can be easily discretized
with Yee’s mesh. For arbitrarily shaped structures, modeling requires the use of fine meshes, with
a subsequent increase in the computation time. To alleviate this problem, we can resort to the con-
formal technique for modeling those arbitrarily shaped structures. The nonorthogonal and conformal
LOD-FDTD methods have been developed for modeling those structures with coarse meshes [52–54].
Another interesting approach for dealing with arbitrarily shaped structures is to use the radial point
interpolation meshless method. The implicit unconditionally stable meshless time-domain method
has been developed [55, 56].
Highly accurate LOD-FDTD methods have also been proposed using an iteration technique [57],
an additional procedure [58] and a higher order scheme [59–62], which are effective for large scale
problems. Last but not least, the implementation of an absorbing boundary condition (ABC) is
quite important for practical simulations. The perfectly matched layer (PML) ABC has widely been
employed in the FDTD simulations, the absorbance of which outperforms other ABCs. The PML-
ABC has been optimized for the use in the LOD-FDTD method [63–66], which makes the LOD
simulation reliable.

7. Conclusions
We have reviewed the formulation of the traditional explicit FDTD method widely used for electro-
magnetic problems and presented its application to implicit LOD calculations. One main drawbacks
of the FDTD method is the restriction of the time step size, known as the CFL condition. When the
spatial sampling meshes should be chosen small, the time step size becomes unnecessarily small, giving
rise time-consuming calculations. The use of the LOD scheme completely removes the CFL condition,
in which the time step can be chosen large as long as the accuracy of the calculations is maintained.
In particular, the plasmonic devices consisting of dispersive media, to which small spatial meshes
should be applied, can be efficiently analyzed using the frequency-dependent LOD-FDTD method.
Since the LOD scheme is essentially applicable to any problem involving the FDTD calculations, its
ripple effect is quite large. As a result, the LOD-FDTD method has been improved and extended
by many researchers round the world. The application of the LOD-FDTD method is still spreading
in various fields [67–69]. In addition to conventional electromagnetic problems, the LOD scheme is
expected to be further utilized for computational techniques to analyze other physical phenomena.

12
Acknowledgments
The author would like to thank Prof. Hiroyuki Torikai, a Guest Editor of this Special Section, for
giving the author the opportunity to write this invited paper. Thanks are also due to Prof. Junji
Yamauchi and Prof. Emeritus Hisamatsu Nakano of Hosei University for their continued support of
the author’s research. The studies regarding the LOD-FDTD method were conducted with previous
graduate students, and the author thanks them for their tremendous efforts. This work was supported
in part by JSPS KAKENHI Grant Numbers JP21760266, JP23560415, JP15K06035, JP19K04537 and
JP23K03962.

References
[1] K.S. Yee, “Numerical solution of initial boundary value problems involving Maxwell’s equations
in isotropic media,” IEEE Trans. Antennas Propagat., vol. AP-14, pp. 302–307, 1966.
[2] A. Taflove and S.C. Hagness, Computational Electrodynamics: The Finite-Difference Time-
Domain Method. 3rd Edition, Norwood, MA: Artech House, 2005.
[3] Y. Yang, R.S. Chen, and E.K.N. Yung, “The unconditionally stable Crank Nicolson FDTD
method for three-dimensional Maxwell’s equations,” Microw. Opt. Tech. Lett., vol. 48, no. 8,
pp. 1619–1622, 2006.
[4] T. Namiki, “A new FDTD algorithm based on alternating-direction implicit method,” IEEE
Trans. Microw. Theory Tech., vol. 47, no. 10, pp. 2003–2007, 1999.
[5] F.H. Zheng, Z.Z. Chen, and J.Z. Zhang, “A finite-difference time-domain method without the
Courant stability conditions,” IEEE Microw. Guided Wave Lett., vol. 9, no. 11, pp. 441–443,
1999.
[6] J. Shibayama, M. Muraki, J. Yamauchi, and H. Nakano, “Efficient implicit FDTD algorithm
based on locally one-dimensional scheme,” Electron. Lett., vol. 41, no. 19, pp. 1046–1047, 2005.
[7] V.E. Nascimento, J.A. Cuminato, F.L. Teixeira, and B.-H.V. Borges, “Unconditionally stable
finite-difference time-domain method based on the locally-one-dimensional technique,” Proc.
22nd Simp. Brasileiro Telecomun., Campinas, Brazil, pp. 288–291, 2005.
[8] J. Shibayama, M. Muraki, R. Takahashi, J. Yamauchi, and H. Nakano, “Performance evalua-
tion of several implicit FDTD methods for optical waveguide analyses,” IEEE/OSA J. Lightw.
Technol., vol. 24, no. 6, pp. 2465–2472, 2006.
[9] E.L. Tan, “Fundamental schemes for efficient unconditionally stable implicit finite-difference
time-domain methods,” IEEE Trans. Antennas Propagat., vol. 56, no. 1, pp. 170–177, 2008.
[10] E.L. Tan, “Unconditionally stable LOD-FDTD method for 3-D Maxwell’s equations,” IEEE
Microw. Wirel. Compon. Lett., vol. 17, no. 2, pp. 85–87, 2007.
[11] I. Ahmed, E.K. Chua, E.P. Li, and Z. Chen, “Development of the three-dimensional uncondition-
ally stable LOD-FDTD method,” IEEE Trans. Antennas Propagat., vol. 56, no. 11, pp. 3596–
3600, 2008.
[12] J. Shibayama, R. Ando, J. Yamauchi, and H. Nakano, “A 3-D LOD-FDTD method for the
wideband analysis of optical devices,” IEEE/OSA J. Lightw. Technol., vol. 29, no. 11, pp. 1652–
1658, 2011.
[13] K. Sasaki, T. Takashima, and Y. Suzuki, “A dispersion analysis on the two-dimensional envelope
LOD-FDTD method,” (in Japanese), Trans. IEICE, vol. J93-C, pp. 279–268, 2010.
[14] O. Ramadan, “Improved and efficient unconditionally stable complex-envelope frequency-
dependent FDTD formulations based on the implicit locally one-dimensional scheme,” J. Elec-
tromagn. Waves Appl., vol. 28, no. 3, pp. 334–345, 2014.
[15] R.J. Luebbers and F. Hunsberger, “FDTD for N th-order dispersive media,” IEEE Trans. An-
tennas Propag., vol. 40, no. 11, pp. 1297–1301, 1992.
[16] D.F. Kelley and R.J. Luebbers, “Piecewise linear recursive convolution for dispersive media
using FDTD,” IEEE Trans. Antennas Propag., vol. 44, no. 6, pp. 792–797, 1996.
[17] R. Siushansian and J. LoVetri, “A comparison of numerical techniques for modeling electro-
magnetic dispersive media,” IEEE Microw. Guided Wave Lett., vol. 5, no. 12, pp. 426–428,

13
 1995.
[18] R. Siushansian and J. LoVetri, “Efficient evaluation of convolution integrals arising in FDTD
formulations of electromagnetic dispersive media,” J. Electromag. Waves Appli., vol. 11, no. 1,
pp. 101–117, 1997.
[19] T. Kashiwa and I. Fukai, “A treatment by FD-TD method of dispersive characteristics associated
with electronic polarization,” Microw. Opt. Tech. Lett., vol. 3, no. 6, pp. 203–205, 1990.
[20] R.M. Joseph, S.C. Hagness, and A. Taflove, “Direct time integration of Maxwell equations in
linear dispersive media with absorption for scattering and propagation of femtosecond electro-
magnetic pulses,” Opt. Lett., vol. 16, no. 18, pp. 1412–1414, 1991.
[21] D.M. Sullivan, “Frequency-dependent FDTD methods using Z Transforms,” IEEE Trans. An-
tennas Propag., vol. 40, no. 10, pp. 1223–1230, 1992.
[22] J. Shibayama, R. Takahashi, J. Yamauchi, and H. Nakano, “Frequency-dependent LOD-FDTD
implementations for dispersive media,” Electron. Lett., vol. 42, no. 19, pp. 1084–1086, 2006.
[23] J. Shibayama, R. Takahashi, A. Nomura, J. Yamauchi, and H. Nakano, “Concise frequency-
dependent formulation for LOD-FDTD method using Z transforms,” Electron. Lett., vol. 44,
no. 16, pp. 949–950, 2008.
[24] J. Shibayama, R. Ando, A. Nomura, J. Yamauchi, and H. Nakano, “Simple trapezoidal recursive
convolution technique for the frequency-dependent FDTD analysis of a Drude-Lorentz model,”
IEEE Photon. Technol. Lett., vol. 21, no. 2, pp. 100–102, 2009.
[25] J. Shibayama, R. Takahashi, J. Yamauchi, and H. Nakano, “Frequency-dependent locally one-
dimensional FDTD implementation with a combined dispersion model for the analysis of surface
plasmon waveguides,” IEEE Photon. Technol. Lett., vol. 20, no. 10, pp. 824–826, 2008.
[26] J. Shibayama, A. Nomura, R. Ando, J. Yamauchi, and H. Nakano, “A frequency-dependent
LOD-FDTD method and its application to the analyses of plasmonic waveguide devices,” IEEE
J. Quantum Electron., vol. 46, no. 1, pp. 40–49, 2010.
[27] A. Vial, A.-S. Grimault, D. Macı́as, D. Barchiesi, and M.L. de la Chapelle, “Improved analytical
fit of gold dispersion: Application to the modeling of extinction spectra with a finite-difference
time-domain method,” Phys. Rev. B, vol. 71, pp. 085416-1-7, 2005.
[28] J. Shibayama, R. Ando, J. Yamauchi, and H. Nakano, “Frequency-dependent 3-D LOD-FDTD
method for the analysis of plasmonic devices,” IEEE Photon. Technol. Lett., vol. 23, no. 15,
pp. 1070–1072, 2011.
[29] J. Shibayama, K. Watanabe, R. Ando, J. Yamauchi, and H. Nakano, “Frequency-dependent
formulations of a Drude-critical points model for explicit and implicit FDTD methods using the
trapezoidal RC technique,” IEICE Trans. Electron., vol. E95-C, no. 4, pp. 725–732, 2012.
[30] J. Shibayama, N. Sasaki, Y. Wakabayashi, J. Yamauchi, and H. Nakano, “Frequency-dependent
fundamental LOD-FDTD formulation for a multi-pole Debye model,” IEEE Photon. Technol.
Lett., vol. 25, no. 10, pp. 965–968, 2013.
[31] J. Shibayama, Y. Kusaka, J. Yamauchi, and H. Nakano, “Development of an LOD-FDTD
method for the analysis of the Cole-Cole model,” Progress in Electromagnetics Research Sym-
posium, Shanghai, China, p. 1781, 2016.
[32] J. Shibayama, T. Hirano, J. Yamauchi, and H. Nakano, “Efficient implementation of frequency-
dependent 3-D LOD-FDTD method using fundamental scheme,” Electron. Lett., vol. 48, no. 13,
pp. 774–775, 2012.
[33] J. Shibayama, A. Yokomizo, J. Yamauchi, and H. Nakano, “Reformulation of the ADI-BPM
using a fundamental scheme,” IEICE Electron. Expr., vol. 9, no. 5, pp. 365–370, 2012.
[34] J. Shibayama, A. Yokomizo, J. Yamauchi, and H. Nakano, “Simplified algorithms for the full-
vectorial ADI-BPM using a fundamental scheme,” IEEE Photon. Technol. Lett., vol. 25, no. 2,
pp. 147–150, 2013.
[35] Y. Nito, J. Shibayama, J. Yamauchi, and H. Nakano, “Full-vectorial beam-propagation meth-
ods based on a fundamental scheme -design of a short polarization converter-,” IEEE/OSA J.
Lightw. Technol., vol. 32, no. 21, pp. 3509–3516, 2014.
[36] J. Shibayama, M. Ito, J. Yamauchi, and H. Nakano, “A fundamental LOD-FDTD method in

14
 cylindrical coordinates,” IEEE Photon. Technol. Lett., vol. 29, no. 11, pp. 865–868, 2017.
[37] J. Shibayama, J. Yamauchi, and H. Nakano, “Metal disc-type splitter with radially placed
gratings for terahertz surface waves,” Electron. Lett., vol. 51, no. 4, pp. 352–353, 2015.
[38] J. Shibayama, T. Hara, M. Ito, J. Yamauchi, and H. Nakano, “Frequency-dependent LOD-
FDTD method in cylindrical coordinates,” IEICE Trans. Electron., vol. E101-C, no. 8, pp. 637–
639, 2018.
[39] J.W. Thomas, Numerical Partial Differential Equations: Finite Difference Methods, Berlin,
Germany: Springer-Varlag, 1995.
[40] J. Shibayama, B. Murakami, J. Yamauchi, and H. Nakano, “LOD-BOR-FDTD algorithm for ef-
ficient analysis of circularly symmetric structures,” IEEE Microw. Wirel. Compon. Lett., vol. 19,
no. 2, pp. 56–58, 2009.
[41] J. Shibayama, T. Oikawa, J. Yamauchi, and H. Nakano, “Efficient LOD-BOR-FDTD imple-
mentation based on a fundamental scheme,” IEEE Photon. Technol. Lett., vol. 24, no. 11,
pp. 957–959, 2012.
[42] E. Li, I. Ahmed, and R. Vahldieck, “Numerical dispersion analysis with an improved LOD-
FDTD method,” IEEE Microw. Wirel. Compon. Lett., vol. 17, no. 5, pp. 319–321, 2007.
[43] J. Shibayama, T. Oikawa, T. Hirano, J. Yamauchi, and H. Nakano, “Fundamental LOD-BOR-
FDTD method for the analysis of plasmonic devices,” IEICE Trans. Electron., vol. E97-C, no. 7,
pp. 707–709, 2014.
[44] J. Shibayama, R. Ando, J. Yamauchi, and H. Nakano, “An LOD-FDTD method for the analysis
of periodic structures at normal incidence,” IEEE Antennas Wirel. Propag. Lett., vol. 8, pp. 890–
893, 2009.
[45] Y. Wakabayashi, J. Shibayama, J. Yamauchi, and H. Nakano, “An LOD-FDTD method for the
analysis of a periodic structure at oblique incidence,” Radio Sci., vol. 46, RS0F031-1-9, 2011.
[46] E.L. Tan, “Acceleration of LOD-FDTD method using fundamental scheme on graphics processor
units,” IEEE Microw. Wirel. Compon. Lett., vol. 20, no. 12, pp. 648–650, 2010.
[47] T. Hemmi, F. Costen, S. Garcia, R. Himeno, H. Yokota, and M. Mustafa, “Efficient parallel
LOD-FDTD method for Debye-dispersive media,” IEEE Trans. Antennas Propag., vol. 62, no. 3,
pp. 1330–1338, 2014.
[48] T.T. Zygiridis, N.V. Kantartzis, and T.D. Tsiboukis, “Parallel LOD-FDTD method with error-
balancing properties,” IEEE Trans. Magn., vol. 51, no. 3, pp. 1–4, 2015.
[49] J. Tan, Z. Li, Q. Guo, and Y. Long, “An efficient parallel implicit solver for LOD-FDTD
algorithm in cloud computing environment,” IEEE Antennas Wirel. Propag. Lett., vol. 17,
no. 7, pp. 1209–1212, 2018.
[50] Z. Li, J. Tan, Q. Guo, Z. Su, and Y. Long, “A parallel CE-LOD-FDTD model for instrument
landing system signal disturbance analyzing,” IEEE Trans. Antennas Propag., vol. 67, no. 4,
pp. 2503–2512, 2019.
[51] T. Nakazawa, D. Wu, S. Kishimoto, J. Shibayama, J. Yamauchi, and S. Ohnuki, “Error-
controllable scheme for the LOD-FDTD method,” IEEE J. Multiscale Multiphysics Comput.
Tech., vol. 7, pp. 135–141, 2022.
[52] K. Fujita and T. Namiki, “Numerical analysis of three-dimensional conformal LOD-FDTD
scheme for solving time dependent Maxwell’s equations,” (in Japanese), Proc. Japan Society
for Simulation Technology 2009, pp. 353–356, 2009.
[53] M.M. Rana and A.S. Mohan, “Nonorthogonal LOD-FDTD method for EM scattering from
two-dimensional structures,” IIEEE Trans. Electromagn. Compat., vol. 55, no. 4, pp. 764–772,
2013.
[54] H. Liu, X. Zhao, X.-H. Wang, S. Yang, and Z. Chen, “An unconditionally stable conformal
LOD-FDTD method for curved PEC objects and its application to EMC problems,” IEEE
Trans. Electromagn. Compat., vol. 64, no. 3, pp. 827–839, 2022.
[55] Y. Yu and Z. Chen, “Towards the development of an unconditionally stable time-domain mesh-
less method,” IEEE Trans. Microw. Theory Tech., vol. 58, no. 3, pp. 578–586, 2010.
[56] N. Khakzar, A. Ali Heidari, and M. Movahhedi, “Unconditionally stable meshless method for

15
 solving 3D transient electromagnetic problems based on LOD scheme,” IET Sci. Meas. Technol.,
vol. 13, no. 6, pp. 818–823, 2019.
[57] K.-Y. Jung and F.L. Teixeira, “An iterative unconditionally stable LOD-FDTD method,” IEEE
Microw. Wirel. Compon. Lett., vol. 18, no. 2, pp. 76–78, 2008.
[58] Z. Yang, E.L. Tan, and D.Y. Heh, “Second-order temporal-accurate scheme for 3-D LOD-FDTD
method with three split matrices,” IEEE Antennas Wirel. Propag. Lett., vol. 14, pp. 1105–1108,
2015.
[59] F. Liang, G. Wang, and W. Ding, “Low numerical dispersion locally one-dimensional FDTD
method based on compact higher-order scheme,” Microw. Opt. Tech. Lett., vol. 50, no. 11,
pp. 2783–2787, 2008.
[60] Q.-F. Liu, Z. Chen, and W.-Y. Yin, “An arbitrary-order LOD-FDTD method and its stability
and numerical dispersion,” IEEE Trans. Antennas Propagat., vol. 57, no. 8, pp. 2409–2417,
2009.
[61] A.K. Saxena and K.V. Srivastava, “A three-dimensional unconditionally stable five-step LOD-
FDTD method,” IEEE Trans. Antennas Propagat., vol. 62, no. 3, pp. 1321–1329, 2014.
[62] A.K. Saxena and K.V. Srivastava, “Higher order LOD-FDTD methods and their numerical
dispersion properties,” IEEE Trans. Antennas Propagat., vol. 65, no. 3, pp. 1480–1485, 2017.
[63] V.E. do Nascimento, B.-H.V. Borges, and F.L. Teixeira, “Split-field PML implementations for
the unconditionally stable LOD-FDTD method,” IEEE Microw. Wirel. Compon. Lett., vol. 16,
no. 7, pp. 398–400, 2006.
[64] O. Ramadan, “Unsplit field implicit PML algorithm for complex envelope dispersive LOD-
FDTD simulations,” Electron. Lett., vol. 43, no. 5, pp. 267–268, 2007.
[65] I. Ahmed, E. Li, and K. Krohne, “Convolutional perfectly matched layer for an unconditionally
stable LOD-FDTD method,” IEEE Microw. Wirel. Compon. Lett., vol. 17, no. 12, pp. 816–818,
2007.
[66] O. Ramadan, “Efficient LOD-SC-PML formulations for electromagnetic fields in dispersive me-
dia,” IEEE Microw. Wirel. Compon. Lett., vol. 22, no. 6, pp. 297–299, 2012.
[67] E.L. Tan and D.Y. Heh, “Stability analyses of nonuniform time-step LOD-FDTD methods
for electromagnetic and thermal simulations,” IEEE J. Multiscale Multiphysics Comput. Tech.,
vol. 2, pp. 183–193, 2017.
[68] D.Y. Heh and E.L. Tan, “Multiple LOD-FDTD method for multiconductor coupled transmission
lines,” IEEE J. Multiscale Multiphysics Comput. Tech., vol. 5, pp. 201–208, 2020.
[69] P. Wu, H. Yu, Y. Xie, H. Jiang, and T. Natsuki, “Computationally efficient locally one-
dimensional algorithm for open region ground penetrating radar problem with improved ab-
sorption,” IEEE Access, vol. 9, pp. 88759–88766, 2021.

16