Effective permittivities with exact second-order

accuracy at inclined dielectric interface for

the two-dimensional finite-difference
time-domain method

Takuo Hirono,1,2,* Yuzo Yoshikuni,1,3 and Takayuki Yamanaka1

1
NTT Photonics Laboratories, NTT Corporation, 3-1, Morinosato Wakamiya,
Atsugi-shi, Kanagawa 243-0198, Japan
2
Present address: NTT Electronics Corporation, 6700-2, To, Naka-shi,
Ibaraki 311-0122, Japan
3
Present address: School of Science, Kitasato University, 1-15-1, Kitasato,
Sagamihara-shi, Kanagawa 228-8555, Japan
*Corresponding author: hirono‑takuo@ntt‑el.com

Received 11 November 2009; revised 20 January 2010; accepted 20 January 2010;

posted 22 January 2010 (Doc. ID 119469); published 22 February 2010

Accuracy degradation at a dielectric interface in simulations using the finite-difference time-domain

method can be prevented by assigning suitable effective permittivities at the nodes in the vicinity of
the interface. The effective permittivities with exact second-order accuracy at the interface inclined
to the Yee-lattice axis are analytically derived for what we believe to be the first time. We discuss
two interfaces with different inclined angles between their normal and the Yee-lattice axis in the case
of two-dimensional TE polarization. The tangent of the angle is 1 for one interface and 1=2 for the
other. With the derived effective permittivities, reflection and transmission at the interface are simulated
with second-order accuracy with respect to cell size. The accuracy is demonstrated by numerical
examples. © 2010 Optical Society of America
OCIS codes: 260.2110, 000.4430, 130.3120, 230.7370.

1. Introduction [3], nonorthogonal FDTD method [4], subgridding

The finite-difference time-domain (FDTD) method method [5], immersed interface method [6], and effec-
has been extensively used for electromagnetic field tive permittivity method [7–13], have been proposed
simulation and photonic device analyses [1,2]. The to overcome this problem. All of them, except for the
FDTD method for homogeneous media is second- effective permittivity method, accept additional
order accurate in space. The accuracy may be de- complexity of the algorithm in exchange for accuracy
graded in simulations including a material interface. improvement.
In practical FDTD applications, accurate modeling of In the effective permittivity method, specially de-
the material interface is important for saving compu- termined permittivities are assigned to the nodes in
tational time and memory resources. We focus on the the vicinity of the dielectric interface to prevent ac-
dielectric interface in this paper. Many methods, curacy reduction. The same algorithm and the same
such as the contour-path integral FDTD method Cartesian grid are used not only for the nodes in an
homogeneous region, but also for those in the vicinity
of the interface.
0003-6935/10/071080-17$15.00/0 In many situations, the average permittivity in the
© 2010 Optical Society of America cell surrounding the electric field component node is

1080 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

used for the effective permittivity at the node. For a on ε0 1=2 and μ0 1=2 and subsequently scale time on
plane interface, the use of the arithmetic mean for ðε0 μ0 Þ−1=2 . The value of the speed of light in a vacuum
the tangential electric field component node and har- is 1 with this scaling. The value of angular frequency
monic mean for the perpendicular electric filed com- ω and the value of the wave-vector magnitude in free
ponent node has been proposed [7]. space k0 are the same.
For more general cases with a curved interface, Basic FDTD difference equations for a lossless and
several methods have been proposed. In [8], the ar- sourceless medium are
ithmetic means of the permittivities in the square
sections of the cell were calculated. The sections were H x n ðj; kÞ
H x n−1 ðj; kÞ
normal to the direction of the electric field component Δt
at the node. The harmonic mean of the arithmetic
means was used for the effective permittivity. In [9], Ez n
1=2 ðj þ 1=2; kÞ
Ez n
1=2 ðj
1=2; kÞ
¼

one of the effective medium approximations, the in- Δ
verted Maxwell–Garnett rule, was used to obtain the Ey n
1=2 ðj; k þ 1=2Þ
Ey n
1=2 ðj; k
1=2Þ
effective permittivity. In [10], the capacitance of the þ ; ð1Þ
cell was computed by solving numerically the La- Δ
place equation of static electric potential. The effec-
tive permittivity was given as the ratio of the
capacitance to that of the region filled with air. In Ey nþ1=2 ðj; k þ 1=2Þ
Ey n
1=2 ðj; k þ 1=2Þ
εj;kþ1=2
an approach called the volume-average polarized ef- Δt
fective permittivity (VP-EP) scheme [11], the H x n ðj; k þ 1Þ
H x n ðj; kÞ
weighted average of the arithmetic mean and the ¼ ; ð2Þ
harmonic mean of the permittivities in the cell Δ
was proposed for the effective permittivity. The
weight was determined depending on the angle be-
tween the interface and the electric field component Ez nþ1=2 ðj þ 1=2; kÞ
Ez n
1=2 ðj þ 1=2; kÞ
εjþ1=2;k
at the node. Δt
These effective permittivities were proposed on the
H x n ðj þ 1; kÞ
H x n ðj; kÞ
basis of phenomenological or physical considera- ¼
; ð3Þ
tions. The accuracy improvements obtained with Δ
these effective permittivities were confirmed from
where Δt is the time increment, Δ is cell size,
the results for numerical examples. For reliable ap-
H x n ðj; kÞ is the component of the magnetic field in
plication of the effective permittivity method, analy-
the x direction at position ðy; zÞ ¼ ðjΔ; kΔÞ on time
tical derivation of the effective permittivity becomes
nΔt , Ey is the component of the electric field in the
important.
y direction, Ez is the component of the electric field
Recently, we analytically showed that, for the in-
in the z direction, and ε is the permittivity. With
terface orthogonal to the Yee-lattice axis, the appli-
the scaling, the permittivity ε equals the square of
cation of the arithmetic mean of the permittivities
the refractive index.
to the tangential electric field component node and
To derive the effective permittivities, we investi-
the harmonic mean to the perpendicular electric field
gate the situation in which a plane wave with angu-
component node is the condition for second-order ac-
lar frequency ω is incident on the interface. We
curacy with respect to cell size [12]. For the interface
obtain the stationary solutions of Eqs. (1)–(3) for this
orthogonal to the Yee-lattice axis, effective permittiv-
situation as
ities with third-order accuracy for the fourth-order
FDTD method have been analytically obtained [13].
In this paper, effective permittivities with second- H x n ðj; kÞ ¼ expðinΔt ωÞH sx ðjΔ; kΔÞ; ð4Þ
order accuracy with respect to cell size for the inter-
face inclined to the Yee-lattice axis are analytically
derived. Ey nþ1=2 ðj; k þ 1=2Þ ¼ expðiðn þ 1=2ÞΔt ωÞ
× Esy ðjΔ; ðk þ 1=2ÞΔÞ; ð5Þ
2. Derivations of Effective Permittivities
with Second-Order Accuracy
We focus on the simulation in the two-dimensional Ez nþ1=2 ðj þ 1=2; kÞ ¼ expðiðn þ 1=2ÞΔt ωÞ
(2D) TEx case on the y–z plane. The cells of the
Yee lattice are square. We assume that the perme- × Esz ððj þ 1=2ÞΔ; kΔÞ: ð6Þ
ability is μ0 throughout the simulation domain. We
represent the direction as the components ðny ; nz Þ The variations of the stationary solutions in the di-
of the unit vector, for example, “in the ð2−1=2 ; 2−1=2 Þ rection orthogonal to the interface are expressed by a
direction.” transfer matrix. The reflection and transmission
To simplify the description of expressions, we scale coefficients at the interface are obtained by connect-
the electric field and the magnetic field, respectively, ing the stationary solutions in the regions separated

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1081



by the interface. The transfer matrix is used for the jþk jþk ð1∶1Þ
connection. The conditions for the effective permit- H sx Δ; Δ ¼ H x ðj þ kÞ; ð12Þ
2 2
tivities are derived by matching the reflection and
transmission coefficients to the exact ones with
second-order accuracy with respect to cell size. Here-
after, in this paper, the order of accuracy means that
with respect to cell size.
We define the inclined angle of the interface as


 
that between the normal of the interface and the jþk 1 jþk 1
Yee-lattice axis. The effective permittivities with Esy þ Δ; þ Δ
2 4 2 4
second-order accuracy for the interface orthogonal

to the ð2−1=2 ; 2−1=2 Þ direction are derived in ð1∶1Þ 1
¼ Ey jþkþ ; ð13Þ
Subsection 2.A, and those for the interface orthogo- 2
nal to the ðð4=5Þ1=2 ; 5−1=2 Þ direction are derived in
Subsection 2.B. The tangent of the inclined angle
is 1 in Subsection 2.A and is 1=2 in Subsection 2.B.
A. Effective Permittivities for the Interface Orthogonal


 
jþk 1 jþk 1
to the ð2−1=2 ; 2−1=2 Þ Direction Esz þ Δ; þ Δ
2 4 2 4
Symbols used in Subsection 2.A but not in

Subsection 2.B are followed by a superscriptð1∶1Þ . ð1∶1Þ 1
¼ Ez jþkþ : ð14Þ
We assume that the identical permittivity is as- 2
signed to a column of Ey and Ez nodes aligned in
the ð−2−1=2 ; 2−1=2 Þ direction as From Eqs. (1)–(14), H x ð1∶1Þ ðm þ 1Þ, Ey ð1∶1Þ ðm þ 1=2Þ,
ð1∶1Þ and Ez ð1∶1Þ ðm þ 1=2Þ are expressed by H x ð1∶1Þ ðmÞ,
εjþ1=2;k ¼ εm ðm ¼ j þ k þ 1=2Þ; ð7Þ Ey ð1∶1Þ ðm − 1=2Þ, and Ez ð1∶1Þ ðm − 1=2Þ as
ð1∶1Þ
εj;kþ1=2 ¼ εm ðm ¼ j þ k þ 1=2Þ: ð8Þ

We assume that the incident plane wave has the ð1∶1Þ 


iWεmþ1=2 Δ 2
wave-vector component kt in the ð−2−1=2 ; 2−1=2 Þ direc- ð1∶1Þ
H x ðm þ 1Þ ¼   iWΔ þ ð1∶1Þ
tion. The stationary solutions also have the wave- 2 cos kpt Δffiffi 2
iWεmþ1=2 Δ
vector component kt in that direction. Thus, the

stationary solutions satisfy the following equations: ð1∶1Þ ikt Δ
× H x ðmÞ þ exp pffiffiffi

 2 2
pffiffiffi k
j

H sx ðjΔ; kΔÞ ¼ exp
i 2kt Δ ð1∶1Þ
ikt Δ
2 × Ey ðm
1=2Þ
exp pffiffiffi

 2 2
jþk jþk 
s
× Hx Δ; Δ ; ð9Þ ð1∶1Þ
× Ez ðm
1=2Þ ; ð15Þ
2 2

 

 
1 pffiffiffi k
j 1
Esy jΔ; k þ Δ ¼ exp
i 2kt þ Δ
2 2 4 



 1
ikt Δ
jþk 1 ð1∶1Þ
Ey ðm þ 1=2Þ ¼
 i exp pffiffiffi
× Esy þ Δ; 2 2
2 4 2 cos kpt Δffiffi2

 
jþk 1

þ Δ ; ð10Þ 2
2 4 × WΔ
ð1∶1Þ
Wεmþ1=2 Δ


 

 


1 pffiffiffi k
j 1
Esz jþ Δ; kΔ ¼ exp
i 2kt
Δ 2 cos ffiffi2 exp 2ikpt Δffiffi2 
kpt Δ
2 2 4


 þ
ð1∶1Þ
H x ðmÞ
jþk 1 Wεmþ1=2 Δ
ð1∶1Þ
× Esz þ Δ;
2 4


  ð1∶1Þ
ikt Δ
jþk 1 þ Ey ðm
1=2Þ
exp pffiffiffi
þ Δ : ð11Þ 2
2 4
ð1∶1Þ
× Ez ðm
1=2Þ ; ð16Þ
Equations (9)–(11) are derived in Appendix A. We
abbreviate the stationary solutions as

1082 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 

ð1∶1Þ
1 ikt Δ In the region of uniform permittivity of ε, the station-
Ez ðm þ 1=2Þ ¼
 i exp p ffiffiffi ary solutions are superpositions of plane waves.
2 2
2 cos kpt Δffiffi2 Since the FDTD method has second-order accuracy

 in an homogeneous region, the following equation
2 is valid for a plane wave:
× WΔ
ð1∶1Þ
Wεmþ1=2 Δ pffiffiffi


 s≈ ε sin θ; ð21Þ
2 cos kpt Δffiffi2 exp
ik ptffiffiΔ 
2 2 ð1∶1Þ where θ is the angle between the wave vector of the
þ ð1∶1Þ
H x ðmÞ plane wave and the ð2−1=2 ; 2−1=2 Þ direction.
Wεmþ1=2 Δ

 Now, we discuss the situation shown in Fig. 1. The
ikt Δ ð1∶1Þ permittivities are ε1 for region 1, y þ z < dΔ, and ε2
þ exp pffiffiffi Ey ðm
1=2Þ for region 2, y þ z > dΔ, where d is the offset ratio.
2
The interface is expressed as
ð1∶1Þ

Ez ðm
1=2Þ ; ð17Þ
y þ z ¼ dΔ: ð22Þ

The permittivities ε1 and ε2 and effective permittiv-




ities εi1 ð1∶1Þ and εi2 ð1∶1Þ are assigned to the electric
ωΔt Δt field component nodes as
W ¼ sin = : ð18Þ
2 2
ð1∶1Þ
εm ¼ ε1 ðm <
1Þ; ð23Þ
Equations (15)–(18) are expressed by the transfer
matrix Gð1∶1Þ as ð1∶1Þ ð1∶1Þ
ε
1=2 ¼ εi1 ; ð24Þ
2 3 2 3
ð1∶1Þ ð1∶1Þ
6 H x ðm þ 1Þ 7 6 H x ðmÞ 7
6 Eð1∶1Þ ðm þ 1=2Þ 7 ¼ Gð1∶1Þ ðεð1∶1Þ Þ6 Eð1∶1Þ ðm
1=2Þ 7: ð1∶1Þ ð1∶1Þ
4 y 5 mþ1=2 4 y 5 ε1=2 ¼ εi2 ; ð25Þ
ð1∶1Þ ð1∶1Þ
Ez ðm þ 1=2Þ Ez ðm
1=2Þ
ð1∶1Þ
ð19Þ εm ¼ ε2 ðm > 1Þ: ð26Þ

Hereafter in this subsection, we manipulate the In this subsection, the goal is to find the effective per-
expressions with second-order accuracy, since our mittivities εi1 ð1∶1Þ and εi2 ð1∶1Þ by which the reflection
targets are effective permittivities with second-order and transmission coefficients are equal to the exact
accuracy. We express second-order accuracy with re- ones with second-order accuracy.
spect to Δ by the symbol “≈.” We used commercial In region 1, the incident angle θi of the plane wave
computer algebra software packages to obtain series incident on the interface satisfies the following
expansions with respect to Δ. The elements of Gð1∶1Þ expression:
are described in Appendix B with second-order pffiffiffiffiffi
accuracy. kt ≈ k0 ε1 sin θi : ð27Þ
In Appendix B, we use the ratio s as
The following stationary solution F1;f ð1∶1Þ ðmÞ
s ¼ kt =k0 : ð20Þ describes the incident plane wave:

2 3
ð1∶1Þ
H x ðmÞ 7

6 ð1∶1Þ pffiffiffiffiffi k0 Δm
ð1∶1Þ
F1;f ðmÞ ¼ 6E 7 ≈ 1 − i ε1 cos θi pffiffiffi
4 y ðm − 1=2Þ 5 2
ð1∶1Þ
Ez ðm − 1=2Þ
2 3 ð28Þ
6 1 pffiffiffiffiffiffiffi 7
pffiffiffiffiffi
×6
4 −ðcos θ i þ sin θ i Þð1 þ i ε cos θi k0 Δ=23=2 Þ= 2ε1 7
p ffiffiffiffiffiffiffi 5:
pffiffiffiffiffi1
ðcos θi − sin θi Þð1 þ i ε1 cos θi k0 Δ=23=2 Þ= 2ε1

The following stationary solution F1;b ð1∶1Þ ðmÞ describes

the reflected plane wave:

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1083

 2 3
ð1∶1Þ
H ðmÞ

6 ð1∶1Þ x 7 k0 Δm
F1;b ðmÞ ¼ 6
ð1∶1Þ
E ðm
1=2Þ 7 ≈ 1 þ ipffiffiffiffi ε
ffi
cos θ p ffiffiffi
4 y 5 1 i
2
ð1∶1Þ
Ez ðm
1=2Þ
2 3 ð29Þ
6 1 pffiffiffiffiffiffiffi 7
pffiffiffiffiffi
×6 1 cos θi k0 Δ=2
4 ðcos θi
sin θi Þð1
i pεffiffiffiffi
3=2
Þ= 2ε1 7
pffiffiffiffiffiffiffi 5:
ffi

ðcos θi þ sin θi Þð1
i ε1 cos θi k0 Δ=23=2 Þ= 2ε1

For a specified value of m, the vectors F1;f ð1∶1Þ ðmÞ and

F1;b ð1∶1Þ ðmÞ are the eigenvectors of Gð1∶1Þ ðε1 Þ. The In region 2, the following stationary solution
transfer matrix Gð1∶1Þ ðε1 Þ has another eigenvector. F2;f ð1∶1Þ ðmÞ describes the transmitted plane wave:
However, we can neglect it, since its eigenvalue is 0.

2 3
ð1∶1Þ
H x ðmÞ 7

6 ð1∶1Þ pffiffiffiffiffi k0 Δm
ð1∶1Þ
F2;f ðmÞ ¼ 6E 7 ≈ 1
i ε2 cos θt pffiffiffi
4 y ðm
1=2Þ 5 2
ð1∶1Þ
Ez ðm
1=2Þ
2 3 ð30Þ
6 1 pffiffiffiffiffiffiffi 7
pffiffiffiffiffi
×6
4
ðcos θt þ sin θt Þð1 þ pi ε2 cos θt k0 Δ=23=2 Þ= 2ε2 7
p ffiffiffiffiffiffiffi 5;
ffiffiffiffiffi
ðcos θt
sin θt Þð1 þ i ε2 cos θt k0 Δ=23=2 Þ= 2ε2

Fig. 1. Schematic drawing of the analyzed y–z plane for the effective permittivities used for the dielectric interface orthogonal to the
ð2−1=2 ; 2−1=2 Þ direction.

1084 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 rffiffiffiffiffi
ε1 1
1þd
þ 1
d
sin θt ≈ sin θi : ð31Þ ¼
ε1 ε2
: ð41Þ
ε2 εhm 2
The exact reflection coefficient Rexact and the exact
Equations (38)–(41) give effective permittivities with
transmission coefficient T exact are
second-order accuracy for the dielectric interface
cos θi cos θ
orthogonal to the ð2−1=2 ; 2−1=2 Þ direction.
pffiffiffiffi
pffiffiffiffit
ε ε When d equals 0, simply ε1 for εi1 ð1∶1Þ and ε2 for
Rexact ¼ cos θ1 i cos θ2 t ; ð32Þ εi2 ð1∶1Þ are the effective permittivities with second-
pffiffiffiffi þ pffiffiffiffi
ε 1 ε 2 order accuracy. In other words, staircase approxima-
tion gives the second-order accuracy by adjusting the
θ
pffiffiffiffii
interface position in the grid as d ¼ 0.
2 cos
ε1
T exact ¼ cos θi cos θ
: ð33Þ B. Effective Permittivities for the Interface Orthogonal to
pffiffiffiffi þ pffiffiffiffit
ε 1 ε2 the ðð4=5Þ1=2 ; 5−1=2 Þ Direction
Symbols used in Subsection 2.B but not in 2.A are
The following stationary solution F1 ð1∶1Þ ðmÞ de-
followed by a superscriptð1∶2Þ . We assume that the
scribes the superposition of the incident plane wave
identical permittivity is assigned to a column of Ey
and the reflected plane wave with the exact reflection
nodes and a column of Ez nodes aligned in the
coefficient.
ð−5−1=2 ; ð4=5Þ1=2 Þ direction as
ð1∶1Þ ð1∶1Þ ð1∶1Þ
F1 ðmÞ ¼ F1;f ðm
dÞ þ Rexact F1;b ðm
dÞ: ð34Þ ð1∶2Þ
εjþ1=2;k ¼ εm ðm ¼ 2ðj þ 1=2Þ þ kÞ; ð42Þ
When the following equation is valid for any incident
angle θi , the exact reflection and transmission coeffi-
ð1∶2Þ
cients are achieved with second-order accuracy by εj;kþ1=2 ¼ εm ðm ¼ 2j þ ðk þ 1=2ÞÞ: ð43Þ
the effective permittivities εi1 ð1∶1Þ and εi2 ð1∶1Þ :
We assume that the incident plane wave has the
ð1∶1Þ
T exact F2;f ð2
dÞ wave-vector component kt in the ð−5−1=2 ; ð4=5Þ1=2 Þ
direction. The stationary solutions also have the
ð1∶1Þ ð1∶1Þ ð1∶1Þ
≈ Gð1∶1Þ ðε2 ÞGð1∶1Þ ðεi2 ÞGð1∶1Þ ðεi1 ÞF1 ð
1Þ: wave-vector component kt in that direction.
ð35Þ Thus, the stationary solutions satisfy the following
equations:
Equation (35) is easily calculated by commercial

algebraic software. The necessary and sufficient con- pffiffiffi
j þ 2k
H sx ðjΔ; kΔÞ ¼ exp
i 5kt Δ
ditions for Eq. (35) are 5


4j þ 2k 2j þ k
ð1∶1Þ
εi1 þ εi2
ð1∶1Þ
¼ ð1 þ dÞε1 þ ð1
dÞε2 ; ð36Þ × H sx Δ; Δ ; ð44Þ
5 5

 

1 1 1 1 1 pffiffiffi
j þ 2k þ 1
þ ¼ ð1 þ dÞ þ ð1
dÞ : ð37Þ Esy jΔ; k þ Δ ¼ exp
i 5kt Δ
εi1
ð1∶1Þ ð1∶1Þ
εi2 ε1 ε2 2 5


4j þ 2k þ 1
The coupled Eqs. (36) and (37) are solved for εi1 ð1∶1Þ × Esy Δ;
5
and εi2 ð1∶1Þ as 
2j þ k þ 1=2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Δ ; ð45Þ
5
ð1∶1Þ ðε2
ε1 Þ ð1
d2 Þεam εhm
εi1 ¼ εam
pffiffiffiffiffiffiffiffiffi ; ð38Þ
2 ε1 ε2


 

1 pffiffiffi
j þ 2k
1=2
Esz jþ Δ; kΔ ¼ exp
i 5kt Δ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 5


ð1∶1Þ ðε2
ε1 Þ ð1
d2 Þεam εhm 4j þ 2k þ 2
εi2 ¼ εam þ pffiffiffiffiffiffiffiffiffi ; ð39Þ × Esz Δ;
2 ε1 ε2 5

2j þ k þ 1
Δ : ð46Þ
5
ð1 þ dÞε1 þ ð1
dÞε2
εam ¼ ; ð40Þ
2 We abbreviate the stationary solutions as

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1085



4j þ 2k 2j þ k ð1∶2Þ Equations (50)–(52) are expressed by the transfer
H sx Δ; Δ ¼ H x ð2j þ kÞ; ð47Þ matrix Gð1∶2Þ as
5 5
2 ð1∶2Þ 3

 H x ðm þ 1Þ
4j þ 2k þ 1 2j þ k þ 1=2
Esy Δ; Δ 6 ð1∶2Þ 7
5 5 6 Ez ðm þ 1Þ 7  
6 7 ð1∶2Þ ð1∶2Þ ð1∶2Þ

 6 ð1∶2Þ  7 ¼ Gð1∶2Þ εm
1=2 ; εm ; εmþ1=2
ð1∶2Þ 1 6 Ey m þ 12 7
¼ Ey 2j þ k þ ; ð48Þ 4 5
2 ð1∶2Þ
Ez ðmÞ

 2 3
4j þ 2k þ 2 2j þ k þ 1 ð1∶2Þ Hx
ð1∶2Þ
ðmÞ
Esz Δ; Δ ¼ Ez ð2j þ k þ 1Þ:
5 5 6 7
6 Ezð1∶2Þ ðmÞ 7
6 7
ð49Þ × 6 ð1∶2Þ  7: ð53Þ
ð1∶2Þ
From Eqs. (1)–(6) and (42)–(49), H x ðm þ 1Þ, 6 Ey m
1 7
ð1∶2Þ ð1∶2Þ
4 2 5
Ez ðm þ 1Þ, and Ey ðm þ 1=2Þ are expressed by ð1∶2Þ
ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ðm
1Þ
H x ðmÞ, Ez ðmÞ, Ey ðm − 1=2Þ, and Ez ðm − Ez
1Þ as Hereafter in this subsection, we manipulate the ex-


ð1∶2Þ
3ikt Δ ð1∶2Þ pressions with second-order accuracy. The elements
H x ðm þ 1Þ ¼ exp pffiffiffi H x ðmÞ of Gð1∶2Þ are described in Appendix C with second-
5

 order accuracy. In the region of uniform permittivity
ð1∶2Þ
ikt Δ ð1∶2Þ ε, the ratio sð¼ kt =ko Þ is expressed as

iWεm Δ exp pffiffiffi Ez ðmÞ
2 5 pffiffiffi

 s≈ ε sin θ; ð54Þ
ð1∶2Þ
2ikt Δ

iWεm
1=2 Δ exp pffiffiffi
5 where θ is the angle between the wave vector of the
ð1∶2Þ
× Ey ðm
1=2Þ; ð50Þ plane wave and the ðð4=5Þ1=2 ; 5−1=2 Þ direction.
Now, we discuss the situation shown in Fig. 2. The


permittivities are ε1 for region 1, 2y þ z < ð1=2 þ dÞΔ
ð1∶2Þ
ikt Δ and ε2 for region 2, 2y þ z > ð1=2 þ dÞΔ, where d is
Ez ðm þ 1Þ ¼
i exp pffiffiffi WΔ
2 5 the offset ratio. The interface between region 1
pffiffiffi  and region 2 is orthogonal to the ðð4=5Þ1=2 ; 5−1=2 Þ
expð
5ikt ΔÞ
1 ð1∶2Þ
þ ð1∶2Þ
H x ðmÞ direction. The interface is expressed as
Wεmþ1=2 Δ


ð1∶2Þ
 1
εm
3ikt Δ ð1∶2Þ 2y þ z ¼ þ d Δ: ð55Þ

ð1∶2Þ exp pffiffiffi Ez ðmÞ 2
εmþ1=2 5


 ð1∶2Þ
The permittivities ε1 and ε2 and effective permittiv-
ikt Δ εm
1=2 ities εi1 ð1∶2Þ , εi2 ð1∶2Þ , and εi3 ð1∶2Þ are assigned to the

exp pffiffiffi þ ð1∶2Þ electric field component nodes as
2 5 εmþ1=2




9ikt Δ ð1∶2Þ ð1∶2Þ 1
× exp pffiffiffi Ey ðm
1=2Þ εm ¼ ε1 m≤
; ð56Þ
2 5 2



ik Δ ð1∶2Þ
þ exp pffiffiffit Ez ðm
1Þ; ð51Þ ð1∶2Þ ð1∶2Þ
5 ε0 ¼ εi1 ; ð57Þ
ð1∶2Þ ð1∶2Þ
ε1=2 ¼ εi2 ; ð58Þ


 ð1∶2Þ ð1∶2Þ

4ik
ε1 ¼ εi3 ; ð59Þ
exp pffiffit Δ
exp ik Δ
ptffiffi
ð1∶2Þ 5 5
Ey ðm þ 1=2Þ ¼

ð1∶2Þ 3
iWεmþ1=2 Δ ð1∶2Þ
εm ¼ ε2 m≥ : ð60Þ
ð1∶2Þ
 2
ð1∶2Þ εm
3ikt Δ
× Hx ðmÞ
ð1∶2Þ
exp pffiffiffi
εmþ1=2 2 5 In this subsection, the goal is to find the effective per-
mittivities εi1 ð1∶2Þ , εi2 ð1∶2Þ , and εi3 ð1∶2Þ , by which the
ð1∶2Þ

ð1∶2Þ
εm
1=2
3ikt Δ reflection and transmission coefficients are equal
× Ez ðmÞ
ð1∶2Þ exp pffiffiffi to the exact ones with second-order accuracy.
εmþ1=2 5 In region 1, the incident angle θi of the plane wave
ð1∶2Þ incident on the interface satisfies the following
× Ey ðm
1=2Þ: ð52Þ expression:

1086 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 pffiffiffiffiffi 2 3
kt ≈ k0 ε1 sin θi : ð61Þ Hx
ð1∶2Þ
ðmÞ
6 ð1∶2Þ 7
6 Ez ðmÞ 7
F1;pp ðmÞ ¼ 6 7
ð1∶2Þ
6 ð1∶2Þ 7
4 Ey ðm
1=2Þ 5
The following stationary solution F1;f ð1∶2Þ ðmÞ
ð1∶2Þ
describes the incident plane wave: Ez ðm
1Þ

pffiffiffi
pffiffiffiffiffi 
3þ 5 m ε1 sin θi k0 Δm
≈ − 1−i p ffiffiffi
2 5
2 ð1∶2Þ
3 2 iε1pkffiffi0 Δ 3
Hx ðmÞ
6 7 6 5
7
6 ð1∶2Þ
Ez ðmÞ 7 6 1 7
ð1∶2Þ 6 7 6   7
F1;f ðmÞ ¼6 7 6 pffiffi pffiffiffiffi 7
6 Eð1∶2Þ ðm
1=2Þ 7 6 pffiffi 5− 5 ε1 sin θi k0 Δ 7
4 y 5 6
×6 −1þ 5
þi 7: ð64Þ
7
Ez
ð1∶2Þ
ðm
1Þ 6 2
 
20
7
6 pffiffi pffiffiffiffi 7

pffiffiffiffiffi  6 pffiffi −5þ3 5 ε1 sin θi k0 Δ 7
ε1 cos θi k0 Δm 4 − 3− 5 − i 5
≈ 1−i pffiffiffi 2 10
5
2 1 3
6 2 cos θi −sin θi
pffiffiffiffiffi 7 The following stationary solution F1;pn ð1∶2Þ ðmÞ
6 7 describes the parasitic solution that exponentially
6

5ε1
 7
6 pffiffiffiffi 7 decays in the ðð4=5Þ1=2 ; 5−1=2 Þ direction:
×6
6− 2 sinpθiffiffiffiffiffi
þcos θi
1þi
ε1 cos θi k0 Δ 7: ð62Þ
pffiffi 7
6 5ε1 2 5 7 2 3
6
 7 ð1∶2Þ
4 pffiffiffiffi 5 Hx ðmÞ
ε cos θ k Δ
2 cospθiffiffiffiffiffi

sin θi
1 þ i 1 pffiffi5 i 0 6 7
6 ð1∶2Þ
Ez ðmÞ 7
5ε1
ð1∶2Þ 6 7
F1;pn ðmÞ ¼6 7
6 Eð1∶2Þ ðm
1=2Þ 7
4 y 5
ð1∶2Þ
Ez ðm
1Þ
ð1∶2Þ
The following stationary solution F1;b ðmÞ
pffiffiffi
pffiffiffiffiffi 
describes the reflected plane wave: −3 þ 5 m ε1 sin θi k0 Δm
≈ 1−i pffiffiffi
2 5
2 3
iε1pkffiffi0 Δ
6 − 7
2 ð1∶2Þ
3 6 5 7
Hx ðmÞ 6 1 7
6 7 6   7
6 ð1∶2Þ 7 6 pffiffi pffiffiffiffi 7
ð1∶2Þ 6 Ez ðmÞ 7 ×6 pffiffi
6 −1− 5 5þ 5 ε1 sin θi k0 Δ 7:
7 ð65Þ
F1;b ðmÞ ¼6 7 6 2 −i 7
6 Eð1∶2Þ ðm
1=2Þ 7 6 20 7
4 y 5 6   7
pffiffi pffiffiffiffi
ð1∶2Þ 4 pffiffi 5þ3 5 ε1 sin θi k0 Δ 5
Ez ðm
1Þ − 2 −i
3þ 5

pffiffiffiffiffi  10
ε1 cos θi k0 Δm
≈ 1þi pffiffiffi
5 For a specified value of m, the vectors F1;f ð1∶2Þ ðmÞ,
2 1 3 F1;b ð1∶2Þ ðmÞ, F1;pp ð1∶2Þ ðmÞ, and F1;pn ð1∶2Þ ðmÞ are the ei-
6 7 genvectors of the transfer matrix Gð1∶2Þ ðε1 ; ε1 ; ε1 Þ.
6
2 cospθiffiffiffiffiffi
þsin θi
7
6 5ε1
7 In region 2, two plane waves, F2;f ð1∶2Þ ðmÞ and
6
pffiffiffiffi
 7 F2;b ð1∶2Þ ðmÞ, and two parasitic solutions, F2;pp ð1∶2Þ
×66
2 sin θ
pffiffiffiffiffi
i þcos θ i
1
i
ε1 cos θi k0 Δ 7:
pffiffi 7 ð63Þ ðmÞ and F2;pn ð1∶2Þ ðmÞ, which have the wave-vector
6 5ε1 2 5 7 component kt in the ð−5−1=2 ; ð4=5Þ1=2 Þ direction, exist.
6
 7
4 pffiffiffiffi 5 They are expressed by means of the angle θt as

2 cos θi
sin θi ε1 cos θi k0 Δ
pffiffiffiffiffi 1
i pffiffi
5ε1 5
rffiffiffiffiffi
ε1
sin θt ≈ sin θi : ð66Þ
ε2

Two parasitic solutions with the wave-vector compo- F2;f ð1∶2Þ ðmÞ and F2;b ð1∶2Þ ðmÞ are plane waves traveling
nent kt in the ð−5−1=2 ; ð4=5Þ1=2 Þ direction exist in re- in the direction that makes angles θt and π − θt against
gion 1. The following stationary solution F1;pp ð1∶2Þ ðmÞ the ðð4=5Þ1=2 ; 5−1=2 Þ direction, respectively. F2;pp ð1∶2Þ
describes the parasitic solution that exponentially ðmÞ and F2;pn ð1∶2Þ ðmÞ are the parasitic solutions that
decays in the ð−ð4=5Þ1=2 ; −5−1=2 Þ direction: exponentially decay in the ð−ð4=5Þ1=2 ; −5−1=2 Þ and

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1087

Fig. 2. Schematic drawing of the analyzed y–z plane for the effective permittivities used for the dielectric interface orthogonal to the
ðð4=5Þ1=2 ; 5−1=2 Þ direction.

ðð4=5Þ1=2 ; 5−1=2 Þ directions, respectively. Detailed de-


scriptions of F2;f ð1∶2Þ ðmÞ, F2;b ð1∶2Þ ðmÞ, F2;pp ð1∶2Þ ðmÞ, 1
and F2;pn ð1∶2Þ ðmÞ are obtained by replacing ε1 and θi m1 ¼ m
þd ; ð70Þ
2
with ε2 and θt in those of F1;f ð1∶2Þ ðmÞ, F1;b ð1∶2Þ ðmÞ,
F1;pp ð1∶2Þ ðmÞ, and F1;pn ð1∶2Þ ðmÞ, respectively. 2 3
a2;f
The stationary solution in region 2, F2 ð1∶2Þ ðmÞ, is 6 a2;b 7
6 7 ð1∶2Þ
generally described as


 4 a2;pp 5 ¼ QðmÞF2 ðmÞ: ð71Þ
ð1∶2Þ ð1∶2Þ 1 a2;pn
F2 ðmÞ ¼ a2;f F2;f m
þd
2



ð1∶2Þ 1 The elements of QðmÞ are described in Appendix D.
þ a2;b F2;b m
þd Here, we assume that the plane wave F1;f ð1∶2Þ ðm −
2


 ð1=2 þ dÞÞ is incident on the interface from region 1.
ð1∶2Þ 1 Then, the stationary solution F1;i ð1∶2Þ ðmÞ in region 1
þ a2;pp F2;pp m
þd
2 is the superposition of plane waves and a parasitic


 solution and is described as follows:
ð1∶2Þ 1
þ a2;pn F2;pn m
þd ; ð67Þ


2 ð1∶2Þ ð1∶2Þ 1
where a2;f , a2;b , a2;pp , and a2;pn are the amplitudes of F1;i ðmÞ ¼ F1;f m
þd
2
the plane waves and parasitic solutions. The ampli-


tudes a2;f , a2;b , a2;pp , and a2;pn are obtained by multi- ð1∶2Þ 1
þ RF1;b m
þd
plying the following matrix QðmÞ by F2 ð1∶2Þ ðmÞ as 2



ð1∶2Þ 1
QðmÞ ¼ PðmÞ
1 ; ð68Þ þ a1;pp F1;pp m
þd ; ð72Þ
2

PðmÞ where R is the reflection coefficient at the interface

h i and a1;pp is the amplitude of the excited parasitic so-
ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
¼ F2;f ðm1 Þ F2;b ðm1 Þ F2;pp ðm1 Þ F2;pn ðm1 Þ ; lution F1;pp ð1∶2Þ ðm − ð1=2 þ dÞÞ. The parasitic solution
F1;pn ð1∶2Þ ðm − ð1=2 þ dÞÞ is not excited, since its abso-
ð69Þ lute value grows exponentially in the ð−ð4=5Þ1=2 ;

1088 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

−5−1=2 Þ direction. In this situation, the stationary so- ð1∶2Þ
½Qð2ÞLF1;i ð0Þ2 ¼ 0; ð76Þ
lution F2;t ð1∶2Þ ðmÞ in region 2, which is compatible
with F1;i ð1∶2Þ ðmÞ, is the superposition of the trans-
mitted plane wave and the parasitic solution that de- ð1∶2Þ
cays in the ðð4=5Þ1=2 ; 5−1=2 Þ direction as ½Qð2ÞLF1;i ð0Þ3 ¼ 0: ð77Þ


 The equivalent expression with explicit descriptions
ð1∶2Þ ð1∶2Þ 1 of R and a1;pp for Eqs. (76) and (77) is described by
F2;t ðmÞ ¼ TF2;f m
þd Eq. (E1) in Appendix E. After obtaining R and
2


 a1;pp by solving Eqs. (76) and (77) simultaneously,
ð1∶2Þ 1 we get T as
þ a2;pn F2;pn m
þd ; ð73Þ
2
ð1∶2Þ ð1∶2Þ
T ¼ ½Qð2ÞF2;t ð2Þ1 ¼ ½Qð2ÞLF1;i ð0Þ1
where T is the transmission coefficient at the inter-  

pffiffiffiffiffi
face and a2;pn is the amplitude of the parasitic solu- ε2 cos θt k0 Δ 32
d 
tion. The parasitic solution F2;pp ð1∶2Þ ðm − ð1=2 þ dÞÞ ≈ 1þi pffiffiffi ð1∶2Þ
e1 LF1;i ð0Þ: ð78Þ
is not excited, since its absolute value grows expo- 5
nentially in the ðð4=5Þ1=2 ; 5−1=2 Þ direction.
F1;i ð1∶2Þ ð0Þ and F2;t ð1∶2Þ ð2Þ are related by the trans- The reflection coefficient R and transmission coef-
fer matrix Gð1∶2Þ as ficient T are expanded in the series of Δ as

ð1∶2Þ ð1∶2Þ
F2;t ð2Þ ¼ LF1;i ð0Þ; ð74Þ R ¼ Rexact þ R1 Δ þ OðΔ2 Þ; ð79Þ

T ¼ T exact þ T 1 Δ þ OðΔ2 Þ: ð80Þ

ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
L¼ Gð1∶2Þ ðεi2 ; εi3 ; ε2 ÞGð1∶2Þ ðε1 ; εi1 ; εi2 Þ:
The conditions for effective permittivities with sec-
ð75Þ ond-order accuracy are derived from the requirement
F2;t ð1∶2Þ ðmÞ does not contain F2;b ð1∶2Þ ðm − ð1=2 þ dÞÞ that T 1 and R1 be zero regardless of the incident an-
or F2;pp ð1∶2Þ ðm − ð1=2 þ dÞÞ. This determines R and gle θi . They are expressed by the following three
a1;pp . The condition is described by the matrix Qð2Þ as equations:

 pffiffiffi
5
3 5 ðε1
ε2 Þd  pffiffiffi ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
ð1 þ 5Þðεi1 εi2 þ εi2 εi3 þ εi3 εi1 Þ
4
ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
pffiffiffi
þ 2½ε1 ðεi2 þ εi3 Þ þ ε2 ðεi1 þ εi2 Þ þ ð
1 þ 5Þε1 ε2
    
pffiffiffi pffiffiffi ð1∶2Þ
pffiffiffi pffiffiffi pffiffiffi pffiffiffi
2 ð1∶2Þ ð1∶2Þ
þ
ε2 ð15 þ 3 5Þε1
2 5ε2 εi1 þ 2 5ðε1 þ ε2 Þ εi2 þ ε1 2 5ε1
ð15 þ 3 5Þε2 εi3 =ð5 þ 5Þ
pffiffiffi
ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ 6 5 ε1 ε2 ðε1 þ ε2 Þ
þ ðε1 þ ε2 Þðεi1 εi2 þ εi2 εi3 þ εi3 εi1 Þ
pffiffiffi ¼ 0;
5þ 5
ð81Þ

ð1∶2Þ ð1∶2Þ ð1∶2Þ

pffiffiffi pffiffiffi ð1∶2Þ ð1∶2Þ
100εi1 εi2 εi3 þ 2½ð
10 þ 3 5Þε1 þ ð
35 þ 8 5Þε2 εi1 εi2
pffiffiffi pffiffiffi ð1∶2Þ ð1∶2Þ
pffiffiffi ð1∶2Þ ð1∶2Þ
þ 2½ð
35 þ 8 5Þε1 þ ð
10 þ 3 5Þε2 εi2 εi3 þ 2ð
10 þ 3 5Þðε1 þ ε2 Þεi3 εi1
 pffiffiffi

5 þ 13 5 pffiffiffi pffiffiffi
ð1∶2Þ ð1∶2Þ
þ fε2 ½ð4 þ 35 5Þε1
41ε2 εi1 þ ½
41ðε21 þ ε22 Þ þ ð
107 þ 17 5Þε1 ε2 εi2
41
pffiffiffi ð1∶2Þ
pffiffiffi
þ ε1 ½
41ε1 þ ð4 þ 35 5Þε2 εi3 g þ ð
25 þ 19 5Þðε1 þ ε2 Þε1 ε2 ¼ 0; ð82Þ

ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ

pffiffiffi ð1∶2Þ ð1∶2Þ
2εi1 εi2
þ 2εi2 εi3
8εi3 εi1 þ ð4
2 5Þðε2 εi1 þ ε1 εi3 Þ
pffiffiffi ð1∶2Þ
pffiffiffi
þ ð
1 þ 5Þðε1 þ ε2 Þεi2 þ 2ð
1 þ 5Þε1 ε2 ¼ 0: ð83Þ

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1089

The coupled Eqs. (81)–(83) are solved for εi1 ð1∶2Þ,
εi2 ð1∶2Þ , and εi3 ð1∶2Þ as
pffiffiffi pffiffiffi pffiffiffi
ð1∶2Þ ð
1 þ 5Þε1 ½
20d2 ðε1
ε2 Þ
10dðε1
ε2 Þ þ ð11 þ 3 5Þε1 þ ð19 þ 7 5Þε2 
εi1 ¼ pffiffiffi pffiffiffi pffiffiffi ; ð84Þ
2½20d2 ðε1
ε2 Þ
10 5dðε1
ε2 Þ þ ð
1 þ 3 5Þε1 þ ð11 þ 7 5Þε2 

pffiffiffi pffiffiffi
ð1∶2Þ
5ð5 þ 5Þε1 ε2 ½4dðε1
ε2 Þ þ ð1 þ 5Þðε1 þ ε2 Þ
εi2 ¼ pffiffiffi pffiffiffi pffiffiffi ; ð85Þ
2½20d2 ðε1
ε2 Þ2 þ 10 5dðε1 2
ε2 2 Þ
ð11 þ 7 5Þðε1 2 þ ε2 2 Þ
ð28 þ 16 5Þε1 ε2 

pffiffiffi pffiffiffi pffiffiffi

ð1∶2Þ ð
1 þ 5Þε2 ½20d2 ðε1
ε2 Þ
10dðε1
ε2 Þ þ ð19 þ 7 5Þε1 þ ð11 þ 3 5Þε2 
εi3 ¼ p ffiffiffi p ffiffiffi p ffiffiffi : ð86Þ
2½20d2 ð
ε1 þ ε2 Þ
10 5dðε1
ε2 Þ þ ð11 þ 7 5Þε1 þ ð
1 þ 3 5Þε2 

Equations (84)–(86) give effective permittivities with 2π

λwg ¼ ; ð87Þ
second-order accuracy for the dielectric interface β
orthogonal to the ðð4=5Þ1=2 ; 5−1=2 Þ direction.
The set of Eqs. (84)–(86) is left invariant when ε1 ,
ε2 , d, εi1 ð1∶2Þ , and εi3 ð1∶2Þ are replaced with ε2 , ε1 , −d, where β is the propagation constant of the mode.
εi3 ð1∶2Þ , and εi1 ð1∶2Þ , respectively. Thus, the reflection With this selection of the length of the diagonal line,
and transmission coefficients for a plane wave travel- the initial electromagnetic fields were periodic at the
ing from region 2 to region 1 are also equal to the boundary of the domain. Thus, the periodic boundary
exact ones with second-order accuracy by Eqs. (84)– condition was used for the simulation. For the effec-
(86). tive permittivities εi1 ð1∶2Þ, εi2 ð1∶2Þ , and εi3 ð1∶2Þ , a rec-
tangular domain was prepared. The sides along
the z direction were twice as long as those along
3. Numerical Examples and Discussion the y direction. The waveguide was set along the di-
agonal line. The core–cladding interfaces were ortho-
A. Propagation along a Waveguide gonal to the ðð4=5Þ1=2 ; 5−1=2 Þ direction. The length of
The propagations of the guided modes along a wave- the diagonal line was adjusted so that it was a posi-
guide with a step-index profile were simulated in the tive integer multiple of five times the wavelength in
2D TEx polarization case. The phase velocities were the waveguide λwg . This makes it possible to use the
compared to the analytical solutions. The refractive periodic boundary condition. The effective permittiv-
indices of the core ncore and cladding ncladding were 3.2 ities were assigned to the nodes in the vicinity of the
and 1.0, respectively. The core width was 1:5 μm. The core–cladding interfaces.
wavelength in a vacuum λ0 was 1:55 μm. The wave- Figures 3 and 4 show the phase velocity error in
guide supports six guided TM modes. The TM modes the TM2 and TM5 modes as a function of the number
of a slab waveguide were analyzed in the 2D TE po- of cells per wavelength in the core region
larization case of the FDTD method [2,14]. The Cour- (ðλ0 =ncore Þ=Δ. In Figs. 3 and 4, the phase velocity
ant–Friedrichs–Lewy (CFL) number was 0.6. For the error decreased as Δ2. These results prove that
initial condition, the analytically obtained values of the present method gives second-order accuracy.
the electromagnetic components were assigned to
the nodes. B. Comparison with Other Methods
The waveguide configurations in the simulated do- The present method (εi1 ð1∶1Þ and εi2 ð1∶1Þ ) is compared
mains are schematically illustrated in the insets of with the method by Kaneda et al. [8] and with the
Figs. 3 and 4. For the effective permittivities VP-EP scheme [11]. We simulated the propagation
εi1 ð1∶1Þ and εi2 ð1∶1Þ , a square domain whose sides of a guided mode along a waveguide set along the di-
are along the y and z directions was prepared. The agonal line of a square domain. The refractive indices
waveguide was set along the diagonal line. The of the core ncore and cladding ncladding were 5.0 and
core–cladding interfaces were orthogonal to the 1.0, respectively. The core width was 0:5 μm. The
ð2−1=2 ; 2−1=2 Þ direction. The length of the diagonal line wavelength in a vacuum λ0 was 2:42 μm. The present
was adjusted so that it was a positive integer multi- method and the other methods were applied to the
ple of double the wavelength in the waveguide λwg , nodes in the vicinity of the core–cladding interfaces.
which is expressed as The CFL number was 0.6. Figure 5 shows the phase

1090 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 Fig. 5. Phase velocity errors in the simulations of wave propaga-
tion in the TM1 mode of a 2D step-profile waveguide. The method
by Kaneda et al. [8], the VP-EP scheme [11], and the present meth-
Fig. 3. Phase velocity errors in the simulations of wave propaga- od were applied to the nodes in the vicinity of the core–cladding
tion in the TM2 and TM5 modes of a 2D step-profile waveguide. interfaces. The offset ratio at the interfaces is 0.4.
The effective permittivities εi1 ð1∶1Þ and εi2 ð1∶1Þ were assigned to
the nodes in the vicinity of the core–cladding interfaces. The offset
ratio of one core–cladding interface is 0. The offset ratio at the
other interface is described in the legend.
velocity error in the TM1 mode as a function of the
number of cells per wavelength in the core region
ðλ0 =ncore Þ=Δ. The phase velocity errors for the meth-
od by Kaneda et al. and the VP-EP scheme decreased
as Δ. In contrast, the error for the present method
decreased as Δ2.
Numerical examples demonstrate that the present
method gives second-order accuracy for inclined di-
electric interfaces. The field update algorithm of
the present method is the basic Yee algorithm on
the Yee lattice [1,2] expressed by Eqs. (1)–(3). In pre-
vious work, subpixel smoothing [15] and a stable
FDTD algorithm for nondiagonal anisotropic dielec-
trics [16] have achieved second-order accuracy for
inclined dielectric interfaces on the Yee lattice. How-
ever, in subpixel smoothing [15], two components of
the electric displacement at the electric field compo-
nent node in the vicinity of inclined dielectric inter-
faces are required for updating the electromagnetic
field. The electric-displacement component perpen-
dicular to the electric field component is interpolated
from that at the four adjacent electric field compo-
nent nodes. Thus, the algorithm for subpixel smooth-
ing is more complex than the one for the present
method. As for the stable FDTD algorithm [16], both
the stable FDTD algorithm and Richardson extrapo-
lation from two simulations are necessary for achiev-
ing second-order accuracy for inclined dielectric
Fig. 4. Phase velocity errors in the simulations of wave propaga-
interfaces. First-order accuracy in simulation results
tion in the TM2 and TM5 modes of a 2D step-profile waveguide.
The effective permittivities εi1 ð1∶2Þ , εi2 ð1∶2Þ , and εi3 ð1∶2Þ were as-
by the stable FDTD algorithm alone is shown in
signed to the nodes in the vicinity of the core–cladding interfaces. Fig. 6 in [16]. The stable FDTD algorithm gives
The offset ratio of one core–cladding interface is 0. The offset ratio second-order accuracy for continuously varying
at the other interface is described in the legend. dielectrics [16].

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1091

C. Reflection at a Waveguide Facet
The slab waveguide had a step-index profile and core
width of 0:2 μm. The refractive indices of the core
ncore and cladding ncladding were 3.6 and 3.24, respec-
tively. The wavelength in a vacuum λ0 was 0:86 μm.
The core–cladding interfaces were parallel to the y
axis of the Yee lattice. We set the interfaces on Ey
nodes and assigned the arithmetic mean of core
and cladding permittivity to the nodes [12]. The facet
was inclined to the z axis with the angle of
arctanð1=2Þ. We assigned the effective permittivities
εi1 ð1∶2Þ , εi2 ð1∶2Þ , and εi3 ð1∶2Þ to the nodes in the vicinity
of the facet. The whole domain was surrounded by a
16-cell PML absorber. The CFL number was 0.62.
The fundamental mode of the waveguide was inci-
dent from the waveguide to the facet, and the reflec-
tion of the mode at the facet was simulated.
Simulations with staircasing for the facet were also
executed for reference. The simulated domain is
schematically illustrated in the inset of Fig. 6.
Figure 6 shows the simulated reflectivity r as a
function of the number of cells per wavelength in
the core region ðλ0 =ncore Þ=Δ. We define the difference
in the simulated reflectivities Δr;2=3 as Fig. 7. Difference in the simulated reflectivity at a waveguide fa-
cet Δr;2=3 as a function of the number of cells per wavelength in the


2Δ core region.
Δr;2=3 ðΔÞ ¼ rðΔÞ − r : ð88Þ
3
Figure 7 shows Δr;2=3 as a function of ðλ0 =ncore Þ=Δ.
When the simulation technique is second-order The reflectivity difference Δr;2=3 decreased as Δ2 in
accurate, the simulations with the present method.

Δr;2=3 ðΔÞ ¼ OðΔ2 Þ: ð89Þ 4. Conclusion

Effective permittivities with exact second-order ac-
curacy were analytically derived for the inclined di-
electric interface for the FDTD method in the case of
2D TE polarization. The tangent of the inclined angle
of the interface is 1 or 1=2. With the derived effective
permittivities, reflection and transmission can be si-
mulated with second-order accuracy with respect to
cell size. The accuracy was demonstrated by numer-
ical examples.
It is expected that effective permittivities with
exact second-order accuracy at dielectric inter-
faces whose inclined angle has the tangent of
1=3; 1=4; … can be derived similarly. However, their
expressions may be complicated. Comparing
Eqs. (38)–(41) to Eqs. (84)–(86), we suppose that it
would be difficult to find simple formulas for the ef-
fective permittivities with exact second-order accu-
racy for general inclined angles.

Appendix A: Derivation of Equations (9)–(11)

We define nt ð1∶1Þ as the unit vector tangential to the
interface as
 pffiffiffi 
ð1∶1Þ −1=pffiffiffi2
nt ¼ : ðA1Þ
1= 2
Fig. 6. Simulated reflectivity at a waveguide facet as a function of
the number of cells per wavelength in the core region. Here, d is Since the stationary solutions have the wave-vector
ð1∶1Þ
the offset ratio at the facet. component kt in the direction in which nt points,

1092 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 ð1∶1Þ Appendix B: Elements of G ð1∶1Þ ðεmþ1=2 ð1∶1Þ Þ with
H sx ðr þ ct nt Þ ¼ expð−ikt ct ÞH sx ðrÞ; ðA2Þ
Second-Order Accuracy
ð1∶1Þ
Esy ðr þ ct nt Þ ¼ expð−ikt ct ÞEsy ðrÞ; ðA3Þ
2 3
ð1∶1Þ ð1∶1Þ ð1∶1Þ
ð1∶1Þ 6 g1;1 g1;2 g1;3 7
Esz ðr þ ct nt Þ ¼ expð−ikt ct ÞEsz ðrÞ; ðA4Þ ð1∶1Þ 6 ð1∶1Þ ð1∶1Þ ð1∶1Þ 7
Gð1∶1Þ ðεmþ1=2 Þ ¼ 6 g2;1 g2;2 g2;3 7; ðB1Þ
4 ð1∶1Þ ð1∶1Þ ð1∶1Þ
5
where r is a position vector and ct is a coefficient g3;1 g3;2 g3;3
ð1∶1Þ
of nt .
ð1∶1Þ
We also define nn as the unit vector normal to
the interface as
ð1∶1Þ

pffiffiffi  g1;1 ≈ 1; ðB2Þ
ð1∶1Þ 1=p2
ffiffiffi :
nn ¼ ðA5Þ
1= 2 i ð1∶1Þ
ð1∶1Þ
g1;2 ≈ εmþ1=2 k0 Δ; ðB3Þ
ð1∶1Þ ð1∶1Þ 2
Position ðy; zÞ is expressed by nn and nt as
 
y ð1∶1Þ ð1∶1Þ
¼ cn nn þ ct nt ; ðA6Þ ð1∶1Þ i ð1∶1Þ
z g1;3 ≈ − εmþ1=2 k0 Δ; ðB4Þ
2
pffiffiffi
cn ¼ ðy þ zÞ= 2; ðA7Þ

ð1∶1Þ s ik Δ s2
g2;1 ≈ − pffiffiffi ð1∶1Þ þ 0 1 − ð1∶1Þ ; ðB5Þ
2εmþ1=2 2 2εmþ1=2
pffiffiffi
ct ¼ ð−y þ zÞ= 2: ðA8Þ
ð1∶1Þ 1
g2;2 ≈ ; ðB6Þ
Thus, the stationary solutions satisfy the following 2
equations:


ð1∶1Þ 1 isk0 Δ
H sx ðy; zÞ ¼
ð1∶1Þ
H sx ðcn nn þ
ð1∶1Þ
ct nt Þ g2;3 ≈− 1 − pffiffiffi ; ðB7Þ
2 2
ð1∶1Þ
¼ expð−ikt ct ÞH sx ðcn nn Þ
 pffiffiffi

¼ exp −ikt ð−y þ zÞ= 2 ð1∶1Þ s ik Δ s2
g3;1 ≈ − pffiffiffi ð1∶1Þ þ 0 ð1∶1Þ
− 1 ; ðB8Þ
× H sx ððy þ zÞ=2; ðy þ zÞ=2Þ; ðA9Þ 2εmþ1=2 2 2εmþ1=2


ð1∶1Þ 1 isk0 Δ
Esy ðy; zÞ ¼
ð1∶1Þ
Esy ðcn nn þ
ð1∶1Þ
ct nt Þ g3;2 ≈− 1 þ pffiffiffi ; ðB9Þ
2 2
ð1∶1Þ
¼ expð−ikt ct ÞEsy ðcn nn Þ
 pffiffiffi
ð1∶1Þ 1
¼ exp −ikt ð−y þ zÞ= 2 g3;3 ≈ : ðB10Þ
2
× Esy ððy þ zÞ=2; ðy þ zÞ=2Þ; ðA10Þ
Appendix C: Elements of
G ð1∶2Þ ðεm−1=2 ð1∶2Þ ; εm ð1∶2Þ ; εmþ1=2 ð1∶2Þ Þ with Second-Order
Accuracy
ð1∶1Þ ð1∶1Þ
Esz ðy; zÞ ¼ Esz ðcn nn þ ct nt Þ
ð1∶1Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
¼ expð−ikt ct ÞEsz ðcn nn Þ Gð1∶2Þ ðεm−1=2 ; εm ; εmþ1=2 Þ
 pffiffiffi 2 3
¼ exp −ikt ð−y þ zÞ= 2 g
ð1∶2Þ ð1∶2Þ
g1;2
ð1∶2Þ
g1;3
ð1∶2Þ
g1;4
6 1;1 7
6 gð1∶2Þ gð1∶2Þ gð1∶2Þ g2;4 7
ð1∶2Þ ðC1Þ
× Esz ððy þ zÞ=2; ðy þ zÞ=2Þ: ðA11Þ 6 7
¼ 6 2;1ð1∶2Þ
2;2
ð1∶2Þ
2;3
ð1∶2Þ ð1∶2Þ 7;
6 g3;1 g g g3;4 7
Equations (9)–(11) are obtained by substituting 4 3;2 3;3 5
ð1∶2Þ ð1∶2Þ ð1∶2Þ ð1∶2Þ
ðjΔ; kΔÞ, ðjΔ; ðk þ 1=2ÞΔÞ, and ððj þ 1=2ÞΔ; kΔÞ into g4;1 g4;2 g4;3 g4;4
ðy; zÞ in Eqs. (A9)–(A11), respectively.

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1093

 ð1∶2Þ 3sk0 Δ Appendix D: Elements of Q ðmÞ with Second-Order
g1;1 ≈ 1 − i pffiffiffi ; ðC2Þ Accuracy
5
2 h  i 3

pffiffiffiffi
ε2 cosθt k0 Δ m−

2þd
1
ð1∶2Þ
g1;2 ≈
ð1∶2Þ
−iεm k0 Δ; ðC3Þ 6 pffiffi
7
6 1þi e1 7
6 h  i
5 7
6 7
6
pffiffiffiffi  7
6 ε2 cosθt k0 Δ m− 2þd 1
7
6 1−i pffiffi e 7
ð1∶2Þ
g1;3
ð1∶2Þ
≈ −iεm−1=2 k0 Δ; ðC4Þ 6 5 2 7
QðmÞ≈ 6
6

pffiffiffiffi
h  i
 7
7;
6 pffiffi −mþ12þd ε2 sinθt k0 Δ m− 12þd 7
6 − 3þ 5 1þi pffiffi e3 7
6 2 7
6 5
h  i 7
ð1∶2Þ
¼ 0; ðC5Þ 6

p ffiffiffi
ffi  7
g1;4 6 7
4 −3þpffiffi5 −mþ2þd ε2 sinθt k0 Δ m− 2þd
1 1

pffiffi
5
2 1þi 5
e4
pffiffiffi

ð1∶2Þ 5s 3s2 ðD1Þ
g2;1 ≈ − ð1∶2Þ þ ik0 Δ −1 þ ð1∶2Þ ; ðC6Þ
εmþ1=2 εmþ1=2

e1 ¼ ½ e1;1 e1;2 e1;3 e1;4 ; ðD2Þ

ð1∶2Þ

ð1∶2Þ εm 3sk0 Δ
g2;2 ≈− ð1∶2Þ
1 − i pffiffiffi ; ðC7Þ
εmþ1=2 5 e2 ¼ ½ e2;1 e2;2 e2;3 e2;4 ; ðD3Þ

ð1∶2Þ
ð1∶2Þ 
ð1∶2Þ
εm−1=2 sk0 Δ 9εm−1=2 e3 ¼ ½ e3;1 e3;2 e3;3 e3;4 ; ðD4Þ
g2;3 ≈ −1 − ð1∶2Þ þ i pffiffiffi −1 þ ð1∶2Þ ; ðC8Þ
εmþ1=2 2 5 εmþ1=2

ð1∶2Þ sk Δ e4 ¼ ½ e4;1 e4;2 e4;3 e4;4 ; ðD5Þ

g2;4 ≈ 1 − i p0ffiffiffi ; ðC9Þ
5
pffiffiffiffiffi
pffiffiffi 1 ε2 ðcos θt þ 2 sin θt Þk0 Δ
5s 3s2 k0 Δ e1;1 ¼ −i pffiffiffi ; ðD6Þ
ð1∶2Þ
g3;1 ≈− þi ; ðC10Þ 2 4 5
ð1∶2Þ ð1∶2Þ
εmþ1=2 2εmþ1=2

ð1∶2Þ 
ð1∶2Þ εm 3sk0 Δ pffiffiffiffiffi  pffiffiffiffiffi 
g3;2 ≈ − ð1∶2Þ 1 − i pffiffiffi ; ðC11Þ ε2 ε2
εmþ1=2 2 5 e1;2 ¼ pffiffiffi 1 − i pffiffiffi ð2 cos θt − sin θt Þk0 Δ ;
2 5 cos θt 2 5
ð1∶2Þ
 ðD7Þ
ð1∶2Þ
εm−1=2 3sk0 Δ
g3;3 ≈− ð1∶2Þ
1 − i p ffiffiffi ; ðC12Þ
εmþ1=2 5

pffiffiffiffiffi
pffiffiffiffiffi 
ð1∶2Þ
g3;4 ¼ 0; ðC13Þ ε2 1 2 ε2
e1;3 ¼ − pffiffiffi þ i pffiffiffi k0 Δ ; ðD8Þ
2 5 cos θt 5
ð1∶2Þ
g4;1 ¼ 0; ðC14Þ

ð1∶2Þ pffiffiffiffiffi
pffiffiffiffiffi 

g4;2 ¼ 1; ðC15Þ ε2 ε2
e1;4 ¼ pffiffiffi 1 − i pffiffiffi sin θt k0 Δ ; ðD9Þ
2 5 cos θt 2 5
ð1∶2Þ
g4;3 ¼ 0; ðC16Þ

ð1∶2Þ pffiffiffiffiffi
g4;4 ¼ 0: ðC17Þ 1 ε2 ðcos θt − 2 sin θt Þk0 Δ
e2;1 ¼ þi pffiffiffi ; ðD10Þ
2 4 5

1094 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010

 pffiffiffiffiffi  pffiffiffiffiffi  Appendix E: Equivalent Expression with the Explicit
ε2 ε2 Description of R and a 1;pp for Equations (76) and (77)
e2;2 ¼ − pffiffiffi 1 þ i pffiffiffi ð2 cos θt þ sin θt Þk0 Δ ;
2 5 cos θt 2 5
ðD11Þ
   
u1;1 u1;2 R u1;0
¼− ; ðE1Þ
u2;1 u2;2 a1;pp u2;0
pffiffiffiffiffi

ε2 1 2 pffiffiffiffiffi
e2;3 ¼ pffiffiffi − i pffiffiffi ε2 k0 Δ ; ðD12Þ
2 5 cos θt 5 ð1∶2Þ
u1;1 ≈ e2 LF1;b ð−ð1=2 þ dÞÞ; ðE2Þ

ð1∶2Þ
pffiffiffiffiffi
pffiffiffiffiffi  u1;2 ≈ e2 LF1;pp ð−ð1=2 þ dÞÞ; ðE3Þ
ε2 ε2
e2;4 ¼ − pffiffiffi 1 − i pffiffiffi sin θt k0 Δ ; ðD13Þ
2 5 cos θt 2 5
ð1∶2Þ
u2;1 ≈ e3 LF1;b ð−ð1=2 þ dÞÞ; ðE4Þ

pffiffiffi
ð5 þ 5Þ sin θt ð1∶2Þ
u2;2 ≈ e3 LF1;pp ð−ð1=2 þ dÞÞ; ðE5Þ
e3;1 ¼ pffiffiffiffiffi
10 ε2
pffiffiffi pffiffiffi
½−3 5 þ ð5 þ 4 5Þ cosð2θt Þk0 Δ
þi ; ðD14Þ ð1∶2Þ
u1;0 ≈ e2 LF1;f ð−ð1=2 þ dÞÞ; ðE6Þ
50

ð1∶2Þ
u2;0 ≈ e3 LF1;f ð−ð1=2 þ dÞÞ: ðE7Þ
pffiffiffi pffiffiffi pffiffiffiffiffi
3þ 5 ð2 þ 5Þ ε2 sin θt k0 Δ References
e3;2 ¼ −i ; ðD15Þ
10 25 1. K. S. Yee, “Numerical solution of initial boundary value prob-
lems involving Maxwell’s equations in isotropic media,” IEEE
Trans. Antennas Propagat. 14, 302–307 (1966).
pffiffiffi pffiffiffiffiffi 2. A. Taflove and S. C. Hagness, Computational Electrody-
1 1 ð8 þ 5Þ ε2 sin θt k0 Δ namics: the Finite-Difference Time-Domain Method, 3rd ed.
e3;3 ¼ þ pffiffiffi − i ; ðD16Þ (Artech House, 2005).
5 5 25 3. T. G. Jurgens, A. Taflove, K. Umashankar, and T. G. Moore,
“Finite-difference time-domain modeling of curved surfaces,”
IEEE Trans. Antennas Propagat. 40, 357–366 (1992).
pffiffiffi pffiffiffiffiffi 4. P. H. Harms, J. F. Lee, and R. Mittra, “A study of the nonortho-
1 ð1 þ 5Þ ε2 sin θt k0 Δ gonal FDTD method versus the conventional FDTD technique
e3;4 ¼− þi ; ðD17Þ for computing resonant frequencies of cylindrical cavities,”
5 50 IEEE Trans. Microwave Theory Tech. 40, 741–746 (1992).
5. I. S. Kim and W. J. R. Hoefer, “A local mesh refinement algo-
rithm for the time domain-finite difference method using Max-
pffiffiffi well’s curl equations,” IEEE Trans. Microwave Theory Tech.
ð−5 þ 5Þ sin θt 38, 812–815 (1990).
e4;1 ¼ pffiffiffiffiffi 6. K. H. Dridi, J. S. Hesthaven, and A. Ditkowski, “Staircase-free
10 ε2
pffiffiffi pffiffiffi finite-difference time-domain formulation for general materi-
½3 5 þ ð5 − 4 5Þ cosð2θt Þk0 Δ als in complex geometries,” IEEE Trans. Antennas Propagat.
þi ; ðD18Þ 49, 749–756 (2001).
50
7. M. Celuch-Marcysiak and W. K. Gwarek, “Higher-order
modelling of media interfaces for enhanced FDTD analysis
pffiffiffi pffiffiffi pffiffiffiffiffi of microwave circuits,” in Proceedings of the 24th European
3− 5 ð−2 þ 5Þ ε2 sin θt k0 Δ Microwave Conference (Wiley, 1994), pp. 1530–1535.
e4;2 ¼ −i ; ðD19Þ 8. N. Kaneda, B. Houshmand, and T. Itoh, “FDTD analysis of
10 25
dielectric resonators with curved surfaces,” IEEE Trans. Mi-
crowave Theory Tech. 45, 1645–1649 (1997).
9. P. Yang, K. N. Liou, M. I. Mishchenko, and B.-C. Gao, “Efficient
pffiffiffi pffiffiffiffiffi finite-difference time-domain scheme for light scattering by
1 1 ð−8 þ 5Þ ε2 sin θt k0 Δ dielectric particles: application to aerosols,” Appl. Opt. 39,
e4;3 ¼ − pffiffiffi − i ; ðD20Þ
5 5 25 3727–3737 (2000).
10. M. Fujii, D. Lukashevich, I. Sakagami, and P. Russar, “Conver-
pffiffiffi pffiffiffiffiffi gence of FDTD and wavelet-collocation modeling of curved
1 ð−1 þ 5Þ ε2 sin θt k0 Δ dielectric interface with the effective dielectric constant tech-
e4;4 ¼− þi : ðD21Þ
5 50 nique,” IEEE Microw. Wireless Compon. Lett. 13, 469–
471 (2003).

1 March 2010 / Vol. 49, No. 7 / APPLIED OPTICS 1095

11. A. Mohammadi, H. Nadgaran, and M. Agio, “Contour-path symplectic finite-difference time-domain method,” Appl.
effective permittivities for the two-dimensional finite-differ- Opt. 46, 1514–1524 (2007).
ence time-domain method,” Opt. Express 13, 10367–10381 14. D. Marcuse, Theory of Dielectric Optical Waveguides
(2005). (Academic, 1974), pp. 7–17.
12. T. Hirono, Y. Shibata, W. W. Lui, S. Seki, and Y. Yoshikuni, 15. A. Farjadpour, D. Roundy, A. Rodriguez, M. Ibanescu, P.
“The second-order condition for the dielectric interface Bermel, J. D. Joannopoulos, S. G. Johnson, and G. W. Burr,
orthogonal to the Yee-lattice axis in the FDTD scheme,” “Improving accuracy by subpixel smoothing in the finite-
IEEE Microwave Guided Wave Lett. 10, 359–361 difference time domain,” Opt. Lett. 31, 2972–2974 (2006).
(2000). 16. G. R. Werner and J. R. Cary, “A stable FDTD algorithm for
13. T. Hirono and Y. Yoshikuni, “Accurate modeling of dielectric non-diagonal, anisotropic dielectrics,” J. Comput. Phys. 226,
interfaces by the effective permittivities for the fourth-order 1085–1101 (2007).

1096 APPLIED OPTICS / Vol. 49, No. 7 / 1 March 2010