Developing a time-domain finite-element method for modeling of electromagnetic cylindrical cloaks
Jichun Li
a,b,⇑,1, Yunqing Huang
a,2, Wei Yang
a,3aHunan Key Laboratory for Computation and Simulation in Science and Engineering, Xiangtan University, Xiangtan 411105, China
bDepartment of Mathematical Sciences, University of Nevada Las Vegas, Las Vegas, NV 89154-4020, USA
a r t i c l e i n f o
Article history:
Received 9 September 2011
Received in revised form 21 November 2011 Accepted 25 December 2011
Available online 4 January 2012
Keywords:
Maxwell’s equations Invisibility cloak Finite element method
a b s t r a c t
In this paper we propose a time-domain finite element method for modeling of electro- magnetic cloaks. The permittivity and permeability of the cloak model are described by the Drude dispersion model. The model to be solved is quite challenging in that we have to solve a coupled problem with different partial differential equations given in different regions. Our method is based on a mixed finite element method using edge elements with different types of meshes in different regions. Numerical results demonstrate that our algo- rithm is quite effective for simulating cloaks in time-domain. To our knowledge, this is the first cloak simulation carried out by the time-domain finite element method.
Ó2012 Elsevier Inc. All rights reserved.
1. Introduction
The possibility of cloaking an object from detection by electromagnetic waves has recently become a very hot research topic. In 2006, Pendry et al.[25]and Leonhardt[14]independently showed that it is possible to create invisible cloaks for ray optics[14]and electromagnetic waves[25]by guiding light around a region as if nothing is there. In late 2006, a 2-D reduced cloak was successfully fabricated and demonstrated to work in the microwave frequency region[26]. This is the first practical realization of such a cloak, and the result matches with the computer simulation[6]performed using COMSOL mul- tiphysics finite element analysis software. These pioneering work inspired researchers in different disciplines around the world to pursue the human being’s invisibility dream.
Since 2006, many papers have been published on the study of using metamaterials[3,8,18]to construct invisibility cloaks of different shapes (e.g.[1,5,21,22,28]). Also cloaks operating from microwave frequencies to optical frequencies have been achieved, more details and references on cloaking can be found in recent reviews[4,9]. Some mathematical analysis has been carried out for cloaking in frequency domain[19,13].
Numerical simulation plays a very important role in modeling different cloaks and validating theoretical predictions. The time-domain finite difference (FDTD) method is a very popular technique used in this area, readers can find more details about the FDTD method and its applications in cloak simulation in a recent book[10]. Due to the major disadvantage of FDTD method in dealing with complex geometry[30], the finite element method (FEM) based commercial package COMSOL has
0021-9991/$ - see front matterÓ2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcp.2011.12.026
⇑ Corresponding author at: Department of Mathematical Sciences, University of Nevada Las Vegas, Las Vegas, NV 89154-4020, USA. Tel.: +1 (702) 895 0355.
E-mail addresses:jichun@unlv.nevada.edu(J. Li),huangyq@xtu.edu.cn(Y. Huang),yangweixtu@126.com(W. Yang).
1 Supported by National Science Foundation Grant DMS-0810896.
2 This work was supported by National Science Foundation Grant DMS-0810896, and in part by the NSFC Key Project 11031006 and Hunan Provincial NSF project 10JJ7001.
3 Supported by Hunan Education Department Key Project 10A117 and Hunan Provincial Innovation Foundation for Postgraduate (CX2011B243).
Contents lists available atSciVerse ScienceDirect
Journal of Computational Physics
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j c p
been extensively used in frequency-domain cloak simulation by engineers and physicists[6,4]. However, COMSOL cannot be used for time-domain cloak simulation due to its limitation on algorithmic development for time-domain cloak modeling.
On the other hand, the recently designed broadband cloaks[20]make the time-domain simulation more appealing and nec- essary. But little attention has been paid to the time-domain modeling of cloaks. To our best knowledge, the first time-do- main simulation of 2-D cloaking structures was carried out by Zhao et al.[31] in 2008 using FDTD method. Later, they generalized the same idea to 3-D cloak simulation in 2009[32]. It seems that there is no existing work on time-domain finite element (FETD) method developed for cloak simulation yet.
In this paper, we propose a FETD method to simulate a 2-D cylindrical cloak. This problem is quite challenging, since we have to solve a coupled problem with different partial differential equations in different regions, and the material parameters are highly anisotropic and nonhomogeneous. The rest of the paper is organized as follows. We first describe the 2-D cylin- drical cloak modeling equations in Section2. In Section3, we develop a FETD method for solving the model problem. Then in Section4, we present a detailed stability analysis of our scheme. Numerical results showing the cloaking phenomena ob- tained by our method are illustrated in Section5. We conclude the paper in Section6.
2. The modeling equations
The cloak modeling is based on the Faraday’s law and Ampere’s law written as follows:
@B
@t ¼ rE; ð1Þ
@D
@t ¼rH ð2Þ
and the constitutive relations
D¼
e
E; ð3ÞB¼
l
H; ð4ÞwhereEandHare the electric and magnetic fields respectively,DandBare the electric displacement and magnetic induc- tion respectively,
e
andl
are cloak permittivity and permeability, respectively. For the cylindrical cloak, the ideal material parameters in the polar coordinate system are given by[25]:e
r¼l
r¼rR1r ;
e
/¼l
/¼ r rR1;
e
z¼l
z¼ R2R2R1
2
rR1
r ; ð5Þ
whereR1andR2are the inner and outer radius of the cloak. In this case,Ebecomes a 2-D vector, andHis a scalar, i.e., we can writeE= (Ex,Ey)0andH=Hz, where the subindexx,yorzdenotes the component in each direction.
Transforming the polar coordinate system to the Cartesian coordinate system, we can obtain[31]:
e
0e
re
/E¼e
rsin2/þe
/cos2/ ðe
/e
rÞsin/cos/ð
e
/e
rÞsin/cos/e
rcos2/þe
/sin2/" #
D: ð6Þ
Because the material parameters given in(5)can not be used directly to simulate the time-domain cloak, we have to map the parameters using the dispersive medium models. Here we consider the Drude model for the permittivity:
e
rðx
Þ ¼1x
2px
2jxc
; ð7Þwhere
c
P0 andxp> 0 are the collision and plasma frequencies, respectively. Substituting(7)into(6)and using the follow- ing rulesj
x
!@@t;
x
2! @2@t2; ð8Þ
we have (detailed derivation see[31]):
e
0e
/ @2
@t2þ
c
@@tþw2p
! E¼ @2
@t2þ
c
@@tþw2p
!
MADþ
e
/ @2
@t2þ
c
@@t
!
MBD; ð9Þ
where we denoteD= (Dx,Dy)0and MA¼ sin2/ sin/cos/
sin/cos/ cos2/
" #
; MB¼ cos2/ sin/cos/
sin/cos/ sin2/
" #
: Similarly, we map the permeability using the Drude model[31]:
l
zðx
Þ ¼A 1x
2pmx
2jxc
m!
; ð10Þ
whereA¼RR2
2R1, andxpm> 0 and
c
mP0 are the magnetic plasma and collision frequencies, respectively. Substituting(10) into(4), we obtainBz¼
l
ol
zHz¼l
0A 1x
2pmx
2jxc
m! Hz:
Then using rules(8), we have
@2
@t2þ
c
m@@t
!
Bz¼
l
0A @2@t2þ
c
m@@tþ
x
2pm!
Hz: ð11Þ
To carry out the cloak simulation, we have to reduce the infinite domain problem to a bounded domain by using Bereng- er’s perfectly matched layer (PML)[2]to absorb waves leaving the computational domain without introducing reflections.
The two dimensional Berenger PML governing equations can be written as:
e
0@Ex
@t þ
r
yEx¼@ðHzxþHzyÞ@y ; ð12Þ
e
0@Ey
@t þ
r
xEy¼ @ðHzxþHzyÞ@x ; ð13Þ
l
0@Hzx@t þ
r
mxHzx¼ @Ey@x; ð14Þ
l
0@Hzy@t þ
r
myHzy¼@Ex@y; ð15Þ
where the parameters
r
i,r
mi,i=x,y, are the homogeneous electric and magnetic conductivities in thexandydirections, respectively.For easy implementation of our algorithm given below, we rewrite(12) and (13)in the vector form:
e
0@E
@tþ
r
y 00
r
x
E¼rH; ð16Þ
where we used the 2-D vector curl operator rH¼
@H
@y
@H@x
!
; forH¼HzxþHzy:
3. A fully discrete explicit finite element scheme
To design our mixed finite element method, we partitionXby a family of regular meshesThwith maximum mesh sizeh.
To accommodate our problem easily, we use a mesh with mixed types of elements: triangles in the cloak and free space re- gion; rectangles in the PML region, cf.Fig. 1(b) below. For simplicity, currently we implement our algorithm using the low- est-order Raviart–Thomas–Nédélec’s mixed spacesUhandVhgiven as follows[23,24]: for any rectangular elementK2Th, we choose
Uh¼nwh2L2ðXÞ:whjK2Q0;0; 8K2Tho
;
Vh¼n/h2Hðcurl;XÞ:/hjK2Q0;1Q1;0; 8K2Tho
;
whereQi,jdenotes the space of polynomials whose degrees are less than or equal toiandjin variablesxandy, respectively.
Here we denote the spaceH(curl;X) = {v2(L2(X))2;rv2(L2(X))2}. While on a triangular element, we choose Uh¼nwh2L2ðXÞ:whjK is a piecewise constant;8K2Tho
;
Vh¼n/h2Hðcurl;XÞ:/hjK¼spanfkirkjkjrkig; i;j¼1;2;3; 8K2Tho
; wherekidenotes the standard linear basis function at vertexiof elementK.
To impose the perfect conducting boundary conditionnE=0, we introduce the space V0h¼ f/h2Vh;n/h¼0on@Xg:
To define a fully-discrete scheme, we divide the time intervalI= [0,T] intoNuniform subintervalsIi= [ti1,ti] by pointsti=i
s
, i= 0, 1,. . .,N, wheres
=T/N. Furthermore, we denoteun=u(,tn) andunþ12¼u;ðnþ12Þs
, and introduce some operators:dsun¼unun1
s
; d2
sun¼un2un1þun2
s
2 ;d2sun¼unun2 2
s
; un1¼unþ2un1þun2
4 ; bun¼unþun1
2 :
With the above preparation, we now construct a leap-frog type scheme for solving the modeling equations in the cloak re- gion: forn= 1, 2,. . ., findDnþh 12;Enþh 122V0h; Bnþ1h ;Hnþ1h 2Uhsuch that
dsDnþh 12;/h
Hnh;r/h
¼0; ð17Þ
e
0e
/d2sEnþh 12;/~h
þ
ce
0e
/d2sEnþh 12;/~h
þ
x
2pe
0e
/Enh 12;/~h
¼ MAþ
e
/MBd2sDnþh 12;/~h
þ
x
2pMADnh 12;/~h
þ
c
ðMAþe
/MBÞd2sDnþh 12;/~h
; ð18Þ
dsBnþ1h ;wh
þrEnþh 12;wh
¼0; ð19Þ
l
0Ad2sHnþ1h ;w~h
þ
l
0Ac
md2sHnþ1h ;w~h
þ
l
0Ax
2pmHnh;w~h
¼ d2sBnþ1h ;w~h
þ
c
md2sBnþ1h ;w~h
ð20Þ hold true for any/h;/~h2V0h; wh;w~h2Uh.
In order to couple(20)well with the PML Eqs.(14) and (15), we split(20)into
l
0Ad2sHnþ1zx;h;w~h
þ
l
0Ac
md2sHnþ1zx;h;w~h
þ
l
0Ax
2pmHnzx;h;w~h
¼1
2 d2sBnþ1h ;w~h
þ1
2
c
md2sBnþ1h ;w~h
; ð21Þ
l
0Ad2sHnþ1zy;h;w~h
þ
l
0Ac
md2sHnþ1zy;h;w~h
þ
l
0Ax
2pmHnzy;h;w~h
¼1
2 d2sBnþ1h ;w~h
þ1
2
c
md2sBnþ1h ;w~h
: ð22Þ
Similarly, we can construct a leap-frog type scheme for solving the Eqs. (16), (14) and (15) in the PML region: find Enþh 122V0h; Hnþ1zx;h;Hnþ1zy;h 2Uh such that
e
0 dsEnþh 12;/~h
þ
r
y 0 0r
x
bEnþh 12;/~h
¼ Hnzx;hþHnzy;h;r/~h
; ð23Þ
l
0 dsHnþ1zx;h;w1;h
þ
r
mxHbnþ1zx;h;w1;h
¼ @
@xEnþy;h12;w1;h
; ð24Þ
PML PML
wave source
PML PML
PEC
cloak air
Fig. 1.(a) The cloak modeling setup; and (b) a coarse mesh.
l
0dsHnþ1zy;h;w2;hþ
r
myHbnþ1zy;h;w2;h¼ @
@yEnþx;h12;w2;h
ð25Þ hold true for any/~h2V0h; w1;h;w2;h2Uh.
In summary, our mixed finite element time-domain algorithm for modeling the invisible cloak can be performed in the following steps: first, construct a proper meshThofX, choose a proper time step size
s
and proper initial conditions; then at each time stepn, perform theFETD Algorithm:1. Solve(17)forDnþh 12onTh.
2. Solve(18) and (23)forEnþh 12onTh. 3. Solve(19)forBnþ1h onTh.
4. Solve(21) and (24)forHnþ1zx;h onTh. 5. Solve(22) and (25)forHnþ1zy;h onTh.
6. CalculateHnþ1h ¼Hnþ1zx;hþHnþ1zy;h, then go back to step 1 and repeat the above process. Note that in the free space region,Enþh 12 andHn+1are updated using(23)–(25)with
r
x=r
y=r
mx=r
my= 0.4. Stability analysis
In this section, we present the stability analysis for our proposed scheme. Letcv¼1=pffiffiffiffiffiffiffiffiffiffi
0l
0be the wave propagation speed in vacuum,cinv> 0 is a constant from the inverse estimatekrwhk0;X6cinvh1kwhk0;X; 8wh2Vh; ð26Þ wherekk0,Xdenotes theL2norm over domainX.
We start with the easy stability analysis in the PML and free space regionXr: Theorem 4.1.Assume that the time step
s
6 hcvcinv; ð27Þ
then for any nP2 and the solutionEnh 12;Hnh¼Hnzx;h;Hnzy;h
of(23)–(25), we have
0 Enh 122 0;Xr
þ
l
0 Hnh 20;Xr 63
0 E12h2 0;Xr
þ
l
0 H1h 20;Xr
;
where we denote Hnh 20;X
r¼ Hnzx;h 2
0;Xr
þ Hnzy;h 2
0;Xr
.
Proof.Choosing/~h¼2
s
bEnþh 12; w1;h¼2s
Hbnþ1zx;h andw2;h¼2s
Hbnþ1zy;h in(23)–(25), respectively, and using the 2-D curl notation, we obtain 0 Enþh 122 0;Xr
Enh 12
2 0;Xr
þ
l
0 Hnþ1h 20;Xr
Hnh 20;X
r
62
s
Hnh;rEbnþh12rEnþh 12;Hbnþ1h
¼
s
Hnh;rEnh 12Hnþ1h ;rEnþh 12
; ð28Þ
where in the last step we used the identity Hnh;rEnþh12þEnh 12
rEnþh 12;Hnþ1h þHnh
¼Hnh;rEnh 12
Hnþ1h ;rEnþh 12 :
Summing up(28)fromn= 1 toN1, we obtain
0 ENh 12 20;Xr
E12h 2
0;Xr
þ
l
0 HNh 20;Xr
H1h 2
0;Xr
6
s
H1h;rE12hHNh;rENh 12
: ð29Þ
Using the definitioncv¼1=pffiffiffiffiffiffiffiffiffiffi
0l
0, the inverse estimate(26), the Cauchy–Schwarz inequality, and the time constraint (27), we haves
HNh;rENh 12¼ ffiffiffiffiffiffi
l
0p HNh;
s
cv ffiffiffiffiffi 0p rENh 12
61
2
l
0 HNh 20;Xr
þ ð
s
cvÞ20 rENh 122 0;Xr
61
2
l
0 HNh 20;Xr
þ
s
cvcinvh12 0 ENh 12 20;Xr
61
2
l
0 HNh 20;Xr
þ1
2
0 ENh 12 20;Xr
: ð30Þ
By the same technique, we obtain
s
H1h;rE12h 612
l
0 H1h 20;Xr
þ1 2
0 E12h2 0;Xr
: ð31Þ
Substituting the estimates(30) and (31)into(29)completes the proof. h
Remark 4.1. To see how largecinv in(26)can be, let us consider an arbitrary rectangular elementK= [xchx,xc+hx] [ychy,yc+hy], on which the lowest edge element basis functions are:
wh1¼
ðycþhyÞy 4hxhy
0
!
; wh2¼ 0
xðxchxÞ 4hxhy
!
; wh3¼
ðychyÞy 4hxhy
0
!
; wh4¼ 0
xðxcþhxÞ 4hxhy
! : Herewhj; j¼1;2;3;4;start from the bottom edge and orient counterclockwisely.
For simplicity, we assume that the meshThofXis formed byNrectanglesK. In this case, it is easy to check that the 2D curl ofwhj satisfies
Z
X
rwhj
2dx dy¼X
K2Th
Z
K
1 4hxhy
2
dx dy¼ N 4hxhy
andwhj satisfies Z
X
wh1
2
dx dy¼X
K2Th
Z
K
ycþhyy 4hxhy
2
dx dy¼X
K2Th
2hx
ð4hxhyÞ21
3 ðycþhyyÞ3
ycþhy
y¼ychy
¼Nhy
3hx
;
Z
X
wh3
2
dx dy¼Nhy
3hx
; Z
X
wh2
2
dx dy¼ Z
X
wh4
2
dx dy¼Nhx
3hy
;
from which we can see that rwhj
2 0;X
whj
2 0;X
¼ 3
4h2y; j¼1;3:
Similarly, we have rwhj
2 0;X
whj
2 0;X
¼ 3
4h2x; j¼2;4:
Denoteh= max{hx,hy}. Hence we have cinvP
ffiffiffi3 4 r h
hx
or cinvP ffiffiffi3 4 r h
hy
; ð32Þ
which means thatcinvcan be very large for anisotropic meshes. But for the often used shape regular mesh,cinvshould not be that large. Of course, exact estimate ofcinvreally depends on the mesh and the order of the basis functions. To our knowledge, there is no general formular forcinv.
The rest of the section is devoted to the stability analysis on the cloak regionXc=XnXr. For simplicity, in the rest of this section, we usekk0to denotek k0;Xc.
Lemma 4.1
02 ffiffiffiffiffi
/p dsEnþh 12
2 0 ffiffiffiffiffi
/p dsEnh 12
2 0
þ
0x
2p2
ffiffiffiffiffi
/p bEnþh 12
2
0 p bffiffiffiffiffi
/Enh 122 0
þ
sc
0 ffiffiffiffiffi/p dsbEnþh 12
2 0
6
s
ðMAþ/MBÞd2sDnþh 12;dsbEnþh 12þ
sc
d2ðMAþ/MBÞdsDbnþh 12;dsDbnþh 12 þ 14d2
ðMAþ
/MBÞdsbEnþh 12;dsbEnþh 12
þ
sx
2p2 d3
0 ffiffiffiffiffi /p dsbEnþh 12
2
0þ 1
2d3
0Dbnþh 12
2 0þ Dbnh 12
2 0
: ð33Þ
Proof. Choosing/~h¼
s
dsbEnþh 12in(18), we obtains
dsEbnþh12
¼
0 /dsEnþh 12dsEnh 12;dsbEnþh 12
¼
02 ffiffiffiffiffi
/p dsEnþh 12
2 0 ffiffiffiffiffi
/p dsEnh 12
2 0
;
c
0/Enþh 12Enh 32
2
s
;s
dsbEnþh 12 0B@
1 CA
¼
sc
02
/dsEnþh 12þdsEnh 12;dsEbnþh12
¼
sc
0 ffiffiffiffiffi /p dsbEnþh 12
2 0;
x
2p0/Enþh 12þ2Enh 12þEnh 32
4 ;bEnþh 12Ebnh 12 0
B@
1 CA
¼
0x
2p2
ffiffiffiffiffi
/p bEnþh 12
2
0 p bffiffiffiffiffi
/Enh 122 0
: Using the facts thatjMAj6I,jMBj6Iand 1<RR2
2R1
6
e
/, and the arithmetic–geometric mean inequalityjabj6da2þ4d1b2, we havec
ðMAþe
/MBÞd2sDnþh 12;s
dsbEnþh 12
¼
sc
ðMAþe
/MBÞdsDbnþh 12;dsbEnþh 12 6sc
d2ðMAþe
/MBÞdsDbnþh 12;dsDbnþh 12þ 1 4d2
ðMAþ
e
/MBÞdsEbnþh 12;dsEbnþh 12
and
x
2pMADnþh 12þ2Dnh 12þDnh 32 4 ;
s
dsEbnþh 12! 6
sx
2p2 dsEbnþh 12
0
Dbnþh 12þDbnh 12
0
6
sx
2p2 d3
e
0 pffiffiffiffiffie
/dsEbnþh 12 2
0þ 1 2d3
e
0Dbnþh 12 2
0þ Dbnh 12 2
0
; where in the last step, we used the fact that dsbEnþh 12 6 ffiffiffiffiffi
e
/p dsEbnþh 12
0. The proof completes by adding the above inequalities together. h
Lemma 4.2 1
2
l
0AðMAþe
/MBÞdsHnþ1h ;dsHnþ1hðM Aþ
e
/MBÞdsHnh;dsHnhh i
þ
sl
0Ac
mðMAþe
/MBÞdsHbnþ1h ;dsHbnþ1h þl
0Ax
2pm2 ðMAþ
e
/MBÞHbnþ1h ;Hbnþ1hðM Aþ
e
/MBÞHbnh;Hbnhh i
6
sc
m d8ðMAþe
/MBÞdsBbnþ1h ;dsbBnþ1h þ 14d8
ðMAþ
e
/MBÞdsHbnþ1h ;dsHbnþ1h
þ
s
ðMAþe
/MBÞd2sBnþ1h ;dsHbnþ1h :ð34Þ
Proof.An equivalent variable coefficient form of(20)can be written as ðMAþ
/MBÞl
0Ad2sHnþ1h þl
0Ac
md2sHnþ1h þl
0Ax
2pmHnh;w~h
¼ ðMAþ
/MBÞd2sBnþ1h ;w~hh
þ ðM Aþ
/MBÞc
md2sBnþ1h ;w~hÞi: ð35Þ
Choosingw~h¼
s
dsHbnþ1h in(35), we obtainl
0A ðMAþ/MBÞdsHnþ1h dsHnh;dsHbnþ1h
¼1
2
l
0A ðMAþ/MBÞdsHnþ1h ;dsHnþ1hðM Aþ
/MBÞdsHnh;dsHnh
;
l
0Ac
mðMAþe
/MBÞd2sHnþ1h ;s
dsHbnþ1h¼
sl
0Ac
mðMAþe
/MBÞdsHbnþ1h ;dsHbnþ1h;
l
0Ax
2pm4 ðMAþ
e
/MBÞHnþ1h þHnhþHnhþHn1h;Hbnþ1h Hbnh
¼
l
0Ax
2pm2 ðMAþ
e
/MBÞHbnþ1h ;Hbnþ1hðM Aþ
e
/MBÞHbnh;Hbnhh i
;
c
mðMAþe
/MBÞd2sBnþ1h ;s
dsHbnþ1¼
s c
mðMAþe
/MBÞdsbBnþ1h ;dsHbnþ1 6s c
m d8ðMAþe
/MBÞdsbBnþ1h ;dsbBnþ1hþ 1 4d8
ðMAþ
e
/MBÞdsHbnþ1h ;dsHbnþ1h
: Summing up the above estimates concludes the proof. h
Lemma 4.3
s
ðMAþ/MBÞd2sDnþh 12;dsbEnþh 12þ
s
ðMAþ/MBÞd2sBnþ1h ;dsHbnþ1h¼
s
2 ðMAþ
/MBÞdsHnh;rdsEnh12ðM Aþ
/MBÞdsHnþ1h ;rdsEnþh 12h i
þ
s
dsHnh;rðMAþ/MBÞ dsbEnþh 12: ð36Þ
Proof. For an arbitrary functionf(x), multiplying(2)byf(x)/and integrating overX, we can obtain fðxÞ@D
@t;/
¼ ðrH;fðxÞ/Þ ¼ ðH;r ðfðxÞ/ÞÞ; 8/2H0ðcurl;XÞ; ð37Þ
from which we can obtain an equivalent form of(17):
fðxÞdsDnþh 12;/h
Hnh;r ðfðxÞ/hÞ
¼0; 8/h2V0h: ð38Þ
Subtracting(38)from itself withnreduced by 1, then dividing by
s
, we obtain fðxÞd2sDnþh 12;/h
dsHnh;r ðfðxÞ/hÞ
¼0; 8/h2V0h: ð39Þ
On the other hand, it is easy to see that an equivalent form of(19)is:
fðxÞdsBnþ1h ;wh
þfðxÞrEnþh 12;wh
¼0; 8wh2Uh; ð40Þ
from which we obtain fðxÞd2sBnþ1h ;wh
þfðxÞrdsEnþh 12;wh
¼0; 8wh2Uh: ð41Þ
Choosing/h¼
s
dsEbnþh 12; wh¼s
dsHbnþ1h in(39) and (41), respectively, we haves
fðxÞd2sDnþh 12;dsbEnþh 12þ
s
fðxÞd2sBnþ1h ;dsHbnþ1h¼
s
dsHnh;rfðxÞdsEnþh 12
fðxÞrdsbEnþh 12;dsHbnþ1h
h i
: ð42Þ
Using the identity
dsHnh;fðxÞrdsEbnþh 12
fðxÞrdsEnþh 12;dsHbnþ1h
¼1
2 dsHnh;fðxÞrdsEnh 12
dsHnþ1h ;fðxÞrdsEnþh 12
h i
; ð43Þ
the formula
r ðfðxÞuÞ ¼fðxÞruþrfðxÞ u
and choosingf(x) =MA+
/MBin(42), we conclude the proof. hLemma 4.4. For any vector (u, v)0, we have ðu;
v
ÞðMAþ/MBÞ uv
>u2þ
v
2:Proof. Using the definitions ofMAandMB, and the fact that
/> 1, we haveðu;
v
ÞðMAþ/MBÞ uv
¼ ðu2sin2/2u
v
sin/cos/þv
2cos2/Þ þ/u2cos2/þ2uv
sin/cos/þv
2sin2/¼ ðusin/
v
cos/Þ2þ/ðucos/þv
sin/Þ2>ðusin/v
cos/Þ2þ ðucos/þv
sin/Þ2¼u2þ
v
2:Combining the above estimates, we finally have the following stability result on the cloaking region:
Theorem 4.2. Under the time step constraint(27),for any NP1, we have
0ffiffiffiffiffi
/p dsENþh 12
2
0þ
x
2p p bffiffiffiffiffi/ENþh 122 0
þ
l
0A dsHNþ1h 20þ
x
2pm HbNþ1h 20
6C
0 ffiffiffiffiffi/p dsE12h
2
0þ
0x
2p p bffiffiffiffiffi/E12h2
0þ
l
0A ffiffiffiffiffi/p dsH1h 2
0þ
l
0Ax
2pm p bffiffiffiffiffi/H1h 20
: