Professional Documents
Culture Documents
Computational Methods For Electromagnetic and Optical System (Jarem-Banerjee)
Computational Methods For Electromagnetic and Optical System (Jarem-Banerjee)
Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
The publisher offers discounts on this book when ordered in bulk quantities. For
more information, write to Special Sales/Professional Marketing at the headquarters
address above.
Neither this book nor any part may be reproduced or transmitted in any form or by
any means, electronic or mechanical, including photocopying, micro®lming, and
recording, or by any information storage and retrieval system, without permission
in writing from the publisher.
Founding Editor
Brian J. Thompson
University of Rochester
Rochester, New York
Editorial Board
This volume is about neither mathematics for the sake of mathematics nor
electromagnetic theory for the sake of electromagnetic theory. It is about the
important and useful computational methods that need to be applied to the
analysis and hence the design of electromagnetic and optical systems.
Computational Methods for Electromagnetic and Optical Systems presents
the best and most pertinent mathematical tools for the solution of current
and future analysis and synthesis of systems applications without over-
generalization; that means using the best and most appropriate tools for
the problem at hand. Optical design certainly proves that some problems
can be evaluated by ray tracing; others need scalar wave theory; still others
need electromagnetic wave analysis; and, ®nally, some systems require a
quantum optics approach. Thus, rays, waves, and photons have coexisted
in optical science and engineering, each with its own domain of validity and
each with its own computational methods.
Solutions of Maxwell's equations are described that can be applied to the
analysis of diffraction gratings, radiation, and scattering from dielectric
objects and holograms in photorefractive materials. Fundamentally it is
necessary to understand how electromagnetic radiation is transmitted,
re¯ected, and refracted through one- and two-dimensional isotropic and
anistrophic materials. One- and two-dimensional Fourier transform theory
allows for the study of how spectral components are propagated. The alter-
native method of split-step beam propagation can be applied to inhomoge-
neous media.
Other computational methods covered in these pages include: coupled-
wave analysis of inhomogeneous cylindrical and spherical systems, state
variable methods for the propagation of anisotropic waveguide systems,
and rigorous coupled wave analysis for photorefractive devices and systems.
v
vi From the Series Editor
Brian J. Thompson
Preface
vii
viii Preface
John M. Jarem
Partha P. Banerjee
Contents
1. Mathematical Preliminaries 1
Index 429
xi
1
Mathematical Preliminaries
1.1 INTRODUCTION
!
r D i
1:1:1
!
r B 0
1:1:2
!
! @B
r E
1:1:3
@t
!
! ! ! @ D
r H Jc J i
1:1:4
@t
! !
In Eqs. 1.1.1±4, i denotes the impressed charge density, and J c , and J i are
conduction and impressed current densities, respectively. In time-reduced
! !
form (i.e., assuming variations of the form A Re
A exp j!t, Eqs. 1.1.1±4
read
1
2 Chapter 1
!
r D i
1:1:5
!
r B 0
1:1:6
! !
r E j! B
1:1:7
! ! ! !
r H Jc Ji j! D
1:1:8
In Eqs. 1.1.5±8, the electric and magnetic ®eld variables are related through
the constitutive relations as
! ! ! ! ! !
D 0 E B 0 lH J c rE
1:1:9
X
1
f
x Fn exp
jnKx
1:2:3
n 1
Introduction 3
Multiplying Eq. 1.2.3 by exp
jmKx, integrating over the interval
x0 ,
x0 2=K, interchanging the summation and the integral, and using Eqs.
1.2.1 and 1.2.2, we obtain
x0 2=K X
1
x0 2=K
f
x exp
jnKx dx Fn exp
jnKx
x0 n 1 x0
exp
jmKx dx
X1
2 2
Fn m;n F
n 1
K K m
Now, replacing m by n,
x0 2=K
K
Fn f
x exp
jnKx dx
1:2:4
2 x0
Note that if a function fe
x is de®ned as fe
x f
x exp
jx, where is a
constant, then over the interval
x0 ; x0 2=K, it can be written as
X
1
fe
x Fn exp
jkxn x kxn nK
1:2:5
n 1
X
1 X
1
f
x Fn exp
jnKx g
x Gn exp
jnKx
1:2:6
n 1 n 1
over the same interval are multiplied, the product function h
x has a
Fourier series expansion
X
1
h
x Hn exp
jnKx
1:2:7
n 1
over the same interval. We can ®nd the Fourier coef®cients of h
x in the
following way:
4 Chapter 1
X
1 X
1
h
x f
xg
x Fn exp
jnKx Gm exp
jmKx
n 1 m 1
X
1 X
1
Fn Gm exp
j
n mKx
n 1 m 1
1:2:8
X
1 X
1
Fl m Gm exp
jlKx
l 1 m 1
X
1
Hl exp
jlKx
l 1
X
1
Hl Fl m Gm
1:2:9
m 1
X
N
h
x lim Hl exp
jlKx
N!1
l N
!
1:2:10
X
L X
M
lim lim Fl m Gm exp
jlKx
L!1 M!1
l L m M
The de®nitions for the forward and backward transforms are consistent
with the engineering convention for a traveling wave, as explained in [2].
If f
x denotes a phasor EM ®eld quantity, multiplication by exp j!t gives a
collection or spectrum of forward traveling plane waves.
The two-dimensional extensions of Eqs. 1.3.1 and 1.3.2 are
1
1
F
kx ; ky f
x; y exp
jkx x jky y dx dy
1:3:3
1 1
1
1
1
f
x; y F
kx ; ky exp
jkx x jky y dx dy
1:3:4
22 1 1
The usefulness of this transform lies in the fact that when substituted into
Maxwell's equations, one can reduce the set of three-dimensional PDEs to a
6 Chapter 1
X
1
fp
n f
n rN
1:4:1
r 1
X1
N
2
Fp
mK fp
n exp
jmnK K
1:4:2
n0
N
1NX1
fp
n F
mK exp
jmnK
1:4:3
N n0 p
1
Fp
mK F
mK jmKj <
1:4:4
In Eq. 1.4.4, Fp
mK is de®ned, as in Eq. 1.4.2, to be the DFT of fp
n. The
equality holds for the ®ctitious case when the function is both space and
spatial frequency limited.
Introduction 7
Many of the computations in this book are based on determining the eigen-
values and eigenvectors of a matrix A. Therefore this section will brie¯y
review the methods and techniques associated with numerically solving this
problem [4,5]. The matrix A, which is a square matrix, in general transforms
a column vector x that transform into themselves and satisfy
Ax qx 1:5:1
These column vectors are called the eigenvectors of the system. The values q
which satisfy are known as the eigenvalues, the characteristic values, or the
latent roots of the matrix A. Equation 1.5.1 can be written as a linear set of
equations as
where I is the identity matrix. The result of Eq. 1.5.3 is an nth order poly-
nomial called the characteristic equation or eigenvalue equation. The equa-
tion is given by
P
q qn a1 qn 1
a2 qn 2
an 1 q an
1:5:4
The roots of this equation are the eigenvalues of the matrix A. When the
roots are all unequal to one another, the roots or eigenvalues are called
distinct. When the eigenvalue occurs m times, the eigenvalue is a repeated
value of order m. When the root has a real and nonzero imaginary part, the
roots occur in complex conjugate pairs. In factored form Eq. 1.5.4 can be
written as
8 Chapter 1
The coef®cients of the eigenvalue equation can be found directly from the
matrix A. For instance, setting q to zero in Eq. 1.5.5, we ®nd
det A q1 q2 qn 1:5:7
and thus
a1 q1 q2 1:5:9
a1 q1 q2 qn 1:5:10
Let Tk Tr
Ak . Then a useful formula for the coef®cient an of the char-
acteristic equation is
Introduction 9
a1 T1
1
a2
a T T2
2 1 1
1
a3
a T a1 T2 T3
3 2 1
1:5:13
1
an
a T an 2 T2 a1 Tn Tn
n n 1 1 1
For the case when the roots of P
q are distinct, a nontrivial vector xi
can be found for each root that satis®es
The matrix formed of the columns of xi is called the modal matrix M. The
name modal matrix comes from control theory where a dynamical system
can be decomposed into dynamic modes of operation. For EM diffraction
grating problems and also for EM problems which use k-space (spatial
Fourier transform) techniques, the EM ®eld solutions associated with a
state variable analysis can be decoupled into spatial mode solutions.
These modes are analogous to the dynamical modes of operation encoun-
tered in control systems.
If the eigenvalues are distinct, which is mainly the case under consid-
eration in this text, the modal matrix is nonsingular and therefore its inverse
exists. Letting M be the modal matrix, we may write
MQ AM 1:5:15
Q M 1A M 1:5:16
If Q is squared we have
Q2
M 1 A M
M 1 A M
M 1 A2 M
1:5:17
10 Chapter 1
A2 M Q2 M 1
1:5:18
Ap M Qp M 1
1:5:19
1
N
A M N
Q M
1:5:20
1 1
exp
A I A
A2
Ak
1:5:21
2 k!
which is the same in®nite series expansion as is used to de®ne the exponen-
tial function exp
a.
We now review two important aids that help in the solution and
evaluation of an exponential matrix and in fact any function of the matrix
A. These are called the Cayley±Hamilton theorem and the Cayley±Hamilton
technique. These aids will be presented only for the cases of matrices with
distinct eigenvalues.
The ®rst theorem to be reviewed is the Cayley±Hamilton theorem. If
we have a polynomial N
q qn c1 qn 1 cn 1 q cn then using Eq.
1.5.20 we have
2 3
N
q1 0 0
6 0 N
q2 7
6 7 1
N
A M6 0 N
q3 7M
1:5:22
4 5
..
.
Introduction 11
We thus see that the matrix A satis®es its own characteristic equation.
Another important aid in evaluating a function of a matrix, where the
function is analytic over a given range of interest, is provided by the Cayley±
Hamilton technique. We ®rst consider the case where the analytic function is
a polynomial of higher degree than the characteristic polynomial P
q of
order n. Let the polynomial be N
q. We consider the case where the roots
(or eigenvalues) of P
q are distinct. In this case,
N
q R
q
Q
q
1:5:24
P
q P
q
R
q 0 1 q 2 q2 n 1 qn 1
1:5:29
F
q1 R
q1
F
q2 R
q2
1:5:30
F
qn R
qn
F
q R
q
Q
q
1:5:31
P
q
In this expression we note that over the region of interest, the numerator
and denominator of Eq. 1.5.31 have the same zeros. Since in Eq. 1.5.31 all
functions F
q, Q
q, P
q, and R
q are analytic over the range where F
q
is, we may replace q by the matrix A. We have
Thus we have shown that the analytic matrix function F
A can be evaluated
by using a polynomial matrix expression of order n 1 as given by R
A in
Eq. 1.5.29.
Introduction 13
PROBLEMS
1. Derive the wave equation for the electric and magnetic ®elds start-
ing from Maxwell's equations in a homogeneous isotropic source
free region. How does this change if the material is anisotropic?
2. Find from ®rst principles the Fourier series coef®cients for a per-
iodic square wave s
x of unit amplitude and 50% duty cycle. Now
®nd the Fourier series coef®cients of s2
x (a) from ®rst principles
and (b) using the Laurent rule. Plot s2
x vs x by employing the
Fourier series coef®cients you found using (b). Use 5, 10 and 100
Fourier coef®cients. Describe the general trend(s).
3. Find the two-dimensional Fourier transform of a rectangle (rect)
function of unit height and width a in each dimension.
4. Show that the two-dimensional Fourier transform of a Gaussian
function of width w is another Gaussian function. Functions like
this are called self-Fourier transformable. Find its width in the
spatial frequency domain. Can you think of any other functions
that are self-Fourier transformable?
5. Find the DFT of a square wave function using a software of your
choice. Comment on the nature of the spectrum numerically com-
puted as the width of the square wave changes.
0 1
1 20 0
6. Find sin A where A is a matrix given by @ 1 7 1 A using
the Cayley Hamilton theorem [5]. 3 0 2
REFERENCES
2.1 INTRODUCTION
15
16 Chapter 2
(this reduces the number of curl equations from six to four), and ®nally (4)
putting the four remaining equations into a standard 4 4 ®rst-order state
variable matrix equation form. The four transverse components Ex , Ez , Hx ,
and Hz form the components of the 4 1 state variable column matrix. As
shown in Section 2.4, this procedure provides a straightforward method of
analyzing bi-anisotropic material layers whenever oblique and arbitrarily
polarized plane wave radiation is incident on the material layers.
This 4 4 state variable matrix procedure has been ®rst implemented
by Teitler and Henvis [19], and perhaps others, who have reduced Maxwell's
equations in an anisotropic layer to a set of four ®rst-order linear differential
equations and then, assuming an exponential form of solution, have solved
for the normal or eigen modes that describe propagation in the layer. The
method is further developed by Berreman [20], who, starting from
Maxwell's six component equations, puts the general anisotropic equations
into a 4 4 form (where the 4 1 column vector contains the two tangential
electric ®eld components and two tangential magnetic ®eld components),
and then solves, using matrix techniques, for the four eigenvectors and
eigenvalues of the system. Berreman [20] has studied several anisotropic
material examples, including propagation in an orthorhombic crystal, pro-
pagation in an optically active material (described by the Drude model), and
propagation involving Faraday rotation based on Born's model. Berreman
[20] has also considered the state variable method as applied to determining
propagation in media that are anisotropic and longitudinally periodic. Lin-
Chung and Teitler [21], Krowne [22], and Morgan et al. [23] have used the
4 4 matrix method of Berreman [20] to study propagation of plane waves
in strati®ed or multilayer anisotropic media. Weiss and Gaylord [24] have
used the Berreman method to study strati®ed multilayer resonators and
optical ®lters (Fabry-Perot/Solc ®lter) composed of anisotropic materials.
Two recent papers by Yang [25,26] study the important problem of formu-
lating the EM state variable equations so that ef®cient numerical solution of
the equations arises. This problem has also been studied by Moharam et al.
(see Ref. 23 in Chapter 3 of this book).
Dispersion in anisotropic and birefringent materials, and properties of
the EM ®eld propagation in these materials, have been studied by many
other researchers. Yeh [27] has studied EM propagation in layered birefrin-
gent media. Alexopoulos and Uslenghi [28] study re¯ection and transmis-
sion with arbitrarily graded parameters. Graglia et al. [29] study dispersion
relations for bi-anisotropic materials and their symmetry properties. The
book by Lindell et al. [6] also quotes many papers that have studied propa-
gation in bi-anisotropic materials.
Another area where the k-space state variable analysis is useful is in
the problem of characterizing radiation from antennas, dipoles, and metallic
18 Chapter 2
not, must show equality between the power radiated out of the region and
the power dissipated and energy stored in the region. This is a more strin-
gent and useful test than the more standard test of checking conservation of
power from one layer to another. Checking power conservation from one
layer to another is a conclusive test as long as the materials inside the layers
are nonlossy. It is inconclusive if the layers inside are lossy, since in this case
the power transmitted out of a given region will necessarily be less than the
power transmitted into the given region, since some power must be dissi-
pated as heat in the lossy layer. The complex Poynting theorem on the other
hand accounts for not only all power transmitted into and out of a given
region but also all power dissipated and energy stored in the region. In a
given computation, if the surface and volume integrals of the complex
Poynting theorem do not agree precisely, some degree of numerical error
has been made in the computation. If the agreement is too poor, most likely
a signi®cant computational error has been made somewhere in the calcula-
tions, and it is most likely that the computations cannot be trusted.
A second way that the complex Poynting theorem is an aid to EM ®eld
analysis is that it can give insight into the way that energy is stored and
power is dissipated in a given region of space. Often in making EM ®eld
plots, the plots of the individual ®eld components, either electric or mag-
netic, can be deceptive, since, for example, the ®elds can appear large but in
reality be standing waves, which are actually transmitting very little real
power into a system. Plots of the energy stored and power dissipated then
give great insight into how EM radiation is actually interacting with a
material at a given place in space.
In what follows, both the state variable method (in conjunction with k-
space analysis) and the complex Poynting theorem will be applied to study a
wide variety of different EM planar re¯ection and transmission problems.
Section 2.2 will consider one of the simplest possible cases, namely when a
normally incident plane wave impinges on an isotropic lossy material slab.
Section 2.3 will study the case when an oblique incident plane wave impinges
on an anisotropic layer. Section 2.4 will develop the general 4 4 state
variable equations that apply to re¯ection and transmission through a gen-
eral bi-anisotropic layer. The analysis will apply to the case when the inci-
dent radiation is an oblique arbitrarily polarized plane wave. The complex
Poynting theorem will also be applied to this case.
Section 2.5 will consider cases when EM sources that are not plane
waves impinge on an anisotropic layer. In this section k-space theory is used
to decompose the EM source into a plane wave Fourier spectrum from
which a tractable analysis can be carried out. In particular, the cases of a
waveguide±¯ange system that radiates into an anisotropic lossy layer are
considered. The expression for the wave slot admittance is developed. In this
20 Chapter 2
2.2.2 Analysis
To begin the analysis in this section we assume that all propagation is at
normal incidence and that the EM ®elds of the system in Regions 1, 2, and 3
~ y;
in an
x; ~ z
~ coordinate system are given by
Spectral State Variable Formulation 21
!
~ a^ x
E ` Ex`
y
2:2:1
!
~ a^ z
H ` Hz`
y ` 1; 2; 3
! !
r~ E ` j!~ ` H `
2:2:2
! !
r~ H ` j!~` E `
@Ex`
j!~ ` Hz`
@y~
2:2:3
@Hz`
j!~` Ex`
@y~
@Sx`
j ` Uz`
@y
2:2:4
@Uz`
j ` Sx`
@y
Letting
Sx` 0 j`
V A
2:2:5
Uz` j` 0
(and dropping the ` subscript for the moment) we may write Eq. 2.2.5 in the
general state variable from
@V
y
AV
y
2:2:6
@y
AV qV 2:2:7
X
N
V Cn Vn eqn y
2:2:8
n1
d
V eqn y A
Vn eqn y
2:2:9
dy n
or
q2 0 2:2:10
j2 0 j 00 0 j 00 0 2:2:11
!
`
E n Sxn` eqn` y a^ x
2:2:14a
!
` 1
H n Uzn` eqn` y a^ z
0
where
Sxn` 1 n 1; 2
j
` j
`
Uz1` Uz2`
22:14b
` `
` ` j`
Since the medium is linear, a superposition over the modes in Eq. 2.2.14
gives the total ®eld in any region. The total electric and magnetic ®elds
which can exist in Regions 1, 2, and 3 is given by
!
` X
2
!
`
E Cn` E n
2:2:15a
n1
!
` X
2
!
`
H Cn` H n
2:2:15b
n1
where Cn` are general complex coef®cients that need to be determined from
boundary conditions.
As a cross-check of the solution we note that for any region (suppres-
sing the ` subscript and superscript),
1 @Ex 1 j @Ex
Hz
2:2:16
j!~ @y~ 0 @y
Ex C1 exp
y C2 exp
y
Spectral State Variable Formulation 25
which is the same solution as Eq. 2.2.15b when the eigenvectors of Eq.
2.2.14 are used.
In addition to the ®eld amplitudes of the electric and magnetic ®elds,
another important quantity to calculate is the time-averaged power that
passes through any layer parallel to the material interface. This is explained
in detail in the next subsection.
!t
where J is a general electric displacement, conduction and source current
!t
term and M represents the generalized magnetic current. Mathematically
these currents are given by
!t ! !i
J j! ~ 0 j ~ 00 E J
2:2:19
!t ! !i
M j! l~ 0 j l~ 00 H M
2:2:20
!i !i
where J and M are impressed source terms, and we have assumed that the
permittivity and permeability are complex anisotropic quantities. After
some algebra, we obtain from Eq. 2.2.18,
where
! !i ! !i
Ps 1
2 E J H M d V~
~
V
(source power)
! !
Pf 12 E H ^a^ n d S~
S~
where
! !
PIN 1
2 E H ^
a^ y d S~
2:2:24
S~ y~ ~y
! !
POUT 1
2 E H ^
a^ y d S~
2:2:25
S~ ~ y~
y
The minus sign in Eq. 2.2.25 is a result of the fact that the outward normal
on the y~ end cap is a^ y . Using Eqs. 2.2.21 and 2.2.23, we ®nd that the
complex Poynting theorem for the present problem can be written as
are known (see Eq. 2.2.15), with C11 E0 , where E0 is the incident ampli-
tude (volts/m), and C23 0 also, since there is no re¯ected wave from
Region 3. As the coef®cient C21 represents the complex amplitude of the
re¯ected ®eld in Region 1, we let C21 R, and since the coef®cient C13
represents the complex amplitude of the transmitted ®elds in Region 3, we
let C13 T. Using these coef®cients, the ®elds in Regions 1, 2, and 3 are
given by (see Fig. 1).
Region 1
Region 2
Region 3
The Ex
3 and Hz
3 ®elds have been written with a exp
3
y L in order to
refer the phase of the T coef®cient to the y L boundary.
The boundary conditions require that the tangential electric and mag-
netic ®elds match at y 0, L. Matching of the tangential electric and
magnetic ®elds at y 0 and y L leads to four equations in four
unknowns, from which the EM ®elds in all regions can be determined. It
is convenient to use the electric ®eld equations at the boundaries to eliminate
the unknowns in exterior Regions 1 and 3, thus reducing the number of
equations from four to two. When we do so, we ®nd that
2
1
E a11 C12 a12 C22
1 0
2:2:31
0 a21 C12 a22 C22
30 Chapter 2
where
2
1
2
a11 a12 1
2 1 1 2
2
3
a21 exp
2 L a22 2 3 exp
2 L
2 3 2 3
2:2:32
Also
R E0 C12 C22
2:2:33
T C12 exp
2 L C22 exp
2 L
where
where
yout yout > L
y2
L yout < L
Spectral State Variable Formulation 31
where
(
L yout > L
y3
yout yout < L
PWM PWM1 PWM2 PWM3
2
yin
2
PWM1 10 1 E0 exp
1 y R exp
1 y dy
1 0
2
0
2
PWM2 20 2 c12 exp
2 y c22 exp
2 y dy
2 y2
2
L
PWM3 30 3 T exp
3
y L2 dy
3 y3
PIN j 1 E0 exp
1 yin R exp
1 yin E0 exp
1 yin
1
R exp
1 yin
POUT j 2 c12 exp
2 yout c22 exp
2 yout c12 exp
2 yout
2
c22 exp
2 yout
32 Chapter 2
3
POUT j T exp
2
yout L2
3
Figure 2 The Ex electric ®eld (magnitude, real and imaginary parts) plotted versus
the distance y~ from the incident side interface is shown.
leave the sum constant and equal to Re
PIN . Also plotted in Fig. 3 is the
Im
POUT and the energy difference term PWE PWM . One observes from
these plots that the Im
POUT and PWE PWM vary sinusoidally in Region
2 and that the nonconstant portions of these curves are out of phase with
one another by 180 . Thus the sum of Im
POUT and PWE PWM is a
constant equal to Im
PIN . Thus the imaginary part of the power is
exchanged periodically between Im
POUT and PWE PWM so as to
keep the Im
PIN a constant throughout the system. Figure 4 shows plots
of the electric and magnetic energy and power stored and dissipated in the
Poynting box, again versus the distance y~OUT . As can be seen from Fig. 4,
the electric and magnetic stored energy terms PWE and PWM are nearly
equal to each other.
Figure 3 Plots of the real and imaginary parts of PIN and PBOX as a function of the
distance y~ OUT .
2
1
E a11 C12 a12 C22
2:2:35
1 0
where a11 and a12 have been de®ned previously. At y~ L~ the tangential
component of the electric ®eld must vanish due to the presence of the metal.
This leads to the equation
From these equations C12 and C22 can be determined as well as all other
coef®cients in the system.
Figure 5 shows the Re
Ex , Im
Ex , and jEx j plotted versus the dis-
tance y~ y~ from the Region 1±2 interface, using the material parameter
Spectral State Variable Formulation 35
Figure 4 Plots of the electric energy term, magnetic energy term, power stored, and
power dissipated in the Poynting box, vs. the distance y~ OUT .
values of Section 2.2.4. As can be seen from Fig. 5, the presence of the EPC
in Region 3 causes a larger standing wave (SWR) than was observed when a
free space occupied Region 3. One also notices that the presence of the EPC
causes more internal re¯ection within the slab layer, Region 2, as can be seen
by the increased ripple or decaying SWR pattern displayed by the jEx j plot.
Figure 6 shows the various normalized power terms associated with the
complex Poynting theorem of Eq. 2.2.34. Figure 6 uses the same geometry
as Fig. 3. The only difference between Fig. 3b and Fig. 6 is that an EPC is in
Region 3 of Fig. 6, whereas free space was in Region 3 of Fig. 3. As can be
seen in Fig. 6, as in Fig. 3, the complex Poynting theorem is obeyed to a high
degree of accuracy since the real and imaginary part of PIN (solid line) and
POUT (cross) agree with each other very closely. We also notice from Fig. 6
that a higher oscillation of PWE PWM and Im
POUT occurs than in Fig.
2. This higher internal re¯ection in the slab is caused by the high re¯ectivity
of the EPC at the Region 2±3 interface.
Figure 7 shows the plot of normalized re¯ected power (re¯ected
power/incident power, db) of a uniform slab that results when a plane
wave is normal to the slab. Region 3 is an EPC, and in Region 2, 2 7
36 Chapter 2
Figure 5 Plots of the Re Ex , Im Ex , and jEx j plotted versus the distance y~ .
j3:5 and 2 2:5 j0:2. In this ®gure, the normalized re¯ected power is
plotted versus the slab length L. ~ As can be seen from Fig. 7, at a slab
~
thickness of L 0:066 the re¯ectivity of the layer drops sharply (about
21 db down from the re¯ection that would occur from a perfect conductor
alone). At this slab thickness the layer has become what is called a ``radar
absorbing layer'' (RAM), since at this slab thickness virtually all radiation
illuminating a perfect conductor with this material will be absorbed as heat
in the layer and very little will be re¯ected. Thus radar systems trying to
detect a radar return from RAM-covered metal objects will be unable to
detect signi®cant power. It is interesting to note that only a very thin layer of
RAM material is needed for millimeter wave applications. For example, at
millimeter wavelengths (95 GHz), L~ 0:066 0:2088 mm.
Figure 6 Plots of the various normalized power terms associated with the complex
Poynting theorem of Eq. 2.2.34. This ®gure uses the same geometry as Fig. 3.
The material slab, like the layer considered in Section 2.2.2, is assumed to be
bounded on both sides by a uniform lossless dielectric material that extends
to in®nity on each side. For this analysis we locate the origin of the coordi-
nate system at the current source and label the different regions of the EM
system as shown in Fig. 8. Following precisely the same state variable EM
analysis as we followed in Section 2.2.2, we ®nd that the general EM ®eld
solutions in each region are given by
Region 1 0
0
Ex
1 C11 0 exp
1 0
y L C21 0 exp
1 0
y L C11 0 0
2:2:37a
0 0 j
1 0
Uz
1 0 Hz
1 C 0 exp
1 0
y L
2:2:37b
1 0 21
38 Chapter 2
Region 1
Region 2
Region 3
Figure 8 Plots of the Re Ex , Im Ex , and jEx j plotted versus the distance y~ .
1 0 =1 0
1 =1
C11 C21 exp
2
1 L
2:2:41
1 0 =1 0
1 =1
Matching the tangential electric and magnetic ®elds at the Region 2±3 inter-
face and eliminating the C13 coef®cient it is found that
3 =3
2 =2
C12 C22 exp
2
2 L
2:2:42
3 =3
2 =2
j
1 j
2
1C21 1C22 0 J
2:2:44a
1 2
1C21 1C22 0
2:2:44b
where
! !
PS 0 E J s 0 Ex Js y0
~
2:2:46
~
y0
j
1 j
2
0 J
C11 C21
C C22
2:2:47
1 2 12
Thus
j
1 j
Ps
C11 C21
C11 C21 2
C12 C22
2:2:48
1 2
Spectral State Variable Formulation 41
Figure 9 Plots of different power terms that make up the complex Poynting the-
orem of Eq. 2.2.45 plotted versus the distance y~ OUT .
2
POUT j C12 exp
2 yOUT C22 exp
2 yOUT
2
2:2:44b
C12 exp
2 yOUT C22 exp
2 yOUT
when yOUT < L . The other terms in Eqs. 2.2.45 are given in Eq. 2.2.34.
42 Chapter 2
Figure 8 shows the Re
Ex , Im
Ex , and jEx j electric ®elds plotted
versus the distance y from the Region 1±2 interface. As can be seen from
Fig. 8, the presence of the electric current source in a lossy medium causes
the electric ®eld to be greatest at the source location and attenuate as dis-
tance increases from the source. Because the regions are different to the left
and right of the source, the ®elds are not symmetric about the source loca-
tion. In observing Fig. 8 one notices that the Re
Ex , Im
Ex , and jEx j are all
continuous at the different interfaces as they must be to satisfy EM bound-
ary conditions. Figure 9 shows different power terms that make up the
complex Poynting theorem of Eq. 2.2.45 plotted versus the distance y~ OUT .
As can be seen from Fig. 9 the real and imaginary parts of PS PSOURCE
(cross) and PBOX (solid line) agree with each other to a high degree of
accuracy, thus showing that the complex Poynting theorem is being obeyed
numerically for the present example. One also observes that as the distance
y~ OUT increases, the power dissipated PD increases, Re
POUT decreases,
and both change so as to leave the sum constant and equal to Re
PS .
Also plotted in Fig. 9 is the Im
POUT and the energy±power difference
PWE PWM . One observes from these plots that the Im
POUT and the
energy±power difference PWE PWM vary sinusoidally in Region 2 and
that the nonconstant portions of these curves are out of phase with one
another. Thus the sum of Im
POUT and PWE PWM is a constant equal
to IM
PS . Thus the imaginary part of the power is exchanged periodically
between Im
POUT and PWE PWM so as to keep the Im
PS a constant
throughout the system. Although the EM ®elds were excited by an electric
current source in Fig. 9 rather than a plane wave as in Fig. 3, the complex
Poynting numerical results in the two ®gures are similar.
We assume that the plane wave is polarized with its electric ®eld in the
plane of incidence of the EM wave. The dielectric slab is assumed to be
characterized by a lossy anisotropic relative dielectric permittivity tensor
where xx , xy , yx , yy , and zz are nonzero and the other tensor elements
are zero. The geometry is shown in Fig. 10. The slab's relative permeability
is assumed to be isotropic and lossy and characterized by 0 j 00 . The
basic analysis to be carried out is to solve Maxwell's equations on the
incident side (Region 1), in the slab region (Region 2), and on the trans-
mitted side (Region 3), and then from these solutions to match EM bound-
ary conditions at the interfaces of the dielectric slab.
ky3
Ex
3 Sx
3
y exp
jkx x T exp
jky3
y L exp
jkx x
3
Cx13 exp
13 y Cx23 exp
23 y exp
jkx x
2:3:4
kx
Ey
3 Sy
3
y exp
jkx x T exp
jky3
y L exp
jkx x
3
Cy13 exp
13 y Cy23 exp
23 y exp
jkx x
2:3:5
0 Hz
3
3
Uz
y exp
jkx x T exp
jky3
y L exp
jkx x
Cz13 exp
13 y Cz23 exp
23 y exp
jkx x
2:3:6
p
where ky3 3 k2x , T is the transmitted plane wave amplitude, and 3 is
the relative permittivity of Region 3.
In the anisotropic dielectric slab region, Maxwell's equations are given
by
! !
r E jl
0 H
2:3:7
! ! !
r
0 H j D j
E
where we assume that l is a diagonal matrix with xx yy zz . The x
! !
component of D E is given by Dx xx Ex xy Ey xz Ez . The Dy and
Dz are similarly de®ned. In order that the EM ®elds of Region 1 and 3 phase
match with the EM ®elds of Region 2 for all x, it is necessary that the EM
®elds of Region 2 all be proportional to exp
jkx x. (This factor follows
Spectral State Variable Formulation 45
!
E
Sx
ya^ x Sy
ya^ y Sz
ya^z exp
jkx x
2:3:8
!
0 H
Ux
ya^ x Uy
ya^y Uz
ya^z exp
jkx x
Using the fact that the only nonzero EM ®eld components in Region 1 are
Ex , Ey , and Hz , a small amount of analysis shows that in Eqs. 2.3.7 a
complete ®eld solution can be found taking only Sx , Sy , and Uz to be
nonzero with Sz Ux Uy 0. Substituting Eqs. 2.3.8 in Eq. 2.3.7 and
taking appropriate derivatives with respect to x, the following equations
result:
@Sx
jkx Sy jzz Uz
2:3:9
@y
@Uz
jxx Sx jxy Sy
2:3:10
@y
jkx Uz jyx Sx jyy Sy
2:3:11
yx k
Sy Sx x Uz
2:3:12
yy yy
" #
@Sx yx k2x
j kx S j zz U
2:3:13
@y yy x yy z
@Uz xy yx xy
j xx Sx j kx U
2:3:14
@y yy yy z
46 Chapter 2
The above equations are in state variable form and can be rewritten as
@V
AV
2:3:15
@y
where
" #
yx k2x
a11 j kx a12 j zz
2:3:16
yy yy
xy yx xy
a21 j xx a22 j kx
2:3:17
yy yy
where V Sx ; Uz t .
The basic solution method is to ®nd the eigenvalues and eigenvectors
of the state variable matrix A, form a full ®eld solution from these eigen-
solutions, and then match boundary conditions to ®nd the ®nal solution.
The general eigenvector solution is given by
V Vn exp qn y 2:3:18
AVn qn Vn n 1; 2 2:3:19
The longitudinal eigenvector components Syn are given by, using Eq. 2.3.12,
yx k
Syn Sxn x Uzn n 1; 2
2:3:24
yy yy
ky1
E R C1 Sx1 C2 Sx2
2:3:28
1 0
E0 R C1 Uz1 C2 Uz2
2:3:29
ky3
T C1 Sx1 exp
q1 L C2 Sx2 exp
q2 L
2:3:30
3
T C1 Uz1 exp
q1 L C2 Uz2 exp
q2 L
2:3:31
By substituting R and T from Eqs. 2.3.28, 2.3.31 in Eqs. 2.3.29, 2.3.30, the
4 4 system may be reduced to the following 2 2 set of equations
2ky1 ky1 ky1
E0 Uz1 Sx1 C1 Uz2 Sx2 C2
2:3:32
1 1 1
ky3
0 exp
q1 L U Sx1 C1 exp
q2 L
3 z1
2:3:33
ky3
Uz2 Sx2 C2
3
The C1 and C2 can be found from the above in closed form. Using Eqs.
2.3.28, 31, the other coef®cients may be found.
Uz zz00 Uz dy. As can be seen from Fig. 12, the dissipated electric and mag-
netic powers PDE and PDM are zero at y~ out 0 and increase in a monotonic
fashion until y~ out 0:6 where they become constant for y~ out > 0:6. This is
exactly to be expected since the only loss in the system is in Region 2 where
0 y~ out 0:6. We note also that the integrals Pdexy and Pdeyx are complex
and satisfy Pdexy P deyx as expected. Thus Pdexy Pdeyx 2Re
Pdexy . The
integrals Pdexx and Pdeyy are purely real, and thus the electric dissipation
integral PDE is purely real. Note as can be seen from Fig. 12 that although
the total electric dissipation integral is positive, the cross-term contribution
given by Pdexy Pdeyx 2Re
Pdexy is negative. This is interesting as one
would usually associate only positive values with typical power dissipation
terms.
Figure 13 shows plots of normalized energy±power terms as result
from Eqs. 2.2.21±27 using the example of this section. In this ®gure as in
the previous one, the Poynting box has been chosen to extend a half wave-
length into Region 1 (see Fig. 13 inset) and to extend a variable distance y~ out
Spectral State Variable Formulation 51
into Region 2 when y~ out 0:6 and into Regions 2 and 3 when y~ out > 0:6
into Region 3. In this ®gure Pywe xx , Pywe xy , etc. are given by the integrals
0
Pwexx Sx xx Sx dy, Pwexy Sx xy 0
Sy dy, etc. and PWE Pwexx Pwexy
Pweyx Pweyy . Also PWM Pwmzz Uz zz0 Uz dy: As can be seen from
Fig. 13, the stored electric and magnetic energy±powers PWE are nonzero at
y~ out 0 and increase in a monotonic fashion thereafter. As in the case of the
dissipation power integrals, we note that the integrals Pwexy and Pweyx are
complex and satisfy Pwexy Pweyx . Thus Pwexy Pweyx 2Re
Pwexy . The
integrals Pwexx and Pweyy are purely real, so the electric energy±power inte-
gral PWE is purely real. Note that, as can be seen from Fig. 13, although the
total electric energy±power integral is positive, the cross-term contribution
given by Pwexy Pweyx 2Re
Pwexy is also negative.
Figure 14 shows plots of the real and imaginary parts of the complex
Poynting theorem terms as result from Eqs. 2.2.21±27 given the same
Poynting box as was used in Figs. 12 and 13. In this ®gure, since we are
testing the numerical accuracy of the computation formulae, we let PBOX
POUT PDE PDM j
PWE PWM and compare PIN and PBOX . As can
be seen from Fig. 14, the real and imaginary parts of PIN (cross) and PBOX
52 Chapter 2
Figure 14 Plots of the real and imaginary parts of the complex Poynting theorem
terms as results from Eqs. 2.2.21±27 given the same Poynting box as was used in
Figs. 12 and 13 are shown.
(solid line) are numerically indistinguishable from one another, showing that
the numerical computations have been carried out accurately. Figure 14 also
shows plots of Re
POUT , which decrease as y~ out increases, and PD PDE
PDM (PD is purely real), which increase as y~ out increases. As can be seen from
Fig. 14, the sum of these two quantities, namely Re
POUT PD adds to
Re
PIN , which is constant as y~out increases. It makes sense that the
Re
POUT decreases as y~ out increases, due to increased power loss as y~out
increases. Figure 14 shows plots of Im
POUT and the energy difference term
PWE PWM . As can be seen from Fig. 14, within Region 2 the two terms
are oscillatory, with the oscillatory terms out of phase with one another by
180 . The complex Poynting results of this section are similar to those of
Section 2.2.
Spectral State Variable Formulation 53
2.3, except that the state variable analysis in Region 2 the slab region is more
complicated than in Section 2.3. The analysis will be based on the general
formulations of Refs. 18±29.
! ! !
D ~ E m~ H
2:4:1
! ! !
B ~ E l~ H
2:4:2
~ ~ , and l~ in Eqs.
It is assumed that each component of the four dyadics ~ , m,
2.4.1 and 2.4.2 are in general lossy nonzero complex constants. After sub-
! !
stituting D and B of Eqs. 2.4.1 and 2.4.2 into Maxwell's equations, intro-
ducing the dimensionless dyadics
~ =0 0 j 00 ~ 0 l0
l l= jl 00
p ~ k0 p ~ k0 ~
a a0 ja 00 0 0 b b0 jb 00 0 0 m m
! !
! h ! !i
r E j a E lH
2:4:3
! h ! !i
r H j E bH
2:4:4
! ! !
j exp
j r
S exp
j l U a S
2:4:5
! ! !
j exp
j r
U exp
j S b U
2:4:6
! !
E S
y exp
j
2:4:7
! !
0 H U
y exp
j
2:4:8
p
where 0 0 =0 377
:
If we carry out the differentiations as indicated by Eqs. 2.4.5 and 2.4.6,
! !
noting that S and U depend only on y, we ®nd after canceling the expo-
nential factors that
@S @Sx ! !
a^ x j z kz Sy a^ y kz Sx kx Sz a^ z kx Sy j lU a S
@y @y
2:4:9
@Uz @Ux
a^ x j kz Uy a^ y kz Ux kx Uz a^ z kx Uy j
@y @y
! !
S bU
2:4:10
Useful relations may be found from Eqs. 2.4.9 and 2.4.10, if out of the
six equations given, the longitudinal components Sy and Uy can be elimi-
nated, and equations for only the tangential components Sx , Sz , Ux and Uz
be used. This is highly useful because the tangential components can be
matched with other tangential EM ®eld components at the parallel bound-
ary interfaces.
The longitudinal Sy and Uy components can be eliminated from Eqs.
2.4.9 and 2.4.10 in the following way. We equate the y components of Eqs.
2.4.9 and 2.4.10 and after transposing terms ®nd that
2 3
Sx
" # 6 7
Sy 6 Sz 7
6 7
T22 R24 6 7
2:4:12
Uy 6 Ux 7
4 5
Uz
2 3 2 3
Sx Sx
" # 6 7 " #6 7
Sy 6 Sz 7 w11 w12 w13 w14 6 7
6 17 6 Sz 7
T R6 7 6 7
2:4:13
Uy 6 Ux 7 w21 w22 w23 w24 6 7
4 5 4 Ux 5
Uz Uz
Our next step is to substitute Sy and Uy as given by Eq. 2.4.13 into the
x and z components of Eqs. 2.4.9 and 2.4.10. Doing so thus eliminates all
longitudinal Sy and Uy terms from the equations. After performing consid-
erable algebra it is found that the Sx , Sz , Ux , and Uz components can be
placed in the following state variable form:
2 3
A11 A12 A13 A14
6 7
@V 6
6 A21 A22 A23 A24 7
7
6 7V AV
2:4:14
@y 6 A31 A32 A33 A34 7
4 5
A41 A42 A43 A44
where
A11 j zy w21 azx
azy kx w11
A12 j zy w22
azy kx w12 azz
A13 j zx zy w23
azy kx w13
A14 j zy w24 zz
azy kx w14
Spectral State Variable Formulation 57
A21 j xy w21 axx
axy kz w11
A22 j xy w22
axy kz w12 axz
A23 j xx xy w23
axy kz w13
A24 j xy w24 uxz
axy kz w14
A31 j zx zy w11
bzy kx w21
A32 j zy w12 zz
bzy kx w22
A33 j zy w13 bzx
bzy kx w23
A34 j zy w14
bzy kx w24 bzz
A41 j xx xy w11
bxy kz w21
A42 j xy w12 xz
bxy kz w22
2:4:15
A43 j xy w13 bxx
bxy kz w23
A44 j xy w14
bxy kz w24 bxz
Equation 2.4.14 is in state variable form and its solution can be deter-
mined from the eigenvector and eigenvalues of A as was done in Sections 2.3
and 2.2. The solution is given by
X
4
V Cn Vn exp
qn y
2:4:16
n1
2 3
Sxn
6 Szn 7
Vn 6 7
4 Uxn 5
2:4:17
Uzn
! X 4
!
E Cn E n
2:4:18
n1
! X 4
!
H Cn H n
2:4:19
n1
where
!
E n Sxn a^ x Syn a^ y Szn a^ z exp
qn y j
2:4:20
! 1
Hn Uxn a^ x Uyn a^ y Uzn a^ z exp
qn y j n 1; 2; 3; 4
0
2:4:21
and
Syn
w24 Vn
2:4:22
Uyn
! !
E I S I exp
j I
2:4:23
! !!
0 H I UI HI exp
j I
!
I kI!
r kx x ky1 y kz z
2:4:24
where
! !
k I kx a^ x ky1 a^ y kz a^ z r xa^ x ya^ y za^ z
2:4:25
Spectral State Variable Formulation 59
1=2
ky1 1 1 k2x k2z 6 0
2:4:26
It is further assumed that the wave vector values kx , kz are known and given
and that the incident plane wave polarization is speci®ed by known and
given values of SxI and SzI . From Maxwell's equations and the assumed
!
known value of k I , the other ®eld components of the incident wave are
given by
kx k
SyI S z S SxI SzI
2:4:27
ky1 xI ky1 zI
1
UxI ky1 SzI kz SyI
2:4:28
1
1
UyI k S kx SzI
2:4:29
1 z xI
1
UzI k S ky1 SxI
2:4:30
1 x yI
! ! ! !
E R S R exp
j R ; 0 H R U R exp
j R
2:4:31
!
R kR!
r kx x ky1 y kz z
2:4:32
where
!
k R kx a^ x ky1 a^ y kz a^z
2:4:33
If the tangential values of the electric ®eld SxR and SzR can be found, it turns
out from Maxwell's equations that the other ®eld components of the
re¯ected wave are given by
60 Chapter 2
kx kz
SyR S S SxR SzR
2:4:34
ky1 xR ky1 zR
1
UxR k S kz SyR
2:4:35
1 y1 zR
1
UyR k S kx SzR
2:4:36
1 z xR
1
UzR k S ky1 SxR
2:4:37
1 x yR
! ! ! !
E T S T exp
j T 0 H T U T exp
j T
2:4:38
!
T k T
!
r La^ y kx x ky3
y L kz z
2:4:39
where
!
k T kx a^ x ky3 a^y kz a^ z
2:4:40
1=2
ky3 3 3 k2x k2z 6 0
2:4:41
If the tangential values of the electric ®eld SxT and SzT can be found, it turns
out from Maxwell's equations that the other ®eld components of the trans-
mitted wave are given by
kx k
SyT S z S 0 SxT 0 SzT
2:4:42
ky3 xT ky3 zT
1
UxT ky3 SzT kz SyT
2:4:43
3
1
UyT k S kx SzT
2:4:44
3 z xT
1
UzT kx SyT ky3 SxT
2:4:45
3
Now that the general EM ®elds have been found in Regions 1±3 of
space (see Fig. 15), as mentioned earlier, the next step is to match EM
boundary conditions at the Region 1±2 and Region 2±3 interfaces. The
boundary conditions for the present problem require that the tangential
Spectral State Variable Formulation 61
P P P
We letPSAx 4n1 Cn Sxn , SAz 4n1 Cn Szn , UAx 4n1 Cn Uxn , and
UAz 4n1 Cn Uzn , evaluate the equations at y 0 and y 0 , cancel
the exp
j
kx x kz z factor and express the unknowns of Eqs. 2.4.46,
SxR and SzR , in terms of SAx , SAz , UAx , and UAz according to the relations
where
VxI SxI kz SyI kz SzI 2ky1 kz
2:4:49
VzI SxI 2ky1 kx SyI kx SzI kx
The terms VxI , VzI represent the known source terms associated with the
incident plane wave. If we further substitute the sums in SAx , SAz , UAx , and
UAz and collect on the unknown coef®cients Cn in the sums, we ®nd
62 Chapter 2
X
4
VxI Cn Sxn kz Szn ky1 kz 1 Uxn
n1
2:4:50
X
4
VzI Cn Sxn kx ky1 Szn kx 1 Uzn
n1
Substituting
X
4
SxT Cn exp
qn LSxn
n1
2:4:52
X
4
SzT Cn exp
qn LSzn
n1
into Eqs. 2.4.51 and following a procedure very similar to the Region 1±2
interface we ®nd that
X
4
0 Cn exp
qn L Sxn kz 0 Szn ky3 kz 0 3 Uxn
n1
X
4
0 Cn exp
qn L Sxn kx 0 ky3 Szn kx 0 3 Uzn
n1
2:4:53
Spectral State Variable Formulation 63
1 n h! i h ! !i o
0 r U
kx ; y exp
j jb U S exp
j dkx
1
2:5:4
Setting the quantities in the curly brackets of Eqs. 2.5.3 and 2.5.4 to zero
and performing a small amount of algebra it is found that
! ! !
j exp
j r
S exp
j l U a S
2:5:5
! ! !
j exp
j r
U exp
j b U S
2:5:6
These equations are of the same form as Eqs. 2.4.5 and 2.4.6 if we take
kz 0. We thus ®nd in Region 2 that the variable equations given in Section
2.4 represent a general solution of the problem being studied here.
Spectral State Variable Formulation 67
R0
0 HzR exp
jk3
y L
2:5:10
3
Ex
3 ExI ExR
2:5:11
Hz
3 HzI HzR
2:5:12
@V a a12
AV A 11
2:5:13
@y a21 a22
where
" #
yx k2x
a11 j kx a12 j zz
2:5:14
yy yy
xy yx xy
a21 j xx a22 j kx
2:5:15
yy yy
and where V Sx ; Uz t . These are in fact the same exact equations as were
studied in Section 2.3 except that here Sx and Uz represent k-space Fourier
amplitudes rather than spatial EM ®eld components as they did in Section
2.3. The general solution to Eqs. 2.5.13 in Region 2 is
1 "X
2
#
Ex
2 Cn Sxn exp
qn y exp
jkx x dkx
2:5:16
1 n1
1 "X
2
#
Ey
2 Cn Syn exp
qn y exp
jkx xdkx
2:5:17
1 n1
1 "X
2
#
0 Hz
2 Cn Uzn exp
qn y exp
jkx xdkx
2:5:18
1 n1
Spectral State Variable Formulation 69
where
Sxn 1
2:5:19
a qn
Uzn 11
2:5:20
a12
yx k
Syn S xU n 1; 2
2:5:21
yy yn yy zn
and where
1=2
q1 0:5a11 a22 0:5
a11 a22 2 4a12 a21
2:5:22
1=2
q2 0:5a11 a22 0:5
a11 a22 2 4a12 a21
2:5:23
where
ky1 1 1 k2x 1=2 1 1 k2x 0
2:5:27
jk2x 1 1 1=2 1 1 k2x < 0
The minus sign of ky1 (or branch of ky1 ) was chosen on the physical grounds
p
that the integrals converge as y ! 1 when the jkx j > 1 1 .
To proceed it is necessary to match EM boundary conditions at the
Region 1±2 and Region 2±3 interfaces. To facilitate the Region 2±3 EM
boundary matching, it is convenient to represent and replace the waveguide
!
aperture slot with an equivalent magnetic surface current M s backed by an
electrical perfect conductor. The boundary condition equation to determine
!
the equivalent magnetic surface current M s backed by an in®nite ground
plane is
70 Chapter 2
1
!
2 !
3 !
a^ y E E A Ms
2:5:28
y L
y L
where
!
3
E 0
2:5:29
y L
where
EA
x represents the x-component of the electric ®eld in the aperture. Using
Eq. 2.5.30 it is found that the equivalent magnetic surface current is given by
x
1
!
M s a^ z EA
x rect a^ z M
kx exp
jkx xdkx
2:5:32
2B 1
!
The last part of Eq. 2.5.32 expresses M s in k-space. For the present problem
the aperture electric ®eld is given by Eq. 2.5.30 evaluated at y L . Thus
EA is a constant given by EA E0 R0 . Using this value of EA it is found
from Fourier inversion that
BEA sin
kx B
M
kx
2:5:33
kx B
We will now present the boundary value equations at the Region 1±2
and Region 2±3 interfaces. At the Region 1±2 interface, matching the tan-
gential electric ®eld (Ex -component) and the tangential magnetic ®eld (Hz -
component) on the y 0 (in Region 1) and y 0 (in Region 2), and at
the Region 2±3 interface, matching the tangential electric ®eld (Ex -compo-
!
nent) at y L (Region 2) to the magnetic surface current M s , and then
recognizing that the Fourier amplitudes of all the k-space integrals must
equal each other for all values of kx , we ®nd the following equations:
Spectral State Variable Formulation 71
ky1
1 X2
Uz
kx Cn Sxn
2:5:34
1 n1
X
2
Uz
1
kx Cn Uzn
2:5:35
n1
X
2
Cn Sxn exp
qn L M
kx
2:5:36
n1
If we eliminate Uz
1
kx from Eqs. 2.5.34±36 we are left with a 2 2 set of
equations from which to determine C1 and C2 in terms of M
kx . We ®nd
that
T2 M
kx
C1
2:5:37
T1 exp
q2 L T2 exp
q1 L
T1 M
kx
C2
2:5:38
T1 exp
q2 L T2 exp
q1 L
where
ky1
Tn a12 a qn n 1; 2
2:5:39
1 11
The right-hand side of Eq. 2.5.41 integrates after using Eq. 2.5.12 to
B
1 1
0 Hz
3 dx E R0
2:5:42
2B B y L 3 0
72 Chapter 2
Thus
B
1 1
E R0 0 Hz
2 dx
2:5:43
3 0 2B B y L
~ Hz
3
y
Y
y
2:5:44
Ex
3
y
B
If we replace 1=3 E0 R0 by
1=2B B 0 Hz
2 jy L dx using Eq. 2.5.43,
we ®nd that
(
)
1 1 1 B
2
Y~ A` 0 H z dx
2:5:47
0 EA 2B B y L
Spectral State Variable Formulation 73
where
2
B T2 Uz1 exp
q1 L T1 Uz2 exp
q2 L sin
kx B
Y
kx
T1 exp
q2 L T2 exp
q1 L kx B
2:5:50
We remind readers that in the above equation, the quantity in square brack-
ets is a complicated function of kx , and the Uzn , n 1; 2, are eigenvector
components associated with the magnetic ®eld in Region 2. Once the inte-
gral in Eq. 2.5.49 is carried out, YA` is known and then a relation between E0
and R0 can be found through the equation
1 E0 R0
~
YA` 0 Y
y
2:5:51
yL 3 E0 R0
R0 1=3 YA`
r
2:5:52
E0 1=3 YA`
(or even singular) near the points kx k1 , and thus signi®cant numerical
error can occur if a very ®ne numerical integration grid is not used around
these points. In the present section using the quadrature formulas
kx k1 cos
u, 0 u , in the visible region and kx k1 cosh
u,
0 u 1, in the invisible region was employed to integrate the YA` inte-
gral. These formulas provide a very dense grid near kx k1 and thus
provide an accurate integration of the YA` integral.
Harrington [3, p. 183, Eqs. 4-104, 4-105] de®nes an aperture admit-
tance for the present slot radiator problem through the Parseval power
relation
P~
Y~ A
2:5:53
jVj2
where E x
k~x and H z
k~x are the Fourier amplitudes (or k-space pattern
space factors) of the Ex
2 electric ®eld and the Hz
2 magnetic ®eld, respec-
tively. P~ has units of (watt/meter)=(volt amp/meter), so Y~ A has units of
(
meter 1 (or mho/meter). Substituting the EM ®eld solutions derived
earlier in Eq. 2.5.54, it is found that the aperture admittance Y~ A as de®ned
by Eq. 2.5.54 is very closely related to the transmission line load admittance
expression Y~ A` . It is related by the equation
Y~
Y~ A A`
2:5:55
2B~
where
1
Uz
1 A
kx exp
jkx x jky1 ydkx
2:5:58
1
and where
BEA T2 T1 sin
kx B
A
kx 1
ky1 T1 exp
q2 L T2 exp
q1 L kx B
2:5:59
We note in passing that Eq. 2.5.58 for Yz
1 is identical to that given by
Ishimaru [4, Chapter 14] when one (1) lets the dielectric layer be isotropic,
(2) lets the slot waveguide width 2B~ approach zero while holding the voltage
potential difference between the parallel plate conductors constant, and (3)
makes the correct geometry association between Ishimaru's analysis and the
present one.
Ishimaru [4] shows, by using the method of steepest descent, that the
integral in Eq. 2.5.58 asymptotically approaches as ! 1 the value
2 1=2 j
Uz
1 F
k1 sin
'c exp jk1
2:5:60
k1 4
where
where k1 sin
c and k1 cos
c have been substituted for kx and ky1 , respec-
tively, in Eq. 2.5.58. To describe the radiation from the waveguide aperture
and material slab system in the far ®eld
! 1 we plot the normalized
radiation intensity, which here is de®ned as the radiation intensity, ! 1,
divided by the total radiation intensity integrated from c =2 to
c =2. This quantity is called the directive gain D
c . Applying this
de®nition and using Eqs. 2.5.60 and 2.5.61 after cancelling common factors
we ®nd
where xx 2, xy 0:3, yx 0:9 j0:2, and yy 2:1. The value of zz is
immaterial to the present analysis and is not speci®ed here. For all calcula-
tions in this section the slot width has been taken to be 2B~ 0:6.
Figure 19 shows a plot of the Y
kx aperture admittance integrand
when the layer thickness has been taken to L~ 0:6. As can be seen from
Fig. 19 for the values used in the present example, the integrand converges
fairly rapidly for values of jkx j 5k1 5. An inspection of Eq. 2.5.50 for
Y
kx shows that for kx large the integrand approaches 1=k3x and thus is
guaranteed to converge. In an inspection of Fig. 19 one sees also that the
integrand Y
kx is not exactly symmetric with respect to the kx variable.
This is a result of the slot radiating through an anisotropic rather than an
isotropic medium. For the present example, the boundary of the visible and
invisible [1] (i.e., propagating and evanescent) radiation range is at
kx k1 1. One observes from Fig. 19 the effect that the discontinuous
ky1 function of Eq. 2.5.27 has on the Y
kx integrand in the kx regions near
kx k1 1. Figure 19 also lists values of the two lowest magnitude poles
which were associated with the Y
kx integrand. The two pole locations in
Spectral State Variable Formulation 77
the complex kx plane
kxp1 1:541 j0:218 and kxp2 1:567 j0:146
were nonsymmetric because of the anisotropy of the material slab. The
values of the poles were listed as they in¯uence the real kx integration
when the kx integration variable passes close to the poles' location.
Figure 20 shows a plot of the YA` aperture load admittance as a
function of the layer thickness L.~ At a value of L~ 0 the layer does not
exist, and the waveguide aperture radiates into free space. As L~ increases,
the real and imaginary parts of the aperture admittance are oscillatory up to
a value of about L~ 1, where it starts to approach a constant value.
Figure 21 shows a plot of the directive gain as a function of the angle
c . One observes from this ®gure that the radiation pattern is concentrated
in a 90 angle around the broadside direction and one also observes that the
radiation pattern is asymmetric in the angle c , with the peak radiation
value occurring at about angle c 10 . The asymmetry is caused by the
fact that the slot has radiated through an anisotropic material slab.
78 Chapter 2
Figure 20 A plot of the YA` aperture load admittance as a function of the layer
~
thickness L.
~ !
r
!
H a j!~ E a
2:6:1
~ !
r Ea
!
~ a
j!lH
2:6:2
where the subscript ``a'' stands for anisotropic. We express the spatial elec-
tric and magnetic ®elds in a two-dimensional k-space Fourier transform as
1
1
! ! !
Fa ~ exp
j
k~x x~ k~y yd
f a
k ; z ~ k~x d k~y
2:6:3
1 1
! !
where k k~x a~ x k~y a~ y ; F a represents, !
respectively,
! either the electric ®eld
! !
E a or magnetic ®eld H a , and where f a
k ; z ~ represents,
! respectively,
either the spectral amplitude of ! the electric ®eld !e
k ; ~
z or the spectral
! a
amplitude of the magnetic ®eld h a
k ; z. ~ Substituting the Fourier trans-
forms integrals into Maxwell's equations and collecting coef®cients of the
exponential in Eq. 2.6.3 we ®nd that
! ! !
~ j!~ !
D h a
k ; z ~
e a
k ; z
2:6:4
! ! !
D! ~
e a
k ; z j!l~ h a
k ; z
~
2:6:5
where
2 3
0 @=@z~ j k~y
6 7
D 4 @=@z~ 0 j k~x 5
2:6:6
j k~y j k~x 0
We ®nd that Eqs. 2.6.4 and 2.6.5 can be put into the form
d ~
xa
z R U ~
xa
z
2:6:8
d z~ ~
ya
z V W ~
ya
z
If X1 and Y1 are solutions of Eq. 2.6.9 that meet the boundary conditions
X1
0 I2 and Y1
0 0 (I2 is a 2 2 identity matrix), then the solution of
Eq. 2.6.8 is given by
~ X1
z
X
xa
z ~
~ 1 1
dc
2:6:10
~ Y1
z
X
ya
z ~
~ 1 1
dc
2:6:11
~ C0
zI
exp
Az ~ 4 C1
zA ~ 2 C3
zA
~ C2
zA ~ 3
2:6:13
~ i 0; 1; 2; 3; satisfy
where Ci
z,
X
3
~
exp
j z kj Ck
z
~ j 1; 2; 3; 4
2:6:14
k0
det I4 A 4 a1 3 a2 2 a3 1 a4 0 2:6:15
~ and y0
z
2.6.3, and using the auxiliary ®eld quantities x0
z ~ de®ned ana-
logously to Eqs. 2.6.10 and 2.6.11, the free space dipole can be written as
" #
1 0 jsgn
z~ z~ 0 0
~ 2
x0
z
8 jsgn
z~ z~ 0 0 j k~2
~0 1
2:6:15
!
Q exp
j k !
0
~0 jz~ z~ 0 j
" #
1
0
!0
1
0 sgn
z~ z~ 0 k~2
!0 1
~ 2
y0
z
8 0 !0
~0 1 0
2:6:16
!
Q exp
j k~ !
0
~0 jz~ z~ 0 j
and
!0 !0
r~ ~ z~ 0 a^ z
2:6:18
For the ®eld re¯ected from the anisotropic layer (an outgoing wave moving
away from the layer),
" #" #
j
~0 0 F
~
xr
z exp
~0 jz~ ~
dj
2:6:19
0 !0 D
" #" #
0 j
~0 F
~
yr
z exp
~0 jz~ ~
dj
2:6:20
!0 0 D
84 Chapter 2
! ! !
where xr
z ~ are determined from !
~ and yr
z ~ and h r
k ; z
e r
k ; z ~ in a
manner! similar to
! ! the way x0
~
z and y 0
~
z were! determined
! !from
! ~ and h 0
k ; z
e 0
k ; z ~ or xa
z
~ and ya
z~ from ! ~ and h a
k ; z.
e a
k ; z ~
The ®nal step in obtaining the solution is to boundary match the
tangential EM ®elds at z~ d. ~ The total EM ®elds for z~ d~ is the sum of
the incident and re¯ected ®elds, and the total ®elds for z~ d~ is the aniso-
tropic slab ®eld; thus equating these total ®elds (using the three sets of
auxiliary vectors) we have
~ c x0
d
xa
d ~ xr
d
~
2:6:21
~ y
d
ya
d ~ y
d
~
2:6:22
0 r
On substituting Eqs. 2.6.15±20 into Eqs. 2.6.21 and 2.6.22, the following set
of 4 4 equations is obtained, from which all unknown constants of the
system can be found. The 4 4 equations are
" #" #
j
~0 0 F
c ~
x0
d
2:6:23
0 !0 D
" #" #
0 j
~0 F
~
Y1
dX ~ 1
dc ~
y0
d
2:6:24
!0 0 D
The general ferrite tensor l
0 ; 0 and the plasma tensor
0 ; 0 are com-
puted by applying unitary transformations to l
0 0; 0 , and
0
0; 0 , respectively. The expressions for these tensor are referred to in
[32]. For uniaxial media the N^ vector represents the orientation of the
optical axis. The direction of the radiating dipole is determined by the
unit vector p^ and is parallel to one of the unit vectors a^x , a^y , a^ z .
Figure 23 (kindly supplied to us in corrected form by the authors of
Ref. 32), gives results for E and E relative far ®eld amplitudes for a
ceramic Polytetra¯uoroethylene (PTFE) uniaxial substrate for various opti-
cal axis orientations
0 20 , 40 , 60 , and 80 ). The dielectric constants
Figure 23 Radiation patterns jE j, jE j versus in the 0 (180 ) plane for a
uniaxial substrate with xx 10:70 , zz 10:40 , l 0 I3 , d 1 mm, and
f 30 GHz. The primary source is an electric dipole located at the substrate surface
z 0 d, and its orientation is de®ned with the unit vector .
^ (# IEEE, 1985.)
86 Chapter 2
along the principal axes are ~xx ~yy 10:70 and ~zz 10:40 . In this case
the
0 ; 0 is independent of the 0 angle and l 0 I 3 . The substrate
thickness is taken to be d 1 mm. Both vertical ^ z^ and horizontal ^
x^ dipoles are considered assuming the same excitation. The variation of the
radiation diagrams is noticeable only for the horizontal dipoles, while for
the vertical dipoles there is almost no effect of the optical axis orientation.
The radiation diagrams, as in the case of isotropic substrates, retain their
symmetry with respect to the z-axis.
In treating ferrite substrates it is assumed that
0 ; 0 150 I3 and
that a strong magnetic type of anisotropy is used with ~ 11 0:6750 ,
~ 12 0:494 0 , !0 =! 2:35 [32], !
M, 0 M 0:3Wb=m2 (
being the
magnetomechanical ratio). Corresponding to various biasing static magnetic
®eld orientations, the computed radiation patterns on various constant
^
planes are given in Figs. 24±26 for x-directed dipoles. The radiation fre-
quency is taken f 30 GHz, and the ferrite layer thickness is d 1 mm. In
general there is a strong dependence of the far ®eld to 0 orientation. When
the constant observation plane coincides with the 0 plane (i.e.,
0 0) and the dipole axis is also parallel to this plane, the patterns
are axisymmetric. This symmetry is not exhibited for other observation
planes such as in Fig. 25, where patterns are varying from an almost omni-
directional coverage (0 20 ) to a rather directional diagram
0 80 ).
Figure 24 Radiation patterns jE j, jE j versus in the 90 observation plane
of a ferrite substrate for various 0 angles and 0 0 . The material properties of the
ferrite are ~ 11 0:6750 , ~ 12 0:4940 [32], and ~
0 ; 0 0 I3 , d~ 1 mm. and
f 30 GHz. The dipole axis is along the x-axis
a^ a^ x and is located at the
~ (# 1985, IEEE.)
substrate surface
z~ 0 d.
Spectral State Variable Formulation 87
Figure 25 Radiation patterns jE j, jE j versus for the same parameters as Fig. 24
except the observation plane is 0 . The magnetostatic ®eld is inside the 0 0
plane. (# 1985, IEEE.)
Figure 26 Radiation patterns jE j, jE j versus for the same parameters as Fig. 25
except the observation plane is 0 and 0 45 . (# 1985, IEEE.)
88 Chapter 2
2.6.5 Conclusion
In conclusion of this section a general formulation is presented for the
analysis of an EM ®eld originating from an arbitrary oriented dipole source
in the presence of a grounded general anisotropic layer. The Green's func-
tion is determined by using linear algebra techniques without restriction on
the anisotropic permittivity or permeability. Several numerical examples
have been presented.
Figure 27 Radiation patterns jE j, jE j versus in the 0 plane for a grounded
plasma layer with !c =!p 1:8, !=!p 2:4, d~ 1 mm, and f 30 GHz [32]. The
~ and
^ and it is located on the plasma surface
z~ 0 d
dipole is along the x-axis
^ x,
0 0 . (# 1985, IEEE.)
Spectral State Variable Formulation 89
Figure 28 Re¯ection from an in-plane biased ferrite layer. Biased ®eld
H0 1000
Gauss in the a^x direction; magnetization 2500 Gauss. Transverse magnetic incidence
i 30 and i 40 , ~f 12:80 , and d~ 3 cm. (Copyright 1995, IEEE [26].)
@ ~ ~
w
z Aw
z
2:7:1
@z~
where
2 3
k~x H~ x k~y H~ y
6 k~ H~ k~x H~ y 7
~ z 6 7
w
~ 6 y x 7
2:7:2
4 k~x E~ x k~y E~ y 5
k~y E~ x k~x E~ y
where
2 3
~
exp
1 d 0 0 0
6 7
6 0 ~
exp
2 d 0 0 7
~ ~ 6 7 ~ 1
T
d /6 7/
6 0 0 ~
exp
3 d 0 7
4 5
0 0 0 ~
exp
4 d
2:7:4
The electromagnetic ®elds in the air
z~ d~ and z~ 0 ) can be derived
from a set of transverse electric and transverse magnetic vector potential
functions. This result can be shown to be
2 q 3
ja 0 k~2 k20
6 7
6 7
6 ! b 0 7
~ 6 q 7
w
d
6 0
7
6 0 ~2 7
6 jb k k 27
4 05
!0 a 0
2 q 3
2:7:5
jc 0 k~2 k20
6 7
6 7
6 !0 d 0 7
~ 6 7
w
0 6 q 7
6 7
6
4 jd 0 k~2 k20 7 5
!0 c 0
~
w
d ~ w
0
T
d ~ Qinc
2:7:6
q
where k~ k~2x k~2y and where Qinc is related to the incident plane wave.
For the problem with a current source the right-hand side should be the
corresponding spectral current component. The state variable exponential
matrix method described above is rigorously correct. However in numerical
implementation this method may break down. Without loss of generality it
is assumed that Re
1 Re
2 Re
3 Re
4 . In many practical
applications when Re
1 1, the transition matrix de®ned in Eq. 2.7.4
92 Chapter 2
~
T
d exp
1 dA ~
1 exp
2 dA2
2:7:7
where the singular matrices A1 and A2 do not contain any terms that grow
exponentially. We have
2 3
1 0 0 0
6 7
60 0 0 07
6 7~
A1 r~ 6 7r
1
2:7:8
60 0 0 07
4 5
0 0 0 0
and
2 3
0 0 0 0
60 1 0 0 7
6 7~
A2 r~ 6
1
~ 7r
2:7:9
40 0 exp
3 2 d 0 5
0 0 0 exp
4 ~
2 d
Note that A1 is obtained from Eq. 2.7.4 by replacing the terms of exp
2 d,
exp
3 d, and exp
4 d with 0 and replacing exp
1 d with 1. Since A1 is a
singular matrix, it can be shown that
2 3
a1
6 7
6 a2 7
~ 06 7 0
A1 w
0
c d 6 7
2:7:10
6 a3 7
4 5
a4
~ u
c 0 d 0 exp
1 d
2:7:11
~ v
c 0 d 0 exp
2 d
2:7:12
Spectral State Variable Formulation 93
where u and v are the new variables replacing c 0 and d 0 . With the variable
transformations, we have
2 3 2 q 3
a1 j k~2 k20
6 7 6 7
6 a2 7 6 7
6 7 u 6 !0 7
~ w
0
T
d ~ u6 7 A exp
2 ~
1 d6 q 7
6 a3 7 2 6 7
4 5 6 j k~2 k20 7
4 5
a4
!0
2 q 3
j k~2 k20
6 7
6 7
v 6 !0 7
A2 6 q 7
6 6
7
4 j k~2 k2 7 0 5
!0
2:7:13
Upon inspecting Eq. 2.7.13, one observes why the transformation provides a
stable invertible matrix equation from which to determine the unknown
coef®cients a 0 , b 0 , u, and v (and therefore c 0 and d 0 ). The right-hand side
of Eq. 2.7.13 is a sum of an exponential and two nonexponential terms.
When 1 2 , the exponential term becomes much smaller than the
nonexponential terms. In this case, when the left-hand side is then numeri-
cally computed, the exponential term will make a negligible contribution to
the matrix elements of Eq. 2.7.6, and the nonexponential terms alone will
provide a ®nite and numerically correct value for the matrix elements of the
system. As mentioned earlier, without using this transformation, a row of
exponentially small matrix elements exists, leading to numerical singularity
of the matrix equation.
2.7.4 Conclusion
A numerical algorithm was developed for the computation of EM ®elds in a
generalized anisotropic structure. The proposed method using variable
transformation overcomes the dif®culty frequently encountered in the tran-
sition cascade method, without increasing computational time or memory.
The extension of this technique to multilayer structures is given in detail by
Yang [25].
PROBLEMS
1. Using the wave equation for the electric ®eld, write down the EM
®eld solutions in the three regions in Fig. 1. Assume normal
incidence from Region 1. Show that your results are the same
as the state variable solutions of Section 2.2.
2. If the interface between Regions 2 and 3 in Fig. 2 has a perfectly
electrically conducting surface, write down the state variable
solutions in each of the three regions for normal incidence
from Region 1. Using these solutions and the EM boundary
conditions, solve for all the EM ®elds.
3. Extend the state variable solutions developed in Sec. 2.2 to the
case of normal incidence onto 2 layers sandwiched in air. Assume
that the permeabilities of the layers are equal to that of free
space, and that the layer relative permittivities are 2 and 4.
Determine the condition on layer thicknesses to achieve maxi-
mum re¯ection from the sandwich.
4. Verify the complex Poynting theorem for the solutions to the
two-layer sandwich in Problem 3. Assume the Poynting box to
be of unit cross-sectional area and of suf®cient thickness to
enclose both layers.
5. If the electric current source in Fig. 8 is replaced with a magnetic
current source, ®nd the ®eld solutions for the system.
6. Starting from Eq. (2.3.7), develop the state variable solution for
the case where the permeability is anisotropic (xx ; xy ; yx ; yy ,
Spectral State Variable Formulation 95
REFERENCES
3.1 INTRODUCTION
In the past thirty years the study and use of periodic structures and diffrac-
tion gratings has become increasingly important. Diffraction gratings have
been constructed for applications in the frequency ranges of microwaves,
millimeter waves, far infrared, infrared, optics, and x-rays. Diffraction grat-
ings occur in such applications as holography, memory storage, spectro-
scopy, phase conjugation, photorefractives, image reconstruction, optical
computing, transducers, integrated optics, microwave phased arrays, acous-
tooptics, interdigitated, voltage controlled, liquid crystal displays, and many
other areas. Petit [1], Gaylord and Moharam [2], Solymar and Cooke [3],
and Maystre [4] give extensive reviews on the applications of diffraction
gratings. Chapter 7 of this book cites many references on diffraction grat-
ings in photorefractive materials.
We will give a brief description and overview of the physical makeup
of diffraction gratings. Diffraction gratings have been manufactured and
constructed in many different forms and types. Two main classi®cations
of diffraction gratings are those that are metallic and those that are dielec-
tric. Metallic gratings have grooves that are etched or cut from a ¯at metal
surface. These grooves may be rectangular or triangular in shape.
Triangular grooves are referred to as blazed gratings. Metallic gratings are
operated in the re¯ection mode, as the diffracted waves are re¯ected from
the metal surface. Metallic gratings are also examples of surface relief grat-
ings, as the rectangular or triangular groove shape of the grating is cut from
the ¯at metal surface.
Dielectric gratings are constructed of dielectric materials that are
transparent to the electromagnetic radiation that impinges on it.
Dielectric gratings can be classi®ed into two major types: dielectric gratings
that are surface relief gratings and dielectric gratings that are volume grat-
99
100 Chapter 3
ings. Surface relief dielectric gratings tend to have a large periodic modula-
tion but small thickness, whereas volume dielectric gratings tend to have a
small periodic modulation but a large thickness. The large modulation of
the surface relief grating occurs because the grating material from which the
grating is constructed has a large difference in index of refraction compared
to the medium adjacent to the grating. Dielectric gratings may be operated
in either the transmission mode or the re¯ection mode. Transmission grat-
ings have periods on the order of a few wavelengths with the grating vector
parallel to the grating surface, whereas re¯ection gratings have periods on
the order of a half wavelength and grating vectors perpendicular to the
grating surface. Gratings that are neither exactly parallel or not exactly
perpendicular to the grating surface are referred to as slanted gratings.
Scattering from dielectric diffraction gratings depends strongly on
three main factors, namely the type and strength of the periodic variation
of the index of refraction that exists in the grating, the type of material
(anisotropic or isotropic, nonlossy or lossy) the grating is made from, and
the type of EM wave that is incident on the grating. We will now brie¯y
discuss these three factors.
The periodic variation of the index of refraction that induces diffrac-
tion when a grating is illuminated may consist of many different forms. The
periodic variation may be one-dimensional; it may be two-dimensional, in
which case it is referred to as a crossed grating; or the grating may consist of
two superimposed one-dimensional gratings. In addition to the index vary-
ing in one or two dimensions, the periodic variation of the index of refrac-
tion may vary longitudinally throughout the grating. A sinusoidal surface
relief grating and a triangular blaze grating that has air as an interface are
examples of this type of variation. A surface relief grating is longitudinally
inhomogeneous because at a plane where the groove is deeper, more material
will be included in the duty cycle of the grating that at a plane closer to the
homogeneous air half space.
The type of material that makes up the grating may be isotropic and
nonlossy, like glass; it may be anisotropic, like calcite or LiNbO3 (lithium
niobate); it may be either weakly lossy (e.g., BaTiO3 ) or strongly lossy.
Lossy gratings attenuate the diffracted waves as they propagate through
the system. In anisotropic materials the anisotropy tends to couple the
polarization states of the incident wave in the medium and induce new
polarization states in the system. In anisotropic systems the diffracted
waves consist of ordinary and extraordinary waves coupled together
through the grating vector.
The type of EM radiation that is incident on the grating strongly
in¯uences the diffraction that will result from the grating. The EM radiation
may consist of either a plane wave or a collection or spectrum of plane
Planar Diffraction Gratings 101
waves (e.g., a Gaussian beam). Further, each of these types of waves may be
incident on the grating at an oblique angle and possess an arbitrary polar-
ization. Later in this chapter we will show examples of H-mode (magnetic
®eld in plane of incidence) and E-mode polarization (electric ®eld in plane of
incidence) states that may be used to illuminate a diffraction grating.
Particularly for anisotropic gratings, the type of incident wave and its polar-
ization determine strongly how the EM wave will couple and diffract from
the grating.
Many mathematical analyses and numerical algorithms have been
developed so that the diffraction that occurs from planar gratings can be
predicted. Some of the main diffraction grating methods and algorithms are
(1) coupled wave analysis [5±9], (2) rigorous coupled wave analysis (RCWA)
[2,10±53], (3) coupled mode theory [54±61] (Refs. 57±59 have been referred to
as the Australian method), (4) the differential method [1,62±65], (5) the inte-
gral method [66], (6) the ®nite difference method [67±69], (8) the boundary
element method [70], (7) the unimoment method [71] and (9) other methods
[72,73], which are either closely related to or variations of the methods listed
above. References 74±76 list papers on energy and power conservation in
electromagnetic and electromagnetic diffraction grating systems.
Concerning the ®rst three methods, within the last ten years, several
researchers have been concerned with the problem of improving the con-
vergence or increasing the stability (that is, allowing analysis of thicker
grating structures that have increased grating strength) of the coupled
mode and coupled wave algorithms, and they also have been concerned
with the problem of understanding in the ®rst place, for certain polariza-
tions and material types, the coupled mode and coupled wave algorithms
that are unstable and why they do not converge well.
Just about all the above-mentioned algorithms solve the EM grating
diffraction problem in three basic steps: one must (1) express the EM ®elds
outside the diffraction grating region as Rayleigh series of propagating and
evanescent planes waves whose amplitudes are unknown and are yet to be
determined (the series is transversely periodic with the period equal to grat-
ing period of the periodic structure), (2) by an appropriate method, ®nd a
general solution of Maxwell's equations in the diffraction grating region,
and (3) match EM boundary conditions at the diffraction grating and
homogeneous grating interfaces to determine all the unknown coef®cients
of the diffraction grating system. Most of the methods differ in the way that
Maxwell's equations are solved in the diffraction grating region. We will
now give a brief description of all of these algorithms. This chapter will
primarily focus on the rigorous coupled wave approach. Chapter 6,
Section 6.2 will brie¯y describe the coupled mode algorithm and show its
connection to anisotropic waveguide propagation theory as developed by
102 Chapter 3
Gardiol [1, Chapter 6]. The reader may refer to the references for further
details on the other methods.
We will now give a brief description of the above-mentioned algo-
rithms. The description here, in order to simplify the discussion and descrip-
tion, is assumed to apply only to longitudinally homogeneous gratings.
When using coupled wave analysis [5±9] and RCWA [10±53], Maxwell's
equations in the diffraction grating region are solved by expanding the
periodic dielectric in the diffraction grating region in a Fourier series,
expanding the EM ®elds in the diffraction grating region in a set of
Floquet harmonics whose amplitudes are functions of the longitudinal coor-
dinate, and after substituting these expansions in Maxwell's equations, orga-
nizing the resulting equations into state variable form where eigensolutions
to the state variable system can be found. Coupled wave analysis [5±9]
differs from RCWA [10±53] in that in coupled wave analysis only a very
few Floquet harmonics are used in the analysis (two or three), whereas in
RCWA the analysis is made nearly exact by including however many
Floquet harmonics are necessary until convergence of the solution is
obtained. Typical state variable matrix sizes in the rigorous coupled wave
analysis method may range from 10 10 to 100 100.
In coupled mode theory algorithms, the transverse periodic region of
the gratings is divided into homogeneous subregions, and wave equation
solutions in the homogeneous subregions [which are linear combinations of
sinusoids proportional to a longitudinal propagation factor exp
z, where
z is the longitudinal coordinate] are EM boundary matched to the adjacent
homogeneous subregions. After imposing the boundary condition that the
overall EM solution across the grating period repeat itself every grating
period, one derives a nonstandard eigenvalue equation, whose multiple
roots thus determine the propagation constant
of the modes that can
propagate in the system. The propagation constant
can of course be purely
imaginary (nonevanescent), purely real (evanescent or attenuating), or com-
plex if the medium is lossy, propagating with attenuation. By summing the
forward and backward modes in the diffraction grating region, a complete
solution of Maxwell's equations in the grating region is found. This method
is particularly useful for lamellar gratings or step gratings, where there are
just two or just a few uniform layers within one grating period. This method
is called a coupled mode approach because it is based on determining the
propagating modes of the system. In the special one-dimensional case when
the grating period is bounded by perfect conductors and the overall grating
region is uniform, the method reduces to the well-known problem of deter-
mining the propagating modes in a parallel plate waveguide. We would like
to caution readers that the algorithm names, coupled mode analysis and
Planar Diffraction Gratings 103
In this section we are interested in using the rigorous coupled wave analysis
algorithm (RCWA) to study the diffraction case that occurs when a plane
wave is incident on the planar grating shown in Fig. 1. The diffraction
!
grating is assumed to have its grating vector speci®ed by K K~ a^ x , where
K~ 2= ~ and ~ is the grating period or grating wavelength. In this case the
electric ®eld is assumed to be polarized perpendicular to the plane of inci-
!
dence as E Ez a^ z . In this section two RCWA formulations will be pre-
sented. In the ®rst formulation (given in Section 3.2.1), the state variable
equations will be derived directly from Maxwell's equations, whereas in the
following section, Maxwell's equations will be reduced to a second-order
wave equation and then placed in state variable form. The complex
Poynting theorem using the solutions found from the full ®eld RCWA
Planar Diffraction Gratings 105
algorithm will be used to calculate the real and reactive power of the dif-
fraction grating system and thus validate the overall analysis.
! !
r E j
0 H
3:2:1
! !
r
0 H j E
3:2:2
p
where 0 0 =0 377
is the intrinsic impedance of free space,
~ 0 is the relative permeability of Region 2, 0 is the permeability of free
=
~ 0 is the relative permittivity of Region 2, and 0 is the permit-
space, =
tivity of free space. We expand the electric and magnetic ®eld as
! X
1
E Szi
y exp
jkxi xa^ z
3:2:3
i 1
! ! X
1
U 0 H Uxi
ya^ x Uyi
ya^y exp
jkxi x
3:2:4
i 1
p ~
kxi kx0 iKx kx0 1 1 sin
Kx 2= k0
Substituting we have
X1
! @Szi !
r E a^ jkxi Szi a^y exp
jkxi x j
0 H
i 1
@y x
X
1
j Uxi a^ x Uyi a^ y exp
jkxi x
3:2:5
i 1
1
X
! @Uxi
r U jkxi Uyi exp
jkxi xa^ z j
xEz a^ z
i 1
@y
X
1
j
x Szi exp
jkxi xa^z
3:2:6
i 1
" #" #
X
1
ji 00 Kx x
X
1
j
kxo i 0 Kx x
xEz i 00 e Szi 0 e
3:2:7
i 00 1 i 0 1
Planar Diffraction Gratings 107
X
1 X
1
jkxo x j
i 0 i 00 Kx x
xEz i 00 Szi 0 e e
3:2:8
i 00 1 i 0 1
@Szi
jUxi jkxi Szi jUyi
@y
X1
3:2:11
@Uxi
jkxi Uyi j i i 0 Szi 0
@y i 0 1
It is useful to introduce column and square matrices and put the preceding
equations into state variable form. Let Ux Uxi , Uy Uyi , Sz Szi , i
1; . . . ; 1 and let i;i 0 i i 0 , Kx kxi i;i 0 , I i;i 0 ,
i; i 0 1; . . . ; 1, be square matrices. i;i 0 is the Kronecker delta and I
is the identity matrix. We ®nd that
@Sz
jUx jKx Sz jUy
@y
3:2:12
@Ux
jKx Uy jSz
@y
108 Chapter 3
@Sz
0Sz jIUx
@y
3:2:14
@Ux 1
j K K Sz 0Ux
@y x x
These equations may be put into state variable form if we introduce the
super matrices
Sz 0 I
Ve Aj
3:2:15
Ux
Kx Kx = 0
we then have
@Ve
y
A Ve
y
3:2:16
@y
AVn qn Vn 3:2:17
The general solution for the electromagnetic ®eld in the grating region can
be found from the state variable solution. The electric ®eld associated with
the nth eigenvector mode is given by
Planar Diffraction Gratings 109
( )
!e X
MT
En Szin a^ z exp
jkxi x exp
qn y
3:2:18
MT
3:2:20
!
2 X
N
!
2 T
!e
U 0 H Cn U n
n1
( )
X
MT X
NT
Cn Uxin a^ x Uyin a^y exp
qn y exp
jkxi x
m MT n1
3:2:21
1
X
1
jkxi x jky1i y
Ezref ri e
3:2:24
i 1
1 1 X 1
jkxi x jky1i y
Hxref k re
3:2:25
0 i 1 y1i i
where
(
n21 k2xi 1=2 n1 > kxi
ky1i
3:2:26
jk2xi n21 1=2 kxi > n1
1
1
Ez
1 Ezinc Ezref
3:2:28
1
1
Ux
1 0 Hx
1 0
Hxinc Hxref
3:2:29
X
1
Ez
3 ti e jkxi xjky3i
yL
3:2:30
i 1
X
1
Ux
3 0 Hx
3 ky3i ti e jkxi xjky3i
yL
3:2:31
i 1
Planar Diffraction Gratings 111
where
ky3i n23 k2xi 1=2 n3 > kxi
3:2:32
jk2xi n23 1=2 kxi > n3
Ez
1 Ez
2
3:2:34
y0 y0
Hx
1 Hx
2
3:2:35
y0 y0
Substituting Eqs. 3.2.28 and 3.2.29 and keeping orders of jij MT , we ®nd
that
( )
X
MT X
MT X
NT
jkxi x jkxi x
fE0 i0 ri ge Cn Szin e
3:2:36
i MT i MT n1
( )
X
MT
jkxi x
X
MT X
NT
jkxi x
f ky1i i0 E0 ky1i ri ge Cn Uxin e
3:2:37
i MT i MT n1
Ez
2 Ez
3
3:2:38
y L yL
Hx
2 Hx
3
3:2:39
y L yL
112 Chapter 3
( )
X
MT X
NT
qn L jkxi x
X
MT
jkxi x
Cn Szin e e fti ge
3:3:40
i MT n1 i MT
( )
X
MT X
Nt
qn L jkxi x
X
MT
jkxi x
Cn Uxin e e f ky3i ti ge
3:3:41
i MT n1 i MT
In Eqs. 3.2.36 and 3.2.37 and Eqs. 3.2.40 and 3.2.41, in order for the left-
and right-hand side of the equations to agree, it is necessary for the Fourier
coef®cients of e jkxi x to agree for each Floquet harmonic e jkxi x . Thus for the
unknown coef®cients ri , Cn , and ti we have the equations
X
NT
E0 i0 ri Cn Szin
3:2:42
n1
X
NT
ky10 i0 E0 ky1i ri Cn Uxin
3:2:43
n1
X
NT
qn L
Cn Szin e ti
3:2:44
n1
X
NT
qn L
Cn Uxin e ky3i ti
3:2:45
n1
" #
X
NT X
NT
ky10 i0 E0 ky1i E0 i0 Cn Szin Cn Uxin
3:2:46
n1 n1
" #
X
NT X
NT
qn L qn L
Cn Uxin e ky3i Cn Szin e
n1 n1
3:2:47
Planar Diffraction Gratings 113
or altogether
X
NT
Cn fky1i Szin Uxin g 2E0 ky10 i0
3:2:48
n1
X
NT
qn L
Cn e Uxin ky3i Szin 0
3:2:49
n1
where i MT ; . . . ; MT .
The above constitutes a set of NT 2
2MT 1 equations for the NT
unknown coef®cients Cn . Power is excited in the diffraction grating system
through the 2E0 ky10 i0 term on the right-hand side of Eq. 3.2.48. Once the
Cn are determined, the ri and ti can be found form Eqs. 3.2.42 and 3.2.44.
! !
r E j
0 H
3:2:50
! !
r
0 H j
x E
3:2:51
! ! !
rr E jr
0 H
x E
3:2:52
! ! !
r r E rr E r2 E
3:2:53
! @E
r E z 0
3:2:54
@z
114 Chapter 3
Therefore we have
r2 Ez xEz 0 3:2:55
@2 @2 Ez
E z
xEz 0
3:2:56
@x2 @y2
For the present analysis we will let
x 2 cos Kx and take 1. K
is a normalized wave number
K~ 2 2
K
3:2:57
ko ko ~
X
1
Ez Si
y exp
j i
3:2:58
i 1
k~2 p
2 cos 0 2 cos 0 i i x 2 y
3:2:59
ko
p
i 1 sin iK
3:2:60
" #
@2 X1
@2 @
2
E z 2
Si
y 2j2 Si
y 22 Si
y exp
j i
3:2:61
@y i 1
@y @y
@2 X 2
2
E z i Si
y exp
j i
3:2:62
@x i
Planar Diffraction Gratings 115
The term
xEz 2 exp
jKx exp
jKx
2
3:2:63
X
1
Si
y exp
ji x j2 y
i 1
equals
X
1
X1
xEz 2 Si
y exp
j i S
y exp
j
i Kx j2 y
i 1
2 i 1 i
X1
S
y exp
j
i Kx j2 y
2 i 1 i
3:2:64
Similarly
i K i1 3:2:67
X1
T2 S
y exp
j
i Kx j2 y
2 i 1 i
3:2:68
X1
S
y exp
ji 1 x j2 y
2 i 1 i
X 1
T2 S 0
y exp
ji 0 x j2 y
3:2:69
2 i 0 1 i 1
116 Chapter 3
X1
T3 S
y exp
j
i Kx j2 y
2 i 1 i
3:2:70
X 1
S 0
z exp
ji 0 x j2 y
2 i 0 1 i 1
X1
xEz 2 Si
y S
y S
y exp
j i
3:2:71
i 1
2 i1 2 i 1
The only way that the above equation can be zero for all values of x and y is
if the curly bracketed expression is zero. Thus Eq. 3.2.72 describes a series of
coupled modal amplitude equations to determine the EM ®elds of the sys-
tem. At this point it is useful to introduce scaled coordinates into analysis.
We let
Substituting the above scaling into Eq. 3.2.72 we ®nd after algebra that
d 2 Si dS
cos 0 i ii BSi Si1 Si 1 0
3:2:74
82 du2 du
where
~ p
2 2 ~ p
2 1
B sin 0 sin
3:2:75
Planar Diffraction Gratings 117
and
22
3:2:76
~ 2
1 d 2 Si c dSi bi
Si Si1 Si 1 0
3:2:78
a du2 a du a
Making these substitutions we ®nd that the second order Eq. 3.2.78 can be
written as
dS1i
S2i
3:2:81
du
dS2i
aS1i1 bi S1i aS1i 1 cS2i
3:2:82
du
d 2 S1i dS2i
aS1i1 bi S1i aS1i 1 cS2i
3:2:83
du2 du
1 d2 b c dSi
S Si1 i Si Si 0
3:2:84
a du2 i a 1
a du
This is identical to Eq. 3.2.78, thus showing that Eq. 3.2.83 is the correct
®rst-order state variable form of Eq. 3.2.78.
The full matrix form for Eqs. 3.2.80 and 3.2.81 when written out for
MT 2 is
d S1 A11 A12 S1
3:2:85
S
du 2 A21 A22 S2
where
t
S1 S1; 2 S1; 1 S1;0 S1;1 S1;2
3:2:86
t
S2 S2; 2 S2; 1 S2;0 S2;1 S2;2
3:2:87
A11 0 55
3:2:88
A12 I ii 0 55
3:2:89
2 3
b 2 a 0 0 0
6 7
6 a b 1 a 0 07
6 7
6 7
6
A21 6 0 a bo a 0 7
3:2:90
7
6 7
6 0 0 a b a 7
4 1 5
0 0 0 a b2 55
A22 cii 0 55
3:2:91
(
1; i i0
i;i 0
3:2:92
0; i 6 i 0
aii 0
i; i 0
1; . . . ; 10 represent the individual matrix elements of the overall
matrix A. Using the just de®ned matrices, Eq. 3.2.85 can be written in full
state variable form as
Planar Diffraction Gratings 119
d
V AV
3:2:94
du
X
NT
Si
u S1i
u Cn win exp
qn u
3:2:95
n1
where NT 2
2MT 1 and where win represents the ith row of the nth
eigenvector
S1 n . The electric ®eld Ez is given by Eq. 3.2.58 with Si
u
substituted. We have
( )
X
MT X
NT
Ez exp j
i x 2 y Cn win exp jqn y
3:2:96
i MT n1
p
where, as already de®ned, =
4 2 ). To proceed further it is necessary
to ®nd the magnetic ®eld associated with Ez . Using Maxwell's equations, the
tangential magnetic ®eld Hx is found from
1 @Ez
Hx
3:2:97
j0 @y
MT X
X NT
Ez2 Cn win exp ji x
2 qn y
3:2:98
i MT n1
X
MT X
Nt
Ux2 0 Hx2 Cn win
qn 2 exp ji x
2 qn y
i Mt n1
3:2:99
X
NT
2ky10 E0 io Cn win ky1i 2 qn
3:2:100
n1
X
NT
0 Cn en win ky3i 2 qn
3:2:101
n1
where
X
NT
ri Cn win E0 i0
3:2:103
n1
X
NT
ti Cn win en
3:2:104
n1
transverse wave period ~ (see Fig. 7.) Details of the calculation are given in
Section 3.3.
We begin presenting results for the cosine grating. The solid line plots
in Fig. 2 show the transmitted diffraction ef®ciencies DET
% for ®ve orders
i 2; 1; 0; 1; 2 as calculated by the full ®eld method (see Section 3.2.1,
Eqs. 3.2.48 and 3.2.49) using the lossless cosine grating as speci®ed in Fig. 2
inset and heading. These plots show DET
% versus the layer length L~ (in
units of free space wavelength ). As can be seen from Fig. 2, as the layer
length L~ increases from 0 to 9, because is at the Bragg angle (implying
Bragg incidence), power is primarily diffracted from the i 0 order into the
i 1 order with a small amount of power being diffracted into the other
orders i 2; 1; 2. For larger values of L, ~ 9 to 18, power is diffracted
from the i 1 order into the i 0 order with a small amount of power
being diffracted into the other orders i 2; 1; 2. This cycle is repeated
over a long range of L~ values. Because the bulk regions had matched per-
mittivities, the re¯ected diffractions were small and have not been plotted.
Also shown in Fig. 2 is the DET
% as calculated by a differential equation,
the state variable method described in Section 3.2.2 and derived originally in
Ref. 16 (dots, i 0). In this analysis, Maxwell's equations are reduced to a
second-order differential equation for the electric ®eld, and this differential
equation is put in state variable form. The state variable form that results is
Figure 2 The transmitted diffraction ef®ciencies DET
% for ®ve orders i 2;
1; 0; 1; 2 as calculated by Eqs. 3.2.48 and 3.2.49 using a lossless cosine grating.
122 Chapter 3
different from the present one, although as MT ! 1 the two methods are
mathematically equivalent. As can be seen from Fig. 2, a comparison of the
i 0 order plots (typical of all orders) shows that virtually identical results
occur from the use of the two methods.
Figures 3 and 4 show plots of the real and imaginary parts of the
normalized complex power PIN (line) and PBOX (dot) of the complex
Poynting theorem, ®rst introduced in Chapter 2 (using the Poynting box
shown in Fig. 15). For more detail on the application of the Poynting
theorem to gratings, see the next subsection. As mentioned earlier, this
case represents a lossless diffraction grating, bulk dielectric case. In these
plots the complex power is plotted versus the layer length L. ~ As can be seen
from Figs. 3 and 4, excellent agreement in both plots is obtained from the
calculation. Figure 5 shows a plot of the electric and magnetic energies PWE
and PWM versus layer length L~ that results for the example under considera-
tion. As can be seen from Fig. 5, the electric and magnetic energies are very
nearly equal to one another, and in a L~ 1 size slab, the electric and
magnetic energies PWE and PWM are much larger than the peak magnitude
energy difference between the two energies.
Figure 6 shows the Im
PBOX versus layer length L~ when the electro-
magnetic ®elds are computed using MT 3 and MT 6. As can be seen
Figure 3 The real part of the normalized complex power PIN and PBOX of the
complex Poynting theorem.
Planar Diffraction Gratings 123
Figure 4 The imaginary part of the normalized complex power PIN and PBOX of
the complex Poynting theorem.
Figure 5 Plots of the electric and magnetic energies PWE and PWM versus layer
length are shown.
124 Chapter 3
Figure 6 The Im
PBOX versus layer length L~ when the electromagnetic ®elds are
computed using MT 3 and MT 6.
from Fig. 6, extremely good convergence is observed using the two different
truncation sizes.
Figures 7 and 8 show plots of the real and imaginary parts of the
complex power PIN and PBOX versus layer length L~ when the diffraction
grating bulk dielectric 2 is lossy rather than lossless and has a value of
2 1 j0:02. In this ®gure one again observes extremely good agreement
between the real and imaginary parts of PIN and PBOX , again showing that
the complex Poynting theorem is obeyed to a high degree of accuracy. A
comparison of Figs. 3 and 4 (lossless case) with Figs. 7 and 8 (lossy case)
shows a very clear difference in the shapes of the real and imaginary parts
of PIN and PBOX that is being computed in the four ®gures. In the lossy
case, as L~ increases, the envelope of the oscillations of PIN and PBOX
damps out, whereas in the lossless case the envelope maintains a long-
itudinal periodic shape. The damping of the envelope with increasing L~ in
the lossy case is expected, since as the layer length increases, the EM ®elds
in the system attenuate near the exit side of the diffraction grating due to
the lossiness. When the diffraction grating becomes suf®ciently long, the
EM ®elds at the exit side approach zero; therefore PIN and PBOX become
independent of L, ~ and thus there is no oscillation.
Planar Diffraction Gratings 125
Figure 7 Plots of the real part of the complex power PIN and PBOX versus layer
length L~ of the complex Poynting theorem when the diffraction grating bulk dielec-
tric 2 is lossy rather than lossless and has a value of w 1 j0:02.
Figure 8 Plots of the imaginary part of the complex power PIN and PBOX , versus
layer length L~ of the complex Poynting theorem when the diffraction grating bulk
dielectric 2 is lossy rather than lossless and has a value of 2 1 j0:02.
126 Chapter 3
Figure 9 Plot of the power dissipated PD , Re
PIN , and Re
POUT versus layer
length L~ is shown. The ripple in the Re
PIN observed in Fig. 7 is not observed
here because of the scale of the Fig. 9 plot.
Planar Diffraction Gratings 127
Figure 12 The real part of PIN and PBOX versus L~ for the lossless square wave
diffraction cases that were studied in Figs. 10 and 11.
Planar Diffraction Gratings 129
Figure 13 The imaginary part of PIN and PBOX versus L~ for the lossless square
wave diffraction cases that were studied in Figs. 10 and 11.
Ez
2
x; y; z 0
3:2:105
y L
X
NT
Cn ky1i Szin Uxin 2Eo ky1o io
3:2:106
n1
X
NT
qn L
Cn e Szin 0
3:2:107
n1
of the system. Incident and re¯ected power are given by the same formulas
as already given for the transmission grating analysis.
Figure 14 shows the re¯ected diffraction ef®ciency
i 0 and i 1
orders) versus layer length L~ that arises when a plane wave is incident on
a square wave or step pro®le dielectric grating backed by a mirror (or short
circuit plate). In Fig. 14 the same lossless square grating that was studied in
Fig. 10 is analyzed. The square wave grating was taken to have a grating
period of ~ and a transverse groove width of =2 (or duty cycle of
50%). The bulk and groove dielectric values and their orientation in the
diffraction grating and the angle of incidence are speci®ed in the Fig. 14
caption and inset. For the present case, for the angle of incidence used, it
turns out that the i 0; 1 orders are the only orders that are re¯ected,
diffracted propagating plane waves. All the other orders are evanescent.
the value of MT 6 was used to calculate the data of Fig. 14. As can be
seen, power for a small grating thickness is diffracted from the i 0 order
into the i 1
0 L~ 0:6. As the thickness increases, however, power is
transferred back to the i 0 order from the i 1, 1 L~ 1:6. This cycle
is repeated for larger values of L.~ In observing the i 0; 1 plots it is very
interesting to note that the transfer of power between the i 0; 1 orders is
not periodic with increasing L~ but irregular and unpredictable. The nonper-
iodicity is undoubtedly due to interaction of the evanescent and propagating
waves that resultP from the matrix solution. Conservation of incident and
re¯ected power
i DERi DETi 1, DETi 0 was observed to a high
degree of accuracy.
The grating in Fig. 15a is assumed to have its grating vector speci®ed
!
by K K~ a^ x , where K~ 2=~ and ~ is the grating period or grating wave-
length. In this case the magnetic ®eld is assumed to be polarized perpendi-
!
cular to the plane of incidence as H Hz a^ z . In the present study, the
complex Poynting theorem will be applied to a Poynting box whose length
extends over the grating region L,~ whose width extends over a grating period
~ and whose thickness is z~ (the electromagnetic ®elds do not vary in the z-
,
direction, so the thickness of the Poynting box is immaterial to the Poynting
power calculation). Figure 15b illustrates the Poynting box of this section
and that of Sec. 3.2 as well. In Section 3.3.1, we will brie¯y summarize the E-
mode RCWA equations for anisotropic diffraction gratings. In Section
3.3.2, the pertinent equations for the power budget as results from the
complex Poynting theorem will be presented. In Section 3.3.3, illustrative
examples will be given for anisotropic media where the permittivity tensor is
either Hermitian or arbitrary.
In much of the existing diffraction grating literature [1±53], power
conservation is veri®ed by calculating the time-averaged real power trans-
mitted and re¯ected from a lossless grating and then verifying that the sum
of these powers equals the power incident on the grating. Computing the
Figure 15 (a) The geometry of the E-mode diffraction grating system is shown. (b)
The complex Poynting box used for calculations is shown.
Planar Diffraction Gratings 133
power budget using only the time-averaged real power has two large limita-
tions associated with it. First, it cannot be used to verify power conservation
for the very common case of lossy gratings, since in this case some power is
dissipated as heat, and thus the transmitted and re¯ected powers will not
equal the incident power. A second limitation of computing the power
budget using only the time-averaged real power is that information about
the reactive power, evanescent ®elds, electric energy, magnetic energy, and
power dissipated within the grating is left undetermined and therefore
unknown. All of these quantities contain important information about the
nature and behavior of the grating. In the area of near ®eld optics, consider-
able attention has been paid to evanescent waves, since these carry informa-
tion about the diffracting or scattering object. Speci®cally, evanescent wave
monitoring has applications in the area of submicron microscopy.
Evanescent waves may also be excited from sharp discontinuities in the
grating, e.g., corners, blaze tips [95]. A power budget approach that can
study energy and power, both real and reactive, during diffraction from such
gratings, is incorporated in the framework of the complex Poynting theo-
rem. Botten et al. [58] consider the problem of energy balance in isotropic
lossy gratings when both E-mode and H-mode polarized incidence plane
waves impinge on the grating.
Energy ¯ow distributions, the generation of plasmon surface waves,
and the absorption of EM energy by metallic sinusoidal gratings has been
studied by Popov et al. [98±101] for shallow and deep gratings. The nature
of the Poynting vector in a dielectric sinusoidal grating under total internal
re¯ection has been studied by Shore et al. [102].
Our discussion of the Poynting vector is fundamentally different from
that of Popov et al. [98±101] and Shore et al. [102]. In their work they were
concerned with the problem of studying the spatial variation of the Poynting
vector (and energy density) on a point-to-point basis over a region of space
close to the diffraction grating surface. The point of their work was to relate
local variation of the Poynting vector to the diffraction that occurred from
the grating. They studied the physical mechanisms of blazing and antiblaz-
ing and its relation to Poynting vector. In this section, we focus on the
Poynting vector power that has been averaged transversely over a diffrac-
tion grating period and relate this averaged Poynting power to the power
dissipated, transmitted, and re¯ected from the grating [105]. We apply the
complex Poynting theorem for EM incidence on periodic diffraction grat-
ings of arbitrary pro®le and made of anisotropic lossy materials. We expli-
citly show also that the energy dissipated in the grating can result from both
imaginary and real parts of the permittivity and permeability for the case of
anisotropic nonreciprocal grating media.
134 Chapter 3
! !
r E j
0 H
3:3:1
! !
r
0 H j E
3:3:2
p
where 0 0 =0 377
is the intrinsic impedance of free space,
~ 0 is the relative permeability of Region 2, 0 is the permeability of free
=
space, ~ =0 0 j 00 is the relative tensor permittivity of Region 2, and
0 is the permittivity of free space. In this section we consider the important
case when the relative permittivity tensor is anisotropic and has the speci®c
form
2 3
xx xy 0
4 yx yy 0 5
3:3:3
0 0 zz
! X
1
E Sxi
ya^ x Syi
ya^ y exp
jkxi x
3:3:4
i 1
! ! X
1
U 0 H Uzi
y exp
jkxi xa^z
3:3:5
i 1
p ~
kxi kx0 iKx kx0 1 1 sin
Kx 2= k0
3:3:6
X
1
x i exp
jiKx x
3:3:7
i 1
Planar Diffraction Gratings 135
where i represent the Fourier coef®cients of
x. Substituting Eqs. 3.3.4±7
in Maxwell's equations; taking the relative permeability of Region 2 to be
1; introducing column and square matrices, namely, Sx Sxi ,
Sy Syi , Uz Uzi , i 1; . . . ; 1, xx xxi;i 0 xxi i 0 ,
xy xyi;i 0 xyi i 0 yx yxi;i 0 yxi i 0 , and yy yyi;i 0 yyi i 0
(here the underbar denotes a square
i; i 0 matrix), Kx kxi i;i 0 , I i;i 0 ,
i; i 0 1; . . . ; 1 square matrices; i;i 0 the Kronecker delta and I the
identity matrix; eliminating Sy using the equation
@Ve
y
A Ve
y
3:3:9
@y
where
Sx a11 a12
Ve A
3:3:10
Uz a21 a22
where
where the superscript 1 in these equations denotes the matrix inverse. The
above equations have been found by expressing each of the product terms
xx
xEx
x; y, xy
xEy
x; y, etc. in a convolution form (see 3.2.7±10) when
!
the Fourier series expansions of
x and E
x; y are substituted in each
!
of the product terms making up
x E and collecting coef®cients on com-
mon i orders.
Let qn and Vn be the eigenvalues and eigenvectors of the matrix A after
truncation. Summing over the individual eigenmodes we ®nd that the over-
all electric and magnetic ®elds in Region 2 are given by
( )
!
2 X X X
NT MT NT
!e
E Cn E n Cn Sxin a^ x Syin a^ y exp
qn y
n1 i MT n1
where NT 2
2MT 1. Equations 3.3.13 and 3.3.14 represent the sum of
NT 2
2MT 1 forward and backward traveling, propagating and non-
!e !e !e
propagating eigenmodes E n and U n 0 H n , which gives the general elec-
tromagnetic ®eld solution in Region 2, the diffraction grating region.
An important problem that remains is to determine the NT coef®-
cients Cn of Eqs. 3.3.13 and 3.3.14. Up to this point we have speci®ed the
general form of the diffracted ®elds in the grating region. The EM ®elds
on the incident side of the diffraction grating (Region 1 of Fig. 15a), on
the transmission side of the diffraction grating (Region 3 of Fig. 15a),
consist of an in®nite number of propagating and nonpropagating
plane waves whose tangential wave numbers are given by
kxi , i 1; . . . ; 1; 0; 1; . . . ; 1. The electromagnetic ®elds in Region 1
consist of the sum of a single E-mode polarized incident plane wave and
an in®nite number of re¯ected propagating and evanescent plane waves.
The total electric and magnetic ®elds in Regions 1 and 3 after summing the
incident and re¯ected ®elds is given by
Region 1
X
1
Uz
1 0 Hz
1 Eo i;o exp
jky1i y ri exp
jky1i y exp
jkxi x
i 1
3:3:15
1 X
1
Ex
1 ky1i Eo i;o exp
jky1i y ri exp
jky1i y exp
jkxi x
1 i 1
3:3:16
1 X
1
Ey
1 kxi Eo i;o exp
jky1i y ri exp
jky1i y exp
jkxi x
1 i 1
3:3:17
Region 3
X
1
Uz
3 0 Hz
3 ti exp
jky3i
y L exp
jkxi x
3:3:18
i 1
Planar Diffraction Gratings 137
1 X 1
Ex
3 ky3i ti exp jky3i
y L exp
jkxi x
3:3:19
3 i 1
1 X 1
Ey
3 k t exp jky3i
y L exp
jkxi x
3:3:20
3 i 1 xi i
where
( p
r r k2xi 1=2 r r > kxi
kyri p r 1; 3
3:3:21
jk2xi r r 1=2 kxi > r r
X
NT
ky1i 2Eo ky1i
Cn Uzin Sxin io
3:3:22
n1
1 1
X
NT
ky3i
Cn exp
qn LSxin Uzin 0
3:3:23
n1
3
where i MT ; . . . ; MT :
The above constitutes a set of NT 2
2MT 1 equations for the NT
unknown coef®cients Cn . Power is excited in the diffraction grating system
through the right-hand side (RHS) term of Eq. 3.3.22.
where
!
1 !
1
PufIN E U
a^ y dS
3:3:25
S y0
!
3 !
3
PufOUT E U
a^ y dS
3:3:26
S y L
!
2 !
2
PuDE E 00 E dV
3:3:27
V
!
2 !
2
PuDM U 00 U dV
3:3:28
V
!
2 !
2
PuWE E 0 E dV
3:3:29
V
!
2
2
0!
PuWM U U dV
3:3:30
V
In Eqs. 3.3.24±30, PufIN represents the complex power radiated into the
diffraction grating power (it is the sum of the incident power, the re¯ected
power, and the interaction power between the incident and re¯ected power),
PuDE and PuDM represent the electric and magnetic dissipated power loss in
the case when the grating material is isotropic, while PuWE and PuWM denote
the reactive powers proportional to the electric and magnetic energies in the
case when the grating material is isotropic. In the general anisotropic case,
however, all four quantities can be complex. Hence, for instance, energy loss
can result from both the imaginary and the real parts of and l, as in a non-
Hermitian medium, to be discussed later. The superscript u in Eqs. 3.3.24±
30 means unnormalized. These power terms will be later normalized to the
incident plane wave power. We will now be concerned with evaluating these
equations for the E-mode plane wave polarization case under consideration.
Each of the four terms in Eq. 3.3.31 must be substituted into Eq. 3.3.29 and
the subsequent volume energy integrals must be evaluated. The analysis
consists of substituting the Fourier series expansions of the electric ®eld
quantities and dielectric tensor quantities into the energy volume integral,
interchanging sum and integral expressions, carrying out all exponential
integrals exactly and in closed form, and ®nally simplifying all summations.
Letting
X
V
x; y Cn 0 Vi 0 ;n 0 exp
qn 0 y exp
jkxi 0 x
3:3:32
i 0 ;n 0
0 0 0 0
represent xx
x, xy
x, yx
x, or yy
x, and letting
X
W
x; y Cn 00 Wi 00 ;n 00 exp
qn 00 y exp
jkxi 00 x
3:3:34
i 00 ;n 00
represent the electric ®eld Ex
x; y or Ey
x; y, we ®nd that any of the four
terms of the unnormalized energy volume integral PuWE can be expressed in
the general form
P V
x; y
xW
x; y dV
V
( )
X
Cn 0 Vi 0 n 0 expqn 0 y exp jkxi 0 x
V i 0 ;n 0
8 9( )
<X = X 000
i 000 expji Kx x C 00 W 00 00 expqn 00 y exp jkxi 00 x
: 000 ; i 00 ;n 00 n i n
i
dx dy dz
3:3:35
140 Chapter 3
X
V
x; y
xW
x; y dxdydz z Cn 0 Cn 00 Iyn 0 ;n 00
V n 0 ;n 00
X
3:3:36
i 00 i 0 V
i 0n 0 Wi00 n 00
i 0 ;i 00
where
0
Iyn 0 ;n 00 exp
qn 0 qn 00 y dy
3:3:37
L
Substitution of the four terms of Eq. 3.3.31, with each term simpli®ed
according to Eq. 3.3.36, produces a closed form expression (that is, all
integrations have been carried out exactly) from which the normalized
energy volume integral PuWE can be evaluated. The evaluation of the PuDE
is identical to the analysis of the PuWE integral except that the lossy relative
permittivity 00
x tensor is used rather than 0
x
X
1
x i ejix
3:3:38
i 1
where
i 0 j 00 i;0 3:3:39
Following the analysis used to determine PuWE and evaluating the discrete
Kronecker delta found, we ®nd that
Planar Diffraction Gratings 141
X Xh
2
2
i
PuWM z 0 Cn 0 Cn 00 Iyn 0 ;n 00 Uzin 0 Uzin 00
3:3:40
n 0 ;n 00 i
X Xh
2
2
i
PuDM z 00 Cn 0 Cn 00 Iyn 0 ;n 00 Uzin 0 Uzin 00
3:3:41
n 0 ;n 00 i
z X
PufIN k E ri E0 i;0 ri
3:3:42
1 i yli 0 i;0
z
Puinc jE0 j2 ky10
3:3:43
1
If we normalize the power of Eq. 3.3.42 to the incident power, we ®nd that
P
Pu i kyli E0 i;0 ri E0 i;0 ri
PIN fIN
3:3:44
Puinc ky10 jE0 j2
Of interest are the powers re¯ected and transmitted from the diffract-
ing grating at y 0 and y L , respectively, and the relation that these
powers have to the power PfIN radiated into the Poynting box y 0 . The
unnormalized re¯ected and transmitted powers are given by the expressions
!
1 !
1 z X
Puref E ref U ref a^ y dS k r r
S y0 1 i y1i i i
3:3:45
!
3 !
3 z X
PufOUT Putrans E U
a^ y dS k t t
S y L 3 i y3i i i
3:3:46
142 Chapter 3
After expanding the product term of Eq. 3.3.47 in square brackets and after
separating the i 0 term from the i 6 0 we ®nd that
X
T ky1i E02 i;0
ri ri E0 i;0 ri ri
3:3:48
i
X
T ky10 E02
2j Im
r0 E0 r0 r0 ky1i ri ri
3:3:49
i;i60
X
T ky10 E02 2jky10 Im
r0 E0 ky1i ri ri
3:3:50
i
Thus
" #
z X z X
PufIN
ky1i E0 E0 i;0 k r r
1 i
1 i y1i i i
3:3:51
z
2jky10 Im
r0 E0
1
The ®rst and second summation terms of Eq. 3.3.51 represent the unnorma-
lized incident and re¯ected power at y 0 . The third term is an interaction
term between the incident and re¯ected EM wave. We have
z
PufIN Puinc Puref 2jky10 Im
r0 E0
3:3:52
1
We now substitute PufIN of Eq. 3.3.52 into the left-hand side (LHS) of Eq.
3.3.24. We ®nd that
z
Puinc Puref 2jky10 Im
r0 E0 PufOUT PuDE PuDM
1
3:3:53
j
PuWE PuWM
Transposing the re¯ected power term and the interaction power term to the
RHS of Eq. 3.3.53, we ®nd that
Planar Diffraction Gratings 143
z
Puinc Puref 2jky10 Im
r0 E0 PufOUT PuDE PuDM
1
3:3:54
j
PuWE PuWM
Puinc
Pinc 1
Puinc
Puref 1 X
Pref u k r r
Pinc ky10 i y1i i i
PufOUT 1 1 X
POUT Ptrans k t t
3:3:55
Puinc 3 ky10 i y3i i i
PuDE PuDM
PDE PDM
Puinc Puinc
PuWE PuWM
PWE PWM
Puinc Puinc
we now ®nd that the complex Poynting theorem of Eq. 3.3.24 after division
of all terms by Puinc can be written in normalized form as
We remind the reader that by de®nition the quantities PDE , PDM , PWE , and
PWM in the general anisotropic case are not purely real, so that taking the
real and imaginary parts as speci®ed in Eqs. 3.3.58 is necessary as shown.
From Eqs. 3.3.57 and 3.3.58 we will now de®ne three useful relations
from which numerical plots can be made and which give insight into the
diffraction process. We will now give the ®rst relation. From Eq. 3.3.57 if we
transpose the Rej
PWE PWM term we have
PWEM
diffR Rej
PWE PWM
3:3:60
PWEM
diffR PdiffR
3:3:62
Equations for PWEM diffR and PdiffR are in general useful quantities to calculate.
When the medium is reciprocal, PWE , PWM , PDE , and PDM are all purely real
quantities; therefore from Eq. 3.3.61, PWEM diffR 0 and PdiffR 0, and thus
Eq. 3.3.61 represents a conservation relation stating that the incident power
should equal the sum of the transmitted, re¯ected, and dissipated powers.
When the medium is anisotropic, PWE and PWM are in general complex, and
thus PWEM WEM
diffR is not necessarily zero. In this case PdiffR (which should equal
PdiffR ) give a sense of how much the anisotropic nature of the medium is
present in the EM ®eld calculation. The computation of PWEM diffR and PdiffR is
also useful in this case as a cross-check of the numerical calculation. It is
useful since both terms are computed from EM ®eld quantities located in
different regions of space. Numerically if PWEM diffR and PdiffR are not equal or
Planar Diffraction Gratings 145
PWEM
diffI Im j
PWE PWM
3:3:63
PdiffI Im
Pref Ptrans PDE PDM 2Im
r0
3:3:64
PWEM
diffI PdiffI
3:3:65
We will now de®ne a third useful relation that results from Eqs. 3.3.56.
Transposing the terms Im
Pref and Im
Ptrans to the left-hand side of Eq.
3.3.58 and multiplying by 1 we ®nd that
Letting
and letting
Pdif
evan 2Im
r0 Im
PDE Im
PDM Imj
PWE PWM
3:3:68
Pevan Pdiff
evan
3:3:69
Puref 1 X
Pref u k r r
3:3:70
Pinc ky10 i y1i i i
146 Chapter 3
In these equations we note ky10 is real and positive. Because the term ri ri is
purely real, the re¯ected power Pref is purely imaginary only when ky1i is
purely imaginary. Thus we see that Im
Pref is only nonzero when
p
ky1i jk2xi 1 1 1=2 , kxi > 1 1 , which occurs only for those space har-
monics that are evanescent. The transmitted power is evanescent for
Im
Ptrans 6 0, just as for the re¯ected power.
Relations 3.3.67±69 like Eqs. 3.3.60±62 are useful for two reasons.
First, they give the evanescent power, thus they give a measure of how
much power and energy is stored in nonpropagating EM waves near the
diffraction grating interfaces. The larger the evanescent power and energy,
the larger and rougher are the diffraction grating interfaces relative to the
bounding regions. A second reason that they are useful is that they provide
an excellent cross-check of the numerical diffraction solution. Pevan and
Pdiff
evan are computed from EM terms that exist in different regions of the
EM system. Thus equality or very close numerical equality of Pevan and
Pdiff
evan helps show that the computations are being made correctly. We also
note that the sum of evanescent powers tends to be small relative to the
other diffraction power terms in the system. Thus the sum of evanescent
powers tends to be a fairly sensitive test of the EM algorithm.
The type of power budget analysis presented here can be applied to
virtually any type of diffraction grating and any kind of polarization for the
incident wave and may be extended to multilayer grating structures in a
straightforward way. For example, results for H-mode incidence on an iso-
tropic grating have already been presented in the previous section.
pic materials with relatively short grating periods ~ tests the RCWA
and the complex Poynting theorem in a fairly severe way.
We begin presenting results for the isotropic step grating. Figure 16a
shows the transmitted diffraction ef®ciencies DET
% for the i 0; 1 orders
when a lossless and lossy grating is present. The parameters of the grating re
given in the Fig. 16a inset. The plots show DET
% versus the layer length L~
(in units of free space wavelength ). The lossless grating example of Fig.
16a was ®rst studied by Yamakita and Rokushima [54], and the lossless
diffraction ef®ciency results of Fig. 16a are identical to their results [54].
As can be seen from Fig. 16a, as the layer length L~ increases from 0 to 2,
because is at the Bragg angle, power is diffracted from the i 0 order into
the i 1 order. For larger values of L,~ 2 to 4, power is diffracted from the
i 1. This cycle of blazing and antiblazing (see [98±101] for an insightful
discussion of blazing and antiblazing in grating analysis and its relation to
the Poynting vector) is repeated over a long range of L~ values. One also
observes that if 2b is lossy, the diffraction ef®ciency of both the i 0 and
i 1 orders is attenuated as one would expect in a lossy material.
Figure 16 (a) The transmitted diffraction ef®ciency of a lossless and lossy step
diffraction grating are shown. (b±e) Plots of the real and imaginary parts of the
normalized complex power PIN and PBOX as computed by Eq. 56a of the complex
Poynting theorem for the lossless (b and c) and lossy cases (d and e) are shown. (f±g)
Plots of the evanescent power as computed by Eqs. 3.3.67±69 for the lossless (f) and
lossy cases (g) are shown. OSA 1999 [103].
148 Chapter 3
Figure 16 (continued)
Planar Diffraction Gratings 149
Figure 16 (continued)
150 Chapter 3
Figure 16 (continued)
Planar Diffraction Gratings 151
Figures 16b±e show plots of the real and imaginary parts of the normal-
ized complex power PIN and PBOX of the complex Poynting theorem for the
lossless (Figs. 16b and 16c) and lossy (Figs. 16d and 16e) cases as speci®ed by
Eq. 3.3.56a. As can be seen from these four plots, the complex Poynting
theorem is obeyed to a high degree of accuracy as evidenced by the close
®t between the data for PIN (solid line) and the data for PBOX (dots). In
comparing the lossless and lossy cases (using Eq. 3.3.56a), one also notices
a signi®cant difference in the complex Poynting results for these cases. In the
lossy case, as the layer becomes larger, the oscillations of the power results
decrease. This is because in the lossy case, less EM power is re¯ected from the
boundary and interferes with forward traveling waves than when the medium
is lossless. Figures 16f and 16g show plots of the evanescent power as com-
puted by Eqs. 3.3.67±69. As can be seen in these ®gures, excellent agreement
with the complex Poynting theorem is observed in both lossy and nonlossy
cases. As discussed earlier, because the grating width is on the order of a
wavelength, a certain amount of energy is stored in the evanescent ®elds of
Regions 1 and 3. The comparison of the lossy and nonlossy materials shows a
de®nite difference in the evanescent ®eld quantities of the system.
Figures 17a and 17b show plots of the real and imaginary parts,
respectively, of the normalized complex power as computed by Eq.
3.3.56a for the lossless case when a general, nonreciprocal, anisotropic
material occupies Region 2a of the step diffractive region as shown in the
inset of Figs. 17a and 17b. Region 2b of the step was chosen to have
2b 2:5, and Region 2a of the step was chosen to have 2ayy 1:52axx ,
2axx 1 j0:1, 2axy 0:22axx , and 2ayx 0. The example being consid-
ered is non-Hermitian since 2axy 6 2ayx . This situation may be encountered
in, for instance, materials with stimulated Raman scattering, a process
whose governing susceptibility does not exhibit overall permutation symme-
try [96]. As can be seen from these ®gures, the complex Poynting theorem is
obeyed to a high degree of accuracy. Figure 17c shows a plot of the evanes-
cent power as calculated by Eqs. 3.3.67±69 for the nonreciprocal anisotropic
case under consideration.
Figures 17d and 17e show plots of PWEM diffR and PdiffR (calculated in Eqs.
3.3.60±62) and plots of PWEM
diffI and PdiffI (calculated in Eqs. 3.3.63±65) for the
same anisotropic case as was considered in Fig. 17a±c (plots labeled
2ayx 0. Also shown in these ®gures are plots made for the case when
all the permittivity elements are the same as in Fig. 17a±c except that instead
of taking 2ayx 0, 2axy has been taken to be 2axy 2ayx and thus the
medium is Hermitian. These plots are labeled 2axy 2ayx in Figs. 17d
and 17e. As can be seen from all the plots shown in Figs. 17d and 17e,
Eqs. 3.3.62 and 3.3.65 (namely, PWEM WEM
diffR PdiffR and PdiffI PdiffI ) are
152 Chapter 3
Figure 17 (a,b) Plots of the real and imaginary parts, respectively, of the normal-
ized complex power PIN and PBOX as computed by Eq. 3.3.56a of the complex
Poynting theorem are shown. (c) A plot of the evanescent power as calculated by
Eqs. 3.3.67±69 is shown. (d) Plots of PWEM
diffR and PdiffR (calculated in Eqs. 3.3.60±62)
for Hermitian and non-Hermitian step diffraction gratings are shown. (e) Plots of
PWEM
diffI and PdiffI as calculated in Eqs. 3.3.63±65 for Hermitian and non-Hermitian
step diffraction gratings are shown.
Planar Diffraction Gratings 153
Figure 17 (continued)
154 Chapter 3
Figure 17 (continued)
obeyed to a very high degree of accuracy as can be seen by the close agree-
ment between lines and dots as displayed in the ®gures.
In Fig. 17d it is very interesting to compare the power results for the
Hermitian and non-Hermitian cases. In Fig. 17d, for the plots labeled
2ayx 0, the material is non-Hermitian, and thus PWE and PWM are not
necessarily purely real, and thus PWEMdiffR of Eqs. 3.3.60±62 is not necessarily
zero. The nonzero nature of PWEMdiffR is clearly seen in the plot of Fig. 17d. On
the other hand, for the plots labeled 2axy 2ayx (Hermitian case), it is
noticed that
PWEM
diffR Rej
PWE PWM PdiffR 0
3:3:72
or
or
3.4.1 Formulation
We consider the case of an EM wave (H-mode polarization) (see Fig. 18)
incident on a sinusoidally modulated relative dielectric constant (relative
permeability equal to unity in all space)
period. is the phase angle of the modulation and plays a signi®cant role in
the state variable analysis of the system. The grating is assumed to be
bounded by lossless homogeneous regions on either side of the grating.
The normalized electric ®eld in each region expressed in normalized coordi-
~ etc., is
nates x k0 x,
Planar Diffraction Gratings 157
Region 1 (z < 0
X
E1 exp j
0 x 10 z R~ i exp j
i x 1i z
3:4:2
i
X
E2 S~i exp j
i x 2i z
3:4:3
i
Region 3 z > d
X
E3 T~i exp j
i x 3i
z d
3:4:4
i
Here
i k1 sin iK sin
2 p
mi k2m 2i km m
2i k2 cos 0 iK cos
" #
0 1 1 1=2
sin sin
2
i k1 sin iK sin m 1; 2; 3
i 0; 1; 2; . . .
d 2 Si
u dS
u
cos 0 i cos i i
i BS i
u
82 d 2 u du
S i1
u exp
j S i 1
u exp
j 0
3:4:6
where
R~ i , S~i
z, and T~i , the normalized amplitudes of the ith diffraction
orders in regions 1, 2, and 3, respectively, can be written in a form that
shows explicitly their dependence on the arbitrary phase modulation .
Planar Diffraction Gratings 159
Upon substituting
where
R~ i Ri exp
ji
3:4:15
T~i Ti exp
ji
3:4:16
Kx 3:4:22
Now the RHS of Eq. 3.4.21 can be expanded in a Fourier series by a Bessel
identity of the form
X
1
exp
jd cos Ji
d exp
ji
3:4:23
i 1
X
1 X
1
E/ Ji
k0 d exp
ji Ji
k0 d exp
ji exp
jiKx
i 1 i 1
3:4:24
The quantity in square brackets is the amplitude Ti of the ith order dif-
fracted ®eld at the exit plane of the grating (transmission type in this simple
example), and it is readily seen that the phase associated with the ith dif-
fraction order is always i.
When the diffraction ef®ciencies are calculated using RCW analysis
two distinct cases must be considered:
1. Slanted gratings
6 0. In this case the re¯ected (transmitted)
orders in Region 1 (3) have different values of 1i
3i ) for different values
Planar Diffraction Gratings 161
By substituting Eqs. 3.4.15 and 3.4.16 into Eqs. 3.4.25 and 3.4.26, respec-
tively, we obtain
1i
DERi Re jRi j2
3:4:25a
10
DETi Re 3i jTi j2
3:4:26a
10
Region 3 (z > d)
R~ and T~ are the electric ®eld amplitudes of the single re¯ected and trans-
mitted plane waves, respectively, and U
z is the z-dependent amplitude of
the electric ®eld inside the pure re¯ection grating. In addition to satisfying
the wave equation in the three regions, the expressions 3.4.28±30 must also
satisfy the boundary conditions at z 0 and z d. The four boundary
equations are tangential E and z 0:
R~ 1 U 0 3:4:31
tangential H at z 0:
dU
j10
R~ 1
3:4:32
dz z0
tangential E at z d:
T~ U d 3:4:33
tangential H at z d:
dU
j30 T~
3:4:34
dz zd
It is now shown that the RCWA equations can also be derived for
0 and therefore are also valid for pure re¯ection gratings.
Since changing the modulation phase by an integral multiple of 2
~ T,
results in the same grating, the amplitudes R, ~ and U
z in Eqs. 3.4.28±30
Planar Diffraction Gratings 163
where
2i k2 cos 0 iK
By setting 0 in Eq. 3.4.1 and substituting Eqs. 3.4.1, 3.4.29, and 3.4.36,
into the wave equation Eq. 3.4.5, the following equation is obtained:
" #
X d 2 S^i d S^ ^
2j2i i 2
2i 20 k22 S^i Si 1 S^i1
i
dz2 dz 2
Since can have any value, it follows that the coef®cient of each exponential
(the term in square brackets in Eq. 3.4.39) must equal zero:
d 2 S^i d S^i ^
2j2i 2
2i 20 k22 S^i Si 1 S^i1 0
3:4:40
dz2 dz 2
P
Noting that 1 i i0 exp
ji, it follows that
and
X 2
30 ~ 2
DET Re jTj Re 30 T~
3:4:49
10 10 i i
The conservation of power for a lossless pure re¯ection grating dictates that
X X
hDER i hDET i jRi j Re 30
2
jTi j2 1
3:4:54
i
10 i
Figure 20 The diffraction ef®ciencies for a 30 grating with 10. The dif-
fraction ef®ciencies of all re¯ected and transmitted waves not shown in the ®gure are
less than 0.01. Used with permission of OSA, 1981 [16, Fig. 4].
P P
Figure 23 Deviations from conservation of power
j i jRi j2 j i Ti j2 1
obtained with the data of Fig. 21 when (top to bottom) 11, 17, and 23 diffraction
orders were used in the numerical calculations. Used with permission of Optical
Society of America (OSA), 1983 [25, Fig. 4].
Planar Diffraction Gratings 171
P P
Figure 24 Deviations from conservation of power
i jRi j2 j i Ti j2 1
obtained with the data of Fig. 22 when (top to bottom) 11, 17, and 23 diffraction
orders were used in the numerical calculations. Used with permission of Optical
Society of America (OSA), 1983 [25, Fig. 5].
172 Chapter 3
within the thin layer as has been done in earlier sections, and to boundary
match the state variable solutions at all thin layer interfaces and at all
regions exterior to the grating to ®nd an EM solution to the inhomogenous
grating problem. This procedure has been used by numerous researchers to
study longitudinally inhomogeneous diffraction gratings [18,19,21±23,
28,31,33,55,65] and other EM systems.
In observing Fig. 29a one also notices why the grating must be con-
sidered inhomogeneous in the y-direction. The dotted lines of Fig. 29a show
examples of the thin layers that can be used to analyze the grating. In
observing these dotted lines, for the example presented, one notices that if
the thin layer is chosen close to the incident side of the grating, little material
of Region 3 is included in the thin layer, and the Fourier series representing
that thin layer will nearly be that of Region 1. On the other hand, if the thin
layer is chosen close to the transmit side of the grating, then in this case most
Planar Diffraction Gratings 175
of the material in the thin layer is that of Region 3, and the Fourier series
representing this thin layer will nearly be that of Region 3. Clearly, from this
discussion and observing the thin layer in Fig. 29a one can see that the
Fourier series in the thin layer near the incident side are signi®cantly differ-
ent from those at the transmit side. Thus from this discussion, the RCWA
method must re¯ect the change in Fourier series coef®cients as one changes
the y position in the grating.
In the following sections we will carry out the multilayer analysis for
the E-mode case studied in Section 3.3. The analysis for the H-mode case is
similar to the E-mode case.
176 Chapter 3
Each layer is assigned a local coordinate system y` with its local origin as
shown in Fig. 28. The ®rst layer has the coordinate y1 and the last layer to
the right is yN . Enough layers N` are used so that the grating inhomogeneity
`
Planar Diffraction Gratings 177
In these equations kxi kx0 iKx , NT 2
2MT 1, qn` and V` Stx` ; Utz`
t
Sx` Sxi` , Uz` Uzi` are the eigenvalues and eigenvectors of the `th
region. The eigenvector Vn`
y Vn eqn `y is assumed to satisfy the eigenvalue
equation
@Vn`
A` Vn`
3:5:4
@y
where the matrix A` is given by Eq. 3.2.15 with the ss 0 i` of Eq. 3.5.1 used to
de®ne A` .
The EM ®elds in Regions 1 and 3 are the same as in the uniform case
and are given in Section 3.2. The analysis proceeds by matching the tangen-
tial electric ®eld (Ex in this case) and tangential magnetic ®eld (Hz in this
case) at every boundary interface. The Ex and Hz ®elds at Region 1: Layer
` 1 interface are
Ex
1 Ex1 y 0 Hz
1 Hz1 y 0
3:5:5
y0 1 y0 1
178 Chapter 3
Substituting we have
( )
X
MT
ky1i X
MT X
NT
jkxi x qn1 y jkxi x
Eo io Ri e Cn1 Sxin1 e 1 e
i M
1 i MT n1
T y 0
1
3:5:6a
( )
1 X
MT
1 X
MT X
NT
jkxi x
E Ri e jkxi x
Cn1 Uzin1 eqn1 y1 e
0 i M o io 0 i M n1
T T y 0
1
3:5:6b
( )
X
MT X
NT
qn` y jkxi x
Cn` Sxin` e ` e
i MT n1
y S
`
`
( )
X
MT X
NT
qn;`1 y jkxi x
Cn;`1 Sxin;`1 e `1 e
3:5:8a
i MT n1
y 0
`1
( )
1 X
MT X
NT
jkxi x
Cn` Uzin` eqn` y` e
0 i M n1
T y S`
`
( )
1 X
MT X
NT
qn;`1 y jkxi x
Cn;`1 Uzin;`1 e `1 e
3:5:8b
0 i MT n1
y 0
`1
X
MT
ky3i jkxi x
Te
3:5:9a
i M
3 i
T
Planar Diffraction Gratings 179
( )
1 X
MT X
NT
jkxi x
CnN` UzinN` exp
qnN` yN e
0 i M n1
`
T y SN`
N`
1 X
MT
jkxi x
Te
3:5:9b
0 i M i
T
Each of the Fourier coef®cients of each exponential e jkxi x on the left and
right sides of the above equations must be equal in order that the equations
be all satis®ed. Equating Fourier coef®cients after evaluation of the y-depen-
dent terms we ®nd
ky1i X
NT
Eo io Ri Cn1 Sxin1
3:5:10a
1 n1
X
NT
Eo io Ri Cn1 Uzin1
3:5:10b
n1
X
NT X
NT
Cn` Sxin` exp
qn` S` Cn;`1 Sxin;`1 ` 1; . . . ; N` 1
n1 n1
3:5:11a
X
NT X
NT
Cn` Uzin` exp
qn` S` Cn;`1 Uzin;`1 ` 1; . . . ; N` 1
n1 n1
3:5:11b
X
NT
ky3i
Cn;N` SxinN` exp
qn;N` SN` T
3:5:12a
n1
3 i
X
NT
Cn;N` UzinN` exp
qn;N` SN` Ti
3:5:12b
n1
K1 C1 K
2 C2
K2 C2 K3 C3
..
.
KN` 1 CN` 1 K
N` CN`
3:5:14
C1
K1 1
K
2 C2
C2
K2 1
K
3 C3
..
.
3:5:15
CN` 2
KN` 2 1
K
N` CN` 1
1
CN` 1
KN` 1
K
N` CN`
so that
2Eo cos 1
Vi i;o
3:5:20
n1
Letting
1 ky1i
R Uzin1 Sxin1
1
NT =2NT
i MT ; . . . ; 0; MT
n 1; . . . ; NT
V Vi i;o i MT ; . . . ; 0; . . . ; MT
2 3
V
6 7 " #
607 C1
6 7
Vext 6 7 C
3:5:21
607 CN`
4 5 2NT
0 2NT
2 3
R
1 0
6 7
A6
4 M I 7
5
3:5:22
0 R
N` 2NT 2NT
We ®nd the ®nal matrix equation from which C1 and CN` can be found. It is
given by
Vext A C 3:5:23
where Ct Ct1 ; CtN` ] and t refers to the matrix transpose. Inversion of this
equation gives C1 and CN` . Power may be analyzed in the same way as in the
182 Chapter 3
single layer E-mode case. The analysis of the H-mode multilayer case is very
similar to that of the E-mode case. The analysis of the H-mode multilayer
case is very similar to that of the E-mode case.
to the one we present here, almost identical results cam from the two for-
mulations. In Fig. 30 we mention that conservation of the real power was
observed to a high degree of accuracy.
Figure 31 shows transmitted (i 1; 0; 1; 2 and re¯ected (i 0; 1)
diffraction ef®ciencies of a sinusoidal surface relief grating when an E-
mode polarized plane wave is incident on the grating rather an H-mode
polarization. The layer thickness is taken to be L k0 L~ and extend from
peak to trough of the grating as in the previous ®gure. In this example
2a 1 1, 2b 3 4, ~ , N` 5, and 30 . In this example
the full ®eld formulation of Section 3.3 was used to calculate the EM ®elds
of the diffraction grating system in each thin layer. As can be seen from the
plots of Fig. 31, diffraction power is mainly transferred from the zero-order
diffracted power to the ®rst-order with a small amount of diffracted power
being re¯ected and transferred into higher order modes. This example has
been analyzed by Yamakita et al. [55, Fig. 6, p. 156] who used a multilayer
coupled mode method to determine the EM ®eld of the system. In compar-
ing their ®gure to the present one, almost identical results came from the two
formulations. In this paper [55] the numerical value of the diffraction orders
was opposite of that used in Ref. 19 and that used in this section. (That is,
X
1 X
1
~pq 0 pq;ii exp
j
iKx x iKz z
3:6:1
i1 i 1
!
2 X !
E S i;i
y exp
jkxi x jkzi z
3:6:2
i;i
where
where
p
where k~xo ko kxo , k~zo ko kzo , n1 1 1 , and , are the incident angles
of the incoming plane wave. The quantity n1 is the index of refraction of
!
2
Region 1. The magnetic ®eld H in Region 2 can also be expanded in
Floquet harmonics, and it is given by
!
!
2 U 1 X!
H U i;i
y exp
jkxi x jkzi z
3:6:5
o o i;i
! !
r S jU
3:6:5a
! !
r U j S
3:6:5b
! @Fzii
r F ii a^ x jkzi Fyii
@y
h i
a^ y jkzi Fxii jkxi Fzii
3:6:6
@Fxii j
a^ z jkxi Fyii e i;i
@y
Further, if we let f represent any of the dyadic components xx , xy ; . . ., with
fi;i representing the Fourier amplitudes, and let F represent Sx , Sy , Sz , Ux ,
Uy , Uz , we ®nd that
2 3
X
1 X
1 X
1 X
1
fF 4
fi i 0 ;i i 0 Fi 0 ;i 0
5 exp
jkxi x jkzi z
3:6:7
i1 i 1 i 0 1 i 0 1
@Szii
jkzi Syii jUxi;i
3:6:8a
@y
jkzi Sxii jkxi Szii jUyii
3:6:8b
@Sxii
jkxi Syii jUzii
3:6:8c
@y
188 Chapter 3
@Uzii X
jkzi Uyii j
@y i;i 0
h i
xx;i i 0 ;i i0 Sxi 0 i 0 xx;i i 0 ;i i0 Syi 0 i 0 xz;i i 0 ;i i0 Szi 0 i 0
3:6:8d
X
jkzi Uxii jkxi Uzii j
i 0 ;i 0
h i
yx;i i 0 ;i i 0 Sxi 0 i 0 yy;i i 0 ;i i 0 Syi 0 i 0 yz;i i 0 ;i i 0 Szi 0 i 0
3:6:8e
@Uzii X
jkxi Uyii j
@y i 0 ;i 0
h i
zx;i i 0 ;i i0 Sxi 0 i 0 zy;i i 0 ;i i0 Syi 0 i 0 zz;i i 0 ;i i0 Szi 0 i 0
3:6:8f
and
where pqii;i 0 i 0 is the
ii;
i 0 i 0 matrix element of matrix pq and pqi i 0 ;i i 0 is
the Floquet harmonic, which de®nes pqii;i 0 i 0 .
With these matrix de®nitions it is possible to express Eqs. 3.6.8a±f in
matrix form. Doing so we ®nd that Eqs. 3.6.8a±f satisfy
@Sz
jKz Sy jUx
3:6:12a
@y
Kz Sx Kx Sz Uy 3:6:12b
@Sz
jKx Sy jUz
3:6:12c
@y
@Uz
jKz Uy j xx Sx xy Sy xz Sz
3:6:12d
@y
jKz Ux jKx Uz j yx Sx yy Sy yz Sz
3:6:12e
@Ux
jKx Uy j zx Sx zy Sy zz Sz
3:6:12f
@y
2 3 2 32 3
Sx A11 A12 A13 A14 Sx
6 7 6 7 6 7
@ 6 7 6
6 Sz 7 6 A21 A22 A23 A24 7
7
6 Sz 7
6 7
6 76 7 6 7
3:6:14
@y 6 Ux 7 6 A31 A32 A33 A34 7 6 Ux 7
4 5 4 5 4 5
Uz A41 A42 A43 A44 Uz
190 Chapter 3
where
A24 j Kz yy1 Kx
@V
AV
3:6:18
@y
The solution of this equation by the state variable method, as has been
discussed previously, is given by
X
NT
Vn
y Vn eqn y
3:6:20
n1
where
AVn qn Vn 3:6:21
The quantities Vn and qn are the eigenvectors and eigenvalues of the matrix
A and have dimension NT 4
2MT 1
2MT 1. The electromagnetic
®elds in Region 2 are given by
!
X X
NT
Eq
2
x; y; z Cn Sqiin e qn y
exp j
kxi x kzi z
ii n1
and
!
1X X
NT
Hq
2
x; y; z C U e qn y
e j
kxi xkzi z
q
x; y; z
o i;i n1 n qiin
3:6:22
where Sxiin , Ssiin , Uxiin , and Uxiin are eigenvectors obtained from the eigen-
vector Vn . The quantities Syiin and Uyiin are obtained from Eqs. 3.6.13a,b,
using the known eigenvectors Sxiin , Sziin , Uxiin , and Uziin .
Now that the EM ®elds have been determined in Region 2, the next
step is to determine the EM ®elds in Region 1 (incident side) and Region 3
(transmit side) of the grating. The ®elds in Region 1 consist of an obliquely
incident plane wave and consist of an in®nite number of Floquet harmonic
re¯ected waves. Using the coordinates shown in Fig. 32, and assuming that
the incident plane wave has polarization ( and are the incident angles of
the incoming wave),
192 Chapter 3
!I I
E E a^ EI a^ exp
jkxi x jky1ii y jkzi z ii0
3:6:23
where
(
1 k2xi k2zi 1=2 k2xi k2zi 1
ky1ii
3:6:24
jk2xi k2zi 1 1=2 k2xi k2zi > 1
I
and letting ii kxi x kyii y kzi z, and substituting a^ and a^ , we ®nd that
!I
E cos EI cos sin EI a^x sin EI a^y
sin EI cos cos EI a^ z exp
j I00
3:6:26a
!I Xh I I I
i
E Exii a^ x Eyii a^ y Ezii a^ z exp
j Iii ii;00
3:6:26b
i;i
where ii;00 i;0 io and ; is the Kronecker delta. The incident magnetic
®eld can be determined from the second Maxwell curl equation. We have
!I 1 !I
H r E
3:6:27a
j0
!I 1 nh I I
i h
I I
i
H kylii Ezi;i kzi Eyii a^ x kzi Exii kxi Ezii a^ y
0
h i o
I I
kxi Eyii ky1ii Exi;i a^ z exp
j Iii ii;00
3:6:27b
!I Xh I I I
i
I
H Hxii a^ x Hyii a^ y Hzii a^ z exp
j ii ii;00
3:6:27c
i;i
!R X !R R
E E i;i exp
j ii
3:6:28a
i;i
!R X !R R
H H ii exp
j ii
3:6:28b
i;i
where
!R h i
E ii Rxii a^ x Ryii a^ y Rzii a^ z
3:6:28c
R
ii kxi x ky1ii y kzi z
3:6:28d
Notice that, in Eq. 3.6.24 for the case that ky1ii is evanescent, e j
jjky1ii jy
e jky1ii jy ! 0 as y ! 0. We thus see that for the evanescent plane wave
wavenumber
h i1=2
ky1ii j k2xi k2zi 1 ; k2xi k2zi > 1
3:6:29
the minus is the correct root. This is the one used in Eq. 3.6.24. The re¯ected
magnetic ®eld in Region 1 is given by
!R h R R R
i
H ii Hxii a^ x Hyii a^ y Hzii a^ z
1 nh i
ky1ii Rzii kzi Ryii a^x
0
3:6:30
h i
kzi Rxii kxi Rzii a^ y
h i o
kxi Ryii ky1ii Rxii a^ z
!I !R
In Eq. 3.6.26 for E and Eq. 3.6.28a for E the longitudinal y-electric ®eld
component Ey can be expressed in terms of the tangential electric ®elds Ex
and Ez . Using the electric ¯ux density equation in Region 1,
! ! ! !
r D r 1 E 1 r E 0 r E 0
3:6:31
! !I !R
where E represents either E or E . Using this equation we have
194 Chapter 3
" #
!I X !I j I
r E r E ii e ii
ii;00 0
3:6:32a
ii
I I
jkx0 Ex00 jky100 Ey00 jkz0 Ez00 0
3:6:32b
kx0 I k
I
Ey00 Ex00 z0 Ez00
I
3:6:32c
ky100 ky100
!R
and for E
" #
!R X j R
r E 0r Rii e ii
3:6:33a
ii
I 1 I I I I
Hx00 Yxx00 Ex00 Yxz00 Ez00
3:6:34a
0
I 1 I I I I
Hz00 Yzx00 Ex00 Yzz00 Ez00
3:6:34b
0
where
I kz0 kx0
Yxx00
ky100
I k2z0
Yxz00 ky100
ky100
3:6:35
k2
I
Yzx00 ky100 x0
ky100
k k
I
Yzz00 x0 z0
ky100
Planar Diffraction Gratings 195
I
The quantities Ypq00 ,
p; q
x; z may be considered the normalized
surface admittances of the system. They are analogous to the surface aper-
ture admittances used in k-space theory to analyze radiation from inhomo-
geneously cover surface aperture antennas [Chapter 2, this book, [1], Rhodes
and Galejs[7]]. The tangential magnetic ®eld re¯ected modal amplitudes can
also be expressed in terms of the tangential re¯ected electric ®eld modal
amplitudes Rxii and Rzii using Eq. 3.6.30. We have
R 1h R R
i
Hxii Yxxii Rxii Yxzii Rzii
3:6:36a
0
R 1h R R
i
Hzii Yzxii Rxii Yzzii Rzii
3:6:36b
0
R kzi kxi
Yxxii
kylii
R k2zi
Yxzii kylii
ky1ii
R k2xi
Yzxii ky1ii
ky1ii
R kxi kzi
Yzzii
3:6:37
ky1ii
The analysis for the EM ®elds in Region 3 on the transmit side is very
similar to the analysis made in Region 1. In Region 3 the electric and
magnetic ®elds consist of an in®nite number of Floquet harmonic diffracted
plane waves. The electric ®eld in Region 3 is given by
!
3 !T Xh i
j T
E E Txii a^ x Tyii a^ y Tzii a^ z e ii
3:6:39
ii
where
T
ii kxi x ky3ii
y L kzi z
3:6:40
196 Chapter 3
where
8h i1=2
>
< 3 k2xi k2zi k2xi k2zi 3
ky3ii h i1=2
3:6:41
>
: j k2 k2
xi zi 3 k2xi k2zi > 3
Note that when the plane wave is evanescent
k2xi k2zi > 3 , the exponent in
3.6.39 tends to zero as y ! 1. Note that in 3.6.40, Tii has been chosen so
that Tiijy L kxi x kzi z, which simpli®es boundary matching. Using the fact
that
" #
!T !T X !T j T
0 r D 3 r E 3 r E ii e ii
3:6:42
ii
The magnetic ®eld in Region 3 can be found from Maxwell's ®rst curl
equation. We have
"
!
3 !
T Xh i
j T
T T T
H H Hxii a^ x Hyii a^ y Hzii a^ z e ii
3:6:45
ii
!
3 1 Xnh i
H ky3ii Tzii kzi Tyii a^ x
0 ii
h i
kzi Txii kxi Tzii a^y
h i o
j T
kxi Tyii ky3ii Txii a^ z e ii
3:6:46
Using Eq. 3.6.44, Tyii can be expressed in terms of Txii and Tzii . Thus it is
possible to express all the magnetic ®eld components in terms of Txii and
T T
Tzii . The tangential magnetic ®eld modal amplitudes Hxii and Hzii are given
by
Planar Diffraction Gratings 197
T 1h T T
i
Hxii Yxxii Txii Yxzii Tzii
0
3:6:47
T 1h T T
i
Hzii Yzxii Txii Yzzii Tzii
0
where
T kzi kxi
Yxxii
ky3ii
T k2zi
Yxzii ky3ii
ky3ii
3:6:48
T k2xi
Yzxii ky3ii
ky3ii
T kxi kzi
Yzzii
ky3ii
The next step in the analysis is to match the EM ®eld solutions at the y 0
and y L interfaces and determine all the unknown constants of the
system.
Now that the EM ®elds have been de®ned in Regions 1, 2, and 3, the
next step in the analysis is to match the tangential electric and magnetic
®elds at boundary plane y 0 and y L. At y 0 we have
1
2
Ex;z Ex;z
y0 y0
3:6:49
1
2
Hx;z Hx;z
y0 y0
" #
Xh I
i
jkxi x jkzi z
X XNT
jkxi x jkzi z
Exii ii;00 Rxii e Cn Sxiin e
i;i i;i n1
" #
Xh I i
jkxi x jkzi z
X XNT
jkxi x jkzi z
Ezii ii;00 Rzii e Cn Sziin e
i;i ii n1
198 Chapter 3
1 Xh I I I I
Yxxii Exii Yxzii Ezii ii;oo
o ii
" #
i 1X X
NT
R R jkxi x jkzi z jkxi x jkzi z
Yxxii Rxii Yxzii Rzii e C U e
o ii n1 n xiin
1 Xh I I I I
Yzxii Exii Yzzii Ezii ii;oo
o ii
" #
i 1X X
NT
R R jkxi x jkzi z jkxi x jkzi z
Yzxii Rxii Yxzii Rzii e C U e
o ii n1 n ziin
3:6:50
3
2
Ex;z Ex;z
y L y L
3
2
Hx;z Hx;z
y L y L
" #
X X XNT
jkxi x jkzi z qn L jkxi x jkzi z
Txii e Cn Sxiin e e
ii ii n1
" #
X X XNT
jkxi x jkzi z qn L jkxi x kzi z
Tzii e Cn Sziin e e
ii ii n1
" #
1 Xh T i 1X X
NT
T jkxi x jkzi z qn L
Yxxii Txii Yxzii Tzii e C U e e jkxi x jkzi z
o ii o ii n1 n xiin
" #
1 Xh T i 1X X
NT
T jkxi x jkzi z
Yzxii Txii Yzzii Tzii e C U e jkxi x jkzi z
o ii o ii n1 n ziin
3:6:51
X
NT
I
Exii ii;oo Rxii Cn Sxiin
3:6:52
n1
Planar Diffraction Gratings 199
I
X
NT
Ezii ii;oo Rzii Cn Sziin
3:6:53
n1
XNT
I I I I R R
Yxxii Exii Yxzii Ezii ii;oo Yxxii Rxii Yxzii Rzii Cn Uxiin
3:6:54
n1
XNT
I I I I R R
Yzxii Exii Yzzii Ezii ii;oo Yzxii Rxii Yzzii Rzii Cn Uziin
3:6:55
n1
We can eliminate Rxii and Rzii and determine equations for Cn alone. We
have
" #
X
NT
I I I I R I
Yxxii Exii Yxzii Ezii ii;oo Yxxii Exii ii;oo Cn Sxiin
n1
" #
3:6:56
R I
X
NT X
NT
Yxzii Ezii ii;oo Cn Sziin Cn Uxiin
n1 n1
X
NT h i
3:6:57
R R
Cn Yxxii Sxiin Yxzii Sziin Uxiin
n1
X
NT h i
3:6:58
R R
Cn Yzxii Sxiin Yzzii Sziin Uziin
n1
X
NT
qn L
Txii Cn Sxiin e
n1
X
NT
qn L
Tzii Cn Sziin e
n1
X
NT
T T qn L
Yxxii Txii Yxzii Tzii Cn Uxiin e
3:6:59a
n1
T T
X
NT
qn L
Yzxii Txii Yzzii Tzii Cn Uziin e
3:6:59b
n1
X
NT n o
qn L T T
0 Cn e Yxxii Sxiin Yxzii Sziin Uxiin
3:6:60a
n1
X
NT n o
qn L T T
0 Cn e Yzxii Sxiin Yzzii Sziin Uziin
3:6:60b
n1
~ x
~ z =2
2 !I !I
PIc ~x
E H
a^ y d xd
~ z
~
3:3:62
~ z =2
2
or after being put in normalized form and carrying out the a^ y dot product,
z =2
x =2 h i
1 I
PIc I
Ez00 I
Hx00 I
Ex00 Hz00 dx dy
3:6:63
k2o z =2 x =2
Planar Diffraction Gratings 201
x z n I h I I I I
i h
I I I
io
PIc Ez00 Yxx00 Ex00 Y xz00 Ez00
I
Ex00 I
Yzx00 Ex00 Yzz00 Ez00
k20 0
3:6:64
I I
The quantities Ex00 and Ez00 are given in terms of the incident angles and
polarizations by Eq. 3.6.26a.
The re¯ected power from the crossed grating is given by
PR 12 Re
PR
c
3:6:65
where
~ z =2
~ x =2
1 !R !R
PR
c 2 E H
a^ y dx dz
3:6:66
ko z =2 x =2
z =2
x =2 h i
1
PR
c 2 EzR HxR ExR HzR dx dz
3:6:67
ko z =2 x =2
1 R
PR
c Izx R
Ixz
3:6:68
k2o
R R
where Izx and Ixz refer to the ®rst and second terms in Eq. 3.6.67. If we
substitute Ex and EzR into Izx , we ®nd after interchanging summation and
R
integration that
XX h i
R R
Izx Rzii Yxxi 0 i 0 Rxi 0 i 0 Yxzi 0 i 0 Rzi 0 i 0
ii i 0i 0
x =2
z =2
3:6:69
j
kxi kxi 0 x j
kzi kzi 0 z
e dx e dz
x =2 z =2
The ®rst integral (x-integral) equals x ii 0 and the second integral equals
z ii 0 , where ii 0 is the Kronecker delta. Substituting these values in Eq.
3.6.69 we ®nd that
X h i
R R
Izx x z Rzii Yxxii Rxii Yxzi
0 i 0 Rzii
3:6:70
ii
202 Chapter 3
R
Carrying out a similar analysis for Ixz and substituting the expressions into
Eq. 3.6.68, we ®nd that
X
PR
c PR
cii
ii
x z Xnh R R
i
R zii Y R
xxii xii Y R
xzii zii
3:6:71
k2o o ii
h io
R R
Rxii Yzxii Rxii Yzzii Rzii
PT 12 Re PTc 3:6:72
where
z =2
x =2
!T !T
PTc E H a^y d x~ d z~
3:6:73
z =2 x =2
z =2
x =2 h i
1
PTc EzT HzT ExT HzT dx dz
3:6:74
k2o z =2 x =2
Substituting the transmitted electric and magnetic ®elds into Eq. 3.6.74
for PTc , and carrying out an analysis similar to that used to determine PR
c , we
®nd that
X x z Xn h T T
i
PTc PTcii T zii Y xxii Txii Y xzii Tzii
ii
k2o o ii
3:6:75
h io
T T
Txii Yzxii Txii Yzzii Tzii
Re
PR
2 cii
DR
ii
3:6:76
Re
PI
2 c
Re
PT
cii
DTii 2
3:6:77
Re
PI
2 c
For a lossless crossed grating, the re¯ected and transmitted diffraction ef®-
ciencies obey the conservation of power relation
X
DR T
ii Dii 1
3:6:78
ii
are the Fourier coef®cients of the square pro®le used in the square wave
example of Section 3.2. The E-mode square wave case was studied using the
same parameters as the H-mode square wave case except that the polariza-
tion was taken to be EI 0, EI E0 . Figure 34 shows the diffraction
ef®ciency results using the one-dimensional theory of Sections 3.2 and 3.3
and using the crossed diffraction grating theory of this section. As can be
seen from Fig. 34, nearly identical diffraction ef®ciency results from the two
algorithms.
The crossed diffraction grating theory has been also used to calculate
the scattering from the H-mode cosine grating (Figure 2) (Gaylord [16])
described in Section 3.2. After setting the parameters of the crossed grating
algorithm to match those of the H-mode cosine grating, identical diffraction
204 Chapter 3
ef®ciency results were obtained for the one- and two-dimensional RCWA
algorithms for the case also.
Figure 35 shows the diffraction ef®ciency data that results when the
crossed grating theory of this section is applied to study scattering from a
two-dimensional crossed cosine wave grating where 1 2 3 1,
10 , ~ x 2:8747, and ~ z 1:5~ x , L~ 9, MT 3, MT 3,
270 , EI 1, and EI 0. The nonzero relative dielectric permittiv-
ities were taken to be
MT
X X
MT
xx
x; z yy
x; z zz
x; z ii expj
iKx x iKz z
i MT i MT
Figure 35 The diffraction ef®ciency data that results when crossed grating theory
is applied to study scattering from a two-dimensional crossed cosine wave grating is
shown.
In Eq. 3.6.79, 0;0 1 ; 1;1 1;1 1; 1 1; 1 1=4, and all other
Fourier coef®cients i;i in Eq. 3.6.79 are zero. In Fig. 35 transmitted diffrac-
tion ef®ciencies (denoted by Ti;i ) of the T00 , T01 , T10 , and T11 orders was
plotted versus the azimuthal angle , which was varied over the range
180 270 . As can be seen from the Fig. 35 plot, changing the
angle of incidence causes a perceptible change in the diffraction ef®ciency
observed from the grating. In making the Fig. 35 plot conservation of real
power, Eq. 3.6.78 was observed to a high degree of accuracy.
Table 1 shows the transmitted diffraction ef®ciency for the crossed
cosine diffraction grating studied in Fig. 35 (taking 270 ) that results
for ®fteen orders (taking all pair combinations of i 2; 1; 0; 1; 2 and
i 1; 0; 1) when ®ve different matrix truncations MT MT 1; 2; 3;
4, 5 are used. (For those truncations where the i; i) pair order exceeds the
truncation order [for example, when the pair
i; i
2; 1 exceeds the
truncation order, MT MT 1 the diffraction ef®ciency is set to zero.)
A striking and reassuring feature of the diffraction ef®ciencies displayed
in Table 1 is how rapidly the diffraction ef®ciency converges to a ®nal
value that does not change with increasing order. After the value of MT
206 Chapter 3
Figure 36 The transmitted diffraction ef®ciencies (solid line) of the Ti;i orders
when i and i rnge from
1; 0; 1, when the grating thickness is varied from L~
0 to L~ 4, are shown.
where
208 Chapter 3
3 ~ x~ 1 ; jzj
jxj ~ z~1
x; y; z ~ x ; jzj ~z
1 ~
elsewhere in the cell jxj ~
3:6:81
and where x~ 1 ~ x =2 and z~1 ~ z =2. Fourier inversion of Eq. 3.6.1 using
the speci®ed permittivity value given by Eq. 3.6.81 speci®es the two-dimen-
sional Fourier coef®cients of Eq. 3.6.80. Figure 36 shows the transmitted
diffraction ef®ciencies (solid line) of the Ti;i orders where i and i range from
1; 0; 1; when the grating thickness is varied from L~ 0 to L~ 4; when
1 1, 3 2:5, 30 , ~ x , ~ z 1:5
~ x , 270 , EI 1, and
I
E 0; and when MT MT 3. As can be seen from Fig. 36, EM power
is diffracted out of the T0;0 order (pump wave or incident wave) and
is subsequently diffracted into the higher orders. Because of symmetry,
the diffraction ef®ciencies of the T1;1 and T1; 1 orders were the same
and the diffraction ef®ciencies of the T0;1 and T0; 1 orders were the same.
By the same token, for 0 , 90 , and 180 , we should observe similar
symmetry in the diffracted orders. Figure 36 also shows the diffraction
ef®ciency of the T1;0 order (dotted) when the truncation was taken to be MT
MT 2 rather than MT MT 3 as was done for the curves discussed
earlier. As can be seen from the ®gure, very little diffraction ef®ciency
difference exists between the two truncations.
Table 2 shows the transmitted diffraction ef®ciency for the crossed
rectangular diffraction grating studied in Fig. 36 (taking 270 and
L~ 1:7) that results for ®fteen orders (taking all pair combinations of i
2; 1; 0; 1; 2 and i 1; 0; 1) when ®ve different matrix truncations MT
MT 1; 2; 3; 4; 5 are used. (For those truncations where the
i; i pair order
exceeds the truncation order [for example, when the pair
i; i
2; 1
exceeds the truncation order, MT MT 1] the diffraction ef®ciency is
set to zero.) A striking and reassuring feature of the diffraction ef®ciencies
displayed in Table 2, like those of Table 1, is that the diffraction ef®ciency
converges fairly rapidly to a ®nal value that does not change with increasing
order. In comparing Table 2 to Table 1 it is interesting to note that the
convergence to a ®nal value is slightly slower in Table 2 than in Table 1. This
is believable since the grating studied in Table 2 is much smaller in size than
the grating studied in Table 1 and also the grating studied in Table 2 has a
much higher spatial spectral content than the grating studied in Table 1
(cosine grating). Both these factors would cause a slower convergence
with truncation order.
The third diffraction ef®ciency example to be presented consists of the
diffraction ef®ciency data that results when the crossed grating theory of
this section is applied to study scattering from a two-dimensional rectangu-
Planar Diffraction Gratings 209
where
n2o n2e
b
3:6:83
2
Cx sin
c sin
c
Cy cos
c sin
c
Cz cos
c
~ x =2, z~1
where C C 45 , x~ 1 ~ z =2, a 1,
~ x , and
~ ~
z 1:5x .
Using these parameters and functions we de®ne the relative dielectric
permittivity to be
Planar Diffraction Gratings 211
2 3
xxC xyC xzC
x; z 4 yxC yyC yzC 5f
x; z
3:6:84
zxC zyC zzC
Figure 37 Plots of the diffraction ef®ciency of the T10 order
i 1; 0 when the
grating thickness is varied from L~ 0 to L~ 4, are shown.
212 Chapter 3
where
p
3jyj
x1
y
4x
p
3jyj
z1
y
4z
where the parameters of Eq. 3.6.86 are already given in Eq. 3.6.83. Because
the grating is longitudinally inhomogeneous, a two-dimensional multilayer
analysis based on the theory of Section 3.5 was used to calculate the diffrac-
tion ef®ciency.
Figure 38 shows plots of the diffraction ef®ciency of the T00 , T10 , T01 ,
and T11 orders; when the grating thickness is varied from L~ 0 to
L~ 2:5; 3 2:5,
~ x , and ~ z 1:5
~ x ; 30 , 270 , EI 1,
and E 0; when MT MT 2; when the values of the parameters n2o , n2e
I
were taken to be n2o 2, n2e 3; and when ten layers
N` 10) were used to
carry out the two-dimensional multilayer analysis. As can be seen from Fig.
38, for the grating under study, power is diffracted out of the T00 order into
higher orders. Conservation of power as speci®ed by equations was
observed to a high degree of accuracy.
In the ®nal example, a crossed pyramidal diffraction grating is again
studied (same pyramid geometry as Fig. 37), but with a mirror (or a per-
fectly conducting short circuit plate) placed at the Region 2±Region 3 inter-
face at y~ L~ (see Fig. 39). In this case just the re¯ected diffraction
ef®ciency was studied (the transmitted diffraction ef®ciency in Region 3 is
zero). The overall EM analysis in this case requires that the tangential EM
®elds at y~ L~ be zero. Imposing this condition (see Section 3.2.4 for an H-
214 Chapter 3
Figure 38 Plots of the diffraction ef®ciency of the T00 , T10 , T01 , and T11 orders,
when the grating thickness is varied from L~ 0 to L~ 2:5, are shown.
In Section 3.4 and later sections a multilayer analysis was used to solve for
the diffraction from both one-dimensional and two-dimensional diffraction
gratings. The method of analysis was to divide the longitudinally inhomo-
Planar Diffraction Gratings 215
! X
H g` Hy;g` a^ y Uyi`
z exp
jkxi x
i
! X
E g` Sx;g` a^ x Sz;g` a^z j0 Szi`
za^ x Szi`
za^ z exp
jkxi x
i
X
2`
x i` exp
i j x
i
3:7:1
~
where kxi kx0 i, i . . . 2; 1; 0; 1; 2; . . . ; 2=, and k0 , ~
is the period of the diffraction grating. If these space harmonic expansions
are substituted in Maxwell's equations, appropriate derivatives are taken,
and the coef®cients of the exponential terms are equated, it is found that the
following coupled equations result:
Planar Diffraction Gratings 217
Figure 40 Geometry and coordinate system for the diffraction grating problem
under consideration. The column matrices C` and C ` represent the coef®cients of
the forward and backward traveling waves in each thin layer, and 2`
x; z repre-
sents the periodic dielectric function in each thin layer. Used with permission of
OSA, 1995 [23].
@Uyi` X
`;i i 0 Sxi 0 `
@z i0
X
jkxi Uyi` `;i i 0 Szi 0 `
3:7:2
i0
@Sxi`
jkxi Szi` Uyi`
^
@z
@Uy`
` Sx`
3:7:3a
@z
218 Chapter 3
If we invert ` in Eq. 3.7.3b and thus express Sz` j
` 1 Kx Uy` and further
substitute Sz` in Eq. 3.7.3c we ®nd that
@Sx`
jKx
j` 1 Kx Uy` Uy`
3:7:4
@z
Taking the derivative of Eq. 3.7.3a and substituting Eq. 3.7.4 we ®nd that
@2 Uy` h i
` Kx ` 1 Kx I Uy`
3:7:5
@z2
@2 Uy`
E` B` Uy` A` Uy`
3:7:6
@z2
@Uy`
Sx` E` 1
3:7:7
@z
U
yn`
z W`n exp
qn` z
3:7:9
@2 @2
2
Uy`n
z W`n 2 exp
qn` z q2n` W`n exp
qn` z
@z @z
3:7:10
A` Wn` exp
qn` z A` Uyn`
z
EV
X
Uyn`
x; z Win` exp qn`
z D` d` exp
jkxi x
3:7:11
i
@ EV EV
U
x; z 2`
xSxn`
x; z
3:7:12
@z yn`
EV
After expressing 2`
x and Sxn`
x; z in an exponential Fourier series sum,
EV
combining the Fourier sums in the product term `
xSxn`
x; z into a
convolution summation term, differentiating Eq. 3.7.11, and equating coef-
®cients of exp
jkxi x, it is found that
X
qn` Win` i i 0 ;` Sxi 0 n`
3:7:13
i0
0
Using matrix inversion one ®nds the amplitude Sxi 0 n` . It is given by the i th
S 1
xn`
qn` E` Wn` Vn`
3:7:14
The tangential magnetic ®eld associated with the nth mode in the `th
layer associated with the eigensolution Uyn`
z Wn` exp
qn` z is given by
X
Uyn`EV
x; z Win` expqn`
z D` exp
jkxi x
3:7:17
i
@
U EV
x; z 2`
xSxn`EV
x; z
3:7:18
@z yn`
After carrying out the differentiation in Eq. 3.7.18 and equating coef®cients
of exp
jkxi x, it is found that
X
qn` Win` i i 0 ;` Sxi 0 n`
3:7:19
i0
Using matrix inversion one ®nds the amplitude Sxi 0 n` . It is given by the i0 th
component of the column vector Sxn` Vn` , where Vn` has been de®ned
previously. Using this value it is found that
X
Sxn`EV
x; z Vin` expqn`
z D` exp
jkxi x
3:7:20
i
If we sum the forward and backward tangential magnetic and electric ®elds
as given in Eqs. 3.7.11, 3.7.16, 3.7.17, and 3.7.20 we ®nd that a total expan-
sion of these ®elds is given by
X
Tot EV
Uyn`
x; z Cn` Uy`n
x; z Cn` Uy`nEV
x; z
n
X
3:7:21
Tot EV
Sxn`
x; z Cn` Sx`n
x; z Cn` Sx`nEV
x; z
n
or
Planar Diffraction Gratings 221
Tot
XX
Uy`
x; z Win` Cn` exp qn`
z D` d`
i n
Cn` expqn`
z D` exp
jkxi x
XX
3:7:22
Tot
Sx`
x; z Vin` Cn` exp qn`
z D` d`
i n
Cn` expqn`
z D` exp
jkxi x
It is assumed that the incident plane wave amplitude is 1 (V/m). The coef®-
cients Ri represent the amplitudes of the re¯ected, diffracted ®elds in Region
1. The tangential electric ®eld in Region 1 is given by
j0 X
Ex
1 jkzi1 exp
jkzi1 zi0 jkzi1 Ri exp
jkzi1 z exp
jkxi x
1 i
3:7:25
j0 X
Ex
3 jkzi3 Ti exp jkzi3
z DL exp
jkxi x
3:7:28
3 i
1 X
jk jkzi1 Ri Vin1 Cn1 Cn1 exp
qn1 d1
3:7:30
1 zi1 i0 n
where Z1
kzi1 =1 i;i 0 , W1 Win1 , V1 Vin1 , and X1 i;n exp
qn1
d1 where i and n range over the number of space harmonics in the system.
Matching the tangential magnetic ®eld at the Region 2,
` 1th thin
layer interface to the Region 2, `th thin layer located at z D` 1 , where
` 2; . . . ; L, we ®nd that
X
Win;` 1 Cn;` 1 exp
qn;` 1 d` 1 Cn;` 1
n
X
3:7:32
Win;` Cn;` Cn;` 1 exp
qn;` d`
n
Planar Diffraction Gratings 223
X
jkzi3
VinL CnL exp
qnL dL CnL Ti
3:7:36
n
3
2 3
i0
4 jkzi1 5 I W1 W1 X1 C
1
R
3:7:38
jZ1 V1 V1 X1 C1
1 i0
1
C
1
W1 X1 W1 W2 W2 X 2 C
2
3:7:39
C1 V1 X1 V1 V2 V2 X2 C2
so that
2 3
i0 " # " #" # 1
6 7 I W1 W1 X1 W1 X 1 W1
4 jkzi1 5 R
i0 jZ1 V1 V1 X1 V1 X1 V1
1
" #" #
W2 W2 X2 C
2
V2 V2 X2 C2
3:7:40
2 3
i0 " # (" #
6 7 I Y1
L W` W` X `
4 jkzi1 5 R
i0 jZ1 `1 V` V` X`
1
" # 1 91
W` X ` W` =
A
3:7:41
V` X` V` ;
" #" #
WL WL XL C
L
VL VL XL CL
Inverting Eq. 3.7.37 and substituting in the above equation we ®nd that
Planar Diffraction Gratings 225
2 3
i0 " # (" #
6 7 I Y
L W` W` X `
4 jkzi1 5 R
i0 jZ1 `1 V` V` X`
1
" # 1 91" #
W` X ` W` = f L1
A T
V` X` V` ; gL1
3:7:42
1
Using the matrix property
A B B 1A 1
we ®nd that
1
aL WL WL f L1
gL1
3:7:46
bL VL VL
226 Chapter 3
De®ning
" #
fL WL
I XL bL aL 1 XL
TL
3:7:48
gL VL
I XL bL aL 1 XL
We next let
TL aL 1 1 XL 1 TL 1 3:7:50
and let
1
aL 1 WL 1 WL 1 fL
3:7:51
bL 1 VL 1 VL 1 gL
Planar Diffraction Gratings 227
3:7:52
where
" #
fL 1 WL 1
I XL 1 bL 1 aL 1 1 XL 1
TL 1
3:7:53
gL 1 VL 1
I XL 1 bL 1 aL 1 1 XL 1
Repeating this cycle and process until the last layer we ®nd that
2 3
i0
4 jkzi1 5 I f
R 1 T1
3:7:54
jZ1 g1
1 i0
In inspecting Eqs. 3.7.53 and Eqs. 3.7.54 one observes why the present
algorithm is extremely ef®cient and stable. In Eq. 3.7.53, the X` matrix is
diagonal and contains the exponential term exp
qn` d` ). When qn` d` is
large, the exponential term exp
qn` d` is very small, and thus the matrix
X` in this case is near zero. The matrix terms f ` W`
I X` b` a` 1 X` and
g` V`
I X` b` a` 1 X` , which form an important part of the algorithm,
are the only terms that contain exponential terms. Further, the terms in the
matrices making up f ` and g` appear as the sum of matrix element terms
near unity (coming form the identity matrix I) and the exponential terms
(coming from matrix X` b` a` 1 X` ). Thus when the matrix elements of the
matrices X` b` a` 1 X` are exponentially small, the matrix elements making
up f ` and g` are not all near zero (because of the presence of the identity
matrix I). Thus when this procedure is repeated for each layer starting at `
228 Chapter 3
L with the computation of f L and gL and repeated until ` 1 and the last f 1
and g1 is produced, one sees that the ®nal matrix equation for R and T1 ,
which uses the matrices f 1 and g1 , will thus not have ill-conditioned or near-
singular matrices, since f 1 and g1 , which make up this equation, do not
possess all exponentially small terms. It is interesting to note that the prin-
ciple used here is similar to the method used by Yang, Section 2.7, discussed
in the previous chapter. It would be useful to extend this technique to the
case of diffraction from gratings in anisotropic materials.
In this section the enhanced transmittance method has been applied to
the case when the electric ®eld polarization was in the plane of incidence. In
a companion paper [23] written with the paper the present analysis was
based on, the authors present an enhanced transmittance method for H-
mode incidence which deals with the conical plane wave diffraction case.
Figure 42 Geometry for the surface relief grating diffraction problem analyzed
herein. Used with permission of OSA [23, Fig. 2].
230 Chapter 3
Figure 44 Geometry for the nonconical grating diffraction problem for E-mode
TM polarization. The parameters of the grating are: grating thickness L~ :36:36;
grating wavelength ~ :4545; relative permittivity of Region 1, "1 1:0; relative
permittivity of Region 2a (nonmetallic portion of grating), "2a 2:25; relative per-
mittivity of Region 2b (metallic portion of grating), "2b
3:18 j4:412 ; relative
permittivity of Region 3, "3 2:25; and grating duty cycle, 30%. Used with permis-
sion of OSA, 1996 [29, Fig. 1].
The ®eld expansions for this case are given by Eqs. 3.3.4.5. We assume
an isotropic grating in Reg. 2 and in Eq. 3.4.3 we take "xx
x "yy
x
"zz
x; " 0; 6 . With these assumptions we ®nd that Maxwell's equa-
tions in Reg. 2 are given by
@
0 Hz
3:8:1
j"
x Ex
@y
@
0 Hz
j"
x Ey
3:8:2
@x
@Ey @Ex
j
0 Hz
3:8:3
@x @y
@2 U z @ 1 @Uz
"
x "
xUz
3:8:4
@y2 @x "
x @x
232 Chapter 3
Substitution of Eq. 3.3.5 into Eq. 3.8.4 and after collection of coef®cients for
the Fourier exponential term exp( jkxi x, for i MT ; . . . ; 0; . . . ; MT ,
Mt ! 1, the following matrix equation results
@2 U z
E
Kx A Kx IU z
3:8:5
@y2
where E "i;i0 , "i;i0 "i i0 ; A ai;i0 , ai;i0 a i i0 and a i are the Fourier
coef®cients of the reciprocal permittivity function
1 X
M
a i exp
ji x
"
x T MT
The other terms are de®ned in Section 3.3. We note in the limit Mt ! 1
that the matrices E and A are inverses of each other and we thus have E
A 1 and A E 1 . Using this inverse relation we may also express Eq. 3.8.5
as Mt ! 1
@2 U z
A 1
Kx E 1 Kx IUz
3:8:6
@y2
It turns out that the above eigenvalue formulation (Eq. 3.8.6) is super-
ior to the one in Eq. 3.8.5 as far as convergence rates are concerned. As an
example we quote from Lalanne and Morris [29] the case of diffraction from
a chrome (refractive index 3.18-j4.41) lamellar grating deposited on a glass
substrate (see Fig. 44). The diffraction ef®ciency of the zeroth order is shown
in Fig. 45 along with the convergence rates, using Eqs. 3.8.5 (line) and 3.8.6,
(circle) respectively. The results clearly show the superiority of the conver-
gence from the second formulation.
The reason for the improvement in convergence using the second
method above has been shown rigorously by Li [53]. The dif®culty arises
in the convergence because
1. The EM ®elds and the periodic dielectric permittivity are discon-
tinuous at the points in the grating where changes from 2a to
2b , and vice versa yet their product must be a continuous func-
tion.
2. Expressing products of periodic functions as in a Fourier series
(the Fourier coef®cients of the product function are a convolution
of the Fourier coef®cients of each seriesÐthis is called Laurent's
rule) involves a ®nite truncation of the convolution series.
Planar Diffraction Gratings 233
1 ab
h
2
T
0 ab h
0
3:89
1=a 1=b=2 2
Thus, this shows that by (a) taking the ®nite Fourier series fTREC
x; of the
reciprocal function 1=f
x, (b) then inverting fTREC
x to obtain the function
Planar Diffraction Gratings 235
1=fTREC
x, (c) ®nding the ®nite Fourier series expansion of 1=fTREC x, and (d)
multiplying this with the ®nnite Fourier series gT
x, gives the accurate value
of the product h
x at the point of discontinuity. Note that since fTREC
x is a
®nite Fourier series, it is ®nite and continuous at the point of discontinuity
of f
x, and hence the Fourier series of its reciprocal, namely 1=fTREC
x, is
continuous at this point and has the value 1/[(1/a+1/b)/2]. Note further that
we could achieve the same result by interchanging f and g.
Summarizing, the Laurent rule gives an, in general, incorrect value at
the point of discontinuity, whereas the inverse method yields the correct
result. When these types of computations are encountered in grating pro-
blems, it is easy to see why numerical dif®culties encountered with the
Laurent method can be alleviated using the inverse rule.
It is instructive to compute the ®rst and second derivatives of the pro-
duct functions h
iT
x; i 1; 2 around the point of discontinuity x 0. This
gives insight into the way the two approximations vary around the point of
discontinuity (by using these derivatives in a Taylor series). Note that
h
1
T
x fT
xgT
x
c0 DN
x
c0 DN
x
3:8:10
where
X
N
DN
x cn sin nKx; c0
b a=2; cn
2=n
b a;
n1
n odd and cn 0; even;
3:8:11
follows that fTREC
1=abgT , provided the same number of harmonics are
used to expand both functions. Therefore
1
h
2 REC
T
x
1=fT gT
x g ab; for all x:
3:8:12
1=abgT T
PROBLEMS
where 108, "1 1:; "3 2:25; "2a 2:25; "2b 1:; " :5; ~
k0 5; d =5:; L~ :75; k0 2= and is the free space
wavelength. Determine the diffraction ef®ciency of the transmitted
and re¯ected orders of the system if the plane wave is
a) H-mode polarized.
b) E-mode polarized.
c) Verify in Parts a) and b) that conservation of real power is
conserved.
d) What is the approximate number of orders MT needed to
ensure convergence of the EM solution in this problem?
4. In Problem 3, verify the complex Poynting theorem as developed
in Sec. 3.3 for the Poynting box shown in Fig. 15.
5. Determine the diffraction ef®ciency of the transmitted and
re¯ected orders of the system in Problem 3 if an in®nitely thin,
perfectly conducting strips are placed at the interfaces between "2a
and "2b .
6. a) Use the method of [25] to determine the EM ®elds of a pure
re¯ection grating when the permitivity is given by Eq. (3.4.5) and
for the data of Fig. 21 except that " :4 rather than " :2.
b) Find and plot the re¯ection ef®ciencies for data of Problem 5a)
as was done in Fig. 21.
7. Determine the re¯ected and transmitted diffraction ef®ciencies of
an asymmetric diffraction grating when the relative dielectric per-
mitivity is given by
"2a , 0 < x < (=L) y
"
x; y
"2b ,
=Ly < x <
where I 208, I 108, "1 1., "3 2:25, "2 2:25, " :2,
z x k0 ~ x ; x 5:, L~ :75, and k0 2= and is the
free space wavelength. Assume the incident plane wave is circularly
238 Chapter 3
REFERENCES
84. S. L. Chuang and J. A. Kong, Wave scattering from periodic dielectric surface
for a general angle of incidence, Radio Sci. 17, 545±557 (1982).
85. T. K. Gaylord and M. G. Moharam, Thin and thick gratings: terminology
clari®cation, Symbols, Units, Nomenclature, Applied Optics, 20(19), 3271±3273
(1981).
86. T. Jaaskelainen and M. Kuittinen, Inverse grating diffraction problems, SPIE
International Colloquium on Diffractive Optical Elements, 1574, 272±281
(1991).
87. E. Gluch, H. Haidner, P. Kipfer, J. T. Sheridan, and N. Streibl, Form bire-
fringence of surface-relief gratings and its angular dependence, Optics
Commun., 89, 173±177 (1992).
88. J. M. Elson, L. F. DeSandre, and J. L. Stanford, Analysis of anomalous
resonance effects in multilayer-overcoated, low-ef®ciency gratings, J. Opt.
Soc. Am. A, 5, 74±88 (1988).
89. T. Tamir and H. C. Wang, Scattering of electromagnetic waves by a sinusoid-
ally strati®ed half space: I. Formal solution and analysis approximations, Can.
J. Phys., 44, 2073±2094 (1966).
90. T. Tamir, Scattering of electromagnetic waves by a sinusoidally strati®ed half
space: II. Diffraction aspects at the Rayleigh and Bragg wavelengths, Can. J.
Phys., 44, 2461±2494 (1966).
91. S. T. Peng, T. Tamir, and H. L. Bertoni, Theory of periodic dielectric wave-
guides, IEEE Trans. Microwave Theory Tech., MTT-23, 123±133 (1975).
92. T. Tamir, H. C. Wang, and A. A. Oliner, Wave propagation in sinusoidally
strati®ed dielectric media, IEEE Trans. Microwave Theory Tech., MTT-12,
323±335 (1964).
93. R. V. Johnson and R. Tanguay, Optical beam propagation method for bire-
fringent phase grating diffraction, Opt. Eng., 25, 235±249 (1986).
94. H. Haus, Waves and Fields in Optoelectronics, Prentice-Hall, New York, 1984,
Section 11.1.
95. S. Samson, A. Korpel, and H. S. Snyder, Conversion of evanescent waves into
propagating waves by vibrating knife edge, Internat. J. Imaging Systems
Technology, 7, 48±53 (1996).
96. P. N. Butcher and D. Cotter, The Elements of Nonlinear Optics, Cambridge,
UK, 1990, p. 127.
97. T. L. Zinenko, A. I. Nosich, and Y. Okuno, Plane wave scattering and absorp-
tion by resistive-strip and dielectric strip periodic gratings, IEEE Trans.
Antennas Propagation, 46(10), 1498±1505 (1998).
98. E. Popov, L. Tsonev, and D. Maystre, GratingsÐgeneral properties of the
Littrow mounting and energy ¯ow distribution, J. Mod. Optics, 37(3), 367±377
(1990).
99. E. Popov, L. Tsonev, and D. Maystre, Losses of plasmon surface waves on
metallic grating, J. Mod. Optics, 37(3), 379±387 (1990).
100. E. Popov, and L. Tsonev. Total absorption of light by metallic gratings and
energy ¯ow distribution, Surface Science, 230, 290±294 (1990).
244 Chapter 3
In the previous chapter, we have discussed the RCWA in detail and also
shown applications where plane waves are transmitted and re¯ected upon
incidence on a layer of arbitrary permittivity and from ®xed gratings. In this
chapter we discuss an alternative method to determine the propagation of a
beam in a semi-in®nite region that may contain a certain optical inhomo-
geneity, whether ®xed (such as in a grating) or induced (due to the nonlinear
change in refractive index). The extension of this method to analyze pulse
propagation as well has been performed but will not be treated here for the
sake of simplicity.
1 @2 E
r2 E 0
4:1:1
v2 @t2
and substitute
245
246 Chapter 4
@Ee 2
2jk0 r? Ee
4:1:4
@z
2
where r? denotes the transverse Laplacian. Equation 4.1.4 describes the
propagation of the envelope Ee
x; y; z starting from the initial pro®le
Ee jz0 Ee0
x; y.
Equation 4.1.4 can be solved readily using Fourier transform techni-
ques. Assuming Ee to be Fourier transformable, we can employ the de®ni-
tion of the Fourier transform
1
E~ e
kx ; ky ; z F x;y fEe
x; y; zg Ee
x; y; z exp
jkx x jky y dx dy
1
4:1:5
d E~ e j
k2 k2y E~ e
4:1:6
dz 2k0 x
where E~ e0
kx ; ky is the Fourier transform of Ee0
x; y. We can interpret Eq.
4.1.7 in the following way: Consider a linear system with an input spectrum
of E~ e0
kx ; ky at z 0 where the output spectrum at z is given by
E~ e
kx ; ky ; z. The spatial frequency response of the system, which we will
call the paraxial transfer function for propagation, is then given by
E~ e 2 2
z
H
kx ; ky ; z exp j kx ky
4:1:8
E~ e0 2k0
This, when convolved with the initial beam pro®le, yields the pro®le of the
diffracted beam in the spatial domain directly. This convolution integral is
in fact the Fresnel diffraction formula.
@Ee 1 2
r E jnk0 Ee
4:2:1
@z 2jk0 ? e
The quantity n is the change in the refractive index over the ambient
refractive index n0 c=v, where c is the velocity of light in vacuum.
Equation 4.2.1 is a modi®cation of Eq. 4.1.4 and can be derived from the
scalar wave equation when the propagation constant, or equivalently the
velocity of the wave, is a function of
x; y; z explicitly, as in gratings or
®bers, or implicitly, such as through the intensity-dependent refractive
index.
The paraxial propagation equation (4.2.1) is a partial differential
equation that does not always lend itself to analytical solutions, except for
some very special cases involving special spatial variations of n, or when,
as in nonlinear optics, one looks for a particular soliton solution of the
resulting nonlinear PDE using exact integration or inverse scattering meth-
ods. Numerical approaches are often sought to analyze beam (and pulse)
propagation in a complex system such as optical ®bers, volume diffraction
gratings, Kerr and photorefractive media, etc. A large number of numerical
methods can be used for this purpose. The pseudospectral methods are often
favored over ®nite difference methods due to their speed advantage. The
248 Chapter 4
@Ee ^ e
D^ SE
4:2:2
@z
^
Ee
x; y; z z exp
D^ SzEe
x; y; z
4:2:3
^
exp
Dz ^
Sz ^
exp
Dz ^
exp
Sz
4:2:5
which implies that in Eq. 4.2.4 the diffraction and the inhomogeneous
operators can be treated independently of each other.
The action of the ®rst operator on the RHS of Eq. 4.2.5 is better
understood in the spectral domain. Note that this is the propagation opera-
tor that takes into account the effect of diffraction between planes z and
z z. Propagation is readily handled in the spectral or spatial frequency
domain using the transfer function for propagation written in Eq. 4.1.8 with
z replaced by z. The second operator describes the effect of propagation in
the absence of diffraction and in the presence of medium inhomogeneities,
either intrinsic or induced, and is incorporated in the spatial domain. A
schematic block diagram of the BPM method in its simplest form is
shown in Fig. 1. There are other modi®cations to the simple scheme, viz.,
the symmetrized split-step Fourier method and the leap-frog techniques.
These are discussed in detail elsewhere [2].
The Split-Step Beam Propagation Method 249
In this section we will illustrate various cases where the BPM can be used to
analyze propagation in inhomogeneous media. While most of the examples
will be connected with beam propagation, we must point out to readers that
the method can be used to analyze pulse propagation as well, simply by
replacing z in Eq. 4.2.2 with t (time) and making the linear spatial transverse
differential operator a similar differential operator in z. With this modi®ca-
tion, Eq. 4.2.2 can model the propagation of one-dimensional longitudinal
pulse through an optical ®ber with arbitrary group velocity dispersion. For
details, we refer the readers to Agrawal [2].
Figure 2 Diffraction of a Gaussian beam during free space propagation. (a) Pro®le
at z 0 (plane wave fronts assumed); (b) pro®le at z zR , where zR is the Rayleigh
length of the original Gaussian beam.
where n0 denotes the intrinsic refractive index of the medium and n
2 is a
measure of the gradation in the refractive index. In this case, the operator S^
becomes
where C is an interaction constant (for details, see Korpel [4]) and s
x; z; t is
the real sound amplitude given by
where Se is the complex amplitude of the sound ®eld that interacts with the
light beam and is traveling in the x-direction and c:c: denotes the complex
conjugate. The quantities K and
are the propagation constant and the
angular frequency of the sound ®eld. Following Korpel [4,5], a snapshot of
the sound ®eld is used at t 0, so that using Eqs. 4.3.3 and 4.3.4,
^
exp
Sz exp
jk0 nz 1 jk0 nz
1 1
2 jk0 zCSe
x; z exp
jKx Se
x; z exp
jKx
4:3:5
The main propagation loop of the algorithm is modi®ed from Fig. 1 and is
shown in Fig. 5. The boxes marked ``Shift K'' are used to facilitate the
operation shown in Eq. 4.3.6 in the spatial frequency domain.
Figure 6 shows problem geometry of a Gaussian beam incident nom-
inally at Bragg angle on a sound column of width z L. The simulated
evolution of the Gaussian beam is shown in Fig. 7 and is taken from Ref. [4].
The peak phase delay of the light traveling through the acousto-optic cell
is taken equal to , and the Klein±Cook parameter Q K 2 L=k0 13:1. We
would like to point out that the same answers could be derived by using the
transfer function for acousto-optic interaction, as given in Refs. 4 and 6.
Figure 5 Flow diagram for the modi®ed split-step technique to analyze acousto-
optic interaction.
254 Chapter 4
Figure 7 Simulation plot of the intensity of the angular spectrum of the total ®eld
at different positions along interaction length [Ref. 4].
The Split-Step Beam Propagation Method 255
@Ee 2
2j k0 r? Ee 2n2 k20 jEe j2 Ee
4:4:1
@z
n n0 n2 jEe j2 4:4:2
In writing Eq. 4.4.1, we have taken the linear refractive index n0 equal to
unity for the sake of simplicity. For a medium with n2 > 0, one can observe
self-focusing of a Gaussian beam traveling through a medium, while self-
defocusing is observed fora medium with n2 < 0. The nonlinear operator
^
expfSzg exp jk0 n2 jEe j2 z. Typical plots showing self-focusing and
self-defocusing of initial Gaussian pro®les in one transverse dimension are
shown in Figs. 8 and 9, respectively. Note that in Fig. 8, the initial power in
the Gaussian beam is taken to be higher than the so-called critical power
required for self-focusing. For this reason, one observes periodic focusing
during propagation through the medium. The physical reasoning behind
self-focusing is as follows. The Gaussian beam induces a positive lens in
the nonlinear material for n2 > 0 because where the beam intensity is high
The Split-Step Beam Propagation Method 257
1=2
8 x
Ee
x sec h
4:4:3
n2 k0
2k0 1=2
This analysis has been done for a thin sample, meaning one where diffrac-
tion of the beam is neglected during its travel through the PR material, and
by using a linearized theory to determine the induced refractive index pro-
®le. We have recently extended the linearized approach to the case of a thick
sample, and have included the transient effects, and are in the process of
determining the effects of transient DBF when a reading beam is used to
illuminate a previously stored hologram in the PR material [14].
Another school of thought is that beam ``fanning'' results from light
scattering from the random distribution of space charges in the PR material.
However, a larger contribution to random beam fanning (RBF) is the so-
called ampli®ed noise [15] that may arise from the couplings between the
plane wave components scattered from crystal defects.
In this section, we examine steady state DBF in a diffusion dominated
PR material by deriving a closed form expression for the induced refractive
index change from the nonlinearly coupled Kukhtarev equations. We also
assess the role of propagational diffraction in DBF by determining the
similarities and differences between the thin and thick sample models.
It can be shown that the coupled set of simpli®ed Kukhtarev equations
[9] (see Chapter 7 for details) for a diffusion dominated PR material can be
decoupled in the steady state to yield an ordinary differential equation for
the space charge electric ®eld [13]. In denormalized form, we can express this
electric ®eld Es
x; y; z as
2 eNA e rI
r Es E
4:4:4
s r Es kB T s =s I
kB T rI
Es Esx a^ x Esy a^ y
4:4:5
e =s I
if
kB =T=e=W2 eNA =s , where W is the characteristic width of the com-
plex envelope Ee
x; y; z of the optical ®eld. The quantities a^ x and a^ y refer to
unit vectors in the x- and y-directions, respectively. Now this electrostatic
®eld induces a refractive index change next
x; y; z for extraordinary polar-
260 Chapter 4
ized (say along x, see Fig. 11) plane waves of light in the PR material,
assumed BaTiO3 from now on, through the linear electro-optic effect,
given by
next
x; y; z; y Ex
x; y; z f
y
1 3
f
y 2 ne
y cos y
r13 sin y r33 cos2 y 2r42 sin2 y
n2e
y
sin2 y=n2o cos2 y=n2e 1
4:4:6
where no and ne are the linear ordinary and extraordinary refractive indices
and the rij are the linear contracted electro-optic coef®cients [9]. The angle
in Eq. 4.4.6 is de®ned in Fig. 11. Note that f
y is a slowly varying function
of over the spectral content of the optical ®eld. It can be readily shown
that, in general, propagation through the PR material under the slowly
varying envelope approximation can be modeled by means of the PDE [13]
@Ee 2
jk0 next Ee j1=2ne
k0 r? Ee
@z
kx
next
x; F x 1 F x Ee
xf 0
ne
k0
kB T @jEe
xj2 =@x
Esx
x
4:4:7
e =s jEe
xj2
x2 y2
Ee
x; y; 0
I0 1=2 exp
4:4:8
W2
with I0 2P=W 2 , where I0 denotes the on-axis intensity and P is the beam
power, is phase modulated owing to the induced refractive index pro®le. The
resulting output ®eld is Ee
x; y; L Ee
x; y; 0 exp
jk0 next
xL, where L
is the thickness of the PR material. Such a phase modulation results in a
shift of the far ®eld pattern with respect to the axis
z of propagation of the
optical beam, and in the appearance of asymmetric side lobes, the so-called
fanning of the beam. Numerical simulations for BaTiO3 with parameters
n0 2:488, ne 2:434, r42 1640 pm/V, r13 8 pm/V, r33 28 pm/V,
NA 2 1022 m 3 , s 3:28 10 8 F/m, s 2:6 10 5 m2 =J, b 2 s 1 ,
T 298 K [13], and L 1 cm and using an incident wavelength of 514.5 nm
show a monotonic increase in the shift of the far ®eld main lobe from the z
axis with increase in I0 (implying either an increase in power P or a decrease
in width W). In Figs. 12a and b, kx is the spatial frequency variable corre-
sponding to x and is related to the far ®eld coordinate xf by kx k0 xf =d, d
being the distance of propagation from the exit of the crystal to the far ®eld.
However, the amount of DBF (de®ned by the relative amount of power in
the side lobes) varies nonmonotonically with intensity, initially increasing as
the intensity is increased from low levels to attain a maximum, and then
decreasing with further increase in intensity.
Note that our results are different from those of Feinberg [10], in that
the latter, based on a linearized two-beam coupling theory that neglects
coupling of the angular plane wave components of the Gaussian with any
!
background illumination, yields E s / rI=I0 , where I0 is the quiescent inten-
sity (to be compared with our Eq. 4.4.7). For a Gaussian intensity pro®le,
the locations of the extrema of E in Feinberg's formulation are ®xed w.r.t.
to the incident pro®le and hence can be shown to predict a monotonic
increase in DBF with a decrease in W. In our nonlinear formulation, how-
ever, for decreasing W, the extrema of E move out with respect to the
incident pro®le, so that the pro®le essentially sees a linear induced refractive
index for suf®ciently small W, resulting in reduced DBF.
262
Figure 12 Normalized far ®eld intensity pro®les for the thin sample model. (a) P 1:5 mW; (b) W 40 microns. Used with
permission of North-Holland, 1993 [13].
Chapter 4
The Split-Step Beam Propagation Method 263
Before comparing the thin sample results with the ®ndings for the
thick sample case, we will, at this point, provide a simple alternate explana-
tion for the observed behavior of DBF when monitored as a function of the
intensity. Our explanation is based on the examination of the spectrum of
the phase modulation exp
ik0 next
xL. The far ®eld pattern is the con-
volution of the above spectrum with that of the input pro®le. Since next
x
is an odd function of x (see Eq. 4.4.6), it can be expanded in a power series
of the form ax3 bx, where a and b are given by
Note that the coef®cients of this expansion hold for all values of the ratio
=sI0 . The spectrum H
kx of exp
ik0 next
xL is then
2 k bk0 L
H
kx 1=3
Ai x
4:4:10
3a
3a1=3
Once again, kx above has the same implication as in the discussion on Fig.
12. We comment that if d is replaced by f , where f is the focal length of a
lens at the exit plane of the crystal, kx , and hence xf , would be representative
of the spatial coordinate on the back focal plane of the lens. Ai is the Airy
function [17]. The ith zero, i , of H
kx is related to the ith zero,
i
< 0, of
Ai by i bk0 L
3a1=3
i . It then follows that the spatial extent of the
Airy pattern for kx < bk0 L, up to say the ith zero, and normalized by the
spectral width 2=W of the incident Gaussian pro®le, varies nonmonotoni-
cally with I0 . Figs. 13a and b show, for instance, the variations of i bk0
L
3a1=3
i =
2=W for i 1 with W and P, respectively. The shift in the
Airy pattern, bk0 L, however, increases with an increase in I0 . For large I0 , it
can be shown that the shift is proportional to 1=W 2 , in agreement with the
trend in Fig. 12a. The resulting far ®eld pattern, which is the convolution of
the Gaussian spectrum and the Airy pattern, generally exhibits decreased
DBF when the Airy pattern has a (denormalized) width much smaller than
that of the Gaussian spectrum (which may occur, for instance, for both
small and large W). This is in agreement with our numerical simulations
in Fig. 12. Appreciable DBF occurs in the region where the normalized
bandwidth (see Figs. 13a and b) is greater than unity. As an example, for
P 1:5 mW, maximum beam fanning, de®ned by the maximum of the ratio
of the peak value of the side lobe and that of the main lobe, occurs when
W 30 microns.
Figure 13 Normalized bandwidth of the induced PR phase modulation (a) for P
1:5 mW, plotted as a function of W, and (b) for W 40 microns, plotted as a
function of P. Used with permission of North-Holland, 1993 [13].
The Split-Step Beam Propagation Method 265
In the remainder of this section, we will present the results for the far
®eld beam pro®les using a thick sample model for the PR material and point
out the similarities and differences with the thin sample approach.
Numerical simulations for the thick sample model were performed on the
basis of Eq. 4.4.7 by employing a split-step beam propagation technique [7].
In this simulation, we track both the phase and the amplitude modulation of
the beam within the crystal due to the combined effects of propagational
diffraction (along x; y) and induced refractive index (along x) arising from
the PR effect. Figs. 14a and b show the normalized far ®eld intensity pat-
terns with W and P as parameters. By W we now mean the beam waist that
would be expected at z L=2 (i.e., the location of the center of the sample)
in the absence of any electro-optic effect (rij 0) (see inset in Fig. 14a). The
results are qualitatively similar: DBF is seen to reduce at very low (high) and
very high (low) values of P
W. Quantitatively, for a ®xed power P (viz.,
1.5 mW), we can predict the absence of DBF for suf®ciently large values for
W (viz., 70 microns) which are independent of the model (thin or thick
sample) used for simulation. Physically, this makes sense, since the thin
and thick sample models must agree if the diffraction effects in the crystal
are suf®ciently small. On the other hand, the reason for the absence of DBF
for a suf®ciently small value of W in the thick sample approach is that
effectively, the beam width, if monitored over most of the sample, is large
(due to a large diffraction angle), implying a reduced PR effect. This in turn
implies that propagation through the crystal is predominantly diffraction
limited. For small W, the thick sample model therefore is more accurate
than the corresponding thin sample model for the same value of W, since
the latter model overestimates the amount of cumulative PR effect. For the
thick sample model, for the same value of P as above, we see negligible DBF
for W less than 25 microns. On the other hand, the thin sample model
predicts a value of W less than 5 microns for negligible beam fanning.
The reason for the disappearance of DBF in the thin sample approach
has been presented above using the Airy function argument and the move-
ment of the extrema of E w.r.t. the incident optical ®eld. Maximum DBF for
P 1:5 mW occurs for W 40 microns, in close agreement with the thin
sample computations and the Airy function approach. However, the shift in
the position of the main lobe in the thick sample model is much smaller as
compared to the thin sample case due to the effective decrease in the PR
effect for a small waist size, as explained above. Referring to Fig. 14a, we
note that for W 40 microns, P 1:5 mW and f 10 cm, and the spatial
shift in the back focal plane of a lens of focal length f located at the exit
plane of the PR material is about 0.2 mm. We would like to comment that
for the above parameters, DBF was also numerically observed at the exit
face of the thick PR sample.
Figure 14 Normalized far ®eld pro®les for the thick sample model. (a) P 1:5
mW; (b) W 40 microns. Used with permission of North-Holland, 1993 [13].
The Split-Step Beam Propagation Method 267
kB T rI
next /
4:4:11
e =s I
The interaction between two focused Gaussian beams incident on the mate-
rial can be effectively studied numerically using the split-step method. In this
case, two-beam coupling results in energy exchange between the two beams
after interaction through the photorefractive material. The problem geome-
try is shown in Fig. 15. The two Gaussian beams are focused in the center of
the photorefractive material, and the angle between them is 2. The
Gaussian beams are expressed in terms of their q-parameters at the entry
face of the material. The split-step algorithm is used to determine the inter-
action and energy exchange between the two beams. The induced refractive
index n written above is used to construct the operator representing the
induced inhomogeneity in the material. The results on two-wave mixing are
shown in Fig. 16. The dot-dashed lines show the far ®eld intensity pro®les of
the two Gaussian beams in the absence of the photorefractive material. The
dashed lines show the beams after energy transfer due to the induced refrac-
tive index. The initial pump-to-signal power is 3. The peak intensity of the
Figure 16 (a) Dotted and dashed lines are respectively the far ®eld signal and
pump intensities with the absence of any PR material, and chain dots and chain
dashes represent the resulting far ®eld intensities after the beams have propagated
through a 5 mm BaTiO3 sample. Incident beams are focused to the center (z L=2)
of the Pr crystal, and the waist of each beam at wavelength 0.632 microns is 100
microns. Signal-to-pump ratio is 3, and semi-angle of crossing is 0.5 degrees. Note
that w1 w2 1:0 10 4 . (b) Interference pattern at center (z L=2) of the PR
crystal for the beams described in (a). (c) Space charge ®eld (V/m) at the center
(z L=2) of the crystal for the beams of (a). Used with permission of North-
Holland, 1994 [18].
pump and signal beams are 63 and 21 W/cm2 , respectively, before the inter-
action. The beams are coupled by a 5 mm BaTiO3 photorefractive material.
The output beams do not show any effect of beam fanning at this power, but
with larger beam powers, distortion of the beams due to beam fanning is
observed. The results have been used to ®nd the two-beam coupling strength
The Split-Step Beam Propagation Method 269
and their dependence on the intensities of the two participating beams. The
results, shown in Ref. 18, depict that the coupling strength depends on the
power ratio between the two beams, a fact that is ignored in perturbation
calculations of two-wave mixing in photorefractive materials. Later, in
Chapter 7, we will analyze this effect in more detail with participating
plane waves and using rigorous coupled wave theory.
The previous examples illustrated the use of the split-step method in calcu-
lating the beam pro®les during diffraction in space or during propagation
through a guided (externally or internally induced) medium. If a Gaussian
beam is assumed, however, the split-step method can be reformulated in
terms of a differential equation that shows the evolution of the Gaussian
beam's parameters, e.g., width, during propagation. The ensuing equation
can be exactly solved in some cases, e.g., for a Kerr-type material, and is
therefore physically more transparent than the results obtained using the
split-step method. The differential equation for the parameter(s) may not be
simpler to solve than the split-step method, but having an analytical solution
(Gaussian beam) adds a tremendous insight into the actual propagation of
the wave through the material, whereas the split-step method only presents
simulation results. When a Gaussian beam travels a distance z in an n2
medium, the q-parameter [19] change using the split-step method can be
written as
q2
q z
4:5:1
find
z
where find is the nonlinearly induced focal length of the slice z [19]. The
above equation shows that the q of a Gaussian beam changes due to pro-
pagational diffraction and due to the induced nonlinearity of the material.
In LiNbO3 the photovoltaic effect is responsible for breaking the circular
symmetry of an incident focused extraordinarily polarized Gaussian beam.
Therefore the propagation model is based on the propagation of an elliptical
Gaussian beam.
As discussed in the last section, beam fanning in photorefractive crys-
tals has received considerable attention for its possible implications in holo-
graphic information recording [10,20±23]. Light-induced scattering resulting
270 Chapter 4
q2x q2y
qx z qy z
4:5:3
findx findy
Since
!
2 2 x2 y2
n ne n2 jEe j ne 2n2 a
z 2 2
4:5:4
wx wy
ne w2x ne w2y
findx findy
4:5:5
4n2x a2
zz 4n2y a2
zz
Substituting Eq. 4.5.5 into 4.5.3 and taking the limit as z ! 0 we obtain
the system of equations
Using the well known relationship 1=q 1=R j=ne w2 , where R is a
radius of Gaussian beam curvature, 1=R
1=w
dw=dz, and is the wave-
length in vacuum, we obtain
d 2 wx 2 4n2x a2
dz2 n2e 2 w3x ne w x
d 2 wy 2 4n2y a2
4:5:8
dz2 n2e 2 w3y ne wy
d 2 wx 2 8n2x P
2
2 2 3
dz ne wx ne w2x wy
d 2 wy 2 8n2y P
4:5:9
dz2 n2e 2 w3y ne w2y wx
Assuming n2x n2y (true for photorefractive lithium niobate), the variation
of the widths wx and wy of an elliptic Gaussian beam propagating through a
The Split-Step Beam Propagation Method 273
d 2 wx 2 8n2 P
dz2 n2e 2 w3x ne w2x wy
d 2 wy 2
2
2 2 3
4:5:10
dz ne wy
The case when n2x n2y has been studied [13] by employing the q-transfor-
mation approach to ®nd the widths of an elliptic Gaussian beam in a non-
linear medium in the presence of diffraction. Equation 4.5.10 assumes that
the nonlinearity is highly inhomogeneous and only affects the width along
the x-axis (which coincides with the c-axis of our crystals) due to the large
electron mobility along that axis [25]. The effective n2 can be written as [11]
1 3 k
R NA
nw n r
4:5:11
2 e 33 mebND
Figure 17 z-scan setup for a thick sample. The thick lines represent the path of the
rays, described as the locus of the 1=e points of the Gaussian beam. The thin lines
show the ray path in the absence of the medium. Circular symmetry of the Gaussian
beam is assumed throughout the sample. Used with permission of OSA, 1998 [26].
274 Chapter 4
above equation, we have made the assumption b sI, where s is the ioniza-
tion cross section per quantum of light and I is the optical intensity.
For a sample length L assumed to be much larger than the Rayleigh ranges
zRy and zRx along z for the elliptic beam, the evolution of wx can be approxi-
mated as
!
z2 ne w20 4ne n2 P
w2x
z w20 1 2 where zRx 1
zR x 0 20
4:5:13
It is clear that in the x-direction, the beam spread is more than that in the
linear diffraction-limited case when n2 < 0 and less when n2 > 0. As seen
from Eq. 4.5.12, the nonlinearity does not affect the beam width along the y
direction, which leads to elliptic beam cross section pro®le at the exit of the
crystal and, in general, in the far ®eld.
For more general geometry, where the incident beam does not have a
planar wave front, we have solved Eqs. 4.5.10 numerically. Figure 18 shows
typical z-scan graphs plotted for four different values of power for the initially
circularly symmetric Gaussian beam. In the calculations we have used the
following parameters: crystal thickness L 10 mm, lens focal length
f0 10 cm, 0 514 nm, initial beam width w0 1:0 mm, ne 2:20,
n2 1:4 10 12 m2 =V2 , P 1 mW, crystal exit plane to observation
plane distance D 1 m. A simple explanation of the behavior in the limiting
case (s much smaller or larger than f0 ) seen in Fig. 18 can be given by referring
to Fig. 17. When the distance s, the lens-to-sample separation, is much smaller
than the lens focal length f0 , the incident beam is weakly focused and therefore
the beam widths lie close to their linear values leading to semilinear diffrac-
The Split-Step Beam Propagation Method 275
Figure 18 Typical z-scan graph drawn by solving Eq. 4.5.10 and propagating the
Gaussian beam a distance D behind the sample. Used with permission of OSA, 1998
[26].
Note that the pattern is approximately symmetric (along x and y). This
symmetry arises because the refractive index changes that are due to photo-
voltaic (and thermal) effects are symmetric and because there is little con-
tribution from diffusion. Experimental results based on the measurement of
ellipticity, as shown in Fig. 21, show the same trend as the theoretical pre-
dictions superposed on the same ®gure. The ellipticity was calculated from
experimental observations by ®rst determining the extent wx , wy of the
bright or gray region along x and y, respectively, from pictures such as
Fig. 20 and taking the ratio of the two. Note that Fig. 21 is in fact a
blow-up of Fig. 19 over the interval 9.5±10.5 cm. The theoretical graph in
Fig. 21 was drawn after examining the experimental results shown in the
same ®gure and choosing that value n2 for the analytical graph that mini-
mizes the sum of the differences between the experimental points and the
corresponding theoretical data.
The Split-Step Beam Propagation Method 277
Figure 20 Typical beam pattern at D 0:5 m for P 0:05 mW, f0 20 cm, and
s 19:5 cm for Fe doped LiNbO3 crystal. Used with permission of OSA, 1998 [26].
Figure 21 Experimental (points) and theoretical (line) variation of the beam ellip-
ticity on the observation plane as a function of scan distance. here, P 0:2 mW, D
0:5 m, f0 10 cm. Upon comparison, n2 1:4 10 12 m2 =V2 . Used with permis-
sion of OSA, 1998 [26].
278 Chapter 4
As a ®nal note, we would like to point out that each time the crystal
was displaced along the longitudinal direction for a fresh z-scan ellipticity
measurement, we also made a transverse movement of the crystal in order to
make sure that we were starting out from a virgin location in the crystal for
each data point. In other words, we always started out from an initially
unexposed region of the crystal and exposed it to the incident illumination
until the steady state was achieved.
Figure 22 (a) Theoretically and (b) experimentally obtained P-scan graph for D
0:5 m for f0 20 cm and s 19:5 cm for the Fe doped LiNbO3 crystal. Fig. (a) was
drawn with n2 4 10 11 m2 =V2 to provide the best match with the experimental
results in (b) for lower powers. Used with permission of OSA, 1998 [26].
from these crystals when holograms were stored in them with a 90 s expo-
sure time at 514 nm using an incident power of approximately 200 mW.
Given a crystal with unknown nonlinearity, the signal can be deter-
mined as follows. We place the crystal around the back focal plane of the
lens. Upon moving it toward the lens, if the ellipticity reduces to less than
unity, nw < 0. If, upon moving the crystal away from the lens, the ellipticity
280 Chapter 4
is less than unity, n2 > 0. The magnitude of n2 can now be evaluated using
the P-scan technique. Details of this can be found in Ref. 26.
In summary, a model for beam propagation through a nonlinear mate-
rial that takes into account inhomogeneous induced refractive index changes
due to the nonlinearity was developed. The theory based on this model can
be used to analyze the propagation of Gaussian beams through PR LiNbO3 .
A focused Gaussian beam of circular cross section incident on the sample
emerges as an elliptic Gaussian after interaction in this material. The P-scan
method can be used to evaluate the effective nonlinearities (resulting from
the photovoltaic effect) of lithium niobate samples doped with different
materials such as Fe, Co, Cr, Rh, Mn, etc. The value of the nonlinear
coef®cient can then be used to determine the acceptor-to-donor ratio of
dopants in the photorefractive samples.
This method can be used to characterize any optically nonlinear mate-
rial that has an induced intensity-dependent refractive index. We would like
to point out that this method is very general and in principle can be applied
to any nonlinear electromagnetic material and at any frequency.
PROBLEMS
fronts at z 0. Find expressions for the width W
z and the radius
of curvature of the wavefronts R
z after an arbitrary distance z of
propagation.
2. A Gaussian beam of width W and having wavefront with a radius
of curvature R is normally incident on the interface between air
and glass of refractive index n. Find the width and radius of
curvature:
(a) immediately after transmission through the interface, and
(b) immediately upon re¯ection at the interface.
[Hint: Calculate the off-normal ray re¯ection and transmission
angles.]
3. A one-dimensional Gaussian beam symmetric along the x-direc-
tion of waist W0 is incident on a thin slice of dielectric material of
thickness z with a graded refractive index n
x n0 n
2 x2 for
small x. Find the effective focal length of the induced lens as the
Gaussian beam propagates through the material. Hence ®nd the
approximate location of the focal point beyond the thin sample
where the Gaussian beam would have minimum waist.
4. A one-dimensional Gaussian beam symmetric along the x-direc-
tion of waist W0 is incident on a thin slice of dielectric material of
thickness z with a graded refractive index n
x n0 n1 cos Kx;
W0 2=K. Calculate the far-®eld diffraction pattern.
5. Use the split step beam propagation technique to analyze propa-
gation along z of a one-dimensional Gaussian beam of W0 100
( is the free-space wavelength) incident onto a lamellar grating
bounded by a material with refractive index n0 . The lamellar grat-
ing has a thickness of 100 with a refractive index pro®le
n
x n0 n1 sgn(cosKx), K 2=; 5, where sgn(y 1
if y > 0 and 1 if y < 0. Take n0 1:5, n1 0:15. Calculate the
pro®le at the exit plane of the grating and in the far ®eld. Repeat
the problem for the case where the thickness of the lamellar
grating is 1000 and characterize the differences between the
two cases.
6. Use the split step method to analyze the propagation of a
Gaussian beam of waist W0 100 through a material of thick-
ness 100 having a refractive index pro®le n
x n0
x=W0 ,
jxj < 5W0 . Let n0 1:5, 0:015. Assume that the material is
bounded by a material of refractive index n0 . Determine the far-
®eld intensity pro®le.
7. A Gaussian beam of waist W0 100 symmetric about x 0 is
incident from air onto a nonlinear material slab of thickness 100
and of refractive index n
x 1 n2 I
x where I
x is the intensity
282 Chapter 4
REFERENCES
5.1 INTRODUCTION
5.1.1 Background
In Chapters 2, 3, and 4 the RCWA method and spectral domain techniques
were used extensively to treat the solution of Maxwell's equations for planar
dielectric systems, which were isotropic, anisotropic, or bianisotropic.
Chapter 2 concentrated on the case when the dielectric layers were trans-
versely homogeneous and the source and EM ®elds of the system could be
effectively represented as a Fourier or K-space integral (i.e., waveguide slot,
dipole antenna, etc.). Chapter 3 concentrated on the case when the dielectric
layers were periodic diffraction gratings and the source of the system were
Rayleigh plane waves. In Chapter 4 the split-step beam propagation method
was used to solve the dif®cult problem that occurs when the dielectric layer
is a diffraction grating and the source (i.e., Gaussian beam) is represented by
a Fourier or K-space integral.
In this chapter we will deal with the problem of using the RCWA or
exponential matrix method to solve Maxwell's equations in circular and
spherical systems that may be inhomogenous in the radial and angular
coordinates and which may be isotropic or anisotropic. In cylindrical or
spherical systems the RCWA method is applied by expanding all EM ®eld
and source quantities in Floquet harmonics that are periodic in the angles
or . The method is similar to that used when applying the RCWA method
to planar diffraction gratings systems where all EM ®eld and source quan-
tities are expanded in Floquet harmonics, which are periodic in the grating
periods x or z .
285
286 Chapter 5
examples of such a system. In this section we treat the cases (1) where the '-
inhomogeneity and its excitation are only periodic over the '-period of '
2 (for example, plane wave scattering off a '-inhomogeneous dielectric
cylinder, see Fig 1a) and (2) where the '-inhomogeneity and its excitation
possesses a higher symmetry than the ®rst case (for example, Fig. 1b with a
centered line source) in which the '-period can be taken to be ' 2=p,
p 2, where p is an integer. The '-inhomogeneous cylindrical dielectric
system that is being studied in this paper can also be viewed as a circular
diffraction grating that has been placed in a circular region.
The solution of the problem just stated may have applications, for
example, to scattering from circular frequency-selective surfaces and scatter-
ing from circular surfaces covered with periodically spaced radar absorbing
material (RAM). It can also be used as a numerical cross-check of other
numerical algorithms that concern scattering from dielectric systems.
The solution method to be proposed in this chapter to solve the cir-
cular diffraction grating problem will be based on a recently developed
algorithm called rigorous coupled wave theory (RCWA), which has been
288 Chapter 5
Moharam and Gaylord [12] to solve planar surface relief diffraction grat-
ings.
We next use RCWA to determine the radiation and scattering that
arises from inhomogeneous anisotropic cylindrical dielectric and permeable
systems that have arbitrary radial and azimuthal spatial variation. The
RCWA state matrix equations and the associated boundary matrix equa-
tions (derived from a multilayer ladder analysis) are presented and solved
for the ®rst time for the cases when a plane wave (TM polarization, electric
®eld parallel to the cylinder axis) or electric line source is incident on a
cylinder that possesses an inhomogeneous permittivity pro®le
; ' and
possesses inhomogeneous anisotropic permeability pro®les
; ',
'
; ', '
; ', and ''
; '. In this chapter, radiation and scattering
from three inhomogeneous examples were studied using the cylindrical
RCWA method.
Finally, we present a rigorous coupled wave analysis of the electro-
magnetic radiation that occurs when a centered electric dipole excites power
and energy in a general three-dimensional inhomogeneous spherical system.
The formulation consists of a multilayer state variable (SV) analysis of
Maxwell's equations in spherical coordinates (the SV analysis used trans-
verse-to-r spherical EM ®eld components) as well as a presentation of the
EM ®elds that exist in the interior and exterior regions that bound the
inhomogeneous spherical system. A detail description of the matrix proces-
sing that is involved with ®nding the ®nal EM ®elds of the overall system is
given and three numerical examples of the RCWA method are studied
X
1
2
; ' i
eji' 0 ' 2 ab
5:2:1
i 1
X
1
I
2 p
Ez
1 c
1
i Ji
X1 c0 H0
1 j~ ~s j
5:2:2a
i 1
p
j 1 @Ez
1
H'
1
5:2:2b
0 @X1
p p p
where cI0 !0 I=4, ~s s a^ x , 0 0 =0 , X1 1 k0 ~ 1 ,
0 a, and I is the electric current line source. The EM ®elds in the
region s < a can be expressed [5,6] as
1 h
X i
Ez
1 ci
1 Ji
X1 cI0 Ji
X1s Hi
2
X1 eji'
i 1
X
1
szi
1
X1 eji'
5:2:3a
i 1
j X
1
ph
1 0 0
i
H'
1 1 ci Ji
X1 cI0 Ji
X1s Hi
2
X1 eji'
0 i 1
j X
1
u
1
'i
X1 e
ji'
5:2:3b
0 i 1
p p
where J 0
X dJ
X=dX, etc., X1s 1 k0 ~ s 1 s , and 1 is the relative
permittivity of Region 1. In the exterior region, the EM ®elds arepa sum of
an incident plane wave (electric ®eld given by E I E0I e j 3 x a^ z ,
~ and a general EM scattered wave. The exterior EM ®elds in
x k0 x)
Region 3 are given by
Inhomogeneous Cylindrical and Spherical Systems 291
1 h
X i X
1
Ez
3 ci
3 Hi
2
X3 EiI Ji
X3 eji' szi
3
X3 eji'
5:2:4a
i 1 i 1
j X1
ph
3
2 0 i
H'
3 3 ci Hi
X3 EiI Ji0
X3 eji'
0 i 1
j X1
u
3
'i
X3 e
ji'
5:2:4b
0 i 1
p p
where EiI E0I j i , X3 3 k0 ~ 3 , and 3 is the relative permittivity in
Region 3.
In Region 2, the middle cylindrical dielectric region, we divide P the
dielectric region into L thin shell layers of thickness d` , b a L`1 d` ,
and solve Maxwell's equations in cylindrical coordinates by a state variable
approach in each thin layer. The layers are assumed to be thin enough so
that the -dependence of 2
; ' and the scale factors can be treated as a
constant in each layer. Letting k0 ,~ we ®nd that Maxwell's equations in
a cylindrical shell of radius are given by
@Ez
j0 H'
5:2:5a
@
@0 H' 1 @2 Ez
j2 Ez
5:2:5b
@ j @'2
To solve Eq. 5.2.5, we expand Ez , 2
; ', and 0 H' in the Floquet har-
monics:
X
1
Ez szi
eji'
5:2:6a
i 1
X1
0 H' u'i
eji'
5:2:6b
i 1
" #
X
1 X
1
2
; 'Ez i i 0 szi 0 eji'
5:2:6c
i 1 i 1 0
These expansions are substituted into Eqs. 5.2.5a and 5.2.5b and we let
sz
szi
and uu
u'i
be column matrices and i i 0 ,
K Ki;j 0 , K 2= 0 (0 is the circular grating period and i;i 0 is the
Kronecker delta) be square matrices. We ®nd then after a small amount
of manipulation that
292 Chapter 5
" #
@V sz
A V; V
@ uu
" #
0
I=
Aj
5:2:7
K K= 0
@V`
s`
A` V`
s` ` 1; . . . ; L
5:2:8
@s`
where qn` and V`n are the nth eigenvalue and eigenvector of the constant
matrix A`
A` V`n qn` V`n . Using the V`n
s` eigenvector solution, the gen-
eral EM ®elds in the `th thin shell region are given by
MT X
X NT
Ez` cn` szin` exp
qn` s`
5:2:10a
i MT n1
MT X
X NT
0 H'` cn` u'in` exp
qn` s`
5:2:10b
i MT n1
h i
V t`n stzn` ut'n`
5:2:10c
Inhomogeneous Cylindrical and Spherical Systems 293
V t V1 ; V2 ; V3 ; V4 ; V5 ; V6
stz sz; 1 ; sz;0 ; sz;1 V1 ; V2 ; V3
ut' u'; 1 ; u';0 ; u';1 V4 ; V5 ; V6
5:2:11
X
NT X
NT
cn` szin` exp
qn` d` cn;`1 szin;`1
5:2:12a
n1 n1
X
NT X
NT
i MT ; . . . ; Mt
cn` u'in` exp
qn` d` cn;`1 u'in;`1
5:2:12b
n1 n1
` 1; . . . ; L 1
D` C` E` C`1 5:2:13a
or
C`1 E` 1 D` C` F` C`
5:2:13b
294 Chapter 5
At the a boundary, if we match the Ez
1 solution with the EzL solution
(the ` L thin layer is assumed adjacent to Region 1, and the ` 1 thin
layer is assumed adjacent to Region 3) and solve for the Region 1 ci
1
coef®cient, we ®nd
X
NT
c
1
i cI0 Ji
X1s Hi
2
X1a cnL szinL exp
qnL dL =Ji
X1a
5:2:15
n1
p
where X1a 1 a:
If the 0 aH'
1 solution is matched with the 0 aH'L solution the ci
1
coef®cient is substituted, and the well-known Wronskian equation for Bessel
functions is used, it is found that
X
NT
Ji
X1a p Ji0
X1a
cI0 Ji
X1s cnL exp
qnL dL ja 1 s u'inL
n1
2 Ji
X1a zinL
i MT ; . . . ; 0; . . . ; MT 5:2:16
At the b boundary, after matching the tangential electric ®eld Ez
3 from
Region (3) to the electric ®eld Ez1 from the ` 1 layer and solving for the
Region 3 c
3
i coef®cient, we have
" #
X
NT
c
3
i EiI Ji
X3b cn1 szin1 =Hi
2
X3b
n1
5:2:17
i MT ; . . . ; 0; . . . ; MT
p
where X3b 3 b. If the 0 bH'
3 is matched with the 0 bH'1 ®eld solution
of the ®rst layer, the ci
3 coef®cient is substituted, and again a Wronskian
Bessel function relation is used, it is found that
" #( )
X
NT
2 0
Hi
2
X3b p Hi
X3b
EiI cn1 jb 3
2 szin1 u'in1
n1
2 Hi
X3b
i MT ; . . . ; 0; . . . ; MT
5:2:18
Inhomogeneous Cylindrical and Spherical Systems 295
jc
3
i j
2
PNi
1
5:2:21
jcI0 j2
For the plane wave scattering case
cI0 0, E0I 6 0), it is useful to
calculate the normalized scattered power ({(power/meter)/wavelength)}/
{Poynting power intensity (watts/m2 )}) in each order. The normalized scat-
tered power at b is given by
296 Chapter 5
( )
PScat
b= p jc
3 j2
2
2 0
PScat
Ni
b i b 3 Re j i I 2 Hi
X3b Hi
X3b
5:2:22
SINC jE0 j
PScat
i
1= 2 jc
3
i j
2
PScat
Ni
1
5:2:23
SINC jE0I j2
In Eqs. 5.2.22 and 5.2.23, SINC is the power per unit area (watts/meter2 ) of
the incident plane wave, and PScat
i is the scattered power per unit length
(watts/m) of the ith order.
1
0
20 200 200
2
1 sin i
i
20 200 jij 1
5:2:24
2 i
where 0 2 and 1 .
Figure 2 shows a comparison of the '-periodic semicylindrical shell
~ normalized
plane wave scattered power in each order (calculated at ~ b,
according to Eq. 5.2.22, using MT 20 and L 30 layers) with that of a
uniform dielectric shell as a function of order i when b k0 b~ 10. In
Region 2 the uniform shell dielectric value was taken to be 2 2:5 (this
value is the average or bulk dielectric value used for the semicylindrical
shell). The scattering from the uniform dielectric shell (dashed curves in
Fig. 2) was calculated both by the current state variable algorithm (using
MT 20 and L 30 layers) and by solving Maxwell's equations in
Regions 1, 2, and 3 in terms of Bessel and Hankel functions and then
Inhomogeneous Cylindrical and Spherical Systems 297
Figure 3 The total plane wave scattered power from the semicylindrical half shell
~ solid line curve, normalized, see Eq. 5.2.23, labeled
(see Fig. 1a) (calculated at ~ b,
periodic) when a k0 a~ 5 and where b~ ranges from b~ 0:8, to b~ 1:6. The
value of MT 10, was used to make this plot. For comparison Fig. 1 also shows
the total plane wave scattered power (dashed curve, labeled Uniform) that results
when a plane wave is incident on the same dielectric system that has already been
described except that Region 2 is taken to be a uniform dielectric shell whose dielec-
tric value is 2 2:5. Used with permission of VSP BV, [15, Fig. 3].
Figure 4 A comparison of the uniform and phi-periodic half shell dielectric sys-
tems (same case as described in Figs. 2 and 3) for the i 0 and i 7 orders when
a k0 a~ 5 and b k0 b~ is varied from values of b 5 to b 10. Used with permis-
sion of VSP BV, 1997 [15, Fig. 4].
shell centered at ' 0
j'j 45 are 20 3:25 for 22:5 22:5 and
200 1:75 for 22:5 < j'j 45 . The Region 2 quarter shell regions centered
at ' 90 , 180 , and 270 repeat the ' 0 centered pattern. The Regions 1
and 3 dielectric values are the same as those used in the ®rst example. In this
case because of the centered location of the line source, the grating period of
the system can be taken to be 0 =2. Using this grating period and the
Region 2 dielectric values already given, the Fourier coef®cients are given by
Eq. 5.2.24 with 1 45 . Also shown in Fig. 5, for comparison, is the
radiation that results when a line source radiates through a uniform dielec-
tric shell (Regions 1 and 3 have the same values as previous examples) and
Region 2 has a bulk dielectric value of 2 2:5. In Fig. 5, a k0 a~ 5 and
the radiated power (normalized in Eq. 5.2.20) is plotted versus the outer
radius b k0 b~ with b k0 b~ varying from 5 to 10 (MT 4 and L 50
layers). The solid lines of Fig. 5 show a comparison of the total power
(normalized) radiated by the uniform and quarter shell periodic dielectric
systems. For the uniform shell, the m 0 order also represents the total
power radiated by the system, since the line is centered on the cylinder axis.
Figure 5 also shows the i 0 and i 4 order powers. The i 4 order is
based on a 0 2 full circle grating period. It is the i 1 order if based on
a 0 =2 quarter circle grating period. The uniform shell radiated power
was calculated both by the current algorithm and by solving Maxwell's
equations in Regions 1, 2, and 3 in terms of Bessel and Hankel functions
and matching EM boundary conditions at the interfaces. The methods gave
300 Chapter 5
Figure 5 The radiated power that results when a line source located at the origin
radiates into the quarter phi-periodic shell shown in Fig. 1b. Here a k0 a~ 5 and
the radiated power is plotted versus the outer radius b k0 b~ with b k0 b~ varying
from 5 to 10. Used with permission of VSP BV, 1997 [15, Fig. 5].
nearly identical results. The Bessel function matching method was used to
make the plot of Fig. 4. As can be seen, the quarter shell dielectric system
causes signi®cantly different radiation than did the uniform shell system,
although both systems had the same bulk dielectric value in Region 2.
The total radiated power shown in Fig. 5 was determined by calculat-
ing the ith order power at ~ b~ (Eq. 5.2.20) and summing these order
powers to obtain the total scattered power. The power was calculated at ~
1 and found to be almost exactly equal to that found at ~ b. ~ The ith
order power was also calculated at ~ a~ (Eq. 5.2.19). It was found that the
order power of the higher orders at ~ a~ was almost exactly zero for jij 1
and that the i 0 power at ~ a~ almost exactly equaled the total radiated
power calculated at ~ b.~ We thus see that conservation of power was
obeyed to a high degree of accuracy. It is interesting that almost no
power was radiated and diffracted into higher orders in the interior region
of the cylindrical system.
Figure 6 The geometry of a uniform cylindrical shell system when a plane wave is
incident on the cylindrical system and when an electric line source excites EM ®elds in
the system is shown. The polarization of the electric ®eld of the plane wave is parallel
to the cylinder axis. Used with permission of EMW Publishing 1998, [23, Fig. 1].
Inhomogeneous Cylindrical and Spherical Systems 303
where
represent '-exponential Fourier coef®cients. The anisotropic
permeability tensor is assumed to be given in rectangular and cylindrical
coordinates by [21,22]
where
xx cos2
'
xy yx sin
' cos
' yy sin2
'
' xy cos2
'
xx yy sin
' cos
' yx sin2
'
' yx cos2
'
xx yy sin
' cos
' xy sin2
'
'' yy cos2
'
xy yx sin
' cos
' xx sin2
'
5:3:3
X
1
rs
; ' rs
eji' 0 ' 2
r; s
; '
5:3:4
i 1
@Sz
j U j' U'
5:3:5
@'
@Sz ''
j' U j U
5:3:6
@ '
@U' @U
jSz
5:3:7
@ @'
To solve Eqs. 5.3.5±5.3.7, we expand Sz
; ', U
; ', U'
; ',
; ', and rs
; ',
r; s
; ', in the Floquet harmonics:
X
1 X
1
Sz
; ' szi
eji' U
; ' ui
eji'
i 1 i 1
" #
X
1 X
1 X
1
U'
; ' u'i
eji'
; 'Ez i i 0 Szi 0 eji'
i 1 i 1 i 0 1
" #
X
1 X
1
rs
; ' F
; ' rs ;i i 0 fi 0 eji'
r; s
; '
5:3:8
i 1 i 0 1
where F
; ' represents either U
; ' or U'
; ' in Eq. 5.5.8. If these
expansions are substituted in Eqs. 5.3.5±5.3.7, and after letting
sz
szi
, u
ui
, and u'
u'i
be column matrices
and
i i 0
, lrs
rsi i 0
,
r; s
; ', K Ki;i 0 , K 2=
' (' is the circular grating period and i;i 0 is the Kronecker delta) be
square matrices, we ®nd after manipulation that
@V s A11 A12
AV V z A
5:3:9
@ uu A21 A22
where
j j
A11 luq lqq 1 K A12 luq lqq 1 lq' l''
1 j
A21 j K lqq 1 K A22 K lqq 1 lq'
(A` NT NT ]. The solution of this equation is given by Vn
Vn exp
qn ,
where qn and Vn are the nth eigenvalue and eigenvector of the constant
matrix An . The quantities An , Vn , and qn satisfy An Vn qn Vn . The general
EM ®elds in the `th thin shell region are given by
MT X
X NT MT X
X NT
Ez cn szin exp
qn 0 H' cn u'in exp qn
i MT n1 i MT n1
5:3:10
where Vtn stn ; utn and where t represents the matrix response.
Although a large matrix equation exists from which the overall solu-
tion of the problem can be obtained, a more ef®cient solution method is to
use a ladder approach [19] (that is, successively relate unknown coef®cients
from one layer to the next) to express the cnL coef®cients of the Lth last layer
in terms of the cn1 coef®cients of the ®rst layer, and then match boundary
conditions at the a and b interfaces to obtain the ®nal unknowns of
the system. From [19] we obtain the following overall matrix equation
CL FL 1
FL
F1 C1 M C1
2
5:3:11
X
NT
I Ji
X1a p Ji0
X1a
c0 Ji
X1s cnL exp
qnL dL ja 1 s u'inL
n1
2 Ji
X1a zinL
5:3:12
" #( )
X
NT
2 0
Hi
2
X3b p Hi
X3b
EiI cn1 jb 3
2 szin1 u'in1
5:3:13
n1
2 Hi
X3b
p
where i MT ; . . . ; 0; . . . ; MT , J 0
X dJ
X=dX, etc., X1s 1 s ,
p p
X1a 1 a, and X3b 3 b. Equation 5.3.12 represents a set of 2MT 1
equations, Eq. 5.3.13 represents a set of 2MT 1 equations, and the matrix
equation Eq. 5.3.11 represents a set of NT 2
2MT 1 equation. Thus
Eqs. 5.3.11±5.3.13 represent a set of 2NT 4
2MT 1 equations to calcu-
late the 2NT set of unknowns represented by C 1 and C L . Once these quan-
tities are known, all other unknown coef®cients in the system can be found.
An important quantity to calculate is the normalized power of each
order. We consider the important cases when either the power is radiated
from the line source in Region 1
cI0 6 0, E0I 0) or the power is scattered by
a plane wave from Region 3 (cI0 0, E0I 6 0). In the case when cI0 6 0 and
E0I 0, the normalized power in each order is given by PNi PRAD i =PINC ,
INC RAD
where P is the incident power of the line source and Pi is the radiation
Inhomogeneous Cylindrical and Spherical Systems 307
at a radial distance . For the plane wave scattering case (cI0 0, E0I 6 0), it
is useful to calculate the normalized scattered at p 1:
PScat
i
1= 2 jc
3
i j
2
PScat
Ni
1
5:3:14
SINC jE0I j2
In Eq. 5.5.14, SINC is the power per unit area (watts/meter2 ) of the incident
plane wave, and PScat
i is the scattered power per unit length (watts/m) of the
ith order.
Figure 9 The normalized radiated power that results when a centered line source
excites a uniform dielectric shell (see Fig. 6, 1 1:5, 2 2:5, 3 1, 1) is
shown when determined by RCWA and when determined by a Bessel function
matching solution. Used with permission of EMW Publishing, 1998 [23, Fig. 4].
of the i 0 order and is diffracted into higher orders. One also observes that
unequal order power is radiated into the i 1 and the i 1 orders. This is
to be expected and is a result of the anisotropy of the permeable half shell.
Figure 10 The total radiated power that results when a centered line source radi-
ates through an anisotropic permeable half shell (see Fig. 7, 1 1:5, 3 1,
200 1:75, 200 1:5, 20 3:25, xx 1:5, xy 0:3, yx 0:3, yy 1:7,
MT 10, L 10 layers) when the inner radius is a k0 a~ 5 and when the outer
radius is varied from b a 5 to b 10 is shown. Used with permission of EMW
Publishing, 1998 [23, Fig. 5].
Inhomogeneous Cylindrical and Spherical Systems 309
Figure 11 A comparison of the total plane wave power scattered by the same
uniform dielectric shell example as considered in Fig. 9 (see Fig. 1, centered line
source not present) as determined by the Bessel function matching solution
(MT 15) and as determined by the RCWA method (using L 15 layers,
MT 15) is shown. Plane wave scattering from an anisotropic cylinder is also
shown. Used with permission of EMW Publishing, 1998 [23, Fig. 6].
310 Chapter 5
Figure 12 A three-dimensional plot of the plane wave scattered order power versus
order i when i is varied from i 15 to i 15 and versus the outer radius b~ when b~ is
varied from b~ 0:8 to b~ 1:6 is shown. This is part of the same numerical case as
was studied in Fig. 11. Used with permission of EMW Publishing, 1998 [23, Fig. 7].
Figure 13 The scattered order power that occurs when a plane wave impinges on a
uniform dielectric shell (see Fig. 6,, 1 1:5, 2 2:5, 3 1, 1) rather than an
anisotropic half shell is shown. Used with permission of EMW Publishing, 1998 [23,
Fig. 8].
Inhomogeneous Cylindrical and Spherical Systems 311
Figure 12 is part of the same numerical case as was studied in Fig. 11. As can
be see from Fig. 12, one clearly observes asymmetry of the order power as
the size of the outer radius b~ is increased. The sum of the plane wave order
power at any given b~ gives the total scattered power, which is displayed in
Fig. 11. We again note that this total plane wave scattered power obeys
conservation of power as expected. Figure 13 in comparison with Fig. 12
shows the scattered order power that occurs when a plane wave impinges on
a uniform dielectric shell (1 1:5, 2 2:5, 3 1, 1) rather than an
anisotropic half shell. The uniform dielectric shell has dielectric permittivity
values roughly the same size as that of the anisotropic half shell. As can be
seen from Fig. 13, the three-dimensional shape of the Fig. 13 plot from the
uniform shell is symmetric in the order parameter i and in general has quite
a different shape from that of the anisotropic half shell in Fig. 13.
Figure 14 shows the total radiated power (normalized to the dipole
power of the centered line source) when a line source radiates from an
isotropic square cylinder embedded in an anisotropic permeable half shell
(see Fig. 8, 1 1:5, 3 1, 200 3:5, 200 1, 20 3:25, xx 1:5,
xy 0:3, yx 0:3, yy 1:7, MT 20, L 25 layers). The radiated
power was calculated at a
a~ 1 which is a circle inscribed in the
square cylinder of Region 1 and was calculated at b (b~ 2:5, which
is the outer radius ~
pof
the anisotropic half cylinder. The outer radius b was
varied from b~ 2a~ 1:414 to b~ 2:5. As can be seen from Fig. 14,
extremely good power conservation was observed at a
a~ 1) and at
~ Despite the square shape of the cylinder, no power was observed to
~ b.
be diffracted into higher orders at a (a~ 1. Also show in Fig. 14 are
the i 1, i 0, and i 1 orders radiated at b (outer radius). As in
Fig. 12, one observes that power is depleted from the i 0 order and
radiated into higher orders. Figure 14 shows the increase in the i 1
and i 1 orders, for example, that occurs when b~ is increased. One also
observes in Fig. 14 that the order power is radiated asymmetrically into the
i 1 and i 1 orders. As in Fig. 14 this is expected and is due to the
anisotropy of the permeable half shell.
Figure 15 shows a plot (dotted line) of the relative dielectric permit-
tivity function
; ' when ~ 1:241 for the square cylinder anisotropic
half shell case displayed in Fig. 8. The circular dashed line of Fig. 8 repre-
sents the approximate placement of the ~ 1:241 parameter used to make
the Fig. 15
; ' plots. Also shown in Fig. 15 (solid line) is the Fourier
series representation of the
; ' pro®le when ~ 1:241 and MT 20
(MT 20 was used to make the RCWA method of Fig. 14). As can be
seen from Fig. 14, enough Fourier terms ( 40 2MT i 2MT 40)
were used to model correctly the inhomogeneous region as de®ned by the
square cylinder. (Note: The convolution matrix of Eq. 5.5.8 requires 2MT
312 Chapter 5
Figure 14 The total radiated power (normalized to the dipole power of the cen-
tered line source) when a line source radiates from an isotropic square cylinder
embedded in an anisotropic permeable half shell (see Fig. 8, 1 1:5, 3 1,
200 3:5, 200 1, 20 3:25, xx 1:5, xy 0:3, yx 0:3, yy 1:7, MT 20, L
25 layers) is shown. Used with permission of EMW Publishing, 1998 [23, Fig. 9].
Figure 15 A plot (dotted line) of the relative dielectric permittivity function
; '
when ~ 1:241 for the square cylinder anisotropic half shell case displayed in Fig. 8
is shown. The circular dashed line of Fig. 8 represents the approximate placement of
~ 1:241 parameter used to make
; ') plots here. Also shown (solid line) is the
Fourier series representation of the
; ' pro®le when ~ 1:241 and MT 20.
Used with permission of EMW Publishing, 1998 [23, Fig. 10].
Inhomogeneous Cylindrical and Spherical Systems 313
Figure 16 Plots of the relative permeability function
; '. Exact (dotted line)
and Fourier series representations (solid line) for the same case as described in Fig.
15 are shown. Used with permission of EMW Publishing, 1998 [23, Fig. 11].
314 Chapter 5
Figure 17 Plots of the relative permeability function '
; ' '
; '. Exact
(dotted line) and Fourier series representations (solid line) for the same case as
described in Fig. 15 are shown. Used with permission of EMW Publishing, 1998
[23, Fig. 12].
Figure 18 Plots of the relative permeability function ''
; '. Exact (dotted line)
and Fourier series representations (solid line) for the same case as described in Fig.
15 are shown. Used with permission of EMW Publishing, 1998 [23, Fig. 13].
Inhomogeneous Cylindrical and Spherical Systems 315
1 @U' 1 @U 1 @S' 1 @S
Er Hr
j
r; ; 'r2 @
@' j
r; ; 'r2 @
@'
5:4:1
5:4:2
@S' 1 @ 1 @ 1 @ 1 @
j
r; ; '
2 U j 2 U'
@r r
@'
r; ; ' @' r @'
r; ; ' @
5:4:3
@U
@ 1 @
r; ; '
@ 1 @
j 2 S j 2 S'
@r r @
r; ; '
@'
r @
r; ; ' @
5:4:4
@U' 1 @ 1 @
j
r; ; '
S
@r r2
@'
r; ; ' @'
1 @ 1 @
j S'
5:4:5
r2 @'
r; ; ' @
azimuthal grating wave vector and ' the azimuthal grating period. The
matrix for a general inhomogeneous factor, say
rmid ` ; ; ', for example, is
`
`
i;m;
i 0 ;m 0 i i 0 ;m m 0 , where i i 0 ;m m 0 are the two-dimensional
Fourier coef®cients of
rmid
`
` ; ; ', and
i;m;
i 0 ;m 0 represents a typical
matrix element of the overall matrix
` (note that
i; m is an ordered
pair representing a single integer in the
` matrix [same for
i 0 ; m 0 ]). The
matrices for the differential operators @=@ and @=@' are given by the diag-
onal matrices D jii;j 0 m;m 0 and D' jmi;i 0 m;m 0 , respectively, where
i;i 0 is the Kronecker delta, and the matrices describing the modal ®eld
amplitudes are given by column matrices (for example, S
`
` t
S
i;m (t
is transpose). Replacing each inhomogeneous factor, derivative operator,
and ®eld amplitude by the appropriate matrix, the overall system state
variable matrix can be found. The ®rst right-hand term of Eq. 5.4.2, for
2
example, is given by
j=rmid `
c
D
K
`
`
1=c
Du U A1;3 Uh , where c
and K
` mid
1=c matrices represent the factors
and 1=
r` ; ; '
, respec-
tively. The matrix A
` 1;3 , which was just formed, represents a square compo-
nent submatrix of the overall state matrix A
` . All component submatrices
`
Aa;b ,
;
1; 4 of the overall state matrix A
` are de®ned in the same
`
way as was A1;3 . (Since the component submatrices can be de®ned by inspec-
`
tion of Eqs. 5.8.2±5.8.5, it is not necessary to list the Aa;b ,
;
1; 4
matrices speci®cally.) The overall state variable equations, determined from
Eqs. 5.8.2±5.8.5 in the `th thin shell layer are given by
@V
`
A
` V
` ` 1; 2; 3; . . . ; L
5:4:6
@s`
where
2 3
0 0
`
A1;3
`
A1;4 2 3
6 7 S
`
6
` 7 6 7
6 0 0
`
A2;3 A2;4 7 6 S
` 7
6 7 6 ' 7
A
` 6
6
`
7
7 V
` 6
6
` 7
7
5:4:7
`
6 A3;1 A3;2 0 0 7 6 U 7
6 7 4 5
4 5
`
A4;1
`
A4;2 0 0 U
`
'
X
PT
`
X
PT
`1
Cp
` Simp exp
qp
` d` Cp
`1 Simp
5:4:8
p1 p1
X
PT
`
X
PT
`1
Cp
` S'imp exp
qp
` d` Cp
`1 S'imp
5:4:9
p1 p1
X
PT
`
X
PT
`1
Cp
` Uimp exp
q
`
p d` Cp
`1 Uimp
5:4:10
p1 p1
X
PT
`
X
PT
`1
Cp
` U'imp exp
q
`
p d` Cp
`1 U'imp
5:4:11
p1 p1
`
D
` C
` D C
`1
5:4:12
or
h i 1
`
C
`1 D D
` C
` F
` C
` ` 1; . . . ; L 1
5:4:13
320 Chapter 5
C
L F
L 1
F
L 2
F
1 C
1 M C
1
5:8:14
where
8 9
XIT <2IT jmj
X m;0 h i=
1;Scat
1
1
1
1
Sm
EAimn Fmn EBimn AA exp
ji
i I
: njmj
mn ;
T m;0
5:4:17
8 9
XIT <2IT jmj
X m;0 h
1
1 i=
1;Scat
1
S'm
ECimn Fmn EDimn A
1
mn exp
ji
i I
: njmj ;
T m;0
5:4:18
Inhomogeneous Cylindrical and Spherical Systems 321
where
Letting gAmn
1 2 1=2 Pnjmj
, gBmn
1 2 1=2 @=@
Pjmj n
), gCmn
1 2 @=@
Pjmj jmj jmj
n
), and gDmn Pn
, where Pn
are associated
Legendre functions of order n and jmj, the coef®cients gA B C
imn , gimn , gimn , and
D
gimn represent the exponential Fourier series expansion coef®cients for the
terms gAmn
, gBmn
, gCmn
, and gDmn
, respectively, on the interval 1
1
gAmn
i gA imn exp
ji). In the present analysis the Fourier coef-
®cients gA imn , g B
imn , g C
imn , and gD
imn have been determined exactly by calculating
higher order derivatives of the Bessel function integral representation given
in Ref. 27, Eq. 9.1.20, p. 360. The exact calculation of these Fourier coef®-
cients is an important step in order to ensure overall accuracy of the entire
1;INC
1;INC
1;INC
RCWA algorithm. The terms Sm
i Sim exp
ji and S'm
1;INC
i S'im exp
ji represent electric ®eld EM incident waves evaluated
at r a that emanate from Region 1. In this chapter, it is assumed that a
centered electric dipole excites EM radiation in the overall system, and for
this source it is found that Sm
1;INC
j=1 AI0;1 H^ 10
1 a1 2 1=2 m;0 ,
1;INC I
S'm
0, and A0;1 is the strength of the electric dipole source. In
p
Eqs. 5.4.15±5.4.18, 1 1 1 and J^n
1 a are spherical Schelkunoff
Bessel functions [5]; the prime in Eqs. 5.4.15±5.4.18 represent differentiation
w.r.t. the argument. For m 6 0, the lower n limits start at jmj, since the
2;a
2;a
Legendre polynomials are zero when jmj > n. The terms Sm
and S'm
represent the state variable solution in Region 2 in the Lth thin shell layer
region evaluated at r a and are given by
( )
2;a
X
IT X
PT
L
Sm
Cp
L Simp exp
q
L
p sL exp
ji
i IT p1
X
IT
2;a
Sim exp
ji
5:4:19
i IT
( )
X
IT X
PT
L
2;a
S'm
Cp
L S'imp exp
q
L
p sL exp
ji
i It p1
X
IT
2;a
S'im exp
ji
5:4:20
i IT
322 Chapter 5
where
8 9
XIT <2IT jmj
X m;0 h i=
1;Scat
1
1
1
1
Um
HBimn Fmn HAimn Amn exp
ji
i I
: njmj
;
T m;0
5:4:24
8 9
XIT <2IT jmj
X m;0 h
1
1 i=
1;Scat
1
U'm
HDimn Fmn HCimn A
1
mn exp
ji
i I
: njmj ;
T m;0
5:4:25
Inhomogeneous Cylindrical and Spherical Systems 323
where
X
IT
2;a
Uim exp
ji
5:4:26
i IT
( )
2;a
X
IT X
PT
L
U'm
Cp
L U'imp exp
qp
L sL exp
ji
i IT p1
X
IT
2;a
U'im exp
ji
5:4:27
i IT
the C
L coef®cients alone. By inspecting the matrix equation Eq. 5.4.21 and
their de®nitions, we notice that two distinct cases arise, namely the cases
when m 6 0 and the case when m 0. In the case of m 6 0, it turns
out that the matrices E
1
m
and H
1
m are nonsingular; therefore it is straight-
F
1
1 1 1
m Em
S
2;a C
L E
1
m
1;INC
Sm Z
1
m C
L
E
1;INC
m
A
1
m
5:4:29
1
1
The determination of F0:n and A0;n (n 1; 2; 3; . . . coef®cients for
the m 0 case requires special matrix processing. We ®rst note for the m
1
1
0 case that EAim EDim 0 in Eqs. 5.4.17 and 5.4.18 and thus the
matrix equations for F0;n and A
1
1
0;n are decoupled from one another. One
1
also observes from Eqs. 5.4.17 and 5.4.18 that when solving for either F0;n
1
1
1
1
1
1
1
and A0;n , the coef®cients of F0;1 ; F0;3 ; F0;5 ; . . . and A0;1 ; A0;3 ; A0;5 ; . . . are
multiplied by the ®rst-derivative Legendre polynomials @=@
P01
,
1
1
@=@
P03
; . . ., which are even in , whereas the coef®cients of F0;2 ; F0;4 ;
1
1
1
1
F0;6 ; . . . and A0;2 ; A0;4 ; A0;6 ; . . . are multiplied by ®rst-derivative Legendre
polynomials @=@
P02
), @=@
P04
; . . . which are odd in . This means
1
1
that when determining the m 0, Region 1 coef®cients F0;n and A0;n , the
1;INC
best numerical processing in Eq. 5.4.21 is to decompose Sm
,
1;INC
2;a
2;a
S'm
, Sm
, S'm
for m 0 into a sum of even and odd func-
1
tions, and from the even functions in Eqs. 5.4.17 and 5.4.18 determine F0;1
1
1
1
1
1
; F0;3 ; F0;5 ; . . . and A0;1 ; A0;3 ; A0;5 ; . . ., and from the odd functions in Eqs.
1
1
1
1
5.4.17 and 5.4.18 determine F0;2 ; F0;4 ; F0;6 ; . . . and A0;2 ; A
1
1
0;4 ; A0;6 ; . . .. The
1
1
1
speci®c matrix processing that is carried out for say the F0;1 ; F0;3 ; F0;5 ;...
1;INC
2a
coef®cients is as follows. After decomposing S'm
and S'm
for m
1
1
1
0 into even and odd functions of , F0;1 ; F0;3 ; F0;5 ; . . . is determined by
1;INC
2;a
(1) expanding the even function part of S'm
and S'm
for m 0
IT IT
2;a
in a fcos
ivgi0 cosine series (the fcos
igi0 series expansion of S'm
v
L
for m 0 depends on the Cp , p 1; . . . ; PT coef®cients in Region 2, and
IT
1;INC
the fcos
ig i0 series expansion of S'm
for m 0 depends on the
incident EM source waves that emanate from Region 1); (2) expanding the
®rst-derivative Legendre polynomial @=@
P01
, @=@
P03
; . . . in a fcos
IT
ig i0 series; (3) equating common coef®cients of the cosine series
IT
fcos
ig i0 ; and (4) from these equations, developing an
IT 1
IT
1
1
1
1 matrix equation that upon matrix inversion expresses the F0;1 ; F0;3 ; F0;5
L
; . . . coef®cients in terms of the Cp , p 1; . . . ; PT coef®cients of Region 2
and incident EM wave coef®cients of Region 1. The determination of the
Inhomogeneous Cylindrical and Spherical Systems 325
" # " #
F
1 Z0
1;F 0
0
1;A C
L E
1;INC Z
1
0 C
L
E
1;INC
A
1
0 0 Z0
0 0
5:4:30
h i
2;a
Um H
1
m Z
1
m C
L
1;INC
Cm H
1
m Em
1;INC
5:4:31
5:4:32
( )
X
IT X
PT
1
X
IT
2;b
2;b
S'm
Cp
1 S'imp exp
ji S'im exp
ji
i IT p1 i IT
5:4:33
and (4) one sets all Region 3 incident source terms to zero, since EM energy
in this chapter is assumed to emanate only from a Region 1±centered dipole
source. After algebraic manipulation it is found that the Region 3 boundary
equations are
h i
2;b
Um H
3
3
m Zm C
1
U
3;INC
m H
3
m Em
3;INC
5:4:34
h i
2;a
Um H
1
1
m Zm M C
1
1;INC
Um H
1
m Em
1;INC
5:4:35
where sgn
X 1 for X > 0 and sgn
X 1 for X < 0. For this pro®le
1. This inhomogeneity pro®le is convenient because if it is integrated
over a spherical surface of radius r, its average or bulk value is always 2 ,
regardless of the value of or ' used. Using this dielectric inhomo-
geneity pro®le, three cases will be studied, namely, Case 1, 0:001,
'
r 0:001; Case 2, 2:8, '
r 0:4; Case 3, 2:8,
'
r 1 r 2 , where '
rjr5 0:6 and '
4jr5:5 1:5 for
a r b, a 5, b 5:5. For all numerical examples of this chapter, the
bulk material parameters will be taken to be 1 1:5, 1 1, 2 7,
2 1:3, 3 1, and 3 1. The ®rst case, which because of the small
values of and ' can be called a homogeneous pro®le case, represents
the application of the RCWA method to the solution of the problem of
determining the EM radiation that occurs when a centered dipole radiates
through a uniform dielectric shell. Since this problem of EM radiation
through a homogeneous dielectric shell can be solved exactly by matching
Bessel function solutions in Regions 1, 2, and 3, comparison of the RCWA
method with the exact Bessel function matching solution represents a
numerical validation of the RCWA method if close numerical results from
the two methods occur.
The second case, which may be designated a
; '-inhomogeneity
pro®le case, represents an inhomogeneous example in which the dielectric
shell is homogeneous in the radial r direction but is inhomogeneous in the
and ' coordinates. This case will be solved by both a single layer RCWA
algorithm and a multilayer RCWA algorithm. The purpose of solving this
second case is to observe in general how much diffraction occurs in higher
order spherical modes when a reasonably large and ' inhomogeneity
material pro®le is present in the dielectric shell. The purpose of comparing
single-layer and multilayer RCWA results is to observe the importance that
the scale factors of Eqs. 5.8.2±5.8.5 have on the overall scattering solution.
The purpose is also to study how well the power conservation law is obeyed
numerically. Power conservation at different radial distances is a good indi-
328 Chapter 5
from Fig. 20, there is excellent numerical agreement between the three
methods used. We also notice that the r a; b power results for each of
the three methods are so close at r a and r b that the two power curves
for each method cannot be distinguished from one another. In Fig. 20 the
RCWA algorithm was calculated using MT 1 and IT 5. Because the
inhomogeneity factor in this case was very close to that of a perfectly homo-
geneous shell, the RCWA algorithm could have calculated the power of this
case using a value of MT 0 and IT 5, which would have meant a sig-
ni®cantly smaller matrix equation than would have resulted from MT 1. A
larger matrix equation than necessary was solved for this case in order to
test the numerical stability of the algorithm and also to test the sensitivity of
the RCWA solution to error in the Fourier coef®cients. (Error in the
Fourier coef®cients would arise for MT 1 because numerical integration
is used to calculate the exp
jm', m 1; 0; 1 Fourier coef®cients, and thus
instead of the m 1 coef®cients being exactly zero, they would have some
small value.) The MT 1 matrix solution showed no ill-conditioned effects
from using a matrix size larger than needed and showed no sensitivity to
error in the Fourier coef®cients. An RCWA method was also carried out
using MT 1 and IT 2. In this case the RCWA algorithm differed per-
ceptibly from the Bessel matching solution. This indicates that for accurate
results, enough Fourier harmonic terms must be included to calculate the
state variable solution of Eqs. 5.4.2±5.4.5. We note ®nally that a case nearly
the same for the purpose of validation, was also studied in Ref. 25 using MT
0 and IT 5. The numerical results between Fig. 20 and Ref. 25 were
almost identical.
Figure 21 shows a comparison of the total normalized powers that
occur when the dielectric shell is taken to be a uniform layer (the power here
is calculated at r b by Bessel function matching) and when the dielectric
shell is taken to be a (; ')-inhomogeneity pro®le with 2:8, '
r
0:4 (Case 2). The power here is calculated at r a; b by a single-layer
analysis and a multilayer analysis. As can be seen from Fig. 21, the presence
of the (; ')-inhomogeneity pro®le causes a marked difference in the total
scattered power of the inhomogeneous shell, although the bulk dielectric
inhomogeneity pro®le was exactly the same as that of the uniform homo-
geneous shell. It is also noticed from Fig. 21 that for both the single- and
multilayer analyses the law of power conservation at r a and r b is
obeyed to a reasonable degree of accuracy. Also plotted in Fig. 21 is the
m 0, n 1 power at r a. It is noticed that the m 0, n 1 (; ')-inho-
mogeneity pro®le power at r a almost exactly equals that of the total
power at r a; b. This indicates that at r a no power has been diffracted
into higher order modes at the r a interior boundary shell interface of the
system. Figure 21 also shows the m 0, n 1 power as calculated at r b.
330 Chapter 5
Figure 21 A comparison of the total normalized powers that occur when the
dielectric shell is taken to be a uniform layer (the power here is calculated at r
a; b by Bessel function matching) and when the dielectric shell is taken to be a (; ')-
inhomogeneity pro®le with 2:8, '
r 0:4 (Case 2). The power here is
calculated at r a; b by a single-layer analysis and a multilayer analysis. The
m 0, n 1 order power is also shown. Used with permission of IEEE, 1997 [26,
Fig. 3].
same result up to about a shell thickness of b a 0:4 rad, but after this
value the multilayer analysis is needed for more accurate results. Figure 23
also shows the m 1, n 1; 2; 3 order power as calculated by a single-layer
analysis. One observes that as the shell thickness b a increases, the m 1,
n 1; 2; 3 order power increases.
Figure 24 shows a plot of the total normalized power that results when
the (r; ; ')-inhomogeneity pro®le of Case 3 was solved using a multilayer
Figure 24 Plot of the total normalized power that results when the (r; ; '-inho-
mogeneity pro®le of Case 3 was solved using a multilayer RCWA method and using
MT 4 and IT 5 is shown. Also shown for comparison is the total power of a
uniform shell system (Case 1 parameters) and the total power when a (; ')-inhomo-
geneity pro®le was used with '
r set to a constant value of '
r 0:375. Used
with permission of IEEE, 1997 [26, Fig. 6].
One of the best areas of research concerning the present chapter would
be to implement the numerical stability algorithms for RCWA [19,20],
which were described in the last section of Chapter 3. This would greatly
increase the size of the spherical or cylindrical scatterer which could be
analyzed by the RCWA algorithm. Problem 3 below suggests a cylindrical
problem for which the RCWA enhanced transmittance matrix method may
be implemented and numerically studied. Problem 10 suggests another
application of this method which applies to an inhomogeneous spherical
scattering system.
PROBLEMS
02 ; 0 < 0 ; a b
l2
l002 ; 0 < 3608; a b
Inhomogeneous Cylindrical and Spherical Systems 335
"02 ; 0 < 0 ; a b
"2
"002 ; 0 < 3608; a b
where 0 1208; "02 1:5; "002 3:5. Assume that free space occu-
pies Regions 1 and 3, which are outside the shell.
(a) From Maxwell's equations in Region 2, ®nd the state variable
equations that describe the EM ®elds in the system. The state
variable equations should relate the nonradial components of
the electric and magnetic ®elds. Be sure to include and account
for all z derivatives in your solution. (Hint: Reduce z derivatives
using @z@ / j:
336 Chapter 5
@V
AV
@r
t
where V Sh t Sr t Uh t Ur t and Sh t Sr t Uh t Ur t are column
matrices holding Floquet harmonics of the EM ®eld variables.
(b) Assuming a centered dipole, solve a numerical example of
your choice using the RCWA method.
10. Apply the RCWA enhanced transmittance matrix method [19,20]
to Problem 7 for cases where numerical instabilities would other-
wise arise.
REFERENCES
341
342 Chapter 6
where h n=b.
These components correspond to the sum of two waves traveling with
the same velocity in the positive and negative y direction. This can only
occur in media presenting symmetry with respect to the y axis, namely media
in which [8]
1
` a ` a
6:1:4a
1
l` a l` a
6:1:4b
Only in such media, which are called cross-tensor media [1], are the ®eld
components obtained by Eq. 6.1.6. The most interesting examples are the
following.
1. Ferrite slabs magnetized in the a^ y direction (the case of intrinsic
anisotropy in which xx 6 zz is included). In this case we have a tensor l`
and scalar ` .
2. Plasma or semiconductor (tensor i ; scalar i ) magnetized along
the a^ y direction (tensor ` and scalar ` ). In this case we also have xx 6 zz .
3. Uniaxial crystals (tetragonal, hexagonal, trigonal) and certain
types of biaxial crystals (orthorhombic) having one of their principal dielec-
tric axes along a^ y .
4. ``Simple anisotropic'' and isotropic materials: these are degenerate
cases of cross-tensor media (diagonal tensors).
It should be noted that longitudinally magnetized ferrites, plasmas,
and semiconductors do not meet these conditions and are not covered by
the present treatment.
We mention at this point that the media studied in Chapter 2, Section
2.4 and Chapter 3, Sections 3.2 and 3.3 are classi®ed as cross-tensor media
as is shown in Section 6.2.4 of this chapter.
6.1.4 Formulation
To begin the analysis we write Maxwell's equation in the `th slab
(` 1; . . . ; L; ` index suppressed in the following equations) in the sym-
metric form [9]
r E~
H
~ 0
6:1:7a
E~ ~ 0
r
H
6:1:7b
Anisotropic Inhomogeneous Waveguide and Periodic Media 345
p
where j 0 =0 j0 and and l are dimensionless relative perme-
ability and permittivity tensors whose nonzero values have been listed ear-
lier. To ®nd a solution to Eqs. 6.1.7a,b we express the electric ®eld E and
magnetic ®eld H as
h i
E~ E^ x
x sin hya^ x E^ y
x cos hya^ y E^ z
x sin hya^ z e
z
6:1:8a
h i
H~
H~ x
x cos hya^ x
H~ y
x sin hya^ y
H~ z
x cos hya^ z e
z
6:1:8b
a^ y
E^ x sin hye
z
rx E^ z sin hye
z
a^ y
H^ x cos hye
z
rx
H^ z cos hye
z
where rx @=@x. Substituting into the lH and E terms of Eqs. 6.1.7a,b,
we ®nd that
346 Chapter 6
2 32 ^ 3
0
h xx 0 xz Ex
6
6 7
6 0 rx 0 yy 0 776 E^ y 7
6 7 6 7
6 h rx 0 zx 0 zz 76 E^ z 7
6 766
7
70
6:1:11
6 h 7 ^x 7
6 xx 0 xz 0
766 H 7
6 7
4 0 yy 0
0 rx 5 6 7
4 H^ y 5
zx 0 zz h rx 0 H^ z
The above matrix equation is still in an awkward form for numerical com-
putation. This analysis will now concentrate on eliminating variables and, in
fact, putting it into a standard state variable form for which a solution can
be found. When reducing the number of ®eld variables in Eq. 6.1.11, the
most logical ones to eliminate are the x-directed variables E^ x ad H^ x . These
variables are normal to the interface of the different anisotropic slabs and
thus are in general discontinuous at the boundary interfaces. On the other
hand, Maxwell boundary conditions require that the EM ®elds tangential to
the interface boundaries should be equal at each boundary interface, and
this implies that the modal amplitudes E^ y , E^ z , H^ y , and H^ z should be
continuous at the boundary interfaces, thus greatly facilitating the boundary
matching operation from layer to layer. The objective now is to express E^ x
and H^ x in terms of E^ y , E^ z , H^ y , and H^ z and thus put Eq. 6.1.11 into a
standard state variable form.
Using Eq. 6.1.10e and 6.1.10b, respectively, we may write
Anisotropic Inhomogeneous Waveguide and Periodic Media 347
1 1
E^ x r E^
H^ y
6:1:12a
x z
yy
yy ^ 1
H^ x E r
H^ z
6:1:12b
y
x
We notice from Eq. 6.1.13c that the only rx term is rE^ z , and thus we may
solve for it alone. Doing so we ®nd that
1 h ^ i
rx E^ z 0Ez
xz xx E^ z xx k2y
H^ y
hxx
H^ z
xx xx
6:1:14a
where k2y
2 yy xx . We also notice from Eq. 6.1.13a that rx H^ z is the
only rx term in Eq. 6.1.13a. Solving for it we ®nd that
1 h i
rx H^ z xx k2x E^ y h
xx E^ z 0
H^ y
xz xx
H^ z
xx xx
6:1:14b
where
xz zx
e zz
6:1:15a
xx
xz zx
e zz
6:1:15b
xx
k2x yy xx
2
6:1:15c
k2y xx yy
2
6:1:15d
The state variable form for the EM ®eld components parallel to each
slab interface can be used to determine the ®nal propagation constant of
modes that exist in the waveguide. We ®rst note that the 4 4 set of ®rst-
order differential equations, because the A matrix in the `th slab is a con-
stant matrix, can be solved by ®nding the eigenvalues and eigenvectors of
the constant matrix A. If we let Ve
x E^ y
x; E^ z
x; H^ y
x, H^ z
xt and
Anisotropic Inhomogeneous Waveguide and Periodic Media 349
A Vn qn Vn 6:1:16
@ e
V
x A Ve
x
6:1:17
@x
where
X
4
Ve `
x` Cn 0 exp
qn
`0 x` Vn
`0
6:1:19
n 0 1
where V
` n 0 are the state variable eigenvectors of the `th slab region (Eq.
6.1.16). The ®eld components of the `th slab region contained in the matrix
Ve `
x` are
h it
Ve `
x` E^ y`
x` ; E^ z`
x` ; H^ y`
x` ; H^ z`
x`
6:1:20
At x s1 s` 1 ; x` 0 for ` 2; . . . ; N`
or
h ih i h ih i
` 1
Vnn 0 exp
qn
`0 1 s` 1 Cn
`0 1 Vnn
`
0 Cn
`0
6:1:24
h ih ih i h ih i
` 1
Vnn 0 exp
q
`n0
1
s` 1 n 0 n Cn
`0 1 Vnn
`
0 Cn
`0
6:1:25
Letting D` 1 exp
q
`
n0
1
`
s` 1 n 0 n , V` Vnn
`
`
0 , and V` Vnn 0 , C` Cn 0 ,
we then have
V` 1 D` 1 C` 1 V` C` 6:1:26
or
C` V` 1 V` 1 D` 1 C` 1 6:1:27
6:1:28
or
Ve N`
sL
VN` DN` VN1`
VN` 1 DN` 1 VN1` 1
V1 D1 V1 1 Ve 1
0
6:1:30
2 2 3 32 3
0 T11 T12 T13 T14 0
6 7 6
6 0 7 6 T21 T22 T23 T24 7
76
6 0 7
7
6 7
6 H^ yN
sN 7 6
4 T31 T32 T33
76 ^
T34 54 Hy1
0 5
7
6:1:32
4 ` ` 5
^
HzN`
sN` T41 T42 T43 T44 H^ z1
0
Since the right-hand side of both these equations is zero, the equations must
be linearly dependent for a nontrivial solution to correspond to them.
Setting the determinant to zero we ®nd that the ®nal propagation equation
for the n 6 0 mode is
TEn0 Modes
To determine the modes of the n 0 case, a separate EM ®eld analysis of
Maxwell's equations must be carried out. In this case we take Ex Ez 0
and Hy 0. After a reduction of equations similar to that carried out for the
n 6 0 case, it is found that the E^ y and H^ z ®eld variables can be put into the
state variable form
" # " #
@ E^ y 1
zx xx e E^ y
6:1:35
@x H^ z xx k2x xz H^ z
or
" #
0 T11 T12 0
6:1:37
H^ zN`
sN` T21 T22 H^ z1
0
or
T11 0 6:1:38
for a nontrivial solution to this equation to hold. Using the same numerical
techniques described earlier, the roots of this equation can be found.
Figure 2 Rectangular waveguide with H-plane ferrite slab. Used with permission
of IEEE-MTT, 1970 [1, Fig. 5].
reverse direction (curves 1 and 2 correspond to the two lowest modes) and
the normalized phase coef®cient k (two modes in both directions) as a
function of frequency. Comparison with values of the attenuation measured
by Bernardi [10] and presented in Fig. 3 show good agreement between
calculations and experiment. It is worthwhile noting that the particular
hump on the right-hand side of the attenuation curve, which corresponds
to the resonance of e (Fig. 5), is quite apparent from the oscillograms of
Bernardi [10].
Comparing Figs. 3 and 4 it appears that coupling takes place between
the empty waveguide and curve 1 in the dispersion diagram (Fig. 4) when
the frequency is lower than the resonant; at higher frequencies coupling
occurs to curve 2 of Fig. 4.
It is also worth noting that in ferrite-loaded waveguides, large values
of attenuation can occur with large values of phase shift. This corresponds
Anisotropic Inhomogeneous Waveguide and Periodic Media 355
6.1.7 Conclusion
The study of a rectangular waveguide containing slabs extending across it
has shown that an exact derivation of the dispersion equation can be made
when the material properties meet certain requirements. This is the case for
cross-tensor media, which include in particular transversely magnetized fer-
rites, plasmas and semiconductors, certain crystal orientations of crystals,
and also dielectrics and resistive materials. A state variable analysis of
Maxwell's equations was made, the dispersion equations of the modes of
the loaded structure were derived, and a numerical solution of the dispersion
356 Chapter 6
the validity of both methods. Second, the present method, because it bound-
ary matches the EM ®elds at different boundaries, should be an ideal
method for describing the internal EM ®elds. In comparison, the RCWA
method uses a large number of Fourier terms to represent a step disconti-
nuity, whereas the present modal method builds step continuities into solu-
tion, therefore giving a very accurate representation of the EM ®elds
internal to the grating system.
~ y k0 y,
tion we will use normalized coordinates in which x k0 x, ~ etc., and
p
k0 ! 0 0 , x~ in meters. In Gardiol's n 0 waveguide formulation
(denoted in this section with a prime in order that the waveguide ®eld
solutions will not be confused with the grading solution), the EM ®elds
are expressed in the form
E~ E^ y0
xa^y e
z
6:2:1a
H~
H x0
xa^ x
H z0
xa^ z e
z
6:2:1b
where
1
zx xx e
A0
xx k2x
zx
Letting V E^ y
x H^ z
xt , a comparison with Eqs. 6.2.1a,b shows that
V 0 V exp
jkx0 x. If V 0 V exp
jkx0 x is substituted into Eq. 6.2.2,
we ®nd after algebra that
" # " #
@ E^ y a11 a12 E^ y
AV
6:2:4
@x H^ z a21 a22 H^ z
where
A
jkx0 I A 0
6:2:5
Anisotropic Inhomogeneous Waveguide and Periodic Media 359
The matrix A refers to any slab layer `, and the subscript ` has been sup-
pressed in the above equations. By carrying out a multilayer matrix analysis
based on eigenvalues and eigenvectors identical to that which has already
been presented, we ®nd that
" # " # " #
E^ yL
sL E^ y1
0 E^ y1
0
TN` TN` 1 T2 T1 T
6:2:6
H^ zL
sL H^ z1
0 H^ z1
0
where
q1` 0
T` Q Q 1
` 0 q2` `
An inspection of Eq. 6.1.3.6 shows that only nontrivial solutions will result
for this case when
This equation then forms an equation from which the propagation con-
stants in the grating can be determined.
Once the propagation constant has been determined, it is possible (up
to a normalization constant) to determine the full form of each electromag-
netic mode that can propagate in the grating system. The ®eld coef®cients
E^ y`n
x` and H^ z`n
x` of the nth mode in the `th layer are found, after
assuming an appropriate normalization constant, by using a matrix cascade
to relate the ®eld amplitudes at one grating boundary to the `th layer. A
small amount of algebra shows that in the `th layer,
1n o
H^ x`n
x` jkx0 a22`n H^ z`n
x` a21`n E^ y`n
x`
6:2:9
n`
360 Chapter 6
X
NT
Ey
2 exp
jkx0 x Cn E^ yn
x exp
n z
6:2:10a
n1
X
NT
Hx
2 exp
jkx0 x Cn H^ xn
x exp
n z
6:2:10b
n1
where
(
n
n
6:2:11
n
8
< E^ yn
x
E^ yn
x
6:2:12a
: E^
x
yn
(
Hxn
x
H^ xn
x
6:2:12b
Hxn
x
where NT 2
2MT 1, where plus and minus denote forward and back-
ward traveling modes, respectively and where n n; n 1; ; 2MT 1
and n n
2MT 1; n 2MT 2; ; NT : In writing Eqs. 6.2.10±
6.2.12, it is assumed that EM ®eld solutions, which were found in terms
of the local coordinates x` of each thin slab, have been properly translated
into the overall transverse variable x, which extends over the periodic grat-
ing interval 0 x .
In Regions 1 and 3 the EM ®elds are given by
Anisotropic Inhomogeneous Waveguide and Periodic Media 361
X
MT
Ey
1 E0 i0 exp
jkxi x jkz1i z ri exp
jkxi x jkz1i z
i MT
6:2:13a
1 X
MT
kz1i
Hx
1 E0 i0 exp
jkxi x jkz1i z
0 i M 1
T
1 X
MT
Hx
3 t exp
jkxi x jkz3i
z L
6:2:13d
0 i M i
T
8 p
< r r k2xi 1=2 r r > kxi
kzri p r 1; 3
6:2:13e
: jk2 r r 1=2 kxi > r r
xi
and
X
MT
Ey
1 jz0 exp
jkx0 x
E0 i0 ri exp
jiKx x
6:2:14b
i MT
X
NT
Ey
2 jz0 exp
jkx0 x Cn E^ yn
x
6:2:14c
n1
or
362 Chapter 6
X
MT X
NT
E0 i0 ri exp
jiKx x Cn E^ yn
x
6:2:14d
i MT n1
Using the orthogonality of Floquet harmonics (or Fourier series) (that is,
after multiplying Eq. 6.2.14d by
1=x exp
ji 0 Kx x, integrating across a
grating period, and replacing i 0 by i in the notation) we ®nd that
X
NT
x =2 X
NT
1
E0 i0 ri Cn E^ yn
x exp
jiKx xdx Cn Eyin
n1
x x =2 n1
6:2:15
where
x =2
1
Eyin E^ yn
x exp
jiKx xdx
6:2:16
x x =2
X
NT
kz1i 1 x =2 ^
E0 i 0 0 ri j Cn H
x exp
jiKx xdx
1 n1
x x =2 xn
X
NT
j Cn H^ xin
6:2:17a
n1
X
NT
ti Cn Eyin
6:2:17b
n1
kz3i X
NT
ti j Cn H^ xin
6:2:17c
1 n1
Elimination of ri and ti from these equations leads to the ®nal matrix equa-
tions for Cn . We have
X N
2kz1i T
k
Eo i0 Cn j H^ xin z1i Eyin
6:2:18a
1 n1
1
Anisotropic Inhomogeneous Waveguide and Periodic Media 363
X
NT
k
0 Cn exp
n L j H^ xin z3i Eyin
6:2:18b
n1
3
2 3 2 32 3
Fx 1 0 0 Fxs
F~ 4 Fy 5 4 0 0 1 54 Fys 5 MFs
6:2:19
Fz 0 1 0 Fzs
364 Chapter 6
Also
2 3 2 32 3
Fxs 1 0 0 Fx
F~s 4 Fys 5 4 0 0 1 54 Fy 5 M 1 F~
6:2:20
Fzs 0 1 0 Fz
We have
or
The permeability tensors ls and l obey the same relations as the permittivity
tensors above. Thus the formulation presented in this section can be used to
solve the diffraction grating problems considered in Sections 3.2 and 3.3,
which were solved by rigorous coupled wave theory.
slab reduce to multiply re¯ected plane waves traveling in the positive and
negative x-directions. Yamakita et al. [13±15] boundary matched expansions
of these plane waves at the slab interfaces, and after imposing periodic bound-
ary conditions at the grating cell boundaries they presented equations that
determined which sets of plane waves can propagate in the diffraction grating
system. The sets of propagating plane waves were termed Bloch wave modal
functions. Once the Bloch wave modal functions were determined, boundary
matching of the incident, re¯ected, and transmitted diffracted waves with
expansions of the Bloch wave modal functions at the Region 1±2 and
Region 2±3 interfaces provided the ®nal matrix equation from which all
EM ®elds in the diffraction grating system could be uniquely determined.
In addition to uniform rectangular groove gratings, Yamakita et al. [15]
have, based on the Bloch wave modal function theory, developed a multilayer
algorithm to study longitudinally inhomogeneous surface relief gratings.
Figure 7 (Yamakita et al. [14], Fig. 5) shows examples of plane wave
E-mode and H-mode diffraction ef®ciency versus groove depth as occurs
from a rectangular groove grating when Bloch wave modal function theory
(and therefore the theory of this section applied to isotropic materials) is
used to determine the EM ®elds of the system. Yamakita et al. [13±15]
number the diffraction ef®ciency orders in the number sequence opposite
to that used in Chapter 3 of the present text. The Yamakita ordering is
shown in the inset of Fig. 7.
PROBLEMS
1 ; 0 x x1
x
2 ; x1 < x a
0zz 3., 0xz 1:5, and 0zx 2:5: Let a~ 2b~ 2 cm. The guide
is operated at a frequency f , which is 15% higher than the cutoff
frequency of the lowest order mode which propagates when free
space occupies the guide.
(a) Find the propagation constants and EM ®elds in the wave-
guide of the lowest four modes. In the numerical analysis be sure
to include a suf®cient number of multilayers to assure proper
convergence of the solution.
(b) Calculate each mode's amplitude coef®cient so that each mode
at z 0 transmits in magnitude 1. watt of power across cross
section of the guide in the z direction.
4. Referring to the lamellar diffraction grating studied by Yamakita
and Rokushima [14, Fig. 5] (and reproduced in Fig. 7 of Sec. 6.2):
(a) Use the method described in Sec. 6.2 (based on the Gardiol
waveguide formulation [1], described in Sec. 6.1) to determine the
four lowest propagating (or possibly evanescent) transverse elec-
tric modes (Ez 0), which are excited in the lamellar diffraction
grating. (Transverse electric is also called H-mode.)
(b) Use rigorous coupled wave analysis as described in Chapter 3
to determine the four lowest transverse electric modes for the case
studied in (a).
(c) Plot the modal ®elds from both methods and compare the
results. Which method does the best job in meeting EM boundary
conditions of the system?
(d) Using the modal ®elds from (a), calculate the diffraction ef®-
ciency of the grating diffraction results shown in Fig. 7, Sec. 6.2
[14, Fig. 5]. Compare the results with that found by RCWA.
(e) Repeat (a) through (d) for transverse magnetic modes.
(Transverse magnetic is also called E-mode.)
5. An anisotropic, lambellar diffraction grating is bounded by free
space on the incident side (Region 1) and is bounded by a homo-
geneous, isotropic dielectric on the transmit side (Region 3),
whose relative dielectric permittivity value is "3 2:5. The permit-
tivity in the diffraction grating region is given by
(
"2a ; 0 x~ x~ 1 ; 0 z~ L~
"2
x
~ 0 z~ L~
"2b I; x~ 1 < x~ a;
REFERENCES
Since its discovery by Ashkin et al. [1] in the mid 1960s, a tremendous
amount of research has been carried out to study the photorefractive (PR)
effect and apply it to real-time image processing [2], beam ampli®cation [3],
self-pumped phase conjugation [4], four-wave mixing [5], and optical com-
puting [6], to name a few applications. A preliminary discussion of beam
propagation through PR materials was presented in Chapter 4. When two
coherent plane waves of light intersect in a PR material, they form an
intensity interference pattern comprising bright and dark regions (see Fig.
1). Assuming that the PR material is predominantly n-doped, the electrons
migrate from bright to dark regions, thus creating an approximately sinu-
soidal charge distribution. This diffusion-controlled PR effect in turn creates
an electrostatic ®eld that is ideally 90 phase shifted from the intensity
pattern and modulates the refractive index of the crystal via the electro-
optic effect (see Fig. 1). The incident plane waves are, in turn, scattered
by the grating so that one wave may have constructive recombination
while the other may encounter a destructive recombination. This effect
leads to energy coupling between the beams through what is commonly
referred to as the two-beam coupling effect [3].
A steady-state nonlinear coupling theory of the two-beam coupling
phenomenon using participating plane waves has been derived and the
results reconciled with numerical simulations of the coupling between two,
in general, focused Gaussian beams [7]. The results indicate that the two-
beam coupling parameter [8] strongly depends on the initial intensity ratio
of the plane waves (or power ratio of the input beams), an effect that has
371
372 Chapter 7
been experimentally observed before [9,10] but for which only an empirical
theory existed [11]. The empirical theory, however, was an improvement
over linearized time-dependent theories that predict that is independent
of the initial intensity ratio [8]. The dif®culty in providing an analytical
nonlinear time-dependent theory stems from the impossibility of exactly
decoupling the Kukhtarev equations [9] except in the steady state, and
numerical simulations of these coupled equations are also rather formidable.
Simpli®ed beam propagation results, based on the approximate steady state
solutions, was presented in Chapter 4.
Over and above the computational dif®culty with the Kukhtarev equa-
tions is that in a typical experimental setup, the PR crystal is ®nite and often
has a large linear refractive index (typically greater than 2), prompting
re¯ections from boundaries. This introduces the added complication of tak-
ing into account forward and backward traveling waves during the two-
beam coupling process and in fact extends the two-beam problem to a
degenerate four-wave mixing problem in a PR material. Recently, a ``uni-
®ed'' method for solution of four-wave mixing problems in PR crystals in
both transmission and re¯ection geometries has been proposed and applied
to problems of double phase conjugation and two-wave mixing with crossed
polarization in cubic crystals [12]. However, the methodology implies lin-
earized interaction equations between the PR grating and the intensity
grating and furthermore has been developed for the steady state.
A rigorous analysis of wave interactions, particularly between ordin-
ary and extraordinary polarizations, coupled through induced material ani-
sotropy, has been developed by Wilson et al. [13]. The material analyzed is
LiNbO3 : Fe where the PR effect is primarily photovoltaic. Arbitrary-direc-
tion two-wave coupling is studied and extended to multiwave coupling. The
analysis is rigorous from an electromagnetics standpoint since it employs the
Induced Photorefractive Gratings 373
@
ND n 1 @J
7:2:1
@t e @x
Induced Photorefractive Gratings 375
@ND
ND ND
sI
R N D n
7:2:2
@t
@n
J eDs enEs
7:2:3
@x
@Es
s e
ND NA n
7:2:4
@x
where
To start the analysis, we differentiate Eq. 7.2.4 with respect to time, and
after noting that @NA =@t 0, we ®nd that
@
ND n s @2 E 1 @J
7:3:1
@t e @x@t e @x
376 Chapter 7
If the last two equation statements of Eq. 7.3.1 are integrated with respect to
x (the constant of integration can be shown to be zero) and J of Eq. 7.2.3 is
substituted, we ®nd that
s @Es @n
nEs Ds
7:3:2
e @t @x
Substituting Es from Eq. 7.3.3 into Eq. 7.3.2, we ®nd the following equation
for :
@ en De @n
s rn40
7:3:4
@t s s @x
sI ND 1 @ND
R n
7:3:5
ND ND @t
The second term on the right-hand side (RHS) of Eq. 7.3.5 is the derivative
of a logarithmic term
1 @ND @ N
ln D
ND @t @t NA
Induced Photorefractive Gratings 377
and can be shown to be small relative to the ®rst term on the RHS of Eq.
7.3.5. It has been ignored for now in this approximate analysis. If ND is
approximated by NA in the denominator of Eq. 7.3.5, we can ®nally relate
the electron density n to the optical intensity I through the linear relations
@ rn40 Ds @I
I
7:3:7
@t
I =s @x
where
R NA s
I
7:3:8
eND
sI
@ @I=@x~
~
I
7:3:9
@t~
I C
where
t rn40 Ds k0
I
x~ k0 x t~ C ~
I
I0 s
I0
7:3:10
and where I0 is the peak intensity of the incident optical ®eld. In Eq. 7.3.10,
I0 denotes a normalizing constant that makes
I and t dimensionless. A
typical range for values of is from 10 5 (e.g., in BSO) to 10 2
(BaTiO3 ). Also, the constant C is of the order of 104 ±106 (W/m2 ) in most
PR materials. We remark also that for small
10 5 ), we have found that
the intensity pro®le I is nearly constant in time; thus Eq. 7.3.9 can be
directly integrated w.r.t. t~ to yield an explicit expression for :
@I=@x~ h ~
I
= ~
i
1 e
7:3:11
CI
378 Chapter 7
where
( )
r13 n2CO 2 2 r33 n2CE 3
Fxx FOE sin c cos c 2 sin c cos c sin c
r42 n2CE r42 n2CO
(
r13 n2CO 2
Fxy FOE sin c cos c cos c cos2 c sin2 c
r42 n2CE
)
r33 n2CE 2
sin c cos c
r42 n2CO
( )
r13 n2CO 3 2 r33 n2CE 2
Fyy FOE sin c 2 sin c cos c cos c sin c
r42 n2CE r42 n2CO
n2CO n2CE
FOE
n2O n2E
n2CO n2O jO00
n2CE n2E jE00
7:3:13
@ @I=@x~ Ds 2 2
~
I n n k r
7:3:14
@t~ CI O E 0 42
~
I is given in Eq. 7.3.10. The optical power intensity I
x; y; t in this
problem is assumed proportional to jEx j2 jEy j2 . The terms n2CO and n2CE
are the relative complex dielectric permittivities of the bulk crystal. The
terms n2O and n2E are real parts of the relative bulk dielectric permittivity
and O00 and E00 are the lossy parts. Note that Exsc , electric space charge ®eld,
is linearly related to the dielectric perturbation or modulation function by
the equation
Induced Photorefractive Gratings 379
The electro-optic constants r13 , r33 , and r42 are speci®ed in Ref. 8, Table 1.2,
pp. 26±29.
or
where F
t; I is the right-hand side of Eq. 7.3.14 and I is the optical intensity
of the light at time t.
380 Chapter 7
The time steps are chosen to be suf®ciently small so that only a very
small change in is observed at any given time step.
A numerical simulation of the nonlinear photorefractive dielectric
equations of Section 7.2 was made in order to study the interaction of the
TE and TM optical electric ®elds with the photorefractive material. The
simulation was performed for a photorefractive crystal located in free
space and was illuminated by two incident plane waves whose separation
angles with the normal to the crystal were 5:71 and 5:71 , respectively,
and whose complex amplitudes were respectively E0 ej0 and E1 ej rand . These
two plane waves caused an electric ®eld interference pattern whose nulls
were 5 apart.
@I=@x~
7:3:18
CI
I0 sin 2x~
7:3:20
C I0 cos2 x~
simulation. As can be predicted from Fig. 4b, the net effect of the nonlinear
function is, in general, to reduce and slow down the transfer of modal
energy in the R1 direction. Similar results occur for the normalized power in
the R0 , T1 , and T0 directions. It appears that the nonlinear does not have a
strong effect on the mode coupling as regards the ®nal steady state
(although the position-dependent case yields a slightly lower steady-state
value [23]); for this reason, a constant has been chosen for all subsequent
calculations in the interest of computation time.
384 Chapter 7
t
In these expressions Ei;inc
0
i 0; 1 is the incident electric ®eld in the ith
t
order direction, Ei
L
i 0; 1 is the transmitted electric ®eld at the ®nal
r
time step transmitted out of the slab in the ith order direction, Ei;inc
0
i
0; 1 is the re¯ected incident electric ®eld in the ith order direction, and Eir
0
i 0; 1 is the re¯ected electric ®eld at the ®nal time step re¯ected out of the
slab in the ith order direction. The gamma mode ratio is a measure that
shows to what degree the photorefractive medium has been able to convert
power from one mode to another.
For the plots of Fig. 6, the bulk dielectric was taken to be 2r 8,
2 10 5 , and the incident wave's refraction ratio p jE1 j=jE0 j was
varied from p 0:1
10 db) to p 10
10 db. The values of E0 and E1
were adjusted to keep the incident power Iinc 9:949 10 4 (W/m2 ) the
same for all values of p used.
The results of Fig. 6 show that the PR medium transfers the most
energy when the incident waves are coherent, and that a gradual decrease
in coupling ef®ciency occurs as the waves become more incoherent as seen
Induced Photorefractive Gratings 385
by the drop in t and r when 15 rand 15 and 30 rand 30 .
The results of Fig. 6 also show that the maximum mode coupling occurs
when the wave amplitudes are equal, p 1 jE1 j=jE0 j. This is to be
expected since the equal amplitude waves cause the greatest interference
pattern in the PR medium and thus cause the largest change in the medium,
which causes the most change in the mode power. The decrease in mode
coupling ef®ciency with increasing incoherence is expected since increasing
incoherence implies less interference of the waves and therefore less change
in the PR medium and thus less mode coupling. Note also that the ¯uctua-
tions in increase with rand indicating possibly reduced coherence of the
signal at the output, in agreement with Ref. 25. A time averaged , which
can be found by numerically repeating the experiment several times, will also
Induced Photorefractive Gratings 387
Figure 7 (a) Plot of the power transmitted in the T0 and T1 directions as functions
of time step in the nearly matched case when 2r 1 j1 10 6 and p 0:01 for the
values of 3:18 10 4 , 3:88 10 4 , and 4:59 10 4 when the PR slab
length=1.5 cm. Also C 2:124 104 (W/m2 ). As increases, the speed and com-
pleteness of mode conversion increases. (b) Plot of the normalized
p0 power re¯ected
and transmitted in the directions R0 , R1 , T0 , T1 when = 2r 4:59 10 4 ,
7 4 2
2r 8 j2:82 10 , C 2:124 10 (W/m ), and p 0:01. Because of the dielec-
tric mismatch, aperiodic variation of the normalized power results. Used with per-
mission of Opt. Engr., 1995 [30].
388 Chapter 7
4:59 10 4 , and fractional ratio of the incident beams has been taken to be
p jE1 j=jE0 j 0:01.
The waves marked T0a and T1a , T0b and T1b , and p0 T0c and T1c show the
diffraction ef®ciencies for the three values of = 2r as indicated in the
®gure insets. Because of index matching (for the case of Fig. 7a) the re¯ected
waves from the slab were nearly zero and thus are p not 0 shown. The main
effect that is observed in Fig. 7a using the values of = 2r that were given is
that a great deal of energy is transferred from the (order 0) transmitted E0
wave (which is large at t 0) to the (order 1) transmitted E1 wave (which is
small at t 0). As can be seen from Fig. 7a,p the conversion of the modal
0
energy depends very strongly on the value = 2r used and affects the value
of power diffracted in the T1 direction and the speed with which the mode
power transfer reaches the steady state. In Fig. 7a the presence of absorption
00
(2r 1 10 6 ) seemed to have a minimal effect on the diffraction except,
of course, to attenuate the T0 and T1 propagating waves.
Figure 7b shows the diffraction p0 that occurs when the real part ofpthe 0
0
bulk dielectric is r2 8 and = 2r 4:59 10 4 . The value of = 2r
used in Fig. 7b is the same as in the ``c'' case of Fig. 7a. In the simulation
case of Fig. 7b, because there is a large bulk dielectric mismatch between
free space and the PR medium, the incident and diffracted optical energy is
multiply re¯ected from the dielectric boundaries on the incident and trans-
mitted sides in a complicated way, causing a standing wave pattern to arise
in the PR medium. The intensity of this standing wave pattern changes
slowly in time, modi®es the PR medium according to Eq. 7.3.14 and thus
further diffracts the incident and re¯ected light. The net effect over time is a
highly inhomogeneous PR medium. The simulations, shown in Fig. 7b,
predict that the optical energy may oscillate or self-pulsate quasi-periodi-
cally between the R0 ; T0 ; T1 , and R1 directions, transferring energy back and
forth between the different orders. This is due to the temporal longitudinal
inhomogeneous PR medium formed by the interfering waves. Similar results
have been reported during self-pumped and mutually pumped phase con-
jugation in PR materials [17±20].
Figure 8 shows a plot of the dielectric perturbation that occurs at
the line y 0:75 cm (this line is midway between the PR medium bound-
aries) as a function of time step and transverse distance x (wavelength) in
0
the casep(1) 0 when the bulk dielectric is matched (2r 1) to free space (Fig.
4
8a, = 2r 3:88 10 ), (2) when p0 the bulk dielectric is mismatched
0
(2r 8) to free space (Fig. 8b, = 2r 4:59 10 4 ) and (3) when the
0
p0 dielectric is 4 mismatched (r2 8) to free space (Fig. 8c,
bulk
= 2r 5:30 10 ). As can be seen from these ®gures, when the bound-
ary is matched, the dielectric modulation approaches the steady state
quickly (Fig. 8a), whereas when the bulk dielectric is mismatched, the
Induced Photorefractive Gratings 389
Figure
p0 8 (a) Plots of versus time step and transverse distance x when
2 3:88 10 4 ,
=p 2r 1 j1 10 6 , p 0:01. (b)p when
0 4 7 0
= p
2r 4:59 10 , p 0:01, and 2r 8 j2:82 10 . (c) = 2r has increased
to = 2r 0 5:30 10 4 . As can be seen, the dielectric mismatch cases have a great
effect on the that forms in the PR medium. Used with permission of Opt. Engr.,
1995 [30].
390 Chapter 7
dielectric modulation does not approach a ®nal steady state but oscillates
in the diagonal ripple pattern shown in Fig. 8b. The temporal pattern
of Fig. 8b, as mentioned earlier is a result of the optical standing wave
pattern slowly changing the PR medium, thus further causing a time
change in the PR medium.
Figure 8c p shows
0 the same case as in Fig. 8b (this isp a mismatched
0 case)
4
except that = 2r has been
p0 increased to a value = 2r 5:30 10 .
Because of the higher = 2r ratio (stronger PR medium), the diffracted
waves more strongly in¯uence the standing wave pattern of the PR medium
and thus more severely change the PR medium. As can be seen from Fig. 8c,
the diffracted waves build up to such a point that almost chaotic behavior
occurs as time proceeds [17]. A simulation was run for the same parameters
00
as given in Fig. 8 except that 2r 0 (no absorption). The resulting was
nearly identical to that seen in Fig. 8c. For this case conservation of power
was observed numerically almost exactly.
Figure 9 shows the PR medium as a function of x and y that results
in the matched p0case
at the time step 200 (last time step
p of Figs. 7 4and 8)
4 0
when (1) = 2r
p0 3:18 10 (Fig. 9a), (2) = 2r 3:88 10 (Fig.
p0
9b), and (3) = 2r 4:59 10 4 (Fig. 9c). The three values
p0 of = 2r
used in Fig. 9a, b, and c correspond to values of = 2r used in Fig. 7
(cases a, b, and c respectively). As can be seen from these plots, the
dielectric modulation is highly inhomogeneous longitudinally, with a
peak maximum and minimum value occurring in the PR medium. The
position
p0 of the peak maximum and minimum values is dependent on the
= 2r value used
p and tends to occur closer to the incident side aspthe 0
0
value of = 2r increases. This occurs because the larger value of = 2r
causes the weak signal to be ampli®ed over a shorter distance in the PR
medium, thus causing the peaks in a maximum and minimum
p to form
0
more quickly. Notice that the higher values of = 2r caused a much
more complete transfer of optical energy as can be seen in Fig. 7a
(cases a, b, c). Note that the diffraction numerical calculation can only
be made correctly using a longitudinally cascaded diffraction analysis algo-
rithm. The plots shown in Fig. 9 are steady state plots. This was veri®ed
by noting that there was no change in the plots shown with plots recorded
at time step 100.
In all our numerical simulations for the TE case, thirty layers were
used to model the inhomogeneity of the PR medium. Because each indivi-
dual layer consisted of many wavelengths (L 1:5 cm/30, 0:5 m), it
was necessary when calculating the intensity I
x; y to average this over a
number of y points distributed over a one-wavelength interval in the center
of each layer. If this were not performed, random widely varying samples of
the standing wave pattern would be obtained, leading to samples that would
Induced Photorefractive Gratings 391
1 @ND 1 ND 1
1:56 102 s 1
ND @t ND
I0
I0
For the ®rst term of the RHS of Eq. 7.3.5 we have
sI ND =ND
5 7 2 1 4 1
s
IC ND =NA
10
1:62 10
10 s 1:62 10 s . The ratio of
Induced Photorefractive Gratings 393
the second term to the ®rst term of the RHS of Eq. 7.3.9 is 0:96 10 2 . We
see that the second term is roughly a hundred times smaller than the ®rst,
which justi®es the approximation used.
The numerical simulation for this case was performed by using Eq.
7.3.14 to calculate at every time step and using RCWA to calculate the
grating diffraction at every time step. The RCWA calculation was per-
formed using NL 160 layers, which gave each layer a length of
L 9:375. The optical power intensity Ipoint
x; y; t /
Ex Ex Ey Ey
(Ex and Ey are the optical electric ®elds) was calculated at 10 equally
spaced points over the layer length L L=NL 9:375 (Ipoint
x; y; t
was sampled every 0:9375). The values of Ipoint
x; y; t were then aver-
aged. This value was used as the average optical power intensity in Eq.
7.3.18.
Averaging the intensity over the layer length L represents an
important part of the interaction of the incident optical light with the
PR medium. Physically averaging the optical intensity over L repre-
sents the way that partially coherent light or light with a ®nite fre-
quency spectrum (or both) would interact with the PR medium. When
perfectly coherent, monochromatic light enters the PR slab, the energy is
multiply re¯ected at the interfaces, and a strong standing wave pattern
is formed in the PR slab. The distance from peak to peak longitudinally
(y-direction) is a few free space wavelengths . When the frequency of
the light is changed by a small amount, the peaks and nulls change
position by a few wavelength, and thus in the crystal a standing wave
pattern is formed whose peaks are in an entirely different position from
the original monochromatic wave from which the frequency was chan-
ged. When many different frequency waves are added together, the
peaks and nulls will tend to average out, and a longitudinally average
®eld will result. Thus spatial averaging simulates the frequency spread of
real optical energy.
Figure 10a shows the dielectric modulation function
x; y; t that
results for the BaTiO3 simulation at time step 100. At this time (see Figs.
10b and 10c) the photorefractive crystal is nearly in a steady state. The
minimum and maximum dielectric modulations that occur are min
0:00605 and max 0:00565. Using the relation r42 n2O n2E Excs , this
min
corresponds to a minimum and maximum space charge electric ®eld of Exsc
4 4
11:1 10 V/m and 9:26 10 V/m. These ®gures are roughly in line (to
a factor of 2) with typical values given in [8, Fig. 3.3b, p. 91]. The simulation
shown in Fig. 10a predicts an interesting feature of the dielectric modula-
tion: two different dielectric slanted gratings have formed as a result of the
photorefractive±optical interaction. The ®rst grating extends from about
394 Chapter 7
Figure 10 The dielectric modulation and diffracted power that results when TM
optical energy illuminates BaTiO3 are shown. I inc 1:79 107 W/m2 ,
I0 6:38
ms, t 0:2
I0 1:27 ms, t 100t, and 0:633 m. Used with permission of
Opt. Engr., 1995 [30].
zero to 300 and has a grating period of 1 120 with grating vector
given by
2
K 1 sin 1 a^ x cos 1 a^ y 1 1:5
7:3:22
1
in the geometry of Fig. 1. The second grating extends from about y 300
to 1500. This grating is more clearly de®ned and has a grating period of
about 2 75.
2
K 2 sin 2 a^ x cos 2 a^ y 2 4:5
7:3:23
2
Induced Photorefractive Gratings 395
Figures 10b and 10c show, respectively, the power transmitted and
re¯ected in different diffraction directions as a function of time. As can be
seen from the power transmission curves of Fig. 10b, the BaTiO3 PR
crystal at the ®rst time step is uniform, and power is transmitted equally
in the T0 and T1 directions. Within a few time steps, a modulation grating
forms and power is rapidly depleted from the T1 direction and transferred
to the T0 direction. As can be seen from Fig. 10b, the T0 modal direction
is completely depleted of power. As time progresses over a period of
approximately 30 time steps, diffracted waves in the T 1 direction build
up, and ®nally a strong mode conversion from the T0 to the T 1 order
occurs and the grating goes into a quasi±steady-state form. The T0 mode
drops to about 10% of the total diffracted power, and the T 1 modal
power builds from 0 to 60%. The ®gures are similar to those in Ref. 26,
which also illustrates the growth of a higher order. The re¯ected power in
the R0 , R1 , and R 1 directions follows a similar time history as did the
transmitted except that the R1 order is not depleted to a zero value by
drops from 18% to above 10% as time increases. In contrast to the results
in Refs. 26 and 28, we predict that the 1 order in the transmitted case
virtually decays to zero. The reason may be that in the unmatched case (a
true experimental possibility), the re¯ected orders also interact. This exam-
ple can also be viewed as a case of six-wave coupling in PR materials, such
as in KNbO3 , which leads to phase conjugation in a two-beam coupling
geometry [33].
The BaTiO3 simulation was performed for four other PR crystal
lengths, which were LiL 1500 iL 18:75, iL 1; 2; 3; 4. An interesting
result of the analysis was that for the lengths of 1518:75
iL 1 and
1556:25
iL 3 simulation showed that diffraction in the T 1 and R1
directions did not occur at all where they did occur for the values of iL
2 and iL 4. Evidently the growth of the T 1 and R1 mode perturbations
depends on a resonant length of the crystal.
@ND =@t is small compared to the source term in Eq. 7.2.2, which depends on
the optical intensity. The effect of this approximation is that the full tem-
poral behavior of the system is not included in the analysis. An important
impact that the above approximations have, over and above the one already
mentioned, is that indirectly both approximations lead to the omission of
second-order transverse spatial derivative terms in the analysis. This is
important for a few reasons. First, any analysis for which a higher order
derivative is ignored is limited to cases where changes in the electric ®eld, the
electron density, etc. in the transverse direction are small. Thus, for example,
if higher order diffraction (which varies rapidly in the transverse direction)
should be excited by the PR system, the effect of this cannot be in general
studied for arbitrary hologram wave numbers because the second-order
derivative terms may contribute more to the analysis than the ®rst-order
derivatives. Thus the analysis could at best be valid only over a range of
hologram wave numbers [29,30]. The purpose of this section is to study an
exact nonlinear time-dependent solution of the Kukhtarev equations in
conjunction with RCWA.
To start the analysis, we substitute Eq. 7.2.3 into Eq. 7.2.1 and ®nd
that
@n @ND @nEs @2 n
Ds 2
7:4:1
@t @t @x @x
s @Es
ND NA n
7:4:2
e @x
or
@ND @2 Es @n
s
7:4:3
@t e @x@t @t
where @NA =@t 0 has been used in Eq. 7.4.3. If the @ND =@t of Eq. 7.4.3 is
substituted into Eq. 7.4.1 and @n=@t is cancelled on the right- and left-hand
sides of the resulting equation, we ®nd that
s @2 Es @
nEs @2
Ds n2 0
7:4:4
e @x@t @x @x
@Es e e @n
nEs Ds
7:4:5
@t s s @x
In Eq. 7.4.5 we have set the integration constant to zero, assuming all
dependent variables and their derivatives tend to zero as jxj ! 1. For a
particular choice of c-axis of the PR crystal, the dielectric permittivity mod-
ulation is related to the electrostatic electric ®eld by
@ e Ds e @n
n r42 n20 n2e
7:4:7
@t s s @x
Up to this point no use has been made of Eq. 7.4.2, the electron rate
production equation. If @ND
=@t of Eq. 7.4.3 is substituted into Eq. 7.4.2, we
®nd that
s @2 Es @n
sI
ND ND
R nND
7:4:8
e @x@t @t
If Eq. 7.4.4 is used to eliminate the
s =e @2 Es =@x@t and Nd of Eq. 7.4.2 is
substituted, we ®nd that
nE @2 n @n s @Es
s Ds 2
sI ND NA n
@x @x @t e @x
s @Es
R n NA n
7:4:9
e @x
where
@ @n~
~
1 n 2
7:4:12
@t~ @x~
from in the PR slab by itself, and (3) by adding these two EM ®elds
together to ®nd the overall EM ®eld in the PR medium and thus ®nd the
overall transmitted and re¯ected EM ®elds in free space.
The algorithm proceeds as follows. The time derivative @=@t~ is
approximated as
~ y; t~ t~
@
x; ~ y; t~
x;
~
@t t~
and after substitution in Eq. 7.4.12 and cross multiplication by t~ we ®nd
that
@n~
x; ~ ~
~ y; t t
x; ~
~ y; t ~
1 n 2 @t~
7:4:13
@x~
P
p 1
p
I
p @
H
p 1
3 4
1 ~
n
p
3
C 1 4
@x~ x
~ x~ p
~
@n
p I
p ND
S
p 3 1 1
7:4:15
@t C NA
Induced Photorefractive Gratings 401
we ®nd for p 1; . . . ; Np
~ 1
n
p ~ n
p
2n
p ~ 1 ~ 1 n
p
n
p ~ 1
2 2
P
p
~
x ~
2x
~ S
p
H
pn
p
7:4:16
~ 1, n
p,
If coef®cients of n
p ~ ~
n
p 1 are collected, Eq. 7.4.16 can be put
in the following form
~ 1 R
pn
p
R
pn
p ~ R
pn
p
~ 1 S
p
7:4:17
where
P
p 2 2 P
p
R
p 2 R
p H
p R
p 2
2x~ x~ 2 x~ 2 2x~ x~ 2
and where p 1; 2; . . . ; Np .
Because the diffraction grating is periodic, we have the important
boundary conditions on the variable that n
0 ~ n
N~ p and
n
N ~
~ p 1 n
1. These equations can be used to eliminate the variables
~
n
0 ~ p 1 in Eq. 7.4.16 and thus give an equation that depends
and n
N
only on the unknowns n
p, ~ p 1; . . . ; Np . We thus observe that Eq. 7.4.17
represents a system of Np equations in Np unknowns n
p, ~ p 1; . . . ; Np .
The system of ®nite difference equations given by Eq. 7.4.17 can be
conveniently expressed in terms of a matrix equation:
2 3
R
1 R
1 0 0 0 0 0 0 0 R
1
6 7
6 R
2 R
2 R
2 0 0 0 0 0 0 0 7
6 7
6 7
6 0 R
3 R
3 R
3 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 7
6 7
L6 7
6 0 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 R
Np 2 R
Np 2 R
Np 2 0 7
6 7
6 7
6 0 0 0 0 0 0 0 R
Np 1 R
Np 1 R
Np 1 7
4 5
R
Np 0 0 0 0 0 0 0 R
Np R
Np
T T
S S
1 S
2 S
Np n n
1 n
2 n
Np
L n~ S 7:4:18
n~ L 1 S
7:4:19
402 Chapter 7
which thus determines the electron density pro®le for a given value of y in
the PR medium.
The algorithm proceeds as follows. Once n
~ y; t~ is determined [n
p
~ x; ~ is
determined from matrix inversion and n
p ~ speci®es n
~ y; t~] and its deri-
~ x;
~ y; t~=@x~ is calculated, n
~ x;
vative @n
~ y; t~ and @n
~ x; ~ y; t~=@x~ are substituted
~ x;
back into Eq. 7.4.13. Once a new value of
x; ~ y; t~ t~ is found from Eq.
7.4.13, this new value of
x; ~ y; t~ t~ is calculated for all values of y (all
discrete layers of the PR slab). Once this step is completed, RCWA is used
to study diffraction from the new value of
x; ~ y; t~ t~, and thus a new
optical intensity value I
x; ~ y; t~ t is found. The new intensity I
xy; ~ t~ t~,
along with
x; ~ y; t~ t~ and its x derivative, is substituted into Eqs. 7.4.17
and 18 and a new value of n
~ x;~ y; t~ t~ is found. By repeating the above
steps for many iterations, the time evolution of the PR material and the
optical diffracted intensity can be found.
Figure 11 (a) The optical power intensity (normalized to the dark current C) in the
grating when the grating period is 1, 2, 5, and 10 as a function of the
normalized grating distance xN x=. The incident power (evaluated at
y L=2) was adjusted in order that the intensity pro®le for each different size
grating period would have the same peak intensity. (b) The steady-state dielectric
modulation function (also evaluated at y L=2) that results when the intensity
pro®les of (a) were used to determine . Because the PR grating was so thin, the
intensity pro®les of (a) were not assumed to change with time as the pro®les
reached a steady state. All grating parameters used in the simulation not listed on the
®gure are given in Section 7.3. Used with permission of JOSA-A, 1996 [34].
404 Chapter 7
pro®le peak is shifted to the left as the grating period becomes smaller. This
is an effect of the anisotropy of the PR crystal. The incident amplitudes of
the interfering waves have been chosen to keep the peak intensity the same
for the different grating periods shown in Fig. 11a. The dielectric modula-
tion function that results from the intensity pro®les of Fig. 12a is shown
in Fig. 11b. In Fig. 11b, the curve marked ``1'' corresponds to a grating
period 1, the curve marked ``2'' corresponds to a grating period
2, and so on. The of each of the curves has been plotted as a
function of the normalized variable xN x= as was the power intensity.
As can be seen in Fig. 11b, the size of the grating period as a large
effect on the value of that results. When the grating period is small (on
the order of 1) the magnitude of is small. As the grating period
increases in size, the maximum magnitude value of the pro®le increases
in size until the grating period is about 5. During this range of
1 5, the pro®le gradually changes shape with the peak
(maximum and minimum) rising more sharply as the grating period
increases. When is increased to the range 5 10, the maximum
magnitude value of the pro®le decreases in size. The pro®le further
gradually changes shape. The increase and then decrease of the maximum
magnitude value of the pro®le results because the equilibrium equation
for n and , namely Eq. 7.4.10, contains zero-, ®rst-, and second-order x
derivatives. Thus when is small
K 2= is large) the second-order x
derivatives terms are large, and when is large the zero-order derivative
term tends to be large. Intermediate to this ( 4 to 5), the largest
magnitude pro®les are reminiscent of the dependence of the linearized
two-beam coupling coef®cient on . We note that for all values of the
grating size the function is shifted to the right of the intensity pro®le,
as can be seen by inspecting Figs. 12a and 12b.
Figures 12a±d display the numerical PR mode diffraction, coupling,
and conversion that occurs (using the Kukhtarev equation and RCWA)
when two interfering plane waves (see Fig. 12a) whose amplitudes are E0
and E1 0:1E0 are incident on an index matched PR crystal of length
L 1530, 0:633 m. The angle of incidence is such as to make the
grating period 5. Figure 12c shows the normalized power transmitted
in the T0 and T1 directions as a function of time step, and Fig. 12d shows the
normalized power re¯ected in the R0 and R1 directions as a function of time
step (t 1 ms). As can be seen from Fig. 12d, because of the index match-
ing, almost zero power ( 10 6 ) is re¯ected from the PR grating in the R0
and R1 directions. Figure 12c shows, for the geometry and material para-
meters of the case under consideration, that a large amount of energy is
transferred from the T0 to the T1 in a period of about 35 ms, at which time
the grating dynamics rapidly approaches the steady state. The power in the
Induced Photorefractive Gratings 405
Figure 12 The dielectric perturbation function (b) and the power transmitted in
the T1 , T0 , (c) R1 , and (d) R0 directions is shown when the regions (Region 1 and 3)
bounding the PR crystal are indexed matched to the PR crystal. All grating para-
meters used in the simulation not listed on the ®gure are given in Section 7.3. Used
with permission of JOSA-A, 1996 [34].
T0 and T1 directions adds to about 90% of the incident power. Because the
grating is assumed to be lossy, the other 10% of the incident optical power is
absorbed as heat in the grating. Figure 12b shows the dielectric modulation
function that results at t 113 ms when the grating has been in a steady
state for a long time. The pro®le has the form of a slanted sinusoidal grating
that grows steadily from a small value at y 0 to a peak-to-peak value of
max min 0:00642, which occurs at about y 1000. It may be
noticed that the grating pro®le is slightly skewed at y 1000. This may
be a slight nonlinearity effect.
The simulation grid used 34 x divisions to solve Kukhtarev's equa-
tions, Eq. 7.4.10, for each layer (thus the matrix equation that was inverted
in Eq. 7.4.18 was 34 34 size), and the grid used NL 160 layers to
describe and simulate the optical wave (RCWA was used to determine the
electromagnetic or optical ®elds of the system) as it propagated and dif-
fracted through the diffraction grating. The optical intensity was averaged
406 Chapter 7
make a very large change in the power transmitted and re¯ected in the
different directions from the crystal. For example in Fig. 13b, the power
transmitted in the T1 direction for the L4 length is small (about 12% in
steady state), whereas when the length is increased to L6 the transmitted
power jumps to a large value (about 50%). A length change of only about
2 L 19 has occurred. A second interesting feature of the plots is that
depending on the length L, the power transmitted (T0 and T1 directions) or
re¯ected (R0 and R1 directions) may go into an oscillatory steady state or a
nonoscillatory steady state. In Fig. 13b, it is observed that the L1 and L5
lengths form oscillatory steady states, whereas the lengths L2 , L3 , L4 and L6
408 Chapter 7
Figure 14 The dielectric perturbation function that results in the index mis-
matched case of Fig. 13 when the PR crystal length is L5 1530 and (a) t 56 ms,
(b) t 90 ms (oscillatory steady state, see Fig. 13, L5 1530). (c) The dielectric
perturbation function that results in the index mismatched case of Fig. 14 when
the PR crystal length is L3 1510:875 and t 90 ms (nonoscillatory steady state,
see Fig. 14, L3 1510:875). Parts (b±d) are drawn to the same scale. (d) The RMS
[
rms 2 is also proportional to the electrostatic energy in a stored a grating
period] as a function of time step when the crystal length is L3 1510:875 and when
the crystal length is L5 1530. Used with permission of JOSA-A, 1996 [34].
form nonoscillatory steady states. It is also interesting to note that the
oscillatory steady state periods depend on the length L. For example the
period of the length L1 is about 50 ms, whereas the length L5 is about 65 ms.
We would like to point out that the results of Fig. 13 did not change
when the number of layers was changed from 160 to 320. The starred line of
length L5 in Fig. 14a was calculated using 320 layers (the power intensity
was averaged over ten points for each of the 320 longitudinal divisions), and
all other plots were made using 160 layers (the power intensity was averaged
over ten points for each of the 160 longitudinal divisions). As can be seen
from Fig. 13a, there was no difference in the numerical results obtained.
Figures 14a and 14b show, respectively, the dielectric modulation
function that results when L L5 1530 (the grating system for this
Induced Photorefractive Gratings 409
Figure 15 Numerical PR mode coupling and diffraction that occur (with the use
of the Kukhtarev equations and RCWT) when two interfering plane waves whose
amplitudes are E0 and E1 0:4E0 are incident on a PR crystal ( 0:633 m,
L 1530 L5 ) that is not index matched to the surrounding medium (free space
is assumed to surround the PR crystal, 1 3 1). (a) The dielectric modulation
function that results when NL 160 layers is used. (b) The results when NL
640 layers is used. (c) The transmitted and re¯ected power diffracted in the zero
and ®rst orders when NL 640. (d) The power transmitted in the second order
when NL 160, 320, and 640 layers. The dashed line shown in (d) shows the
transmitted power that is diffracted in the second order when NL 160 layers
and MT 3. Used with permission of JOSA-A, 1996 [34].
2
I2 sin2
7:4:20
Q
where Q K 2 L=k0 and k0
npp Leff =4. In the above expression K is the
hologram wave number, Leff is the effective grating thickness, and
npp
denotes the peak-to-peak change in induced refractive index.
In our case, with the hologram spacing 5 and
Leff L=2 1530=2, 0:633 m and Q 200. Also from the plots
(Fig. 16b) the peak value of
npp 0:00134 (since (npp
pp =
2n0 , pp max min 0:00654 (see Fig. 16b), n0 2:437)
at t 200 ms implying 1:6. Note that this approximately corresponds
to the condition for maximum diffracted power in the second order.
Equation 7.4.20 shows that as Q increases, a larger value of is required
for enhanced power transfer to the second order.
A numerical simulation was also performed for the mismatched case
(see Fig. 12a, 1 3 1) when the incident amplitudes were E0 and E1
E0 for the cases when L 1500, In these cases very interesting results
occurred. In the case when L 1500; the power in the T0 order was
diffracted and the mode converted into the T1 order, and no other appreci-
able diffraction occurred. The diffraction in the T0 and T1 directions was
observed to be in a nonoscillatory steady state as time increased. For
L 1500, the peak-to-peak
pp max min 3:86 10 3 which
made (npp 0:79 10 3 . Further, it was observed that the peak-to-peak
dielectric modulation function (pp decreased in value, nearly linearly,
from the incidence side to the transmission side, assuming a very small
value at the transmission side. Thus the effective length of the grating was
about Leff L=2 1500=2. In the case when L 1530, power was initi-
ally diffracted from the T0 order to the T1 order and then subsequently at
412 Chapter 7
about t 75mg was diffracted from the T1 order to the T2 second order.
For this case, the peak-to-peak dielectric modulation function
(pp max min 5:33 10 3 , which made
npp 1:09 10 3 .
From this simulation we ®nd the interesting results that the index mis-
matched PR crystal for certain lengths appears to be resonant in the sense
that for certain lengths (in this case L 1530), the optical energy in the PR
crystal can interact with the crystal in such a way that the dielectric mod-
ulation function can build up to relatively large values in the crystal.
Figure 16a shows plots of the normalized electron density n~ n=NA
that result from the simulation shown in the matched case of Fig. 13
(t 113 ms, y L=2, L 1530, 0:633 m, 5) and the mis-
matched case of Figs. 14 and 15 (t 200 ms, y L2, L L5 1530,
0:633 m, 5). As can be seen from these plots, we ®rst notice that
the results of the simulation show that the normalized electron density n~
n=NA assumes a very small value of n~ 10 5 or less. Secondly we note that
the presence of matched or mismatched boundaries makes a signi®cant
difference as to where the peaks of the electron density occur and the
peak-to-peak size of the electron density. Figure 17b shows a plot of the
dielectric modulation function obtained at the same location where the
electron densities of Fig. 17a were obtained. As we can see from these plots,
the presence of matched or mismatched boundaries makes a signi®cant
difference as to position of the peaks and the peak-to-peak amplitude of
the pro®les. We ®nally note that a comparison of Figs. 17a and 17b
shows that in both the matched and the mismatched cases the electron
density is always displaced from, and out of phase with, the pro®le.
This is a result of Eq. 7.4.12 and the fact that n~ and are related by a
®rst space and time derivative equation.
Figure 16 (a) Plot of the normalized electron density n~ n=NA that results from
the simulation shown in the matched case (t 113 ms, y L=2, L 1530,
0:633 m, 5) and the mismatched case (t 200 ms, y L=2,
L L5 1530, 0:633 m, 5). (b) Plot of the dielectric modulation func-
tion obtained in the same location as that of the electron densities of (a). Used
with permission of JOSA-A, 1996 [34].
414 Chapter 7
X
S~ Sxi a^ x Syi a^ y exp
jkxi x ji y
i
X
U~ Uzi a^z exp
jkxi x ji y
7:5:1
i
where S E, 0 377
, kxi p
U 0 H; ~ sin
',
1 sin I i
=
i i
=~ cos
', and 0 . The angle ' [39±41] is the tilt angle of
the DG with respect to the planar interfaces and can be taken to be zero
for a pure RG analysis [41]. In terms of S and U,
Maxwell's equations are
2 3
xx yy 0
r S~ j U~ r U~ jS~ where 4 ys yy 0 5
0 0 zz
7:5:2
@V a a12
11 V
7:5:3
@y a21 a22
@2 n @n I
y @
2 1 3 1 1 4 3 4 1 n
@y2 @y C @y
@n I
y ND I
y @
3 1 1 1
7:5:4
@t C NA C @y
@ @n
1 n 2
7:5:5
@t @y
In these equations I
y is the optical intensity W/m2 and n20 n2e r42 Es
y
is a normalized dielectric modulation function linearly related to the long-
itudinal electrostatic ®eld Es
y (V/m). All other parameters have been
de®ned in the previous sections.
Because a re¯ection grating is being studied, it is useful to expand the
optical ®eld I
y and the material variables n
y and
y in a spatial
Fourier series (the period of the Fourier series is the grating wavelength
~ where the Fourier amplitudes are all spatially varying functions
k0 )
of the longitudinal coordinate y. We have
Xc
iM
I
y Ii
y exp
jiy
7:5:6
i Mc
Xc
iM
n
y ni
y exp
jiy
7:5:7
i Mc
Xc
iM
y i
y exp
jiy
7:5:8
i Mc
@2 ni
y @n
y
2
2ji i 2
i2 ni
y Hi
y i M C ; . . . ; MC
@y @y
7:5:9
where
Induced Photorefractive Gratings 417
1=2
3
1 4
2
and
X
Hi Ti i 0 ni 0 Fi
7:5:10
i0
and where
1 3 Ii 1 4 @i 3 4
Ti ji i jii ni
2 2 C 2 @y 2
7:5:11
X
1 @n 0 @ni I ND
Fi i i 0 i 3 i;0 i 1
2 i0 @y 2 @t C NA
1 X @i i 0 0 Ii 0
j
i i i i 0 i 0 ;0
7:5:12
2 i0 @y C
ni 0 ni L 0 i M C ; . . . ; MC 7:5:13
To proceed further we now for the moment regard the RHS of Eq. 7.5.12 as
a known function y. Equation 7.5.12 along with its boundary conditions, for
each i, is classi®ed as a linear second-order nonhomogeneous differential
equation. The solution to this type of equation is well known and can be
found by using a Green's function approach: (1) setting the RHS of Eq.
7.5.12 to a Dirac delta function
y y 0 ; (2) solving the resulting differen-
tial equation
0 0
where y and y represent locations an in®nitesimal to the left and right of
y 0 ; (3) superposing the Green's function solutions times the nonhomoge-
neous RHS Hi to ®nd the overall response of the system. Regarding Hi
y
as a known function, the solution for ni
y is given by
0
ni
y gi
yjy 0 Hi
y 0 dy 0
7:5:16
L
Although Eq. 7.5.16 is an exact integral for the differential equation Eq.
7.5.12, it is an unnecessarily complicated one for the current analysis. For
typical PR parameters used in this chapter, the constant is on the same
order of magnitude as 2=. Investigation of the Green's function
gi
yjy 0 for the just-described values of and shows that this Green's
function has a signi®cant nonzero value only within a few of the point
y y 0 in the interval. Investigation has further shown that most of the
exponential terms are exponentially small. After analysis it is found that gi
yjy 0 is well approximated by [37]
0 1 exps1
y y 0 L y < y0
gi
yjy
7:5:17
s1 s2 exps2
y y 0 0
y <y0
where
Induced Photorefractive Gratings 419
1
Gi
7:5:19
2
i2
Substituting Eq. 7.5.10 into Eq. 7.5.18 and collecting terms on ni
y, we
derive a matrix equation for the normalized electron density ni
y:
X X
i;i 0 Gi Ti i0 ni 0
y Li;i 0 ni 0
y Gi Fi
7:5:20
i0 i0
X
@i @ni
1 ni i0 i 0 2 jin i
7:5:21
@t i0
@y
If Eq. 7.5.20 is substituted for ni
y, and matrix terms common to i are
collected in Eq. 7.5.21, it can be placed in the usual state variable form:
@i X
Ai;i 0 i 0 fi
7:5:22
@t i0
where Ai;i 0 and fi are de®ned in Ref. 37, where the complete temporal
numerical solution is detailed.
the z 0 -axis. The optical and material analysis was carried out using values of
MT 5 and MC 10. Also L~ 1000 ~ 0:132 mm, =k ~ 0 4:795,
4:858, and I 6 .
In Ref. 37, we have shown the plots of the dielectric modulation
function
y; t [proportional to the electrostatic space charge Es
y; t] as
calculated over a range of two grating wavelengths ~
~ in meters, k0 ,
~ ~
from a position 123 to 125 from the incident side of the grating at the
times of t~ 0:059, 0.477, 4.97, 30.1, and 69.0 ms. Also plotted was the
re¯ected diffraction ef®ciency DER
% as a function of time for the simula-
tion under consideration. We have also shown plots of the terms of the
electron density balance equation Eq. 7.5.20 at a given time t~ 30:1 ms.
These plots were all done only for a tuned layer and only for a thickness of
1000. Also in Ref. 37, no study was done of the convergence of the
numerical solution with different spatial harmonics MC (number of harmo-
nics in the material equations) and Mt (the number of harmonics used in
RCWA analysis).
We will now present additional numerical simulations for the diffrac-
tion ef®ciency and the various terms of the electron density balance equation
for different lengths of the photorefractive material, and for tuned and
detuned cases, respectively. In Fig. 18, we study the convergence of the
solution (diffraction ef®ciency) for a tuned layer of thickness 3000 by
varying the values of MC and Mt from MC 1 and Mt 1 to MC 14
Figure 18 Transient re¯ected diffraction ef®ciency for different optical and mate-
rial harmonics. Used with permission of JOSA-B, 1998 [37].
Induced Photorefractive Gratings 421
Figure 19 Transient re¯ected diffraction ef®ciency for different tuned layer thick-
nesses. Used with permission of JOSA-B, 1998 [37].
422 Chapter 7
Figures 21a and 21b show plots of several terms that make up the
electron density matrix equation as given by Eq. 7.5.20 (the terms making
up Eq. 7.5.20 are speci®ed in Eqs. 7.5.9±7.5.12) when a tuned layer re¯ection
grating is formed (Fig. 21a, L 0) and when a nontuned layer re¯ection
grating is formed (Fig. 21b, L 0:5). These ®gures were formed by
Fourier summing the slowly varying amplitudes according to Eqs. 7.5.6±
7.5.8 and then plotting the sums as a function of yloc as shown in Figs. 21a
and 21b. As can be seen when comparing them, a nontuned layer re¯ection
grating causes a signi®cantly different electron density pro®le from when a
tuned layer re¯ection grating is present. Figure 22 shows plots of the dielec-
tric modulation function (proportional to the electrostatic ®eld inside the
re¯ection grating) when a tuned layer re¯ection grating is formed (same case
as Fig. 21a, L 0) and when a nontuned layer re¯ection grating is
formed (same case as Fig. 21b, L 0:5). As can be seen from Fig. 21,
detuning of the layer causes a shift in the dielectric modulation function .
7.5.4 Conclusion
In this section, we have, after appropriate reformulation of the multilayer
RCWA, analyzed the time evolution of re¯ection gratings in a photorefrac-
tive material of an arbitrary thickness. This has been accomplished by let-
ting all but the last layer be an integral number of wavelengths and then
Induced Photorefractive Gratings 423
Figure 21 Spatial distribution of different terms in Eq. 7.5.20 for a tuned layer (a)
and for a detuned layer (b). Used with permission of JOSA-B, 1998 [37].
letting the last layer have an arbitrary length. The re¯ected diffraction ef®-
ciency has been plotted as a function of time for various layer thicknesses,
viz., from 100 to 6000. The evolution of the diffraction ef®ciency has
also been shown for various detunings L ranging from 0 to 0:5.
Numerical convergence was also analyzed. We have found that a minimum
number of spatial harmonics (viz., MC 6 and Mt 3) should be used for
reasonable accuracy. Examples of the spatial distribution of various physi-
cal quantities such as the electron density, induced permittivity etc., have
been provided.
424 Chapter 7
Figure 22 Typical induced permittivity distributions for a tuned and detuned layer.
Used with permission JOSA-B, 1998 [37].
PROBLEMS
REFERENCES
429
430 Index