by prof. thomas pertsch. - uni-jena.de · by prof. thomas pertsch. the course material in 2019...
TRANSCRIPT
This is the 2018 version of the
Computational Photonics script, written
by Prof. Thomas Pertsch.
The course material in 2019 derives
largely from this source, however it may
deviate in scope, notation etc.
You can use this script as an additional
referential resource for the lecture.
Dr. Thomas Kaiser
May 2019
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 2
Computational Photonics Prof. Thomas Pertsch
Abbe School of Photonics Friedrich-Schiller-Universität Jena
0. Introduction and Motivation ..................................................................... 4
0.1 Why computational photonics? ......................................................................... 4
0.2 Maxwell’s equations .......................................................................................... 5 0.2.1 Maxwell’s equations in time domain ................................................................. 5 0.2.2 Maxwell’s equations in frequency domain......................................................... 6 0.3 Basic numerical operations ............................................................................... 6 0.3.1 Differentiation .................................................................................................... 6
0.3.2 Integration ......................................................................................................... 7
0.3.3 Root Finding & Minimization/ Maximization ...................................................... 8
0.3.4 Linear systems of equations ........................................................................... 11 0.3.5 Eigenvalue problems ...................................................................................... 12 0.3.6 Discrete (Fast) Fourier transform - FFT .......................................................... 12 0.3.7 Ordinary differential equations (ODEs) ........................................................... 13
1. Matrix method for stratified media ......................................................... 16
1.1 Wave equation in volumes of homogeneous media ....................................... 16 1.2 Optical layer systems ...................................................................................... 18
1.3 Derivation of the transfer matrix ...................................................................... 19 1.4 Reflection and transmission problem .............................................................. 21
1.5 Guided modes in layer systems ...................................................................... 26
2. Finite-difference method for waveguide modes ..................................... 29
2.1 Scalar approximation for weakly guiding waveguides with small index differences ...................................................................................................... 29
2.1.1 Stationary solutions of the scalar Helmholtz equation .................................... 29 2.1.2 Matrix notation of the eigenvalue equation ..................................................... 31 2.1.3 Boundary conditions ....................................................................................... 32 2.2 Full vectorial mode solver for waveguides with large index differences .......... 33
3. Beam Propagation Method (BPM) ......................................................... 41
3.1 Categorization of Partial Differential Equation (PDE) problems ...................... 41 3.2 Slowly Varying Envelope Approximation (SVEA) ............................................ 43 3.3 Differential equations of BPM ......................................................................... 44 3.4 Semi-vector BPM ............................................................................................ 46 3.5 Scalar BPM ..................................................................................................... 46 3.6 Crank-Nicolson method .................................................................................. 46
3.7 Alternating Direction Implicit (ADI) .................................................................. 47 3.8 Boundary condition ......................................................................................... 47
3.8.1 Absorbing Boundary Conditions (ABC) ........................................................... 47 3.8.2 Transparent Boundary Condition (TBC).......................................................... 48 3.8.3 Perfectly matched layer boundaries (PML) ..................................................... 50 3.9 Conformal mapping regions ............................................................................ 51 3.10 Wide-angle BPM based on Padé operators .................................................... 53
3.10.1 Fresnel approximation – Padé 0th order .......................................................... 54 3.10.2 Wide angle (WA) approximation – Padé (1,1) ................................................. 54
4. Finite Difference Time Domain Method (FDTD)..................................... 56
4.1 Maxwell’s equations ........................................................................................ 56 4.2 1D problems ................................................................................................... 57 4.2.1 Solution with finite difference method in the time domain for Ez ..................... 58
4.2.2 Yee grid in 1D and Leapfrog time steps .......................................................... 60
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 3
4.3 The full 3D problem ........................................................................................ 61 4.3.1 Yee grid in 3D ................................................................................................. 62
4.3.2 Physical interpretation..................................................................................... 63 4.3.3 Divergence-free nature of the Yee discretization ............................................ 64 4.3.4 Computational procedure ................................................................................ 65
4.4 Simplification to 2D problems ......................................................................... 66 4.5 Implementing light sources ............................................................................. 66 4.6 Relation between frequency and time domain ................................................ 67 4.7 Dispersive and nonlinear materials ................................................................. 69 4.8 Boundary conditions ....................................................................................... 71
4.8.1 Perfectly conducting material boundary .......................................................... 71 4.8.2 Floquet-Bloch periodic boundaries ................................................................. 72 4.8.3 Transparent boundary conditions - 2nd order Mur (TBC) ............................... 75 4.8.4 Perfectly matched layer boundaries (PML) ..................................................... 77
5. Fiber waveguides .................................................................................. 81
5.1.1 The general eigenvalue problem for scalar fields ........................................... 81 5.1.2 Properties of guided modes ............................................................................ 81 5.1.3 Cylinder symmetric waveguides ..................................................................... 83 5.1.4 Bessel’s differential equation .......................................................................... 84
5.1.5 Analytical solutions of Bessel’s differential equation ....................................... 85 5.1.6 Specifying the numerical problem ................................................................... 87
5.1.7 Solving the second order singularity ............................................................... 88 5.1.8 Numerical integration methods ....................................................................... 88 5.1.9 Eigenvalue search .......................................................................................... 91
5.1.10 Calculation examples ...................................................................................... 92
6. Fourier Modal Method for periodic systems ........................................... 95
6.1 Formulation of the problem in 2D for TE ......................................................... 96 6.2 Scalar theory for thin elements ....................................................................... 97
6.3 Rigorous grating solver ................................................................................. 100 6.3.1 Calculation of eigenmodes in Fourier space ................................................. 101
6.3.2 Explicit derivation for 2D problems ............................................................... 105
7. Finite Element Method (FEM) .............................................................. 112
7.1 The basic set up ........................................................................................... 112 7.2 Global vs. local labeling ................................................................................ 114 7.3 Weak formulation .......................................................................................... 114 7.4 Example Results ........................................................................................... 117
8. Outlook ................................................................................................ 120
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 4
0. Introduction and Motivation
0.1 Why computational photonics? it’s a numerical experiment
provides insides to inaccessible domain
permits to interpret and understand experimental results
simplifies the design of functional elements
explores prospective applications presently not realizable
with available large scale computational resources it became an inevitable tool in the world of micro- and nanooptics
What is the basis of our computational model?
„Light is like an odor an emanation of our body“ Epikur (Greek philosopher, 341-271 BC)
Straight „Light-Ray“ as an abstract imagination Euklid in „Elements“ (365 - ca. 300 BC)
Light is an electromagnetic wave J. C. Maxwell 1873 (Propagation and interaction of light with matter)
Light consists of particles (Photons) A. Einstein 1905 (Creation and absorption)
Light is particle and wave De Broglie 1923 (quantum mechanics)
Formulation of the problem:
For a specific geometry and a particular set of boundary conditions Maxwell‘s equations have to be solved (with or without approximations)
Description of the interaction of electromagnetic waves with matter
Initially published in 1873 by James Clark Maxwell
Experimental proof by Heinrich Rudolf Hertz 1884 (speed of radio waves corresponds to the speed of light)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 5
0.2 Maxwell’s equations
0.2.1 Maxwell’s equations in time domain
makr
ext
( , ) ( , )rot ( , ) , rot ( , ) ( , ) ,
div ( , ) ( , ), div ( , ) 0,
t tt t t
t t
t t t
B r D rE r H r j r
D r r B r
(1)
E(r,t) electric field [V m-1]
H(r,t) magnetic field [A m-1]
D(r,t) dielectric flux density [As m-2]
B(r,t) magnetic flux density [Vs m-2]
ext(r,t) external charge
density [As m-3]
jmakr(r,t) macroscopic current
density [A m-2]
(Usually there are no external charges or currents in optics.)
Matter equations in time domain
0
0
( , ) ( , ) ( , ),
( , ) ( , ) ( , )
t t t
t t t
D r E r P r
B r H r M r (2)
P(r,t) dielectric polarization [As m-2]
M(r,t) magnetic polarization (magnetization) [Vs m-2]
0 permittivity of vacuum 2 1 12
0 0 0( c ) 8.85 10 As/Vm
0 permeability of vacuum 12
0 4 10 Vs/Am
in linear, local, isotropic Media in optics
0
0
( , ) ( , ) ( , ) ,
( , ) 0
t R t t t t
t
P r r E r
M r
(3)
with ( , )R tr being the response function
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 6
0.2.2 Maxwell’s equations in frequency domain
Using Fourier transformation to transform into frequency space
1
( , ) ( , )exp( ) , ( , ) ( , )exp( )2
t i t d t i t dt
V r V r V r V r . (4)
Maxwell’s equations in frequency domain by FT it
rot ( , ) ( , ), rot ( , ) ( , ),
div ( , ) ( , ), div ( , ) 0.
i i
E r B r H r D r
D r r B r (5)
Matter equations in frequency domain
( , ) ( , )exp( )R t i t t
r r (6)
with ( , ) r being the material´s susceptibility, which is connected to the
dielectric constant ( , ) r by
( , ) 1 ( , ) r r )
0 0
0
( , ) ( , ) ( , ) ( , ) ( , ) ( , )
( , ) 0 ( , ) ( , )
P r r E r D r r E r
M r B r H r (7)
0.3 Basic numerical operations
0.3.1 Differentiation
Derived from the definition of differentiation, e.g. right-sided/forward difference equation
0
( ) ( ) ( )( ) lim
h
f x f x h f xf x
x h
for finite h
( ) ( )( ) [ ( )]h
f x h f xf x D f x
h
or left-sided/backward
( ) ( )[ ( )]h
f x f x hD f x
h
.
or central operator
( ) ( )[ ( )]
2h
f x h f x hD f x
h
higher order differentiation
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 7
2
2
( ) 2 ( ) ( )( ) [ ( )]h
f x h f x f x hf x D f x
h
0.3.2 Integration
11
0
.i
i
xb N
ia x
A dx f x dx f x
with 1[ , ]i i iI x x where
1i ix x h and ( ) /h b a N with 0 , Nx a x b
Decomposition of the full integration interval ,a b into N equivalent partial
intervals 1[ , ]i i iI x x .
Implementation example: rectangle rule
in each interval the mean value of the function is approximated by
1
2
( )2
i ii
hf x f f x
resulting in an integral approximation
11 1 1
1 1 1
0 0 02 2 2
i
i
xN N N
i i ii i ix
b aA dx f h f f
N
Rectangle rule for approximation of integrals.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 8
0.3.3 Root Finding & Minimization/ Maximization
Root finding of a single isolated root in one dimension.
Secant method for 1D
iterative solution by
11
1
( )( ) ( )
i ii i
i i
x xx x f x
f x f x
until
1 10i i if x f x f x
with determined by the desired accuracy
Illustration of the secant method (The individual points are numbered in the order of the iterations.)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 9
Secant method 1 function [x,i,status] = secant_method(x0,x1,tol1,tol2,no)
2
3 %x0,x1: start value
4 %tol1, tol2: tolerance
5 %no: maximum number iterations
6
7 %graphic output
8 ab = -0.4; x_old=1000;fx_old=1000;
9 axis on;
10 plot( [-5 5], [0 0]); % plotting x-axis
11 hold on;
12 set(findobj(gca,'Type','line','Color',[0 0 1]),'Color',
'black','LineWidth',1)
13 p=(-5):0.1:(5);
14 for i=1:1:(101)
15 f(i) = f1(p(i));
16 end
17 plot(p,f,'LineWidth',1);
18 i = 0; fa = f1(x0); fb = f1(x1);
19 if abs(fa) > abs(fb)
20 a = x0; b = x1;
21 else
22 a = x1; b = x0; tmp = fb; fb = fa; fa = tmp;
23 end
24 % iteration of the secant
25 while i < no
26 s=fb/fa; r=1-s;
27 t=s*(a-b); %ascent between two points oft he secant
28 x=b-t/r; % step towards root
29 fx=f1(x); % converging towards root
30 % output of points of the secant
31 if abs(x_old-x)<0.3 && abs(fx_old-fx)<0.3
32 ab=ab-0.2;
33 end
34 x_old=x; fx_old=fx;
35 plot(x,fx,'rO','LineWidth',1);
36 text(x+ab, fx,num2str(i+1),'HorizontalAlignment','left')
37 if t == 0
38 status = 'Method did not converge.'; return
39 end
40 if abs(fx) < tol1
41 status = 'Method converged and calculated solution.';
return
42 end
43 if abs(x-b) < tol2*(1+abs(x))
44 status = 'Method converged and calculated solution.';
return
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 10
45 end
46 i = i+1;
47 if abs(fx) > abs(fb)
48 a = x; fa = fx;
49 else
50 a = b; fa = fb; b = x; fb = fx;
51 end
52 end
53 status = 'Number iterations exceeded.';
Minima of higher-dimensional functions by minimization along alternating directions
Approach: solution of multiple one-dimensional minimizations in alternating directions.
Starting from point 1P into direction 1u
1 1,P u
and iteratively minimizing
1 1 1 2 1 1 1
2 2 2 3 2 2 2
1
min
min
min n n n n n n
f P u P P u
f P u P P u
f P u P P u
until no further improvement can be obtained along any direction.
Choosing directions along steepest descent (opposite to gradient)
1i i iiP P f P mit min i iit
f P t f P
Examples of easy and difficult converging surfaces.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 11
0.3.4 Linear systems of equations
System of algebraic equations
11 1 12 2 13 3 1 1
21 1 22 2 23 3 2 2
31 1 32 2 33 3 3 3
1 1 2 2 3 3
...
...
...
...
...
n n
n n
n n
m m m mn n m
a x a x a x a x b
a x a x a x a x b
a x a x a x a x b
a x a x a x a x b
in matrix representation
Ax b
with A being a coefficient matrix and b a column vector
11 12 1
21 22 2
1 2 3 4
...
...A
...
n
n
m m m m
a a a
a a a
a a a a
,
1
2
...
m
b
bb
b
.
alternative formulation of the problem ( ) 0f x with 1
n
i ij j i
j
f x a x b
and
1, , ; 1, ,j n i m
Types of problems:
Ax b
calculating the inverse matrix 1A with
-1ˆ ˆ ˆAA E equivalent to ˆ
j jAx b with 1,...,j N and 1jjb , and all other 0 , then
1
1ˆ ,..., NA x x
Important matrix properties:
1. n m same number of equations and unknowns
2. hermite matrix †ˆ ˆA A (komplex conjugate and transposed)
3. positiv definite ˆ 0vAv v
4. band matrix
5. sparse matrix most matrix coefficients are zero
always use addapted solution schemes
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 12
0.3.5 Eigenvalue problems
2
2ˆ ˆ ˆM with M E 0x x x
characteristic polynomial ˆ ˆdet M E ( ) 0P
to have solutions of
ˆ ˆ 0M E x which are not identical zero ( 0)x
0.3.6 Discrete (Fast) Fourier transform - FFT
Starting from a periodic function ( ) ( )f x L f x Periodic original space with set of discrete points with period L and step size a
0, ,2 , ,x a a L a
Periodic frequency space with set of discrete frequencies
2 2 20,1 ,2 , , 1LL L a L
k with L
Na
Definition of discrete Fourier transform (FT)
ikx
x
f k a e f x
with 2, 0La L
N k n n N , x n a
and inverse discrete Fourier transform (FT-1)
1 ikx
L
k
f x e f k
General property: 1( ) FT ( ) ( )f x f k FT f x
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 13
Illustration of decomposition of periodic function into harmonic functions.
Generalization for higher dimensions
with ( x x ; k k ), e.g. for 3 dimensions
3
3
1
ikx
x
ikx
k
f k a e f x
f x e f kL
0.3.7 Ordinary differential equations (ODEs)
General form of ordinary differential equation
( 1), ',..., ,
n
n
n
d f tG f f f t
dt
distinction of initial value problems and boundary value problems
Initial value problems (AWA)
initial values
10 , ' 0 ,..., 0
nf t f t f t
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 14
equivalent to system of coupled first order differential equations
11 1 2
22 1 2
33 1 2
1 2
( ), ,..., ,
( ), ,..., ,
( ), ,..., ,
( ), ,..., ,
n
n
n
nn n
df tG f f f t
dt
df tG f f f t
dt
df tG f f f t
dt
df tG f f f t
dt
Simple solution method
Forward Euler scheme
problem:
,f t G f t t
transformed into difference equation:
'( ) ,t
f t t f tf t D f t G f t t
t
with discretization t : :nf f n t
results in difference equation
1 ,n nn
f fG f t
t
and recursion formula for the solution of the ODE
1 ,n n nf f t G f t
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 15
Limitation: global error
h
Error
Discretization error
Round off error
Global
Limitation: instability
1
2 3
exakte Lösung
Euler Lösung
p
q
Instability scheme of harmonic oscillator equation.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 16
1. Matrix method for stratified media
1.1 Wave equation in volumes of homogeneous media From Maxwell's equation one can derive wave equations in inhomogeneous media for different fields.
Electric field:
2
0 0( , ) ( , ) ( , ) E r r E r (8)
Dielectric flux density (electric displacement):
2
0
0
1( , ) ( , )
( , )
D r D r
r (9)
Magnetic field:
2
0
0
1( , ) ( , )
( , )
H r H r
r (10)
For homogeneous media these wave equations simplify considerably:
( , ) ( ) r (11)
divergence condition
0 0( ) ( , ) ( ) ( , ) 0 E r E r (12)
curl curl grad div – Laplace
for, e.g. the electric field's wave equation:
0
( , ) ( , ) ( , ) ( , )
E r E r E r E r (13)
Helmholtz equation
2
2( , ) ( ) ( , ) 0
c
E r E r with 2
0 0
1c
(14)
To solve the Helmholtz equation one can choose between different ansatz functions, each having advantages for different investigated problems or coordinate systems. Here we are choosing the plane wave ansatz function:
0( , ) ( )exp(i ( ) ) E r E k r with 2
0 0
1c
(15)
The plane wave ansatz function has the following parameters
complex amplitude vector (polarization & phase):
0
0
0
0
( )
x
y
z
E
E
E
E (16)
wave vector (phase velocity & direction of
energy flow (for isotropic homogeneous media)): ( )
x
y
z
k
k
k
k (17)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 17
These six parameters can't be chosen completely independently. They rather have to fulfill the dispersion relation and must form a divergence-free field:
0 ( ) ( , ) 0 E r (18)
The ansatz must solve the Helmholtz equation which results in an eigenvalue problem, the solution of which determines the dispersion relation.
Inserting the plane wave ansatz into Helmholtz equation:
0( , ) ( )exp(i ( ) ) E r E k r
2
2( , ) ( ) ( , ) 0
c
E r E r (19)
with
0
0
0
( , ) exp( ( )) ( ) ( , )
x
y x y z
z
x E
E i k x k y k z iy
E
z
E r k E r (20)
and
2( , ) ( , ) ( ) ( , ) E r E r k E r (21)
Hence the dispersion relation of free space is
2
2
2( ) ( )
c
k (22)
Without loss of generality we define the z-axis as the axis of propagation.
2
2 2
2k ( ) ( ) k ( ) k ( )z x y
c
(23)
We are looking at the different types of solutions which fulfill the dispersion relation:
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 18
Propagating waves
for 2
2 2
2( ) k ( ) k ( )x y
c
22 2
2k ( ) ( ) k ( ) k ( )z x y
c
The sign of the wave vector dictates the direction (forward/backward) of propagation.
source
Evanescent waves
for 2
2 2
2( ) k ( ) k ( )x y
c
22 2
2ik ( ) ( ) k ( ) k ( ) ( )z x y
c
The sign of the wave vector must be chosen to ensure decay.
source
1.2 Optical layer systems Now we are looking for solutions of the wave equation in inhomogeneous media, which are constructed as a stack of homogeneous layers. This results in two categories of physical problems:
Transmission-reflection problem
Bragg mirrors
chirped mirrors for dispersion compensation
z
x
y
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 19
interferometers
Guided modes
multi-layer waveguides
Bragg waveguides
The general idea for the solution of these problems is the following:
Separating the domains in regions for which an analytical solution for the wave equation exists = mode expansion (free space plane wave)
Expanding the field into a superposition of these modes = Adjusting the amplitudes of each mode, such that the boundary conditions are fulfilled (exact or approximately)
General comment:
Modes in which the field is expanded should be adopted to the geometry
Fields in the layer system
Prerequisites:
stationary
layers in y-z-plane
incident fields in x-z-plane
Ansatz:
z( , , ) Re ( )expx z t x ik z i t rE E
z( , , ) Re ( )expx z t x ik z i t rH H
Decomposition in TE and TM fields:
TE
0
, 0
0
H
E
H
E H
x
TE y TE
z
TM
0
, 0
0
E
H
E
H E
x
TM y TM
z
1.3 Derivation of the transfer matrix Transition conditions
z
x
y
z
x
given field
to be calculated field
given field
to be calculated field
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 20
Fields: Et and Ht continuous TE: E(y) and Hz TM: H(y) and Ez
Calculation of tangential components (normal components can be derived from them)
wave vector component kz conserved throughout the layer system (determined by incidence angle)
wave vector component kx constant in a single homogeneous layer but varies from layer to layer
2
2 2
2ix i zk kc
Field calculation of continuous components (TE)
2
2 22
z2 2( ) 0
xi
i y
k
k E xx c
and z
0
( ) ( )y
iH x E x
x
Solution:
i1 2 x
0 1 2
( ) cos sin
( ) ( ) sin cos
i
i i i
y x
z y x x x
E x C k x C k x
i H x E x k C k x C k xx
Determination of 1 2,C C by
1(0)yE C and 2
0iy xE k C
x
TE:
i 0x
0
1( ) cos (0) sin
sin (0) cos
i i
i i i
y x y x y
y x x y x y
E x k x E k x Ek x
E k k x E k x Ex x
TM:
0
0
1( ) cos (0) sin
1 1sin (0) cos
i i
i
i
i i
iy x y x y
x i
x
y x y x y
i i i
H x k x H k x Hk x
kH k x H k x H
x x
kz
kz
kz
kx
kz
kx
kz
kx
kz
kx
kz
kxkz
kx
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 21
Combined TE/TM:
i
i
1( ) cos (0) sin (0)
( ) sin (0) cos (0)
i i
i
i i i
x x
x
x x x
F x k x F k x Gk
G x k k x F k x G
TE: 0 i, , 1y z yF E G i H Ex
TM: 0 i i, , 1/y z i yF H G i E Hx
Summary of matrix method:
1 0 0
ˆm ( ) MN
i i
iD
F F Fd
G G G
with
ix
1cos sin
m
sin cos
i i
i
i i
k d k dkx
k k d k d
ix
ix i
ix i ix ix
TE: i, , 1y yF E G Ex
TM: i i, , 1/y i yF H G Hx
1.4 Reflection and transmission problem
Transmission coefficient (T) und reflection coefficient (R)
in in
T RF FT R
F F
z
x
given: F(0), G(0), kz, i, di
to be calculated fields: F(D), G(D)
2
2 2i z
0
2,
ix zk k k
0
D
z
x
y z
x
yFin FR
FT
Fin FR
FT
substrate, s
cladding, c
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 22
Connecting the incident, reflected, and transmitted fields with the transfer matrix by expressing the fields at the interfaces of the layer system based on incident, reflected, and transmitted fields:
At the substrate side:
in S R S
S S in S R S
(0,z) exp( ) exp( ) exp( )
(0,z) exp( ) exp( ) exp( )
z x x
x z x x
F ik z F ik x F ik x
G i k ik z F ik x F ik x
At the cladding side:
T C
C C T C
( ,z) exp( ) exp ( )
( ,z) exp( ) exp ( )
z x
x z x
F D ik z F ik x D
G D i k ik z F ik x D
Connecting the substrate and cladding side by the transfer matrix:
in RT 11 12
S S in RC C T 21 22
( ) ( )
( ) ( ) xx
F FF M D M D
k F Fk F M D M D
ii
Resulting in reflection/transmission coefficients:
22 11 21 12
2
k M k M M k k MR
N
kT
N
s sx c cx s sx c cx
s sx
i
with
22 11 21 12N k M k M M k k M s sx c cx s sx c cx
i
2
2 2
s/c x s/c 0
0
2zk k
Energy flux
defined by the normal component of the Poynting vectors sx
x T x R
x in x in
s s
s s
z
x
ysin sR sxR
sT sxT
sx in
z
x
y z
x
ysin sR sxR
sT sxT
sx in
substrate, s
cladding, c
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 23
2
2Re
Re
c cx
s sx
R
kT
k
Example: Bragg mirror
Series of alternating dielectric layers with a chosen thickness, such that reflected light interferes constructively
The thickness of each layer is chosen to match 0 / 4i id n .
Example: 10 layers, 1 1.5n ,
2 2.0n , Design 0.6µm
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 24
Introducing a defect will couple light evanescently through the structure
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 25
Field distribution
Aim: calculation of the field distribution ( )F x in the entire structure
Starting point: known shape of transmitted vector
Remark: The field at x D is more easily known than at 0x since it only depends on a single wave (transmitted wave). In contrast at 0x the field is a superposition of the incident and reflected field. Hence to fix the field at
0x one would have to solve the reflection problem first, whereas at x D one can determine the field by setting ( )F D to an arbitrary value, e.g.
( ) 1F D and calculating ( )G D from Maxwell´s equations to ( ) c cxG D i k .
1
T
c cxcDD
FF
FFi kG
x
now: 1TF
1. invert the structure (vector transforms into (1, -ickcx))
2. calculate the field vector up to the next layer boundary
3. calculate towards the current x-value, starting from the layer boundary
4. save the first component of the vector
5. turn back the derived field and structure by inverting the order of the calculated fields
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 26
Calculating the real fields
z( , , ) Re ( )expx z t x ik z i t rE E
z( , , ) Re ( )expx z t x ik z i t rH H
with
TE: yE ex F x
TM: yH ex F x
1.5 Guided modes in layer systems
no y dependence, phase rotation in z direction
waves propagating without diffraction
allows miniaturization of optics
concept is used in optical communication systems
How can waves be bound by the layer system?
principle mechanism is total internal reflection
Physical reasoning of the field distribution:
zx
y
zx
y FinFR
FT
FinFR
FT
zx
y
zx
y
zx
y
zx
y
Fin FR
FT
Fin FR
FT
Fin FR
FT
FinFR
FT
FinFR
FT
FinFR
FT
z
x
nx
substrate
cladding
layer (system)
cladding
substrate
core
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 27
plane wave in propagation direction: exp( )zik z
evanescent wave in substrate and cladding 2
2
,2max ( )k
c
z s c
oscillating solution in core sin( ) cos( )fx fxA k x B k x 2
2
2max ( )i
ik
c
z
general condition for guided waves: 2
2
s,c 2max ( ) max ( )i
ik
c c
z
Modes are resonances of the system
A physical explanation of the correspondence of roots and modes can be seen in the singularities of the reflection and transmission coefficients. Hence there is a field in the layer without an input field.
For comparison this is the reflectivity and transmissivity of the same layer system in the kz-domain corresponding to the reflection/transmission problem addressed in the previous section. Here no singularities are observed.
Dispersion relation of guided waves singularities of R and T
22 11 21 12
22 11 21 12
k M k M M k k MFR
F k M k M M k k M
s sx c cx s sx c cxR
I s sx c cx s sx c cx
i
i
singularity: 22 11 21 12 0D k M k M M k k M s sx c cx s sx c cx
i
0 0.5 1 1.5 20
10
20
30
40
kz/k
0
Inte
nsity
abs(T)
abs(R)
0 0.5 1 1.5 20
1000
2000
3000
4000
5000
6000
kz/k
0
Inte
nsity
abs(T)
abs(R)
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
kz/k
0
Inte
nsity
abs(T)
abs(R)
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
kz/k
0
Energ
y
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 28
Roots of the denominator correspond to modes guided along the layer system.
The problem of finding a guided mode is reduced to finding a root.
with: , ,k i k i sx s cx c
2
2
, ,2( ) 0k
c
s c z s c
TE,TM TE,TM TE,TM TE,TM
22 11 21 12 0M M M M s c c ss c c s
Field distributions of guided modes inside a single high-index layer embedded into low-index substrate and cladding.
12 14 16 18 20-100
0
100
200
300
kz [µm
-1]
D
0 0.5 10
2
4
6
8
lateral distance [µm]
abs(E
) [V
2/m
2]
0 1 2 3-2
-1
0
1
2
lateral distance [µm]
abs(E
) [V
2/m
2]
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 29
2. Finite-difference method for waveguide modes
2.1 Scalar approximation for weakly guiding waveguides with small index differences
(A similar method is documented in J. Riishede, N. A. Mortensen, and J. Laegsgaard, "A ‘poor man´s approach´ to modeling micro-structured optical fibers," J. Opt. A: Pure Appl. Opt. 5, 534-538 (2003))
Starting from the wave equation
2
2rot rot ( , ) (r, ) ( , )
c
E r E r
Neglecting the divergence of the electric field in weakly inhomogeneous media with grad ( , ) 0 r
div ( , ) 0 D r 0 ( , )div ( , ) 0 r E r
We obtain the Helmholtz equation for inhomogeneous media
2
2( , ) (r, ) ( , ) 0
c
E r E r
Neglecting the vectorial properties of the electric field scalar Helmholtz equation
2(r) ( , ) (r) 0v k v r with 2
2
2( , ) ( , )k
c
r r
2.1.1 Stationary solutions of the scalar Helmholtz equation
Now we search for the stationary states (modes) of the problem with ( , ) ( , , )x y r , by taking the following ansatz function
(r) ( , )expv u x y z i
which results in an eigenvalue equation for the propagation constant
(2) 2 2( , ) ( , , ) ( ) ( , ) 0u x y k x y u x y
This eigenvalue problem is to be solved by a finite difference scheme
2 2
2 2
2 2( , ) ( , , ) ( ) ( , ) 0u x y k x y u x y
x y
with the Laplace operator for two dimensions
2 2
2 2
( , ) ( , )u x y u x y
x y
which must be approximated by a discrete operator in the following way
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 30
Discretization of Laplace operator in two dimensions
21 1
2 2
,
( , ) 2 ( , ) ( , )
j k
j k j k j k
x y
u x y u x y u x yu
x h
21 1
2 2
,
( , ) 2 ( , ) ( , )
j k
j k j k j k
x y
u x y u x y u x yu
y h
2 21, 1, , 1 , 1 ,
,2 2 2
,
4( )
j k
j k j k j k j k j k
j k
x y
u u u u uu uu
x y h
Hence the eigenvalue problem in discrete notation reads as
1, 1, , 1 , 1 , 2 2
, ,2
4( ) ( ) 0
j k j k j k j k j k
j k j k
u u u u uk u
h
which is a linear equation for each variable ,j ku coupled to 4 neighboring
equations
( , ) ,( , ) ( )x y j ku x y u
Introducing a computation grid:
quadratic area of size a a
equidistant discretization of the area with N N grid points
-10-5
05
10 -10
-5
0
5
10
0
50
100
150
200
X
y
U(x,y)
xj= yk=h
(xj,k,Uj,k) (xj+1,k,Uj+1,k)
x
y
(xj,k-1,Uj,k-1)
(xj,k+1,Uj,k+1)
(xj-1,k,Uj-1,k)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 31
example at point ( 2, 3)j k :
3,3 1,3 2,4 2,2 2,3
2,3 2
4( )
u u u u uu
h
2.1.2 Matrix notation of the eigenvalue equation
,j ku : originally 2D variable depending on x-direction (j) and y-direction (k)
unfolding of ,j ku into a 1D vector
for each vector component ,j ku results an individual linear equation
matrix dimension: number variables in x-direction times number variables in y-direction
a
N
.
.
11 2 3 N
N
.
.
11 . . N
j
k
U|G = UG
.
.
= 0
U j,k
matrix
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 32
Matrix equation: schematic picture of the matrix formulation of the 2D Laplace operator
matrix: small number of non-zero values 'sparse' matrix
2.1.3 Boundary conditions
Example: metal boundaries (Metal tube with boundaries i )
U 2,2
=
U 3,2
U N-1,2U 2,3
…
U 3,3 0
U N-2,2
U 4,3
U N-2,3
…
U N-1,3U 2,4U 3,4U 4,4
-4 1 0 0 .. 0 1 0 0 0 0 … 1 -4 1 0 .. 0 0 1 0 0 0 …
1 0 0 0 .. 0 -4 1 0 0 0 … 0 1 0 0 0 … 0 1 0 0 .. 0 1 -4 1 0 0 … 0 0 1 0 0 …0 0 1 0 .. 0 0 1 -4 1 0 … 0 0 0 1 0 …
…
discrete Laplace operator
N
.
k
.
11 j N
x iterates first
accounts for the 1D – problem (tridiagonal matrix)
occures for 2D problems
a
N
.
.
1
- boundaries of the grid ?
- 4 U1,3 + U2,3 + U0,3 + U1,4 + U1,2
h2( U ) 1,3 =
example:
1 20 outside the grid
boundary conditions
(compare with theory of
partial differential equation)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 33
2.2 Full vectorial mode solver for waveguides with large index differences
(according to Zhu et al., Optics Express 10, 853 2002)
starting from directly discretizing Maxwell’s equations using finite differences
t
BE and
t
DH
looking for guided modes propagating in the z-direction, all fields are assumed to have a spatio-temporal dependence like
exp i z t
introducing this ansatz into Maxwell's equation, rescaling the electric field by the free space impedance
0 0 0/Z in order to symmetrize the equations
and performing all derivatives with respect to time t and space z gives
0 0
0 0
0 0
z zx y r x y
z zy x r y x
y yx xz r z
E Hik H i E ik E i H
y y
E Hik H i E ik E i H
x x
E HE Hik H ik E
x y x y
G
U|i = Ui = const
N
.
.
11 . . N
grid with metal boundaries (U|G = 0 )
2 j N-1, 2 k N-1
(N-2) (N-2) – equations
(N-2) (N-2) – unknown
- 4 Uj,k + Uj+1,k + Uj-1,k + Uj,k+1 + Uj,k-1
h2=( U )j,k = 0
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 34
now the differential operators are approximated by finite differences which are formulated on a so-called Yee-grid to preserve the inversion symmetry of the original problem also in its discrete mathematical representation
Remarks on the properties of the Yee-grid
Please be aware that electric and magnetic field are evaluated half a unit cell apart! The 'beauty' of the Yee-grid can be seen when illustrating the fundamental laws of electromagnetics. Let's start from Ampere's law by taking the first Maxwell's equation
( , )( , )
tt
t
B rE r
0 ( , ) ( , )i t t H r E r
and integrating over a finite surface F
0 ( , ) ( , )F F
i t t H r f E r f
where f is the surface normal
Applying Stokes law the surface integral on the right hand side is transformed into a boundary integral
0
( )
( , ) ( , )F F
i t t H r f E r s
where s is the tangential element of the boundary
This integral equation can be expressed approximately based on the field components in the Yee-grid as
0 ( , ) x ( 1, ) ( ,( , 1) x ( ,) x )y x y x yz E j l E j l y E j l E jH l lj yi
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 35
Resorting the equation gives
0
( 1, ) ( , ) ( , 1) ( , )( , )
y y x xz
E j l E j l E j l E j li H j l
x y
which is the finite difference approximation for the z-component of the H-field in the Maxwell's curl equation.
Now we discretize the complete Maxwell's equations using this ansatz
0
0
0
( , 1) ( , )( , ) ( , )
( 1, ) ( , )( , ) ( , )
( 1, ) ( , ) ( , 1) ( , )( , )
z zx y
z zy x
y y x xz
E j l E j lik H j l i E j l
y
E j l E j lik H j l i E j l
x
E j l E j l E j l E j lik H j l
x y
0
0
0
( , ) ( , 1)( , ) ( , ) ( , )
( , ) ( 1, )( , ) ( , ) ( , )
( , ) ( 1, ) ( , ) ( , 1)( , ) ( , )
z zrx x y
z zry y x
y y x xrz z
H j l H j lik j l E j l i H j l
y
H j l H j lik j l E j l i H j l
x
H j l H j l H j l H j lik j l E j l
x y
with the spatially averaged material parameters
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 36
( , ) ( , 1)( , )
2
( , ) ( 1, )( , )
2
( , ) ( 1, 1) ( , 1) ( 1, )( , )
4
r rrx
r rry
r r r rrz
j l j lj l
j l j lj l
j l j l j l j lj l
The spatial averaging of material properties is a technique to increase computational convergence by minimizing the so-called 'staircase approximation error' in finite difference schemes.
As a last step we rewrite the discretized Maxwell's equations in matrix notation and use the perfect electric conductor boundary conditions from above
0
0
0
0
0
0 0 0
0 0 0
0 0 0
x y x
y x y
z y x z
rx x y x
ry y x y
rz z y x z
i
ik i
i
ik i
H I U E
H I U E
H U U E
ε E I V H
ε E I V H
ε E V V H
with
I the identity matrix,
, ,x y zE and , ,x y zH the concatenated vectors of x,y,z-directed field
components , , ( , )x y zE j l and , , ( , )x y zH j l , respectively
rxε the diagonal matrix of spatially averaged material parameters.
Furthermore there are the matrices of differential operators in discrete form on the Yee-grid
1 1
1 11x
x
U ( , 1) ( , )A j l A j l
1 1
1 11y
y
U ( 1, ) ( , )A j l A j l
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 37
1
1 11
1x
x
V ( , ) ( , 1)A j l A j l
1
11
1
1
yy
V ( , ) ( 1, )A j l A j l
Please be aware that even though these matrices look asymmetric they are just asymmetric in the notation and not in real space. They represent symmetric finite difference operators since the Yee grid symmetrizes the problem.
For the numerical solution of the eigenvalue problem one can combine the two coupled differential equations of first order for the two fields E and H into a single decoupled differential equation of second order for a single field, i.e. E or H .
For the electric field this results in (quasi TE version):
2xx xy x x
yx yy y y
P P E E
P P E E
with the sub-matrices
2 1 2 1 2
0 0 0
2 1 2 1 2
0 0 0
1 2 2 2 1
0 0 0
1 2 2 2 1
0 0 0
xx x rz y x y x rz x rx y y
yy y rz x y x y rz y ry x x
xy x rz y ry x x x rz x y x
yx y rz x rx y y y rz y x y
k k k
k k k
k k k
k k k
P U ε V V U I U ε V ε V U
P U ε V V U I U ε V ε V U
P U ε V ε V U I U ε V V U
P U ε V ε V U I U ε V V U
These equations might look quite lengthy but finally they consist just of a few matrix operations.
Similarly for the magnetic field this results in (quasi TM version):
2xx xy x x
yx yy y y
Q Q H H
Q Q H H
with the sub-matrices
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 38
2 1 2 2 1
0 0 0
2 1 2 2 1
0 0 0
2 1 2 2 1
0 0 0
2 1 2 2 1
0 0 0
xx x y x rz y ry x x y rz y
yy y x y rz x rx y y x rz x
xy ry x x y rz x x y x rz x
yx rx y y x rz y y x y rz y
k k k
k k k
k k k
k k k
Q V U U ε V ε V U I U ε V
Q V U U ε V ε V U I U ε V
Q ε V U U ε V V U I U ε V
Q ε V U U ε V V U I U ε V
Example 1: Photonic crystal fiber
Solving the eigenvalue problem for obtaining the propagation constant and the field distribution of guided modes.
Absolute value of the H-field in the x-y cross section of the first mode.
Example 2: Silver nano wire in vacuum
Ag-cylindrical waveguide (r=λ/3) at λ=630nm
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 39
Absolute value of the E-field in the x-y cross section of the first mode.
Intensity of the E-field in the x-y cross section of the second mode.
Dispersion relation of a metallic waveguide with 0 effk n , where
effn is the
so-called effective index of the mode.
Example 3: Silver nano wire on substrate
Ag-cylindrical waveguide (r=λ/3) on a substrate with 1.5n at λ=840nm
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 40
Intensity of the E-field in the x-y cross section of the first mode. The light is mainly confined at the interface between the silver nanowire and the substrate.
Dispersion relation of a metallic waveguide on a substrate.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 41
3. Beam Propagation Method (BPM) up to this point we only dealt with eigenmodes
This required invariance of the structure in the third dimension.
What happens if light propagation occurs in a medium where the index distribution changes weakly along the propagation direction?
need for accurate model of very wide range of devices
linear and nonlinear light propagation in axially varying waveguide systems, as e.g.
o curvilinear directional couplers,
o branching and combining waveguides,
o S-shaped bent waveguides,
o tapered waveguides
ultra short light pulse propagation in optical fibers
implementation
finite difference BPM solves Maxwell's equations by using finite differences in place of partial derivatives
computational intensive
entirely in the frequency domain only weak nonlinearities can be modeled
use of a slowly varying envelope approximation in the paraxial direction the device is assumed to have an optical axis, and that most of the light travels approximately in this direction, (paraxial approximation) allows to rely on first order differential equations
3.1 Categorization of Partial Differential Equation (PDE) problems
general second-order Partial Differential Equation
2 2 2
2 20
f f f f fp q r s t u f v
x x y y x y
(23)
The following different types of partial differential equations are distinguished
1. 2 4q pr : elliptic PDE
2. 2 4q pr : parabolic PDE
3. 2 4q pr : hyperbolic PDE
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 42
Elliptic PDE
Example: ( , ) ( , )
( , )u x y u x y
x yx y
2 2
2 2+ = Poisson equation
Boundary Value Problem (BVP) limited mainly by computing memory
Hyperbolic PDE
Example: ( , ) ( , )u x y u x y
vt y
2 22
2 2 1D wave equation
Parabolic PDE
Example: ( , ) ( , )u x t u x t
Dt x x
= diffusion equation (IVP)
both are Initial Value Problem (IVP) limited mainly by computing time and stability issues
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 43
3.2 Slowly Varying Envelope Approximation (SVEA)
Assumptions
Each component of the optical electromagnetic field is primarily a periodic (harmonic) function of position.
The field changes most rapidly along the optical axis z (with a period on the order of the optical wavelength ).
Slowly Varying Envelope Approximation (SVEA)
replace the quickly varying component, , with a slowly varying one, , as
0( , , ) ( , , )exp( )x y z x y z ikn z with 2 /k
introduction of a reference index 0n
light is travelling mostly parallel to the z axis (paraxial approximation) and is monochromatic with wavelength
SVEA Advantages
requirements on the mesh to represent derivatives by finite differences are relaxed choose fewer mesh points higher speed of the calculation without compromising accuracy
accurate BPM calculations using step sizes z
SVEA Problems
if part of the light strongly deviates from the direction of the axis z improved solution: wide angle BPM
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 44
if structure has large index contrast no accurate global choice of
0n finer mesh needed
solution: if variation is in z direction 0( )n z
quality of choice of 0n can be checked by evaluating the speed of phase
evolution of along z in the numerics
3.3 Differential equations of BPM starting from Maxwell’s equations in the frequency domain with an inhomogeneous distribution of the dielectric properties ( , , )x y z
0
0 ( , , )
i
i x y z
E H
H E
+ no charges [ ( , , ) ] 0
0
x y z
E
H
magnetic field can be eliminated by taking curl of curlE -equation
2 ( , , )k x y z E E with 0 0k
using the vector identity 2( ) one gets the wave equation
2 2 ( , , ) ( )k x y z E E E
since BPM is biased for the z -axis, it is natural to treat z -components of the field E and the operator separately from the transverse components x and y
ˆt zE E E z and ˆ
tz
z
the transverse component of the wave equation becomes
2
2 2
2( , , ) ( )t z
t t t t t t
Ek x y z
z z
EE E E (23)
which is an inconsistent problem since it contains transverse and longitudinal components.
Splitting also the divergence equation in the transverse and longitudinal components
( , , )
( , , ) ( , , ) 0zt t z
x y z Ex y z E x y z
z z
E
and neglecting the second term ( is assumed to change only slowly in z )
( , , ) ( , , ) 0zt t
Ex y z x y z
z
E
one can eliminate the longitudinal term in the right hand side of the wave equation (23)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 45
22 2
2
1( , , ) ( ( , , ) )
( , , )
tt t t t t t t tk x y z x y z
z x y z
EE E E E
applying the chain rule on the second divergence term on the right hand side, the first divergence term is canceled out
22 2
2
1( , , ) ( ( , , ))
( , , )
tt t t t t tk x y z x y z
z x y z
EE E E (23)
up to now: Due to the propagation direction, which points mainly along z , the field
tE is varying slowly in x and y , but rapidly in z
now: SVEA is introduced
0( , , ) ( , , )exp( )t tx y z x y z in kz E e
the new field ( , , )t x y ze is slowly varying in all coordinates (compared to )
Inserting SVEA ansatz into wave equation (23) gives
22 2 2
0 02
12 ( ( , , ) ) ( ( , , )) 0
( , , )
t tt t t t t tikn k x y z n x y z
z z x y z
e ee e e
If reference index 0n was chosen correctly, the first term will be much smaller
than the second and hence can be neglected (physics: neglect coupling to backward propagating waves and introducing paraxial approximation)
results in a first-order differential equation for te in the z coordinate and
changes the problem from an elliptic PDE (boundary value problem) to a parabolic PDE (initial value problem)
Next step: collect the transverse 2. order operators in a matrix form and shift them to the right hand side
02x xxx xy
yx yyy y
e eP Pikn
P Pe ez
with the components of the operator matrix being
22 2
02
2
2
22 2
02
1( , , ) ( ( , , ) )
( , , )
1( , , )
( , , )
1( , , )
( , , )
1( , , ) ( ( , , ) )
( , , )
xx
xy
yx
yy
P x y z k x y z nx x y z x y
P x y zx x y z y x y
P x y zy x y z x y x
P x y z k x y z ny x y z y x
Summary of the results
paraxial vector wave equation for the optical electric field
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 46
initial value problem: knowledge of the electric field in some transverse plane ( const.z ) is enough to calculate the field in the entire space
reflections of light are neglected
in a finite differencing scheme the operator P is a large sparse matrix
3.4 Semi-vector BPM up to now: full vector character of the electromagnetic field is included
now: if the modeled device operates mainly on a single field component, the other component’s contribution can often be neglected
semi vector TE equation for E-field being mainly transversely polarized
02 xxx x
eikn P e
z
or semi vector TM equation for H-field being mainly transversely polarized
02y
yy y
eikn P e
z
Properties
correctly models differences of TE and TM wave propagation
neglects coupling to other field component
3.5 Scalar BPM if structure has very low index contrast, the x and y term can be
neglected operators commute xxP and yyP reduce to scalar operator
2 2
2 2
02 2( ( , , ) )P k x y z n
x y
3.6 Crank-Nicolson method formal solution of BPM equation can be written as
t 1 t 0
0
( ) exp ( )2
zz z
in k
Pe e with
1 0z z z
however, a rational function is needed to approximate the exponent of the operator
a simple approximation would be
1 (1 )
exp1
xx
x
The case =0.5
called Padé(1,1) approximation
accurate for small x small step size z (higher order Padé approximations can be used, see below as supplementary material)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 47
generates the first three terms of the MacLaurin series expansion for
exp x
leads to Crank-Nicolson scheme
1 0
0 0
( ) ( )4 4
t t
z zz z
in k in k
I P e I P e
since the operator P is applied to the unknown t 1( )ze it is an implicit method
requires solution of set of linear equations
Variable is called scheme parameter, for 0.5 the method is stable and energy conserving, since it corresponds to the discrete operator which preserves the inversion symmetry in z.
3.7 Alternating Direction Implicit (ADI) Solving the implicit problem in 2D leads to a huge set of linear equations. time consuming numerical solution
However, 1D problems result only in a simple tridiagonal matrix which can be solved very fast.
ADI approximation
splitting the n-dimensional operator into n subsequent 1-dimensional operators
each operation only on a single dimension
fast algorithm: periodic application of 1D operators in x and y direction
3.8 Boundary condition
Easiest boundary condition: reflecting boundaries (electric walls)
assuming that the field is zero at the boundary perfect reflection
Advanced boundary conditions: fields can radiate out of simulation area
fields radiating out, e.g. from waveguide, should not disturb the evolution by being reflected back from the boundaries let the radiation fields out
do not create any additional effects for the fields propagating inside the computation domain (e.g. instability, dissipation etc.)
Typical advanced boundary condition
absorbing boundary conditions (ABC)
transparent boundary conditions (TBC)
perfectly matched layer boundaries (PML)
3.8.1 Absorbing Boundary Conditions (ABC)
problem:
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 48
field at the boundary of the computational window is not wanted
solution:
remove field at boundaries by multiplication with factor <1 can be introduced by ( ) 0 at the boundary
however:
All inhomogeneities of the optical properties induce reflections themselves.
This can be shown by computing Fresnel reflection at interface where just the imaginary part of the dielectric function changes.
improvement:
soft onset of the absorbing layer
technical realization:
multiplication of the field in each iteration step with a filter function, e.g. in 1D:
abs
abs abs
abs abs
1 exp expp p
p p Pe e A A
N N
with
abs 3...5N
drawback:
absorption strength absA and absorption width
absN are difficult to adjust for
the smallest reflection
individual optimal parameters for each problem
3.8.2 Transparent Boundary Condition (TBC)
goal:
simulates a nonexistent boundary
Radiation is allowed to freely escape from the simulated area without reflection.
Flux of radiation back into the simulated area is prevented.
method:
assuming that the field in the vicinity of the virtual boundary consists of an outgoing plane wave and does not include any reflected wave from the virtual boundary wave function for the wave traveling to the right with the x -directed wave number
xk is expressed as
( , ) ( )exp(i )xx z A z k x
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 49
Derivation
projecting the field onto the following lattice
( )p px
lattice points 1p and p m are outside the computational window
At the left boundary the relation of the field at neighboring mesh points can be expressed as
1
0
exp( )xjk x
with 1 0x x x
from which we can calculate the x -directed wave number xk as
1
0
1lnxk
j x
if ( ) 0xk plane wave is traveling outwards
( ) 0xk plane wave is traveling inwards
since inward waves should not exist
( )xk must be positive ( ) 0xk
Implementation
xk is calculated at the boundary
another mesh point at 1p is artificially added to the mesh
the field at 1p is assumed to be determined by the same plane wave
function
0
1
exp( )xjk x
with
1 0 0 1x x x x x
assuming that the xk -vector is preserved by the wave
1 01 1 0 0
0 1
exp( ) 0xjk x
kx-Vektor boundary
(x)=0exp(ikxx)
x
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 50
trick: calculate 1n
xk one step z before applying it to the boundary
takes into account that the wave energy is traveling also forwards
(keep in mind, that we are in the paraxial approximation vor which all
waves propagate mainly forward)
1 1
1 1 0 00 n n n n
3.8.3 Perfectly matched layer boundaries (PML)
Definition
perfectly matched layer = artificial absorbing layer for wave equations used to truncate computational regions in numerical methods to simulate problems with open boundaries
Property
waves incident upon the PML from a non-PML medium do not reflect at the interface strongly absorb outgoing waves from the interior of a computational region without reflecting them back into the interior (impedance matching required)
Different formulations
nice one: stretched-coordinate PML (Chew and Weedon)
coordinate transformation in which one (or more) coordinates are mapped to complex numbers analytic continuation of the wave equation into complex coordinates, transforming propagating (oscillating) waves into exponentially decaying waves
Technical description
to absorb waves propagating in the x direction, the following transformation is applied to the wave equation:
all x derivatives / x are replaced by
1
( )1
i xx x
with being the angular frequency and some positive function of x
Hence, propagating waves in x direction ( 0k ) become attenuated
exp ( ) exp ( ) ( ') 'xk
i kx t i kx t x x
which corresponds to the following coordinate transformation (analytic continuation to complex coordinates)
( ') 'xi
x x x x
or equivalently (1 / )x x i
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 51
Properties
for real valued the PML attenuate only propagating waves
purely evanescent waves oscillate in the PML but do not decay more quickly attenuation of evanescent waves can be accelerated by including a real coordinate stretching in the PML corresponds to complex valued σ
PML is reflectionless only for the exact wave equation discretized simulation shows small numerical reflections can be minimized by gradually turning on the absorption coefficient σ from zero over a short distance on the scale of the wavelength of the wave (e.g. with quadratic spatial profile)
3.9 Conformal mapping regions used to simulate curved optical waveguides
solving bends by BPM directly leads to large errors due to the paraxial approximation
can be used to treat losses in curved waveguides
radii of curvature need to be restricted to large values, when first order approximation of the conformal mapping is used (corresponding to linear index gradient)
Method
apply a conformal mapping in the complex plane to transform a curved waveguide ( , )x y in coordinates ( ,x y ) into a straight waveguide described
by a modified epsilon distribution ( )u in new coordinates ( ,u v )
conformal mapping is an angle-preserving transformation in a complex plane
Remarks
This concept led to the emergence of an entire field of optical science, the so-called "transformation optics", where instead of interfaces between
different materials, like lenses, prisms, etc., continuous variations of (x,y)
and (x,y) are exploited for the manipulation of optical fields. However the mathematical apparatus just works in 2D and there is no formulation for 3D.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 52
typical example for local angle preserving transformation: world maps
Demonstration for 2D scalar wave equation
2 2
2 2( , ) ( , ) 0k n x y x z
x z
general transformation
( ) ( )W u iv f x iz f Z
for f being an analytical function in the complex plane
To straighten a bend with radius R in the ( ,u v ) plane, ( )f Z should be taken
as
( ) ln( / )W f Z R Z R
applying the transformation to the wave equation gives
2
2 2
2 2( , ), ( , ) 0
dZk n x u v y u v
dWu v
with the Jacobian of the transformation being
exp( / )dZ
u RdW
a bended waveguide is transformed into straight waveguide with modified refractive index
( , )dZ
n x ydW
limitations imposed by the paraxial approximation are avoided
in first order approximation a linear index gradient is added to account for the bending
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 53
3.10 Wide-angle BPM based on Padé operators supplementary material, not covered by the lecture
Principle
expansion via Padé is more accurate than Taylor expansion for the same order of terms larger angles / higher index contrast / more complex mode interference can be analyzed as the Padé order increases
Derivation
starting from scalar wave equation without neglecting second order z derivative, the equation can be formally rewritten as
0 0
0 0
/ 2
1 ( / 2 )( / )
P k nj
z j k n z
(*)
which can be reduced to
N
jz D
with N and D being polynomials determined by the operator P
applying a finite difference scheme we get to iteration equation
1 (1 )l lD j z N
D j z N
with being a control parameter of the finite difference scheme ranging between 0 and 1 ( 0 fully implicit scheme; 1 fully explicit scheme;
0.5 Crank-Nicolson scheme)
the numerator (1 )D j z N can be factorized as
1 2
Nom Nom Nom 2 1( ...) (1 )...(1 )(1 )N N NNA P B P C P c P c P c P
with coefficients 1 2, ,..., Nc c c which can be obtained from the solution of the
algebraic equation
1 2
Nom Nom Nom( (1 ) ) ( ...) 0N N ND j z N A P B P C P
similarly the denominator D j z N can be factorized as
1 2
Den Den Den 2 1( ...) (1 )...(1 )(1 )N N NNA P B P C P d P d P d P
with coefficients 1 2, ,..., Nd d d which can be obtained from the solution of the
algebraic equation
i
0
( ) 0N
i
i
D j z N d P
with 0
0 0 1c d P
Thus, the unknown field 1l at z z is related to the known field
l at z as
1 2
1 Nom Nom Nom1 2
Den Den Den
...
...
N N Nl l
N N N
A P B P C P
A P B P C P
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 54
or
1 2 1
2 1
(1 )...(1 )(1 )
(1 )...(1 )(1 )
l lN
N
c P c P c P
d P d P d P
3.10.1 Fresnel approximation – Padé 0th order
starting again from wave equation (*) we account for the z-derivative by the recursion equation
0 0
0 01
/ 2
1 ( / 2 )n
n
P k nj
zj k n
z
(**)
which is used to replace the z-derivative in the denominator of (**)
in paraxial approximation 0
1
/
1
P a Pj j
jz a
a z
with 0 02a k n and 1
0z
comparison to the expansion equations gives
1D and /N P a
and therefore the nominator becomes
Nom(1 ) 1 (1 ) 1P
D j z N j z A Pa
and the denominator analogously
Den1 1P
D j z N j z A Pa
with
Nom
(1 )zA j
a
Den
zA j
a
thus the unknown field 1l at z z is related to the known field
l at z as
1 Nom
Den
1
1
l lA P
A P
or
1Den Nom1 1l lA P A P
which is the standard Fresnel formula
3.10.2 Wide angle (WA) approximation – Padé (1,1)
using again the recursion equation
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 55
1
0
/
1
P aj
jz
a z
and inserting the previous result
0
Pj
z a
we get
1
2
/
1
P aj
Pz
a
comparison to the expansion equations gives
2
1P
Da
and /N P a
and therefore the nominator becomes
Nom2 2
1 1(1 ) 1 (1 ) 1 1
P PD j z N j z j z P A P
a aa a
and the denominator analogously
Den2 2
11 1
P PD j z N j z j z P A P
a aa a
with
Nom 2
1 (1 )zA j
aa
Den 2
1 zA j
aa
thus the unknown field 1l at z z is related to the known field
l at z
again as
1Den Nom1 1l lA P A P
which includes now higher order corrections
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 56
4. Finite Difference Time Domain Method (FDTD) ab initio, direct solution of Maxwell‘s equations
probably the most often used numerical technique, as its implementation is straight forward (but what is cumbersome: implementation of proper boundary conditions)
requires excessive computational resources for reasonable problems in 3D (implementation requires cluster computers)
implementation is absolutely general but often doesn‘t take explicit advantage of symmetries
in principle all kinds of materials are treatable (dispersive or nonlinear materials) but one has to be careful to preserve the stability of the method
4.1 Maxwell’s equations
makr
ext
( , ) ( , )rot ( , ) , rot ( , ) ( , ),
div ( , ) ( , ), div ( , ) 0,
t tt t t
t t
t t t
B r D rE r H r j r
D r r B r
mater equations
0
0
( , ) ( , ) ( , ),
( , ) ( , ) ( , )
t t t
t t t
D r E r P r
B r H r M r
E(r,t) electric field
H(r,t) magnetic field
D(r,t) dielectric flux density
B(r,t) magnetic flux density
P(r,t) dielectric polarization
M(r,t) magnetic polarization (magnetization)
ext(r,t) external charge density
jmakr(r,t) macroscopic current density
MWEQ in linear isotropic, dispersionless & non-magnetic dielectric Media
0
0
( , ) ( , ),
( , ) ( , ),
t t
t t
D r r E r
B r H r
0 0 makr
0
( , ) ( , )rot ( , ) , ,
div ( , ) 0, di
rot ( ,
,
)
v ( , ) 0
t tt t
t
t
t
t
r E r
H r E rE r H
r
r r
H
j
curl-equations for individual components in Cartesian coordinates
0
( , ) 1(a) rot ( , )
tt
t
H rE r
field sources
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 57
0
0
0
1,
1,
1
yx z
y xz
yxz
EH E
t z y
H EE
t x z
EEH
t y x
0
( , ) 1(b) rot ( , )
tt
t
E rH r j
r
0
0
0
1,
1,
1
yx zx
y x zy
y xzz
HE Hj
t y z
E H Hj
t z x
H HEj
t x y
r
r
r
We will assume source-free conditions ( 0j ) for the moment and add the
sources later.
4.2 1D problems Assuming that there is no dependence on y and z all dynamics in x
0 0
1 1( ) 0, ,
y yx z zH EH E H
at t x t x
0 0
1 1( ) 0, ,
y yx z zE HE H E
bt t x t x
r r
grouping together all non-mixing transverse electromagnetic components (TEM)
z-polarized E-field 0 0
1 1,
y yz zH HE E
t x t x
r
y-polarized E-field 0 0
1 1,
y yz zE EH H
t x t x
r
Without loss of generality we concentrate on the first case (zE ).
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 58
4.2.1 Solution with finite difference method in the time domain for Ez
Discretization of derivative operators
symmetric discretization is second order accurate
21 1
1 1
1 1
( )n N NN N
N N N
f f fO x x
x x x
while the asymmetric discretization is just first order accurate
11
1
n N NN N
N N N
f f fO x x
x x x
always use symmetric discrete operators if possible / feasible!
introduce discrete space-time index notation: ( , ) ( , )i n i x n t
upper index = time(n )
lower index = space( i )
21 1
1 12
21 1
1 12
( , ) ( )2
( )2
( , ) ( )2
( )2
( )
n n nn i i i
z i
n n n
i i i
n n nn i i i
y i
n n n
i i i
i
E E EE i x n t E O x
x x
E E EO t
t t
H H HH i x n t H O x
x x
H H HO t
t t
i x
Maxwell’s equations
0 0
1 1
1 1
0
1 1
1 1
0
0 0
1 1
1 1
0
1 1
1
0
1 1
1
2 2
1
1 1
1
2 2
1
n ny i iz
i
n n n n
i i i i
i
n n n n
i i i i
i
n ny i iz
n n n n
i i i i
n n n
i i i i
H E HE
t x t x
E E H H
t x
tE E H H
x
H H EE
t x t x
H H E E
t x
tH H E E
x
r
1
n
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 59
FDTD Discretization for the H field:
FDTD Discretization for the E field:
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H 5
4
3
2
1
1 2 3 4 5 6 x(i)
t(n)
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
5
4
3
2
1
1 2 3 4 5 6 x(i)
t(n)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 60
4.2.2 Yee grid in 1D and Leapfrog time steps
Equations for a single space-time step
1 0.5 0.5
0.5 0.5
0.5 0.5
0.5 0.5 1
0
0
1
1
n
n n n n
i i i i
i
n n n
i i i i
tE
tH H E
E H
Ex
Hx
Properties
Divergence MWEQs?
The condition of no divergence of the fields is always fulfilled in 1D, since fields are always transverse polarized to the direction of field change.
Resolution of discretization x and t (from physical arguments)?
5
4
3
2
1
1 2 3 4 5 6 x(i)
t(n)
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
E
H
2
1
1 2 3 x(i)
t(n)
E
H
E
H
E
H
E
H
E
H
E
H
1.5 2.5 3.5
2.5
1.5
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 61
spatial grid resolution x must be fine enough to display the finest structures of the distribution and the fields rule of thumb (heuristically):
max/ (20 )x n , with maxn being the
highest refractive index in the simulation domain
temporal stepsize t is limited by the speed of light, i.e. the interaction in space can reach only up to the next neighbor in one time-step sets upper limit to the phase velocity /t x c (see notes on stability in section 4.3.4)
Boundaries?
finite size of treatable simulation domain requires truncation of space most simple approach: perfectly conducting boundary ( 0E at the boundary)
Sources?
either initial field distribution or sources in the simulation domain
initial field is difficult in higher dimensions since field must have zero divergence (would require prior knowledge of solution)
sources as e.g. currents:
0.5
0.5
0
1 0.5 0.5
0.5 0.5
0
1n n n n
i i i i
n
i
i i
tE E H
x
tjH
4.3 The full 3D problem space grid
( , , , ) ( , , , )i j k n i x j y k z n t
( , , , )
( , , )
n
x x ijk
ijk
E i x j y k z n t E
i x j y k z
Finite differencing in space and time
1 1 2
1 1
2
( )2
( )2
n nn
x xx i j k i j kijk
n n n
x x xijk ijk ijk
E EEO y
y y
E E EO t
t t
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 62
4.3.1 Yee grid in 3D
discretizing MWEQ 0
1x
x yz jt y
H
z
E H
r
1 12 2
1 1 1 12 2 2 2
1 1 1 12 2 2 2
1 12 2
1 12 2
, , , ,
, 1, , , , , 1 , ,
, ,0 , ,
n n
x xi j k i j k
n nn n
y yz zi j k i j k i j k i j k n
x i j k
i j k
E E
H HH Htj
y z
other equations: 1 12 2
1 1 1 12 2 2 2
1 1 1 12 2 2 2
1 12 21 1
2 2
, 1, , 1,
, 1, 1 , 1, , 1, 1, 1,
, 1,0 , 1,
n n
y yi j k i j k
n n n n
x x z z ni j k i j k i j k i j k
y i j ki j k
E E
H H H Htj
z x
1 12 2
1 1 1 12 2 2 2
1 1 1 12 2 2 2
1 12 2
1 12 2
, , 1 , , 1
, , 1 1, , 1 , 1, 1 , , 1
, , 10 , , 1
n n
z zi j k i j k
n n n n
y y x x ni j k i j k i j k i j k
z i j k
i j k
E E
H H H Htj
x y
1 11 12 22 2
3 31 1 1 1 1 12 2 2 2 2 2 2 2
1 12 2
, 1, , 1, , , 1 , , 11
, 1, 1 , 1, 1
0
n n n n
y y z zi j k i j k i j k i j kn n
x xi j k i j k
E E E EtH H
z y
Ez
Ey
Ex
Hy
Hy
Hz
Hz
Hz Hx
Hx
Hy
Hx
x(i)
y(j)
z(k)
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 63
1 11 12 22 2
31 1 1 1 1 1 12 2 2 2 2 2 2 2
1 12 2
1 , , 1 , , 1 , , , ,
, , 1 , , 10
n nn n
x xz zn n i j k i j k i j k i j k
y yi j k i j k
E EE EtH H
x z
1 11 1
2 22 2
3 1 1 1 1 1 1 12 2 2 2 2 2 2 2
1 12 2
, , , , , 1, , 1,1
, 1, , 1,
0
n nn n
y yx xi j k i j k i j k i j kn n
z zi j k i j k
E EE EtH H
y x
4.3.2 Physical interpretation
3D Yee grid & Amper’s law
0
( )
( ) (, ) , )(F F
tt
t
r f sH rE r
Hx(i-1/2,j,k)
Hx(i-1/2,j+1,k)
Hy(i,j+1/2,k)
Hy(i-1,j+1/2,k)
Ey(i-1/2,j+1,k-1/2)
Ey(i-1/2,j+1,k+1/2)
Ez(i-1/2,j+3/2,k)
Ez(i-1/2,j+1/2,k)
1 12 2
1 1 1 12 2 2 2
1 12 2
1 11 12 22 2
, , , ,
0 , ,
, , , 1,, , 1, ,
n n
i j k i j k
i j k
n nn n
i j k i jx ki j k i j ky
z
y
z
x
x yt
x y x
E
H H y
E
H H
solving for the unknown component:
1 12 2
1 1 1 12 2 2 2
1 1 1 12 2 2 2
1 12 2
, , , ,
, , 1, , , , , 1,
0 , ,
y y
n n
i j k i j k
n n n n
i j k i j k i j
z z
k i j k
i j k
x xH H H
E E
t
x y
H
3D Yee grid & Faraday’s law
0
( )
( , ) ( , )F F
tt
t
E rf sH r
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 64
Hx(i-1/2,j,k)
Hx(i-1/2,j+1,k)
Hy(i,j+1/2,k)
Hy(i-1,j+1/2,k)
Ey(i-1/2,j+1,k-1/2)
Ey(i-1/2,j+1,k+1/2)
Ez(i-1/2,j+3/2,k)
Ez(i-1/2,j+1/2,k)
4.3.3 Divergence-free nature of the Yee discretization
0 0 makr
0
( , ) ( , )rot ( , ) ,
div
r ,
,( , ) 0, div
ot ( , )
( , ) 0
t tt t
t
t
t
t
r E r
H r E rE r H
r
r r
H
j
0div ?t
E 0 0
Yee cell Yee cell
div dV dt t
E E f by Gauss's theorem
1 1 1 12 2 2 2
1 1 1 12 2 2 2
1 1 1 12 2 2 2
0 0 , , 1, ,
Yee cell
Term1
0 , 1, , ,
Term 2
0 , , 1 , ,
Term 3
x xi j k i j k
y yi j k i j k
z zi j k i j k
d E E y zt t
E E x zt
E E x yt
E f
substitute Term 1 with curl equation 0
1 yx zHE H
t y z
r
1 1 1 12 2 2 2
1 1 1 12 2 2 2
, 1, , , , , 1 , ,
1, 1, 1, , 1, , 1 1, ,
Term 1y yz zi j k i j k i j k i j k
y yz zi j k i j k i j k i j k
H HH H
y z
H HH H
y z
collecting the contributions from Term 1, Term 2 and Term 3 results in vanishing of all contributions:
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 65
0
Yeecell
Term 1 Term 2 Term 3 0d y z x z x yt
E f
hence, if the field was divergence-free at some time it will conserve this property and stay divergence-free!
0 0 0
Yee cell Yee cell
div ( 0) 0 & div 0 div 0t dV dVt
E E E
However, it is important to ensure that sources do not introduce artificial divergence!
4.3.4 Computational procedure
Using the spatial differences of the E Field that are known for the time step n Δt to calculate the H field at the time step (n+1/2) Δt
Using the spatial differences of the H Field that are known for the time step (n+1/2) Δt to calculate the E field at the time step (n+1) Δt
Using the spatial differences of the E Field that are known for the time step (n+1) Δt to calculate the H field at the time step (n+3/2) Δt
…
Properties of the algorithm
Leapfrog algorithm discretization in all components, explicit formulation in time
Close to the physical world as the spatial and temporal propagation is exactly simulated
Yee grid can ensure inherently divergence-free solution.
Method is second order accurate in all spacetime dimensions (from central differences).
To ensure stability, the Courant-Friedrichs-Lewy condition has to be fulfilled (published very early in [R. Courant, K. Friedrichs, H. Lewy, “Über die partiellen Differenzengleichungen der mathematischen Physik”, Mathematische Annalen 32, (1869)]. This condition links temporal and spatial resolution (as in the 1D case). In three dimensions, the stability condition will be
1/2
2 2 2
1 1 1 1
3
ht
c x y z c
(for a cubic grid).
Note that the temporal and spatial discretization cannot be chosen independently. This means that if you want to double the spatial resolution in a 3D FDTD, your memory consumption rises by a factor of 8 (23), while your computing time rises at least by a factor of 16 (24)!
sou
rce:
urb
an-f
itn
ess.
de
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 66
4.4 Simplification to 2D problems Problems are often invariant in one spatial direction, e.g. taking z (as e.g.
for gratings, cylindrical objects)
Derivations of the field along this directions are zero
0
1 yxEH
t z
0
1 yz x zx
HE E Hj
y t y z
r
0
1y z xH E E
t x z
0
1y xy
E Hj
t z
r
0 0
1 1
z
y yz x z xz
H
x
E HH E E Hj
t y x t x y
r
Maxwell can be decoupled into 2 sets of each 3 differential equations
0 0
0
1 1
1
y yz x z xz
yx
H EE H H Ej
t x y t y x
EH
t z
r
0
1 yz x zx
HE E Hj
y t y z
r
0
1y z xH E E
t x z
0
1y xy
E Hj
t z
r
zH
x
TE polarization TM polarization
4.5 Implementing light sources arbitrary light sources can be modeled simply by adding the source field to
the field in the computational domain
a physical model for sources is a macroscopic current density
makr
( , ) ( , )rot ( , ) , rot ( , ) ( , )
t tt t t
t t
B r D rE r H r j r
Simplified implementation by adding a source term to the electric field, which corresponds to an externally driven dipole polarization
Examples for temporal variation of the light source
x-polarized cw-source , , , ,
sin( )n n
x xi j k i j kE E n t
x-polarized impulse ,, , , ,
n n
x x n ni j k i j kE E
Examples for spatial variation of the light source
x-polarized Gaussian wave
22
, , , ,exp sin( )
n n
x xi j k i j kx y
i x j yE E n t
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 67
Example calculations (2D-configuration, TM, Hy plotted)
source type: plane wave
homogeneous medium Cylinder with refractive index 2n and diameter 2D
source type: point source
homogeneous medium Cylinder with refractive index 2n and diameter 2D
4.6 Relation between frequency and time domain If one would like to determine the response of an optical system for different excitation wavelengths, one could perform many calculations for different cw excitations with a single frequency. In this case one has to wait for a long computing time until the stationary state of the calculation is reached, since the calculation starts with a step function in the excitation.
Alternatively, on can perform this calculation in the frequency domain:
The frequency spectrum used for the illumination is given by the Fourier-transformation of the time dependent incident field.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 68
With a single calculation with sufficient spectral width, e.g. a delta distribution, we can calculate the entire frequency response of a particular optical structure by detecting the temporal evolution of the field behind the structure.
The contributions of the different frequencies are then separated by taking the Fourier transform of the time series of the field.
For a high resolution in the wavelength domain, we have to record the temporal evolution of the field for a long time (
tN = total number of time
steps) long computing time
Example of a grating waveguide coupler
Periodically structured slab waveguide corresponds to one-dimensional Photonic Crystal waveguide (n1=1.58, n2=1.87, d1=d2=165 nm, TE).
Transmission spectrum - The dips are waveguide resonances that are excited if the transverse k-vector (i.e. the momentum) provided by the grating matches the propagation constant of a waveguide mode.
t
EInc
f
EInc
Tra
nsm
issio
n?
d1
n1
d2
n2
x
z
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 69
Field distributions obtained for calculations with cw-excitation at frequencies of the dips in the spectrum, corresponding to guided modes in the grating waveguide.
4.7 Dispersive and nonlinear materials FDTD is not directly applicable for materials with 0 as e.g. metals
Material properties depend strongly on the wavelength (dispersion)
Inclusion of nonlinear response (instantaneous or non-instantaneous) of the material
There exists a great diversity of approaches, but usually all of them require the simulation of the dynamics of an additional quantity
So far we have taken into account E and H
Now we include: j current density, P polarization, and D displacement
Metals
Maxwell's equations
0 0
( , ) ( , )rot ( , ) , rot ( , ) ( , )
t tt µ t t
t t
H r E rE r H r j r
in addition we need an equation relating current and electric field (Drude model to describe mean velocity of electrons driven by an electric field)
2
0 Pt
jj E
with P the plasma frequency and 1 / is governed by the relaxation time
This is usually solved in the frequency domain as
2
P( ) 12 (1 2 )
i
i
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 70
Epsilon function of a metal described by the Drude model
Dielectrics
Assuming a 2D geometry (y-z plane) with TM polarization
ˆxE xE
The dielectric material response results in an induced polarization
ˆxP xP
which can be approximated by an Lorentz model similar to Drude model but with non-zero resonance frequency
Lorentz dispersion in time domain: 2
2 2
0 0 0 L2
x xx x
P PP E
t t
Solution can be derived in frequency domain.
Lorentz dispersion in frequency domain: 2
0 0 L, ,2 2
0
x xP Ej
(see also R.W. Ziolkowski et al., JOSA A, Vol. 16, No. 4, 980)
Nonlinear materials
Maxwell's equations
makr
ext
( , ) ( , )rot ( , ) , rot ( , ) ( , ),
div ( , ) ( , ), div ( , ) 0,
t tt t t
t t
t t t
B r D rE r H r j r
D r r B r
Linear material response: D E
Nonlinear material response: 0 D E P
with (1) (2) 2 (3) 3
0( ...) P E E E
Example: instantaneous Kerr (3) nonlinearity
0 D E with intensity dependent refractive index
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 71
2 22 2 2
0 2 0 0 2( ) 2n n n E n n n E
Refractive index depends on the square of the E-Field (for 2 0n n )
Problem: implicit equation and solution by Newton iterative procedure necessary
4.8 Boundary conditions Problems appear if the fields at the boundary have to be evaluated.
For keeping the discretized mesh treatable on a computer, we have to limit its size.
But for a proper determination of the field components that are positioned directly at the boundary of the computational domain, we need actually information about field components outside.
Choosing proper boundary conditions
4.8.1 Perfectly conducting material boundary
The easiest boundary condition is again to assume a perfectly conducting material (for E or H ), which surrounds our computational domain. Then the field cannot penetrate the structure.
Setting the field values outside the structure equal to zero
1 12 2
1 1 1 1 1 1 1 12 2 2 2 2 2 2 2
1 1 1 12 2 2 2
; , ; , ; , ; , ; , ; ,
, ,
n n n n n n
x j k x j k y j k y j k z j k z j k
j k j k
t tH H E E E E
z y
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 72
Grid size for Perfect Electric Conductor (PEC)
12
12
12
, 1, 1 1, ,
1, , 1 , 1,
1, 1, , , 1
n n
x x y z x x y z
n n
y x y y x y z
n n
z x y z x y z
E N N N H N N N
E N N N H N N N
E N N N H N N N
Index notation for Perfect Electric Conductor boundaries
:,1,: 0, :,:,1 0
:, 1,: 0, :,:, 1 0
x x
x y x z
E E
E N E N
1,:,: 0, :,:,1 0
1,:,: 0, :,:, 1 0
y y
y x y z
E E
E N E N
1,:,: 0, :,1,: 0
1,:,: 0, :, 1,: 0
z z
z x z y
E E
E N E N
4.8.2 Floquet-Bloch periodic boundaries
Floquet-Bloch periodic boundaries have to be applied for periodic objects, like e.g. gratings, photonic crystals.
Floquet-Bloch boundaries in the frequency domain:
( , , ) ( , , )exp(i )exp(i )x y x x y yx m y n x y k m k n
For 1m and 0n this results in
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 73
( , , ) (x, , )exp(ik )x xE x y t E y t
Incident plane wave (arbitrary propagation direction):
Incid. exp( )exp( )exp( )exp( )x y zE ik x ik y ik z i t
Example: Floquet-Bloch periodic boundaries applied for calculating the band
structure of a Photonic Crystal (PC)
Left: Periodic geometry of a 2-dimensional photonic crystal slab, consisting of air holes in an otherwise homogeneous material. The red lines indicate the
periodic boundary with a particular xk and yk . The spatial x and y coordi-
nates are measured in numbers of the lattice period a of the photonic crystal. Right: Unit cell of the photonic crystal, which is the size of the computational domain being considered in the numerical simulations.
Lunching an arbitrary field distribution and recording the evolving pattern
on some discrete points in the space
Initial field distribution with random spatial distribution to avoid selective excitation of states with particular symmetry.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 74
Time evolution of the field at some specific spatial position which is recorded during the numerical simulation. The underlying photonic crystal structure was
taken as in the previous picture with air holes of radius 0.3R a in a material
with dielectric constant 13 .
All the frequencies which do not satisfy the periodic boundaries are annihilated and only the modes that are allowed to propagate persist
Spectra obtained by an Fourier Transform of the recorded time series from the previous figure.
Scanning the k-space and tracing the frequencies that persist as modes
delivers the band structure via FDTD
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 75
Left: Calculated temporal frequency spectrum for one particular Floquet-Bloch periodic boundary condition. Right: Band structure computed by tracking the peak positions in the temporal frequency spectrum over a variation of the wave vectors defined by the boundary conditions.
4.8.3 Transparent boundary conditions - 2nd order Mur (TBC)
For simplicity of the explanation we neglect the vectorial aspect of the field equations. Then each field component obeys approximately the scalar wave equation:
2 2 2 2
2 2 2 2 2
10
f f f f
x y z c t
In an operator notation this wave equation can be written as
0Lf
with the operator
2 2 2 22 2 2 2 2
2 2 2 2 2
1x y z tL c
x y z c t
This operator can be decomposed into a product of operators
0Lf L L f
If the two operators describe the waves propagation in the forward ( x ) and backward ( x ) direction they read as
1 21x tL c S wave propagating in the x -direction
1 21x tL c S wave propagating in the x -direction
with
X
M
irreducible Brillouin zone
k
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 76
2 2
2
1 1
y z
t t
Sc c
Then the equation
0L f
would describe the backward wave propagation in the x -direction.
Reference: Engquist-Madja exact ABC [Mathematics of computation 31, 629, (1977)]
The direct implementation of the operator is not possible, but the square root
can be expanded as a Taylor-series:
2 1/2 2 412
(1 ) 1 ( )S S O S
First order approximation
2 2
2
1 10
y z
t t
Sc c
nearly plane wave propagating in the x-direction as:
1 0x tL f c f
Second order approximation
2 1/2 212
(1 ) 1S S
resulting in a backward operator
2 212
tx y z
t
cL
c
and eventually in a wave equation in backward (-x) direction
2 2 2 1 212
0xt yy zz ttL f f c f f c f
Writing the differential operators as finite differences [G. Mur, IEEE Trans. Electromagnetic Compatibility 32, 377 (1981)]
Discretizing the operator a half spatial step in front of the boundary (example of the boundary at x=0)
1 1 1 1
1, , 0, , 1, , 0, ,2
1/2, ,
1
2
n n n nn j k j k j k j k
xt j k
f f f ff
t x x
Averaging the second order time derivative at 0x and x x
1 1 1 1
0, , 0, , 0, , 1, , 1, , 1, ,2
2 2
2 21
2
n n n n n n
j k j k j k j k j k j k
t
f f f f f ff
t t
first order approx. second order approximation
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 77
Same holds for the second order derivatives in y and z direction
Inserting all those difference schemes leads to
1 1 1 1
0, , 0, , 1 1, , 0, , 2 1, , 0, ,
3 0, 1, 0, , 0, 1, 1, 1, 1, , 1, 1,
3 0, , 1 0, , 0, , 1 1, , 1 1, , 1, , 2
2 2
2 2
n n n n n n
j k j k j k j k j k j k
n n n n n n
y j k j k j k j k j k j k
n n n n n n
z j k j k j k j k j k j k
f f k f f k f f
k f f f f f f
k f f f f f f
with
1
c t xk
c t x
, 2
2 xk
c t x
, and
2
3 2
( )
2 y ( )y
c t xk
x t x
The example above was derived for the boundary at 0x . Similar equations are used for the other boundaries.
Fields have to be stored for 2 different time steps
Simplification by using only the first order Taylor approximation (skipping the derivatives along the y and z directions)
1 1 1 1
0, , 0, , 1 1, , 0, , 2 1, , 0, ,
n n n n n n
j k j k j k j k j k j kf f k f f k f f
Properties
Only the fields components that are evaluated at this (most outer) boundary have to be updated with this equation (the tangential components of the E-field e.g. at the boundary 0x )
Reflection coefficients are in the order of 10-2
Easy to implement
4.8.4 Perfectly matched layer boundaries (PML)
Most efficient currently known boundary conditions
Published by Berenger in 1994 with reflections about 3x103 times less than with 2nd order Mur boundary
Basic idea: constructing a media which absorbs light and at whose interfaces with the region of interest no reflections take place Impedance matching of the medium with the surrounding
Impedance of a medium having both electric and magnetic conductivity
l
i
i
with the electric conductivity
e
and the magnetic conductivity
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 78
m
Let the loss-less region being region 1, characterized by and
Impedance:
If the condition e m
holds, the impedances are equal:
l
Plane waves propagating in each of the regions are characterized by their
propagation constants, given as (by choosing and )
kc
(dispersion relation)
resulting in
e m1 1k i ic
And the velocity of propagation is given by
ek ic
the same as in free space but light is additionally absorbed
The reflections coefficients of the interface are given generally by
r2 i 1 t
i
2 i 1 t
cos cos
cos cos
ER
E
and
r
2 t 1 i
i
2 t 1 i
cos cos
cos cos
ER
E
with
1 i 2 tsin sinn n
reflection is zero, only if
angle of incidence = angle of transmittance
The following derivation is done as an example for 2D FDTD in TE polarization [Jean-Pierre Berenger, "A perfectly matched layer for the absorption of electromagnetic waves," Journal of Computational Physics 114, 185-200 (1994) (2D)]
e
e
x zx
y zy
E HE
t y
E HE
t x
m
yz xz
EH EH
t x y
absorb light only in one direction, fields shall propagate parallel to the surface
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 79
Berenger’s idea: splitting the H field into a x and y component
z zx zyH H H
(x-derivative of the E field drives the zxH component and vice versa)
Introduction of an anisotropy of all properties
e
e
zx zyxy x
zx zyy
x y
H HEE
t y
H HEE
t x
m
my
yzxx zx
zy xzy
EHH
t x
H EH
t y
Important: waves propagating along the y-axis are not absorbed in the x-boundary
Splitting the field in the boundaries
Adding the electrical and magnetic conductivity to the equations as material parameters
Only the components propagating normal to the boundary are absorbed by adjusting the proper absorption for each component
It can be shown, that the impedance of the Berenger medium equals the impedance of the free space, regardless of the angle of propagation
Problem with sudden change in the material properties: electric and magnetic field are evaluated at different spatial positions (staggered grid)
asymmetric absorption
Choosing an appropriate absorption profile for ex and
mx
Polynomial scaling
e max
m
ex
x
d
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 80
Practical considerations: 2 << m << 6 and 10d x 1610R
Electric and magnetic fields are evaluated at half discretization step apart, hence the absorption profile is evaluated likewise at different spatial coordinates
Extensions for TM polarization and 3D is straight forward
[Jean-Pierre Berenger, "Three dimensional perfectly matched layer for the absorption of electromagnetic waves," Journal of Computational Physics 127, 363-379 (1996) (3D)]
e m
e m
e m
e
y xz zx zy y xy zx zy
z xz yx yz z xz yx yz
x yx zx zy x yx zx zy
z yz xy xz
E H H H E Et y t x
E H H H E Et z t z
E H H H E Et x t x
E H Ht z
m
e m
e m
z yz xy xz
x zx yx yz x zx yx yz
y zy xy xz y zy xy xz
H E Et z
E H H H E Et x t x
E H H H E Et y t y
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 81
5. Fiber waveguides
5.1.1 The general eigenvalue problem for scalar fields
Restrictions:
weakly guiding structure ( 1n ln ( , ; ) 0x y )
linearly polarized field
Scalar wave equation for a scalar mode field u propagating in the z-direction
2 2
t ( , ) ( , , ) ( ) ( , ) 0u x y k x y u x y
The mode solution ( , )u x y can be assumed to correspond to the power
density u EH and it has two important properties:
u and 'u must be continuous and bound, since the second derivative of u must be finite due to the finite refractive index discontinuities.
u and 'u converge to zero as ,x y , since physically relevant
fields should have finite energy.
These properties determine the eigenvalue problem for the propagation
constant of guided modes.
The discrete solutions are called guided modes.
5.1.2 Properties of guided modes
Orthogonality and normalization
Two solutions au and
bu of the scalar wave equation with the respective
propagation constants a and
b read as
2 2
t ( , , ) ( ) ( , ) 0a ak x y u x y
2 2
t ( , , ) ( ) ( , ) 0b bk x y u x y
By multiplying the first equation with bu and the second with
au and
subtracting the results one obtains
2 2
t ta b a b b a a bu u u u u u
Integration over the infinite cross section of the waveguide A
2 2
t ta b a b b a a bA A
u u dA u u u u dA
Using Green’s law to transform the surface integral on the right side into a line integral, one can show that the right side is equivalent to zero, since the mode field is assumed to tend to zero at infinity.
Hence only the part at the left hand side remains:
2 2 0a b a bA
u u dA
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 82
To fulfill the equation for different the integral must be zero
,a b a bA
u u dA
This requires the normalization of the fields according to
2 1
Au dA
For cylinder symmetric problems we will see later that this reads as
,( ) ( )a b a br
Y r Y r r dr
Phase velocity
The propagation velocity of the phase fronts is called phase velocity
p
2 cv
with as the vacuum wavelength and c as the speed of light in vacuum
Group velocity
The propagation velocity of the energy is called group velocity
g 2
2 cv
which can be calculated very accurately by the following approach:
Take two solutions at two different wavelength and
2 2
t ( ) ( ) ( ) ( ) 0k u
2 2
t ( ) ( ) ( ) ( ) 0k u
Do the same as done for deriving the orthogonality relation: multiply the first equation with ( )u and the second with ( )u and subtracting the results one
obtains
2 2 2 2
t t( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )k k u u u u u u
As before we integrate over the infinite waveguide cross section A and transform the right side into a line integral using Green’s law. Similar to the demonstration of the orthogonality of the individual modes above, we can show that the right side is equivalent to zero. Then the remaining left side reads as
2 2 2
2 2
( ) ( )( ) ( ) ( ) ( ) 4 ( ) ( )
A A
u u A u u A
Dividing by ( ) results in
2 2 2
2 2
( ) ( ) 4 ( ) ( )( ) ( ) ( ) ( )
A A
u u A u u A
Now we can introduce the limit or ( ) 0 and obtain
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 83
2 2 2
2
1 ( ) ( )( ) 4 ( )
( )A A
u A u A
Using this expression to substitute in our definition for the group
velocity we obtain
22
g 22
2
( )4 ( )
( )( )
A
A
u Acv
u A
Group velocity dispersion
The dependence of the group velocity on the wavelength is expressed by the Group velocity dispersion coefficient
g
1
vD
5.1.3 Cylinder symmetric waveguides
In particular for fibers the refractive index profile of the waveguide is very often cylinder symmetric
2( , ) ( ) ( )x y r n r
For fibers the index profile is then usually characterized by special parameters
Fiber parameter
co cl
1/ 22 22 a
V n n
with
a - core radius
con - maximal refractive index in the core
cln - constant refractive index in the cladding surrounding the core
Normalized index difference
co cl
co
2 2
2
1
2
n n
n
Index profile
co
2 2 1 2 ( )n n f r
with
1
( )1
r af r
r a
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 84
Index dispersion
To describe the wavelength dependence of SiO2, which is the standard fiber material, often the Sellmaier equation is used
2
2
21
( ) 1j
i
i i
an
b C
which takes into account electronic transitions in the UV and molecular vibrations in the IR by
2
14
1
10
2
14
1
10
2
1.53714322m
1.968198 10
1.4108754 10
1.790244 10
1.572513 10
C
a
a
b
b
5.1.4 Bessel’s differential equation
Using the Laplace operator in cylindrical coordinates
2
t 2 2
1 1( , )
u uu r r
r r r r
the scalar Helmholtz equation reads as
2
2 2
2 2
1 1( ) 0
u ur k r u
r r r r
Applying the product Ansatz
( , ) ( ) ( )ru r u r u
the scalar cylindrical Helmholtz equation reads as
222 2
2 2 2
1 1( ) 0r r
r r
uu uu u u k r u u
r r r r
multiplying with 2 /( )rr u u the equation can be separate in terms depending
only on r and terms depending only on
22 22 2 2
2 2
1( )r r
r r
ur u r ur k r
u r u r u
Since both sides of the equations are independent, the equation can be
solved non-trivially only when both sides are equivalent to a constant 2m ,
which splits the equation in two independent equations
2 22 2
2 2
2
2
2
1( ) 0
0
r rr
u u mk r u
r r r r
um u
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 85
The r -equation is Bessel’s differential equation, the solution of which will be discussed in detail below.
The -equation can be solved analytically very easily as
0cos sin cos( )u A m B m C m .
To obtain a continuous field the so-called azimuthal mode constant m must obey
m
Otherwise unphysical azimuthal discontinuities of the field would result.
Bessel’s differential equation can be simplified further by introducing the dimensionless modal parameters
co
1/ 22 2 2U a k n in the core
cl
1/ 22 2 2W a k n in the cladding
which are connected by
2 2 2V U W
With co
2 2 1 2 ( )n n f r we can express 2 2 2k n as
co co
2 2 2 2 2 2 2 22 ( )k n k n k n f r and hence 2 2
2 2 2
2
( )U V f rk n
a
When introducing additionally the normalized radius /r a we can rewrite
Bessel’s differential equation for ( ) ( )m rY u r in its final form
2 2
2 2
2 2
( ) 1 ( )( ) ( ) 0m m
m
Y Y mU V f Y
5.1.5 Analytical solutions of Bessel’s differential equation
Bessel’s differential equation can be solved analytically for some specific profiles ( )f or in segments where ( )f is constant. Even though we are
interested in the numeric solution of the general problem, we need particular analytic solutions to derive some general properties of the numeric solution, as e.g. the boundary conditions.
Introducing
2 2 2( ) ( )G U V f
the analytic solution in an interval with constant refractive index (here taken as the index in the center of the core ( ) (0) const.f f ) reads as
( ) ( (0))m mY AJ G
with A being an arbitrary scaling constant and ( )mJ being the Bessel
function of the first kind and mth order.
Besides ( )mJ there are other mathematical solutions of Bessel's differential
equation. However these other solutions can be neglected in the
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 86
representation of the physical fields in the core area. They can be ruled out by physical reasoning due to their infinite value at the origin or their unboundednes for growing arguments, which would result in infinite fields
( )mY for growing radius .
Similarly the solution in the cladding, where we take ( ) 1f , is based on the
modified Hankel function
2( ) ( )m mY BK W where 1
with B being an arbitrary constant.
Using the requirement that the general solution and its derivative need to be continuous throughout the computed domain, we can derive the following
conditions at the core-cladding interface ( 1 ), which can be written as a set
of linear equations using the arbitrary integration constants A and B
continuity condition for the field: 2( 1) ( ( 1)) 0m mAY BK W
continuity condition for the 1st derivative: 2( 1) ( ( 1)) 0m mAY BWK W
The determinant of this set of linear equations must vanish to obtain unambiguous solutions
2 2 2( 1) ( ( 1)) ( 1) ( ( 1)) ( ) 0m m m mWY K W Y K W F U
This is very elegant, since it transforms the physical boundary condition, which says that the field has to decay to zero at an infinite radius ( ), to
a finite radius at the core cladding interface ( 1 ).
This characteristic equation 2( )F U is only solved for discrete values of the
parameters W or U (W and U are connected) which selects a finite number
of modes. Usually this equation is solved as a function of 2U which can vary
in the interval
2 20 U V .
Example of 2( )F U in the interval (0, )U V for a multimode fiber with large
core and step index profile with azimuthal mode number 0m .
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 87
The modes are often categorized in their linearly polarized form as LPmp -
modes, where the index m corresponds to the azimuthal mode number and index p selects the modes according to order of individual solutions of the
above equation. p is therefore often called the radial mode number.
LPmp -modes: m - azimuthal mode number 0,m m
p - radial mode number 1,p p
5.1.6 Specifying the numerical problem
Here we would like to solve Bessel's differential equation for inhomogeneous index profiles using numerical methods
2 2
2 2
2 2
( ) 1 ( )( ) ( ) 0m m
m
Y Y mU V f Y
The numerical problem consists in finding discrete modal solutions for a given azimuthal mode index m with an unknown eigenvalue U .
First we need to define the boundary conditions. Starting from the core's
center at 0 we assume that solution ( )Y for 0 behaves like the
Bessel function ( ) ( (0))m mY AJ G of the analytic solution in the previous
section. Since the Bessel function can be expressed as a series like
2
0
( 1)( )
!( 1) 2
n mn
m
n
Jn n m
we can introduce an approximation for small arguments ( 0 ) and use it as
an ansatz function:
( ) ( )m
mY X
where ( )X is an analytic function which is finite valued for finite arguments
. From this ansatz we can also determine the derivative of the solution as
( 1)( ) ( ) ( )m m
mY m X X
Hence the boundary conditions in the center read as
0 : (0) (0)
(0) (0)
1: (0) 0
(0) (0)
1: (0) 0
(0) 0
m Y X
Y X
m Y
Y X
m Y
Y
As was demonstrated in the “Analytic solution” section, the boundary conditions at infinity ( ) can be transformed to the position of the core
cladding interface as
2 2 2( ) ( 1) ( ( 1)) ( 1) ( ( 1)) 0m m m mF U WY K W Y K W
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 88
5.1.7 Solving the second order singularity
The governing Bessel’s differential equation
2 2
2 2
2 2
( ) 1 ( )( ) ( ) 0m m
m
Y Y mU V f Y
possesses a second order singularity at 0 which makes it very hard to
integrate numerically.
Using again the ansatz function
( ) ( )m
m mY X
we obtain a new differential equation for ( )X as
2
2 2
2
( ) 2 1 ( )( ) ( ) 0m m
m
X m XU V f X
(24)
which possesses only an ordinary 1st order singularity at 0 .
To use this equation for ( )X instead of the previous equation for ( )Y , for
identifying the modes, the boundary conditions, which we had derived above for ( )mY must be transformed to boundary conditions for ( )X .
Since ( )X is an analytic function it can be expressed as a power series
following the series representation of the Bessel function as
0
( ) n
m n
n
X c
Hence the new boundary conditions in the center of the core, i.e. at 0 , are
independent of m
0(0)X c and (0) 0X
Similarly the boundary condition at the core-cladding interface, i.e. at 1 ,
becomes
2 ( )( ) ( 1) ( 1) 0
( )
mm m
m
K WF U W m X X
K W
with 2( )F U being a characteristic function, which has to be solved for its
roots.
However, we should keep in mind that, once found, the new solution ( )mX
has to be transformed back to ( )mY to derive some physical meaning.
5.1.8 Numerical integration methods
Up to now the problem is a boundary value problem which would require e.g. a relaxation method for its numerical solution in order to find the desired
unknown parameter 2U .
Using the shooting method the problem can be transformed into an initial
value problem starting at 0 which depends on the parameter 2U . Different
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 89
2U result in different solutions at 1 and hence also different 2( )F U . The
aim must be to find parameters *U U with
*2( ) 0F U
Then the solutions of the initial value problem would also solve the boundary value problem.
When selecting a specific method for the numerical integration of the differen-tial equation (24) for ( )mX we have to take into account the following
properties of the problem:
At 0 the coefficient in front of function ( ) /mX has a first-order
singularity, which depending on m can lead to instabilities of the numeric solution.
The profile function ( )f can possess discontinuities in the integration
interval, which can reduce the order of the integration methods.
The coefficient in front of ( )mX can go through zero in the integration
interval, which can change the continuity properties of the solution.
Hence to ensure stability of the numerical method one should refer to implicit integration methods, since they are often unconditionally more stable than explicit integration methods.
Furthermore we should reformulate the problem in a notation which is compa-tible to standard numerical schemes. Hence, we transform the second order differential equation into a set of coupled first order differential equations with
( )( )
( )( )( )
I
m
IIm
XZ
XZ
Z
as
2 2
( )( )
( ) 2 1( ) ( ) ( )
III
III II
ZZ
Z mV f U Z Z
In matrix representation the differential equation reads as
2 2
0 1( )
( ) ( , )2 1( ) jm
V f U
ZZ L Z (25)
Now we have to discretize the integration interval [0,1] , e.g. with very
simple equidistant spacing h as:
1
| , 0(1) ,h j j jh j N hN
then the discrete approximation of the solution can be written as
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 90
( ),j j j h Z Z
Implicit Runge-Kutta-Midpoint-Method of 2nd order
The midpoint-method, which is a second order accurate Runge-Kutta-method, has the following Runge-Kutta parameter table (see e.g. the explanation of the Butcher tableau in the Wikipedia article on Runge-Kutta methods)
0.5 0.5
1 1
This particular Runge-Kutta method is a single step implicit method. Using this method a single integration step in the solution of an differential equation reads like
1 ( , , )j j j jh h Z Z Φ Z , ( 0,1,..., )j N
where ( , , )j j hΦ Z approximates the ascend of the solution as
1
( , , ) ( , , ),2 2
j j j j j j
hh h h
Φ Z L Z Φ Z
when L is the matrix operator, describing the differential equation.
Using this method to solve our particular problem, where L is defined as in equation (25), gives for the approximation of the ascend of the solution
12 2
2
0 1
22 1( , , )2
22
I I
j
j j jII I
jj
hZ
h mh V f Uhh
Z
Φ Z .
Since we are dealing with a linear problem we can avoid the otherwise necessary iterative numerical solution of the above problem and solve it explicitly
2 2
1 22 2
2 2 2 2
1 22
2 2
2 11
2 4 2
2
2 1
2 2 2
2
2 11
2 4 2
2
I II
j j j
I I
j j
j
j
I II
j j j j
jII II
j j
j
j
h hV f U Z Z
Z Z hh m h h
V f Uh
h h h mV f U Z V f U Z
h
Z Z hh m h h
V fh
2U
The solution accuracy of the method is second order in h , as was already said in the introduction of the midpoint method. The discretization grid
h
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 91
should be chosen such that the discontinuity points D of the profile function
( )f coincide with grid points of h
D h
This ensures that the differential equation is only evaluated in continuous intervals at
D 0.5x h . Consequently the second order solution accuracy is
preserved also for discontinuous index profiles ( )f .
Implicit Runge-Kutta-Method of 4th order
To improve the accuracy one could refer to more accurate higher order numerical integration schemes. One such example is the Gaussian type fourth-order method. We are not describing the details here, but give only the respective Butcher scheme
1 11 12
2 21 11
1 2
1 3 1 1 3
2 6 2 2 6
1 3 1 3 1
2 6 2 6 4
1 11
2 2
5.1.9 Eigenvalue search
Since we have transformed the original boundary value problem into an initial
value problem using a shooting method, the search for the eigenvalues 2U is
a simple search for the roots of 2( )F U for parameter m . However, for each
calculation of 2( )F U we have to compute the full differential equation
problem in the interval [0,1] .
The eigenvalue interval for guided modes is restricted to
2 20 U V
containing a finite number of roots. To find all roots the given interval can be
scanned for sign changes of 2( )F U with a fixed step size 2U . The resulting
subintervals can then be refined using e.g. a converging secant method.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 92
5.1.10 Calculation examples
Example of 2( )F U in the interval (0, )U V . This particular example was
calculated for a fiber with a step index profile having a large diameter. This
results in a large fiber parameter V with a value around 52. The calculation was
done for 0m , which restricts to the modes having no azimuthal dependence of the fields.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 93
Radius-dependent field distribution for azimuthal index m=0 (=804nm)
Radius-dependent field distribution for azimuthal index m=1 (=804nm)
Radius-dependent field distribution for azimuthal index m=70 (=804nm)
Examples of calculated radial dependence of the mode field for a ring index
fiber for different azimuthal mode indices m , which means different numbers of
oscillation periods of the fields around the azimuthal angle . The particular
ring index fiber which was simulated in these graphs, is similar to a strongly guiding step index fiber with a radius of 30 µm. Additionally, it has a ring-shaped higher index area extending from a radius of 21 µm to 25 µm. The wavelength for the calculation was set to 804 nm.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 94
Examples of calculated intensity distribution of the mode field for a ring index fiber as in the previous figure. Here the full radial and azimuthal dependence of the intensity is shown for two particular modes at two different wavelengths.
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 95
6. Fourier Modal Method for periodic systems
Aim
rigorous solution of the diffraction of light at periodic structures (gratings)
Methods
A) thin element approximation
B) formulating the eigenvalue problem in the Fourier space and deriving boundary conditions imposed onto the modes in order to be matched properly to the homogenous space
Examples of diffraction gratings
1D grating 2D grating (biperiodic)
Gratings are characterized by a periodic variation of the dielectric constant in x and y direction with periods x and
y
Illustration of the problem by understanding a grating as a periodic waveguide array embedded between two homogenous media
Dielectric waveguide with rectangular cross section Periodically arranged dielectric waveguide with rectangular cross section
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 96
6.1 Formulation of the problem in 2D for TE
Rigorously calculated field distribution around a binary grating for plane wave TE excitation and sketch of the diffraction problem
Incident field
inc, 0 Iexp (sin cos )yE ik n x z with being the angle of incidence
Diffracted field in different diffraction orders i
I, inc, I,exp ( )y y i xi zi
i
E E R i k x k z for 0z
II, II,exp ( ( ))y i xi zi
i
E T i k x k z d for z d
Remaining question: What is the direction and strength of the reflected and transmitted amplitudes?
The wave number of each wave is given by
2 2
0
2x zk n k k
The incident wave is characterized by
I
0
2coszk n
I, I
0
2sinxk n
I
The grating provides a momentum for each diffraction order of
2
xik i
with i being the index of the diffraction order
Hence the diffracted waves are characterized by
0 I
2sinxik i k n
,
2 2 2
I/II, 0 I/IIzi xik k n k
The propagation directions of the diffraction orders are only determined by the grating period , the angle of incidence and the refractive indices
I,IIn of the surrounding media.
The character of the diffraction orders is determined by 2 2 2
0 I/IIxik k n propagating wave
2 2 2
0 I/IIxik k n evanescent field
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 97
Illustration of the diffraction orders on the Ewald sphere with:
red xik , green I,zik , blue ,II zik . This illustration is for the simplifying case of
normal incidence.
Illustration of the diffraction orders on the Ewald sphere for oblique incidence.
6.2 Scalar theory for thin elements The field after the structure is given by the incident field multiplied by the transfer function
T 0, , ,U x y T x y U x y with ˆU E y
kx
kz
incident
01
2
-1-2
-3
01
2
3
-1-2
-3
-4
nI
nII
G
kx0
x
z
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 98
, , exp ,T x y T x y i x y
For a pure phase grating the amplitude transmission function equals unity and the phase is given by
0 II I( ) ( )x k n n f x
Hence for perpendicular plane wave illumination the field after the structure is given by
exp i x
The amplitudes of the plane waves in the direction of the diffraction orders are given by the Fourier transform of the transmission function
0 II I0
exp ( ) exp expn xnT FT i x ik n n f x ik x dx
Physical interpretation of diffraction efficiency
The diffraction efficiency corresponds to the energy transferred into a diffraction order normalized to the incident energy (given by the Poynting vector) S E H
Introducing the plane waves into the equation leads to
2II,I
T,
II 0
zi
n n
kpT
p k
with
I/II 1p for TE and I/II I/IIp for TM
2I,
R,
0
zi
n n
kR
k
Energy must be conserved for loss-less materials!
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 99
Examples for the validity of the thin element approximation
Rigorous simulation of wave propagation through a sinusoidal long period
grating ( 10 ), which is illuminated with a plane wave. Since the grating
period is much larger than and the depth of the grating is much smaller than its period, diffraction effects within the grating have only a negligible influence. This can be verified since no transverse redistribution of the wave's intensity within the volume of the grating is observed in the rigorous simulation. Hence the thin element approach is justified.
Rigorous simulation of wave propagation through a sinusoidal short period
grating ( 2.1 ), which is illuminated with a plane wave. Since the grating
period is in the range of and the depth of the grating is also in the order of its period, diffraction effects within the grating cannot be neglected. This can be verified since a transverse redistribution of the wave's intensity within the volume of the grating is observed in the rigorous simulation. This is a clear signature of diffractive effects inside the grating volume. Hence the thin element approach would fail.
Amplitude Phase
nI=1
nI=2
nI=1
nI=2
nI=1
nI=2
nI=1
nI=2
Amplitude Phase
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 100
Limitations of the scalar method
Phase of the field directly after a sinusoidal grating with different grating periods
(grating profile with depth h , I 1n ,
II 2n , TE illumination).
Scalar theory fails if the period gets comparable to the wavelength and the thickness of the grating is significant.
Proper description of the field inside the grating (0 z d ) is necessary.
Maxwell’s equations must be solved also inside the grating!
6.3 Rigorous grating solver
Ansatz
expanding the fields in the different domains in terms of modes (mode = eigensolution of a wave equation)
Modes are just plane waves in the incident and transmitted region.
In grating region the modes are waveguide modes propagating along z , that fulfill the periodic boundary conditions in the transverse direction.
The mode profile must be determined numerically (analytical solutions can be used in special cases).
deriving the proper boundary conditions and solving for the unknown amplitudes of each mode
comparable to the plane interface problem
tangential electric and magnetic field are continuous + transversal wavevector component
Outline of the algorithm
calculate all wave vector components of interest
Fourier transforming the permittivity distribution
calculating the eigenvalues and the eigenvectors of the eigenmodes as supported by the structure in Fourier space
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 101
solving the system of linear equations that provide the amplitudes of all relevant field components
calculating derived quantities of interest, such as diffraction efficiency and/or field distributions in the plane of interest
Comment: Note that the algorithm thus far requires invariance of the structure in the propagation direction. However this restriction can be overcome by decomposing a longitudinally varying structure into a system of thin layers, where in each individual layer the structure can be assumed to be longitudinally invariant. But keep in mind that the structures in each layer must be periodic sharing the same period.
6.3.1 Calculation of eigenmodes in Fourier space
starting from Maxwell’s equations for a time harmonic field with the magnetic field scaled by the impedance
0 0( ) ( ) ( ) ( ) ( )ik ik E r H r H r r E r
plugging equations together and eliminating the z-component of the field (unambiguously determined by the divergence equations)
0 0 ( )y z x y x zE E ik H H H ik Ez y x y
r
results in the following dynamic equations
0
0
1 1y y x xE H H ik H
z ik y x y
0
0
1 1x y x yE H H ik H
z ik x x y
0
0
1y y x xH E E ik E
z ik y x y
0
0
1x y x yH E E ik E
z ik x x y
(For details see Li et al., Phys. Rev. E 67, 046607 (2003).)
Fourier expansion of all quantities
To solve the eigenvalue problem derived from the above equations all quantities are transformed into Fourier space: ( ) ( )E r E k , ( ) ( )H r H k ,
( ) ( ) r k , and also 1 1( ) ( ) r k .
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 102
, ,
, ,
1 1
( ) exp
( ) exp
( ) exp
( ) exp
ij ij x ij y
ij
ij ij x ij y
ij
ij ij
ij
ij ij
ij
z i k x k y
z i k x k y
i
i
E r E
H r H
r G r
r G r
with , , 0 0 1 2( , ) ( , )ij ij x ij y x yk k k k i j k b b and
1 2ij i j G b b
The integers i , j run from to , which has to be truncated for numerical
calculations.
Now these Fourier expansions are inserted into the dynamic field equations, here for the following example
0
0
1x y x yH E E ik E
z ik x x y
the individual terms are substituted
, , ,expx mn x m x n y
mn
H H i k x k yz z
, , ,
, , , ,
exp
exp
y mn y m x n y
mn
m x mn y m x n y
mn
E E i k x k yx x
ik E i k x k y
, , , , ,exp expy pq p x q y mn y m x n y
pq mn
E i G x G y E i k x k y
now all terms are multiplied with , ,exp i x j yi k x k y
and integrated over x
and y across one unit cell as
, , , ,exp expi x j y m x n y im jni k x k y i k x k y dxdy
This results in the following example
Real space Reciprocal space
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 103
, , , , , ,
,
exp expmn x i x i y m x n y mn x im jn
mn mn
ij x
H i k x k y i k x k y x y Hz z
Hz
which has to be done with all terms
, , , , , ,
, , , , , ,
exp exp exp
exp ( ) ( )
m x n y p x q y i x j y
m x p x i x n y q y j y i m p j n q
i k x k y i G x G y i k x k y dxdy
i G G G x G G G y dxdy
This results in the following set of equations
, 1
, , , , , , 0 ,
0
, 1
, , , , , , 0 ,
0
,
, , , , , 0 , ,
0
,
,
0
ij x
ij x i m j n m x mn y n y mn x ij y
mn
ij y
ij y i m j n m x mn y n y mn x ij x
mn
ij x
ij x im jn m x mn y n y mn x i m j n mn y
mn mn
ij y
ij y
ikE k H k H ik H
z k
ikE k H k H ik H
z k
ikH k E k E ik E
z k
ikH
z k
, , , , 0 , ,im jn m x mn y n y mn x i m j n mn x
mn mn
k E k E ik E
.
These equations can be conveniently written in matrix form as
1 2z z
E TH H T E
where
, ,
, ,
ij x ij x
ij y ij y
E H
E H
E H
and
1 1 2
, ; , , ; , 0 ;;
1 1 2 1
, ; , 0 ; , ; ,0
2 1
, ; , , ; , 0 ;;
2 2 1
, ; , 0 ;0
ij x ij mn mn y ij x ij mn mn x ij mnij mn
ij y ij mn mn y ij mn ij y ij mn mn x
ij x ij mn mn y ij x ij mn mn x ij mnij mn
ij y ij mn mn y ij mn i
k k k k kiT
k k k k kk
k k k k kiT
k k k kk
, ; ,j y ij mn mn xk
This set of two coupled 1st order differential equations for E and H can be combined into a single 2nd order equation for E
2
1 22z
E TT E
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 104
For z-invariant geometries this differential equation is transformed into an eigenvalue equation using the ansatz
(z) exp( )i z E E
This results in the eigenvalue equation which has to be solved
2
1 20 TT E E
Eigenvalues
This eigenvalue equation can be solved for its eigenvalues with standard routines.
The infinite Fourier expansions are truncated up to 0 , ,0, ,N N N for
numerical solution.
This results in 02N eigenvalues 2
i with 0i which correspond to
forward or backward propagating solutions.
Eigenvectors
Eigenvectors are associated with the eigenvalues and are given in a matrix
aS with size 0 02 2N N .
They provide the Fourier components of the guided eigenmodes.
Eigenfields are given by the forward and backward propagating eigenmode
a a a
E S E E
with
, ,( ) ( )a a i a a iE z E z
E E
where
, ,( ) exp ( ) expa i i i a i i iE z E i z E z E i z
Here iE and
iE are unknown (free) amplitudes, since the field amplitudes
can be chosen arbitrarily.
(For details see Li et al., Phys. Rev. E 67, 046607 (2003).)
The magnetic fields can be determined from the 1st order differential equations as
1
1
1
1
1
1
a a a
a a a
a a a
z
z
i
H T E
T S E E
T S E E
T E E
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 105
with 1
1a ai T T S
Examples of eigenfields
Eigenmodes propagate in the waveguide normal to this plane (periodic boundaries)
6.3.2 Explicit derivation for 2D problems
E.g. for TE polarization
0
0
1y y x xH E E ik E
z ik y x y
2
02
0
10 y x xH E ik E
x z ik y
and
TE
TM
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 106
0
0
1 1x y x yE H H ik H
z ik x x y
00 x yE ik Hx z
We substitute the fields by their Fourier series representations.
We start from
0x yE ik Hz
and insert
, ,expx m x m y
m
E E ik yz z
and 0 0 , ,expy m y m y
m
ik H ik H ik y
We get
, , 0 , ,exp expm x m y m y m y
m m
E ik y ik H ik yz
Multiplying with the periodic function ,exp j yik y results in
, , , 0 , , ,exp exp exp expm x m y j y m y m y j y
m m
E ik y ik y ik H ik y ik yz
Integrating over y as
, , , 0 , , ,exp exp exp expm x m y j y m y m y j y
m m
E y ik y ik y ik H y ik y ik yz
gives
, , 0 , ,m x m j m y m j
m m
E ik Hz
, 0 ,j x j yE ik H
z
which are individual equations for the different Fourier components of the fields, as
0, 0 0,
1, 0 1,
, 0 ,
x y
x y
n x n y
E ik Hz
E ik Hz
E ik Hz
0,0, 0
1,1, 0
,, 0
0 0
0 0
0 0
yx
yx
n yn x
HE ik
HE ik
z
HE ik
which in matrix notation reads as
1x yz
E TH with ,
1 0 ,
j n
j nT ik
The other equation we treat equivalently
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 107
2
02
0
1y x xH E ik E
z ik y
We do this term by term separately and the recompose the entire equation.
By substituting
, ,expy m y m y
m
H H ik yz z
and, as above, multiplying with the periodic function and integrating over y
we get directly
,i yHz
We apply the same procedure to the second term by substituting
2 2
2
, , , , ,2 2
0 0 0
exp expx m x m y m y m x m y
m m
i i iE E ik y k E ik y
k y k y k
results in
2
, ,
0
i y i x
ik E
k
For the third term
0 0 , , 0 , ,exp exp expx m m n x n y m n x m n y
m n m n
ik E ik iG y E ik y ik E iG y ik y
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 108
16
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 109
Eigenvalue problem in 2D
(Li et al., Phys. Rev. E, Vol. 67, 046607 2003)
Field expansion inside the grating
The field inside the periodic structure can be written down unambiguously as (here at the example of the x-component of the electric field)
This is the plane wave expansion of the lth eigenmode.
xmnlE are the eigenvectors from the eigenvalue problem.
exp( )li z describes the dynamics of the forward/backward propagating
eigenmodes.
lA and
lB are the amplitudes of the eigenmodes, which are up to now
unknown and which have to be determined from boundary conditions.
The same expansion holds also for the other components of the field
17
TE:
TM:
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 110
Example
1D periodic waveguide array
All possible eigenmodes are calculated using the method
shown is the electric field upon propagating through a waveguide in z-direction
Fundamental mode Higher order guided mode
Guided mode with field localization Evanescent mode between the waveguides
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 111
6.3.2.1 Solving the reflection/transmission problem
see separate presentation file
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 112
7. Finite Element Method (FEM) (Material is derived from "A Tutorial on the Finite Element Method" by Arashi Mafi.)
FEM is a general purpose tool for finding approximate numerical solution to partial differential equations as well as integral equations.
History: Alexander Hrennikoff (1941) and Richard Courant (1942)
(solution of problems on elasticity and structural analysis in aeronautics)
Solution strategy: reduction of partial differential equation to a set of linear equations for steady state problems or to a set of ordinary differential equation for time dependent problems.
Essential ansatz: discretization of a continuous computational domain into discrete sub-domains, which are called elements
Here: demonstration of the basic ideas at the example of a 1D scalar FEM
Professional use of the method: commercial programs are available that employ this computational technique:
Comsol – multiphysics simulations
JCMwave – electromagnetic problems
Strength of the method:
ability to retain various physical models in the simulation simultaneously
takes advantage of a large computational toolbox for the solution of linear equations
applicable to accurate solution of problems involving strongly different scales and where symmetries can't be exploited
Literature:
W. G. Strang and G. J. Fix, "An Analysis of the Finite Element Method," Wellesley Cambridge Press (1973).
O. C. Zienkiewicz, R. L. Taylor, J. Z. Zhu, "The Finite Element Method: Its Basis and Fundamentals," Butterworth-Heinemann (2005).
S. Humphries, "Finite-element Methods for Electromagnetics," CRC Press (1997) now freely available at http://www.fieldp.com/femethods.html
7.1 The basic set up The Finite Element Method (FEM) is a weighted residual method that uses compactly-supported basis functions.
Comparison to other methods
Before: The finite difference method approximates an operator (e.g., the derivative) and solves a problem on a set of points (the grid).
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 113
Now: Finite Element Method (FEM) uses exact operators but approximates the solution basis functions. Also, FEM solves a problem on the interiors of grid cells (and optionally on the gridpoints as well).
Furthermore one should distinquish between spectral methods, which use global basis functions to approximate a solution across the entire domain and FEM, which uses compact basis functions to approximate a solution on individual elements.
Different formulations of FEM
Strong form
Weak form
Galerkin approximation
Matrix form
Introducing to the FEM method by an example for a 1D problem
Differential equation: 2 2( ) ( ) ( ) 0xR x x x
Global solution interval: / 2, / 2
Neumann boundaries: ,
2 2
( ) 0x xx
Analytical solution: ( ) sin( )x A x with n (integer)
Strict solution: ( )R x is everywhere zero
For numerical solution the global computational domain is partitioned into local segments, the so-called elements:
Position of the element's boundaries: ix (nodes)
Value of the function at the nodes: iU
Solution to the PDE is the (linear/quadratic/…) interpolation between the nodes
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 114
7.2 Global vs. local labeling
Local index: The two nodes of the element e are identified as 1
ex and 2
ex .
This labeling is used when discussing a single node.
Global index: The element e is identified in the entire computational domain
as 1,e ex x .
This labeling is used when discussing the entire assemblage of elements.
7.3 Weak formulation Expression of the original problem in a way that the solution has to fulfill less requirements, e.g. smoothness.
The weak form is a variational statement of the problem in which one integrates against a test function. The choice of test function is up to the specific problem. This relaxes the problem since instead of finding an exact solution everywhere, one searches for a solution that satisfies the original problem (strong form) on average over the solution domain.
A solution of the strong form will also satisfy the weak form, but not vice versa.
How is it introduced mathematically?
The global interval / 2, / 2 is partitioned into small local elements.
Introducing a "weight function" ( )i
eQ x for each element such that instead of 2 2( ) ( ) ( ) 0xR x x x in the entire computation domain on requires
2
1
( ) ( ) 0e
e
x
i
e
x
Q x R x x
Inserting 2 2( ) ( ) ( )xR x x x and integration by parts gives
2
1
2( ) ( ) ( ) ( ) 0e
e
x
i i
x e x e
x
Q x x Q x x x (26)
Discretization
Partitioning of the global domain / 2, / 2 into N local elements of equal
size with
Ue-1=U1e
Ue=U2e
xe-1=x1e xe=x2
e
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 115
2ex e
N
Linear approximation of the function ( )x on each local element
in element( )
0 elsewhere
e e
e
p x q ex
Globally the solution is expressed as
1
( ) ( )N
e
e
x x
( )e x has to satisfy 1 1( )e e ex U and
2 2( )e e ex U which determines the
parameters of the local element as
ascent 2 1
2 1
e ee
e e
U Up
x x
and starting value
1 2 1 1
2 1
e e e ee
e e
U x U xq
x x
However, please note that the values eU of the final solution are still
unknown.
Now each local element ( )e x can be rearranged to
2
1
( ) U W ( )i i
e e e
i
x x
with the solution independent coefficients
21
2 1( ) e
e
e e
x xW x
x x
and
12
2 1( ) e
e
e e
x xW x
x x
(in element e and otherwise 0 )
Remarks:
This expression would correspond to the Lagrange linear
polynomials for ( )i j
e e ijW x .
The coefficients 1
eW and 2
eW just depend on the discretization.
Consequently the global solution ( )x can be written as
2
1 1
( ) U W ( )N
i i
e e
e i
x x
where we have to solve for the unknown i
eU .
By choosing the weighting function to be ( ) ( )i i
e eQ x W x , inserting everything
into equation (26), and puling the summations out of the integral we get
1
22
1 1
U ( ) ( ) ( ) ( ) 0e
e
xNj i j i j
e x e x m e m
m j x
W x W x W x W x x
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 116
which gives for each individual element e an individual equation which has to be fulfilled
1 1
2 22
1 1
U ( ) ( ) U ( ) ( )e e
e e
x x
j i j j i j
e x e x e e e e
j jx x
W x W x x W x W x x
This expression can be reformulated by introducing
1
( ) ( )e
e
x
ij i j
e x e x e
x
A W x W x x
and 1
( ) ( )e
e
x
ij i j
e e e
x
B W x W x x
resulting in
2ij j ij j
e e e eA U B U
which is only related to the single local element e .
Assemblage of global system
The solution is partially redundant since 2 1
1e eU U
The final global assembly of the local elements removes this ambiguity.
Example: 3N (Three elements with three matrix equations):
2ij j ij j
e e e eA U B U with 1,2,3e
switch from local to global labeling scheme:
1 2 1 2 1 2
0 1 1 1 2 2 2 3 3 3, , ,U U U U U U U U U U
11 12 11 120 021 1 1 1
21 22 21 221 11 1 1 1
11 12 11 121 122 2 2 2
21 22 21 222 22 2 2 2
11 12 11 122 223 3 3 3
21 22 21 223 33 3 3 3
U UA A B B
U UA A B B
U UA A B B
U UA A B B
U UA A B B
U UA A B B
These are six equations with only four unknowns.
Remark: In this formulation of the problem the eigenvalue is not
an unknown. It must either be determined by the task or it must be numerically determined from the boundary conditions within the FEM algorithm.
Adding up equations containing the same unknowns:
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 117
21 22 2 21 22
1 0 1 1 1 0 1 1
11 12 2 11 12
2 1 2 2 2 1 2 2
21 22 11 12 2 21 11 22 12
1 0 1 2 1 2 2 1 0 2 1 1 2 2
A U A U B U B U
A U A U B U B U
A U A A U A U B U B B U B U
which can be summarized in matrix notation as:
2AU BU
with
11 12
1 1
21 22 11 12
1 1 2 2
21 22 11 12
2 2 3 3
21 22
3 3
0 0
0
0
0 0
A A
A A A A
A A A A
A A
A and
11 12
1 1
21 22 11 12
1 1 2 2
21 22 11 12
2 2 3 3
21 22
3 3
0 0
0
0
0 0
B B
B B B B
B B B B
B B
B
and
0 1 2 3, , ,U U U UU
The construction of the matrices A and B for N elements is straightforward as
2 1 0 0 0
1 4 1 0 0
3
0 1 4 1 0
0 0 1 4 1
0 0 0 1 2
AN
and
1 1 0 0 0
1 2 1 0 0
0 1 2 1 02
0 0 1 2 1
0 0 0 1 1
NB
7.4 Example Results
Example solutions obtained with Mathematica for 10N .
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 118
Problem specific mesh generation
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 119
Computational Photonics – old 2018 script version by Prof. Thomas Pertsch 120
8. Outlook
What is missing in this lecture?
Near-field to far-field transformation technique (and vice versa): Richards-Wolf integrals, Stratton-Chu integrals, or Surface integral techniques
Green's function methods
Finite Integration technique by employing the integral form of Maxwell’s equations
Freely available programs
MPB – Powerful tool for calculating photonic band structures http://ab-initio.mit.edu/mpb
MEEP – Parallel FDTD http://ab-initio.mit.edu/meep
Survey of various scattering codes available at http://www.iwt-bremen.de/vt/wriedt
Code for the Discrete Dipole Approximation http://ascl.net/ddscat.html
Code for the T-Matrix approach http://www.giss.nasa.gov/~crmim/t_matrix.html
Codes published under GNU are available, e.g. FDTD http://www.borg.umn.edu/toyfdtd
Commercial programs
Integrated simulation environments incorporating different methods http://www.rsoftdesign.com http://www.optiwave.com http://www.cst.com
Rigorous grating solver http://www.unigit.com http://www.gsolver.com/
Finite Element Method http://www.comsol.com http://www.jcmwave.com http://www.ansoft.com
Finite Difference Time Domain method http://www.lumerical.com
Multiple Multipole Method http://alphard.ethz.ch/hafner/MaX/max1.htm