Lecture 4b -- Maxwell's Equations on a Yee Grid · í ì l í ì l î ì í õ í ó & ] v ] r ] (...
Transcript of Lecture 4b -- Maxwell's Equations on a Yee Grid · í ì l í ì l î ì í õ í ó & ] v ] r ] (...
10/10/2019
1
Advanced Computation:Computational Electromagnetics
Maxwell’s Equationson a Yee Grid
Outline• Yee Grid• Maxwell’s equations on a Yee grid• Finite-difference approximations of Maxwell’s equations on a Yee grid• Consequences of the Yee grid• Alternative grids
Slide 2
10/10/2019
2
Slide 3
Yee Grid
Kane S. Yee
Slide 4
Kane S. Yee
Kane S. Yee was born in Canton, China on March 26, 1934. He received a B.S.E.E., M.S.E.E and Ph.D. in Applied Mathematics from the University of California at Berkeley in 1957, 1958, and 1963, respectively.
He did research on electromagnetic diffraction while employed by Lockheed Missile and Space Co. (1959-1961). He has been associated with the Lawrence Livermore Laboratory since 1963. At present he is a professor in mathematics at Kansas State University. His main areas of interest are electromagnetics, hydrodynamics and numerical solution to partial differential equations.
K. S. Yee, “A Closed-Form Expression for the Energy Dissipation in a Low-Loss Transmission Line,” IEEE Trans. Nuclear Science, vol. 21, no. 1, pp. 1006-1008, 1974.
10/10/2019
3
3D Grids
Slide 5
A three-dimensional grid looks like this:
One cell from the grid looks like this:
x y
z
, , grid resolution parametersx y z
𝑁=15
𝑁 = 10𝑁 = 10
Collocated Grid
Slide 6
Within the grid cell, where should Ex, Ey, Ez, Hx, Hy, and Hz be placed?
A straightforward approach would be to locate all of the field components at a common point within in a grid cell; perhaps at the origin.
xE
xy
z
xHyEyH
zEzH
10/10/2019
4
Yee Grid
Slide 7
Instead, the field components will be staggered within the grid cell.
K. S. Yee, “Numerical solution of the initial boundary value problems involving Maxwell’s equations in isotropic media,” IEEE Trans. Microwave Theory and Techniques, vol. 44, pp. 61–69, 1966.
Stereo Image of Yee Cell
Slide 8
To view the Yee cell if full 3D, look past the image above so that they appear double. When the double images overlap so that you see three Yee cells, the middle image will be three-dimensional.
10/10/2019
5
Yee Cell for 1D, 2D, and 3D Grids
Slide 9
xy
z
xEyE
zE
xHyH
zH
3D Yee Grid
2D Yee Grids
zE
xHyH
xy
xy
zHyE
xE
E Mode
H Mode
1D Yee Grids
z
xE
yH
yExH
Ey Mode
Ex Mode
z
* These are the same for isotropic materials.
Reasons to Use the Yee Grid
Slide 10
0E
0H
1. Divergence-free3. Elegant arrangementto approximate curl equations
2. Physical boundary conditions are naturally satisfied
10/10/2019
6
Visualizing Extended Yee Grids
Slide 11
y
x
ij
222 Grid
44 Grid (E Mode)
Slide 12
Finite-Difference Approximations of
Maxwell’s Equationson a Yee Grid
10/10/2019
7
Starting Point
Slide 13
0
0
0
0
0
0
yzxx x
x zyy y
y xzz z
yzxx x
x zyy y
y xzz z
EEk H
y z
E Ek H
z xE E
k Hx y
HHk E
y z
H Hk E
z x
H Hk E
x y
Start with Maxwell’s equations in the following form.
Here only diagonally anisotropic material tensors were retained. This will be needed to incorporate a uniaxial perfectly matched layer (UPML) absorbing boundary condition.
These equations are valid independent of the chosen sign convention.
Normalize the Grid Coordinates
Slide 14
0x k x 0y k y 0z k z
yzxx x
x zyy y
y xzz z
yzxx x
x zyy y
y xzz z
EEH
y z
E EH
z xE E
Hx y
HHE
y z
H HE
z x
H HE
x y
The grid is normalized according to
This “absorbs” the k0 term into the spatial derivatives and simplifies Maxwell’s equations to
0
0
0
0
0
0
yzxx x
x zyy y
y xzz z
yzxx x
x zyy y
y xzz z
EEk H
y z
E Ek H
z xE E
k Hx y
HHk E
y z
H Hk E
z x
H Hk E
x y
10/10/2019
8
Starting Point
Slide 15
yz
x
z
xx
y
x
z
xz
yy
zy
EE
E E
E E
y z
z x
x
H
H
yH
xx
y
z
yz
x z
x
y
yz
y
z
x
HH
H H
y z
z x
E
H
y
HE
x
E
Finite-Difference Equation for
Slide 16
x xyz
xy z
EEH
, , ,, ki j k ixx
jxH
10/10/2019
9
Finite-Difference Equation for
Slide 17
x xyz
xy z
EEH
, 1, , ,i j k i j kz zE
y
E
, , ,, ki j k ixx
jxH
Finite-Difference Equation for
Slide 18
x xyz
xy z
EEH
, 1, , ,i j k i j kz zE
y
E
, , ,, ki j k ixx
jxH
, , 1 , ,i j k i j ky yE
z
E
10/10/2019
10
Finite-Difference Equation for
Slide 19
y yx z
yz x
E EH
, , ,, ki j k iyy
jyH
Finite-Difference Equation for
Slide 20
y yx z
yz x
E EH
, , 1 , ,i j k i j kx xE
z
E
, , ,, ki j k iyy
jyH
10/10/2019
11
Finite-Difference Equation for
Slide 21
y yx z
yz x
E EH
, , 1 , ,i j k i j kx xE
z
E
, , ,, ki j k iyy
jyH
1, , , ,i j k i j kz zE
x
E
Finite-Difference Equation for
Slide 22
z zy x
zx y
E EH
, , ,, ki j k izz
jzH
10/10/2019
12
Finite-Difference Equation for
Slide 23
z zy x
zx y
E EH
1, , , ,i j k i j ky yE
x
E
, , ,, ki j k izz
jzH
Finite-Difference Equation for
Slide 24
z zy x
zx y
E EH
1, , , ,i j k i j ky yE
x
E
, , ,, ki j k izz
jzH
, 1, , ,i j k i j kx xE
y
E
10/10/2019
13
Finite-Difference Equation for
Slide 25
xxxyz
H
yE
z
H
, , ,, ki j kx
i jxx E
Finite-Difference Equation for
Slide 26
xxxyz
H
yE
z
H
, , , 1,i j k i j kz z
y
H H
, , ,, ki j kx
i jxx E
10/10/2019
14
Finite-Difference Equation for
Slide 27
xxxyz
H
yE
z
H
, , , 1,i j k i j kz z
y
H H
, , ,, ki j kx
i jxx E
, , , , 1i j k i j ky yH H
z
Finite-Difference Equation for
Slide 28
yyx z
y
H H
xE
z
, , ,, ki j ky
i jyy E
10/10/2019
15
Finite-Difference Equation for
Slide 29
yyx z
y
H H
xE
z
, , ,, ki j ky
i jyy E
, , , , 1i j k i j kx x
z
H H
Finite-Difference Equation for
Slide 30
yyx z
y
H H
xE
z
, , ,, ki j ky
i jyy E
, , 1, ,i j k i j kz zH H
x
, , , , 1i j k i j kx x
z
H H
10/10/2019
16
Finite-Difference Equation for
Slide 31
zzy x
z
H H
yE
x
, , ,, ki j kz
i jzz E
Finite-Difference Equation for
Slide 32
zzy x
z
H H
yE
x
, , ,, ki j kz
i jzz E
, , 1, ,i j k i j ky y
x
H H
10/10/2019
17
Finite-Difference Equation for
Slide 33
zzy x
z
H H
yE
x
, , , 1,i j k i j kx xH H
y
, , ,, ki j kz
i jzz E
, , 1, ,i j k i j ky y
x
H H
Summary of Finite-Difference Approximations of Maxwell’s Equations
Slide 34
x
y
z
yz
x z
y
xx
yy
zz
xx
yy
zz
yz
x z
y
x
x
x
y
z
EE
E E
E E
E
E
E
y z
z x
x y
y z
H
H
H
HH
H H
H H
z x
x y
, , 1, ,
, ,
,
, ,
,
, ,, 1, , ,
, , 1 , , 1, , , ,
1, , , , , 1,
,
,
, ,
, , ,
, ,
1
,
i j kx
i j ky
i j kz
i j k i
i j k i j ki j ki j kxx
i j kyy
i j ky yz z
i j k i j k i j k i j kx x z z
i j k i j k i j k i j ky y x x i j k
j
z
kz z
z
y
E E
y z
E E
E E E E
H
H
H
HH
z x
x
E E E
y
H
y
E
, , , , 1
, , , , 1
, ,
, ,
, ,
, , 1, ,
, , 1, , , ,
, ,
, ,
,, 1
,,
i j kx
i j ky
i
i j k i j ky
i j k i j k i j k i j k
i j kxx
i jx x z z
i j k i j k i j k i j k
kyy
i j kyz
kz
xz
jy x
H
H H H H
H H
z
z x
x y
H
E
E
HE
10/10/2019
18
Slide 35
Consequences of theYee Grid
Field Components are at Physically Different Positions
Slide 36
Even though the field components reside within the same Yee cell, they are out of phase.
Any time that a wave is injected, extracted, or analyzed within a simulation, this must be accounted form.
10/10/2019
19
Field Components are at Physically Different Positions
Slide 37
Even though the field components reside within the same Yee cell, they may reside in different media. This happens when an interface slices through the middle of a Yee cell.
This is the main reason why each field component is assigned its own material properties.
𝜀 assigned to 𝐸 .𝜀 assigned to 𝐸 .𝜀 assigned to 𝐸 .𝜇 assigned to 𝐻 .𝜇 assigned to 𝐻 .𝜇 assigned to 𝐻 .
𝜀
𝜀
𝐸 resides in 𝜀
𝐸 and 𝐸 resides in 𝜀
Dispersion on a Yee Grid
38
Recall the dispersion relation for an isotropic material with parameters r and r.2
2 2 2r r
0x y zk k k
c
The analogous dispersion relation on a frequency-domain Yee grid filled with r and r is22 22
r r
2 2 2sin sin sin
2 2 2y yx x z z
x y z
kk k
v
In this equation, the speed of light c0 is written as v because the velocity is different due to the dispersion of the grid.
10/10/2019
20
Drawback of Structured Grids
Slide 39
Structured grids exhibit highly anisotropic dispersion.
Anisotropic Dispersion
FDM
Compensation Factor g (1 of 2)
40
The numerical dispersion equation is solved for velocity v.1
22 2 2
r r
2 2 2sin sin sin
2 2 2y yx x z z
x y z
kk kv
In the absence of grid dispersion, v should be exactly the speed of light c0. Due to the Yee grid, waves slow down by a factor g.
0v c g
We can calculate this factor by combining the above equations.
22 2
0
r r
2 2 2sin sin sin
2 2 2y yx x z z
x y z
kc k kg
10/10/2019
21
Compensation Factor g (2 of 2)
41
We can write a simpler and more useful expression for g.
22 2
0
00
r r
1 2 2 2sin sin sin
2 2 2
2
y yx x z z
x y z
kk k
k n
k
n
g
Compensating for Numerical Dispersion
42
Given that the wave slows down by factor g in the direction of , it follows that we can compensate for the dispersion by artificially “speeding up” the wave.
We do this by decreasing the values of r and r across the entire grid by a factor of g.
k
r r r r g g
Notes:1. We can only compensate for dispersion for one direction k.2. We can only compensate for dispersion in one set of material
values r and r.3. It is best to choose average or dominant values for these
parameters.4. Choose q = 22.5° if nothing else is known.
22.5q
k
10/10/2019
22
Slide 43
Alternative Grids
Drawbacks of Uniform Grids
Slide 44
Uniform grids are the easiest to implement, but do not conform well to arbitrary structures and exhibit high anisotropic dispersion.
Anisotropic Dispersion (see Lecture 10) Staircase Approximation (see Lecture 18)
10/10/2019
23
Drawbacks of Structured Grids (1 of 2)
Slide 45
Structured grids are the easiest to implement, but do not conform well to arbitrary geometries.
Structured Grid Unstructured Grid
Hexagonal Grids
Slide 46
Hexagonal grids are good for minimizing anisotropic dispersion suffered on Cartesian grids. This is very useful when extracting phase information.
See Text, pp. 101-103.
Yee-FDTD
Hex-FDTD
10x
Phase Velocity as a Function of Propagation Angle
57°0° 115° 172° 229° 286° 344°
10/10/2019
24
Nonuniform Orthogonal Grids (1 of 2)
Slide 47
Nonuniform orthogonal grids are still relatively simple to implement and provide some ability to refine the grid at localized regions.
See Text, pp. 464-471.
Nonuniform Orthogonal Grids (2 of 2)
Slide 48
Uniform Grid Simulation• 80×110×16 cells• 140,800 cells
Nonuniform Grid Simulation• 64×76×16 cells• 77,824 cells
Conclusion: Roughly 50% memory and time savings.
10/10/2019
25
Curvilinear Coordinates
Slide 49
Maxwell’s equations can be transformed from curvilinear coordinates to Cartesian coordinates to conform to curved boundaries of a device.
M. Fusco, “FDTD Algorithm in Curvilinear Coordinates,” IEEE Trans. Ant. and Prop., vol. 38, no. 1, pp. 76-89, 1990.See Text, pp. 484-492.
Structured Nonorthogonal Grids
Slide 50
This is a particularly powerful approach for simulating periodic structures with oblique symmetries.
M. Fusco, “FDTD Algorithm in Curvilinear Coordinates,” IEEE Trans. Ant. and Prop., vol. 38, no. 1, pp. 76-89, 1990.
10/10/2019
26
Irregular Nonorthogonal Unstructured Grids
Slide 51
Unstructured grids are more tedious to implement, but can conform to highly complex shapes while maintaining good cell aspect ratios and global uniformity.
ln x
P. Harms, J. Lee, R. Mittra, “A Study of the Nonorthogonal FDTD Method Versus the Conventional FDTD Technique for Computing Resonant Frequencies of Cylindrical Cavities,” IEEE Trans. Microwave Theory and Techniq., vol. 40, no. 4, pp. 741-746 , 1992.
Comparison of convergence rates
Bodies of Revolution (Cylindrical Symmetry)
Slide 52
Three-dimensional devices with cylindrical symmetry can be very efficiently modeled using cylindrical coordinates.
even odd
0
even odd
0
, , , cos , sin
, , , cos , sin
m
m
E e m e m
H h m h m
q q q
q q q
Devices with cylindrical symmetry have fields that are periodic around their axis. Therefore, the fields can be expanded into a Fourier series in .
Due to a singularity at 𝑟 = 0, update equations for fields on the z axis are derived differently.
See Text, Chapter 12
10/10/2019
27
Some Devices with Cylindrical Symmetry
Slide 53
Bent WaveguidesDipole Antennas
Cylindrical Waveguides
Conical Horn Antenna
Focusing Antennas
Diffractive Lenses