FDTD: An Introduction

Prof. Rakhesh Singh Kshetrimayum

1 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

How do one visualize wave?

How can one observe wave propagation?

Radio waves cann’t be observed in nature

But one can write some EM codes (simulation) and

observe wave behavior

One popular choice for this is Finite Difference Time
Domain method

Also referred to as FDTD

2 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

What is FDTD?

Let us start with 4 Maxwell’s equations
r ρv
1. ∇•E =
ε
r
2. ∇• B = 0
r
r r ∂D
3. ∇× H = J +
∂t
Maxwell curl ur
ur
equations (why?) 4. ∇× E = −
∂B
∂t

3 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Basically, we replace spatial & time derivatives in the two
Maxwell’s curl equations

by central finite difference approximation

What is central finite difference approximation of
derivatives?

Consider a function f(x), its derivative at x0 from central
finite difference approximation is given by

 ∆x   ∆x 
f  x0 +  −f x
 0 − 
= f ' ( x0 ) ≅ 
df ( x0 ) 2   2 
dx ∆x

4 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
r

The time-dependent and source free (J = 0 )
Maxwell’s
curl equations in a medium with ε=ε0εr and µ = µ0 µr
are r
r ∂H
r
r
3 equations
∂E
=
1
∇× H; =−
1
∇× E 3 equations
∂t ε ε ∂t0 r µµ 0 r

For Cartesian coordinate systems,

expanding the curl equations,

equating the vector components,

we have 6 equations

5 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Expanding the first vector curl equation
r
∂E 1 r
= ∇× H
∂t ε 0ε r

we get 3 equations
∂Ex 1  ∂H z ∂H y 
=  − 
∂t ε 0ε r  ∂y ∂z 
∂E y 1  ∂H x ∂H z 
=  − 
∂t ε 0ε r  ∂z ∂x 
∂Ez 1  ∂H y ∂H x 
=  − 
∂t ε 0ε r  ∂x ∂y 

6 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Expanding the second vector curl equation
r
∂H 1 r
=− ∇× E
∂t µ0 µr

We get 3 equations

∂H x 1  ∂E y ∂Ez 
=  − 
∂t µ0 µr  ∂z ∂y 
∂H y 1  ∂Ez ∂Ex 
=  − 
∂t µ0 µr  ∂x ∂z 
∂H z 1  ∂Ex ∂E y 
=  − 
∂t µ0 µr  ∂y ∂x 

7 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
As we know that FDTD is a time-domain solver
The question is how do we solve those 6 equations above?
1. 1-D FDTD update equations

For 1-D case (a major simplification), we can consider

(a) Linearly polarized wave along x-axis

exciting an electric field which has Ex only (Ey = Ez = 0)

(b) propagation along z-axis

no variation in the x-y plane, i.e. ∂/∂x = 0 and ∂/∂y = 0

8 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

∂Ex 1  ∂H z ∂H y  ∂H x 1  ∂E y ∂Ez 
=  −  =  − 
∂t ε 0ε r  ∂y ∂z  ∂t µ0 µr  ∂z ∂y 
∂E y 1  ∂H x ∂H z  ∂H y 1  ∂Ez ∂Ex 
=  −  = −
∂t ε 0ε r  ∂z ∂x   
∂t µ0 µr  ∂x ∂z 
∂Ez 1  ∂H y ∂H x  ∂H z 1  ∂Ex ∂E y 
=  − 
∂t ε 0ε r  ∂x ∂y  =  − 
∂t µ0 µr  ∂y ∂x 

9 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Then, the above 6 equations from 2 Maxwell’s curl
equations

reduce to 2 equations for 1-D FDTD
∂Ex 1  ∂H y  ∂H y 1  ∂Ex 
=−   =−
∂t ε 0ε r  ∂z  ∂t µ0 µr  ∂z 

We want to solve these equations at different locations

and time in solution space (observe fields at different
place and time)

Hence we need to discretize both in time and space

10 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

space discretization size ∆z running index k

1 2 3 4 5
Fig. 1(a) Discretizations in 1-D space
time discretization size ∆t running index n
1 2 3 4 5

Fig. 1(b) Discretization in time

11 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

Notations:

n is the time index and

k is the spatial index,

How to decide space and time discretization size?

How do one determine the values of ∆z and ∆t?

12 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Locate positions z = k∆z

Usually ∆z is taken as 10-15 per wavelength


gives accurate results

times t = n∆t and

∆t (dependent on ∆z) chosen from Courant stability
criterion


gives stable FDTD solution (will be discussed later)

13 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Let us discretize the first equation:
∂Ex 1  ∂H y 
=−  
∂t ε 0ε r  ∂z 

the central difference approximations for both the
temporal and spatial derivatives are obtained at

z = k∆z, (space discretization size and index)

t = n∆t (time discretization size and index)

 1  1  1   1 
E x  k , n +  − Ex  k , n −  Hy k + ,n − Hy k − ,n
 2  2
=−
1  2   2 
∆t ε 0ε r ∆z
14 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction
∂H y 1  ∂Ex 

For the second equation: =−
∂t µ0 µr  ∂z 

the central difference approximations for both the

temporal and spatial derivatives are obtained at

at (z + ∆z/2,
∆ t + ∆t/2)
∆

(increment all the time and space steps by 1/2):

 1   1   1  1
H y  k + , n + 1 − H y  k + , n  Ex  k + 1, n +  − Ex  k , n + 
 2   2 
=−
1  2  2
∆t µ0 µ r ∆z

15 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

The above equations can be rearranged as a pair of
‘computer update equations’,

which can be repeatedly updated in loop,

to obtain the next instant time values of Ex (k, n+1/2)
and Hy (k + ½, n+1)

from the previous instant time values as follows

 1  1 1 ∆t   1   1 
Ex  k , n +  = E x  k , n −  − H
 y k + , n  − H y  k − , n 
 2   2  0 r
ε ε ∆z   2   2 

 1   1  1 ∆t   1  1 
H y  k + , n + 1 = H y  k + , n  − E x k + 1, n +  − E x k , n + 
 2   2  µ0 µr ∆z   2  2  

16 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

E xn −1/ 2 k+2
k-2 k-1 k k+1

H yn
k-5/2 k-3/2 K-1/2 K+1/2 K+3/2 K+5/2

E xn +1/ 2
k-2 k-1 k k+1 k+2

Fig. 2 Interleaving of E and H fields

17 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Interleaving of the E and H fields in space and time in the
FDTD formulation

to calculate Ex(k),

the neighbouring values of Hy at k-1/2 and k+1/2 of
the previous time instant are needed

Similarly,

to calculate Hy(k+1/2), for instance,

the neighbouring values of Ex at k and k+1 of the
previous time instants are needed

18 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

In the above equations,

ε0 and µ0 differ by several orders of magnitude,

Ex and Hy will differ by several orders of magnitude

We also know that ratio of electric field and magnetic
field for plane waves is 120π

Numerical error is minimized by making the following
change of variables as
' ε
E =
x Ex
µ

19 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Hence,

1  1 1 ' 1  1 ∆t   1   1 
Ex'  k , n +  = Ex  k , n −  −  H y k + , n  − H y k − , n 
ε  2 ε  2  ε ∆z   2   2 
µ µ
 1  1 ε 1 ∆t   1   1 
⇒ Ex'  k , n +  = Ex'  k , n −  −  H y k + , n  − H y  k − , n 
 2  2 µ ε ∆z   2   2 
 1  1 1 ∆t   1   1 
⇒ Ex'  k , n +  = Ex'  k , n −  −  H y k + , n  − H y k − , n 
 2  2 µε ∆z   2   2 
 1  1 1 ∆t   1   1 
⇒ Ex'  k , n +  = Ex'  k , n −  +  H y k − , n  − H y k + , n 
 2  2 µε ∆z   2   2 

20 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Similarly,
 1   1  1 1 ∆t  '  1 '  1 
H y  k + , n + 1 = H y  k + , n  − E
 x k + 1, n +  − E x k , n + 
 2   2  ε µ ∆z   2  2  
µ
 1   1  1 ∆t  '  1 '  1 
⇒ H y  k + , n + 1 = H y  k + , n  − E
 x k + 1, n +  − E x k , n + 
 2   2  µε ∆z   2  2  
 1   1  1 ∆t  '  1 '  1 
⇒ H y  k + , n + 1 = H y  k + , n  + E
 x k , n +  − E x k + 1, n + 
 2   2  µε ∆z   2   2 

21 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
Courant stability criteria:

Courant stability criteria dictates the

relationship between the time increment ∆t with
respect to space increment ∆z

in order to have a stable FDTD solution of the
electromagnetic problems

In isotropic media, an electromagnetic wave propagates a
distance of one cell in time ∆t = ∆z/vp,

where vp is the phase velocity

22 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

This limits the maximum time step

This equation implies that an EM wave cannot be
allowed

to move more than a space cell during a time step

Otherwise, FDTD will start diverging

23 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

If we choose ∆t > ∆z/vp,

the distance moved by the EM wave over the time
interval ∆t will be more than ∆z,

the EM wave will leave out the next node/cell and

FDTD cells are not causally interconnected and

hence, it leads to instability

24 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

cell
y

∆y
∆x x
Fig. 3 (a) 2-D discretizations
25 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

∆ Propagation at θ=450
2
∆

Propagation at θ=00

∆
Plane wave
front
Fig. 3 (b) Plane wave front propagation along θ=00 andθ=450
26 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

Propagation at θ=450

Wavefront jumps from one row of nodes to the next
row,

the spacing between the consecutive rows of nodes is ∆
2

∆

Similarly for 3-D case
3

27 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

In the case of a 2-D simulation,

we have to allow for the propagation in the diagonal
direction,

which brings the time requirement to ∆t = ∆z/√2vp

Obviously, three-dimensional simulation requires ∆t∆ =
∆z/√3vp

We will use in all our simulations a time step ∆t of
∆z
∆t =
2 ⋅ vp

28 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

where vp is the phase velocity,

which satisfies the requirements in 1-D, 2-D and 3-D
for all media (√2≈1.414<√3≈1.7321<2)

In 3D case, it may be more appropriate to modify the
above stability criteria as
min ( ∆x, ∆y, ∆z )
∆t =
2 ⋅ vp

Using the above relation, we may simplify for 1-D FDTD

as 1 ∆t ∆t v p ⋅ ∆z 1
= vp ⋅ = =
εµ ∆z ∆z 2 ⋅ v p ⋅ ∆z 2
29 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

Making use of this in the above two equations,

we obtain the following equations
 1  1 1  1   1 
Ex'  k , n +  = Ex'  k , n −  +  H y  k − , n  − H y  k + , n  
 2  2 2  2   2 
 1   1  1  1  1 
H y  k + , n + 1 = H y  k + , n  +  Ex'  k , n +  − Ex'  k + 1, n +  
 2   2  2  2  2 

Hence the computer update equations (for free space
er(k)=1) are

ex[k] = ex[k] + (0.5/er(k))*( hy[k-1] - hy[k] )

hy[k] = hy[k] + 0.5*( ex[k] - ex[k+1] )

30 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

k+1/2 and k-1/2 are replaced by k or k-1

Note that the n or n + 1/2 or n −1/2 in the superscripts
do not appear

FDTD Simulation of Gaussian pulse propagation in free space
(fdtd_1d_1.m)

FDTD Simulation of Gaussian pulse hitting a dielectric medium
(fdtd1_dielectric.m)

From theoretical analysis of EM wave hitting a dielectric
surface 2η 2 ε η −η ε − ε
1 1 2
τ= 2
= Γ= 2 1
=
η 2 + η1 ε1 + ε 2 η 2 + η1 ε1 + ε 2

31 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Fig. EM wave hitting a dielectric surface

32 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction
1 2

For ε1=1.0 and ε2=4.0, we have Γ = − ; τ =
3 3

FDTD simulation of Absorbing Boundary Condition

(fdtd_1d_ABC_boundary.m)

(fdtd_1d_no_boundary.m)

33 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
1 2

For ε1=1.0 and ε2=4.0, we have Γ = − ;τ =
3 3

FDTD simulation of Absorbing Boundary Condition
(fdtd_1d_ABC_boundary.m)

(fdtd_1d_no_boundary.m)

In calculating the E field,

we need to know the surrounding H values;

this is the fundamental assumption in FDTD

At the edge of the problem space,

we will not have the value to one side,

but we know there are no sources outside the problem space
34 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

The wave moves ∆z/2 (=c0 · ∆ t) distance in one time
step,

so it takes two time steps for a wave front to cross one
cell

Suppose we are looking for a boundary condition at the
end where k=1

Now if we write the E field at k=1 as

Ex (1,n) = Ex (2, n-2),

then the fields at the edge will not reflect

This condition must be applied at both ends
35 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction
2. FDTD Solution to Maxwell’s equations in 2-D
Space

In deriving 2-D FDTD formulation, we choose between
one of two groups of three vectors each:

(a) Transverse magnetic (TMz) mode,

which is composed of Ez, Hx, and Hy or

(b) Transverse electric (TEz) mode,

which is composed of Ex, Ey, and Hz

36 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Unlike time-harmonic guided waves,

none of the fields vary with z so that

there is no propagation in the z-direction

But in general propagation along x- or y- directions or
both is possible

37 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

FDTD simulation of TM mode wave propagation
(fdtd_2d_TM_1.m )

Expanding the Maxwell’s curl equations with

Ex = 0, Ey = 0, Hz = 0 and ∂/∂z = 0,

we obtain,

3 equations from the 6 equations

38 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

3 equations as follows
∂Ez 1  ∂H y ∂H x 
=  − 
∂t ε  ∂x ∂y 

∂H x 1  ∂Ez 
= 
∂t µ  ∂y 
∂H y 1  ∂Ez 
= −
∂t µ  ∂x 

39 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

40 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Observations:

Electric fields are calculated at integer space steps

Electric fields are calculated at half integer time steps

Magnetic fields are calculated at half integer space step
and integer space step

Magnetic fields are calculated at integer time steps

For example,

Hy(i-1/2,j), Hx(i,j-1/2), Hy(i+1/2,j), Hx(i,j+1/2)

41 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Central finite difference approximations of the 1st
equation is as follows ∂Ez 1  ∂H y ∂H x 
=  − 
∂t ε  ∂x ∂y 
 1  1
Ez  i, j , n +  − Ez  i, j , n − 
 2  2
∆t
  1   1   1   1 
 H y  i + , j , n  − H y i − , j , n  H x i , j + , n  − H x i , j − ,n
    2 − 
1 2 2   2 
= 
ε 0ε r  ∆x ∆y 
 

42 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Central finite difference approximations of the next equation
is as follows

∂H x 1  ∂E z 
=  
∂t µ  ∂y 

 1   1   1  1
H x  i, j + , n + 1 − H x  i, j + , n  Ez  i, j + 1, n +  − Ez  i, j , n + 
 2   2 
=−
1  2  2
∆t µ0 µ r ∆y

43 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Central finite difference approximations of the next equation
is as follows
∂H y 1  ∂E z 
= − 
∂t µ  ∂x 

 1   1   1  1
H y  i + , j , n + 1 − H x  i + , j , n  Ez  i + 1, j , n +  − Ez  i, j , n + 
 2   2 = 1  2  2
∆t µ0 µ r ∆x

44 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Rearranging the above equations, we can write the final

3 update equations’ expressions as

Electric field update equation
 1
Ez  i, j , n + 
 2
  1   1   1   1 
H i + , j , n − H i − , j , n H i , j + , n − H i , j − ,n
 1  ∆t  y  2 

y
 2


x
 2 
 x
 2 
= Ez  i, j, n −  +  − 
 2  ε 0ε r  ∆x ∆y 
 

45 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Magnetic field update equation

 1  1
Ez  i, j + 1, n +  − Ez  i, j , n + 
 1   1  ∆t  2  2
H x  i, j + , n + 1 = H x  i, j + ,n −
 2   2  µ0 µr ∆y

 1  1
Ez  i + 1, j , n +  − Ez  i, j, n + 
 1   1  ∆t  2  2
H y  i + , j , n + 1 = H x  i + , j , n  +
 2   2  µ0 µr ∆x

46 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
 1
Ez  i , j , n + 
 2
  1   1   1   1 
H i + , j , n − H i − , j , n H i , j + , n − H i , j − ,n
 1  ∆t  y  2 

y
 2

−
x
 2


x
 2 
= Ez  i, j, n −  + 
 2  ε 0ε r  ∆x ∆y 
 

Electric field computer update equation is (i+1/2 and i-1/2

are replaced by i or i-1)

ez(i,j)=ez(i,j)+{dt/(eps_0*eps_r)}*[{Hy(i,j)-Hy(i-
1,j)}/dx-{Hx(i,j)-Hx(i,j-1)}/dy]

47 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
 1  1
Ez  i, j + 1, n +  − Ez  i, j, n + 
 1   1  ∆t  2  2
H x  i, j + , n + 1 = H x  i, j + ,n −
 2   2  µ0 µr ∆y

Magnetic field computer update equation is (j+1/2 is

replaced by j, usually meu_r=1)

Hx(i,j) = Hx(i,j)-{dt/(mu_0*dy)}*{Ez(i,j+1)-Ez(i,j)}

48 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
 1  1
Ez  i + 1, j , n +  − Ez  i, j, n + 
 1   1  ∆t  2  2
H y  i + , j , n + 1 = H x  i + , j , n  +
 2   2  µ0 µr ∆x

Magnetic field computer update equation is (i+1/2 is

replaced by i, usually meu_r=1)

Hy(i,j) = Hy(i,j)+{dt/(mu_0*dx)}*{Ez(i+1,j)-Ez(i,j)}

49 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Some new MATLAB commands:

image: IMAGE(C) displays matrix C as an image

imagesc: data is scaled to use the full color map

Colorbar: appends a vertical color scale to the current axis

Colormap: a matrix may have any number of rows but it
must have exactly 3 columns

Each row is interpreted as a color, with the first element
specifying the intensity of red, the second green and the third
blue

Color intensity is specified on the interval of 0.0 to 1.0

50 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

For example, [0 0 0] is black

[1 1 1] is white

[1 0 0] is red

[0.5 0.5 0.5] is gray

[127/255 1 212/255] is aquamarine

51 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
FDTD simulation of TM wave hitting a dielectric surface (right side),
source at the center (2-D) (fdtd_2d_TM_dielectric.m)

Modify the above program as follows:

Specify the position of the dielectric surface (id=..,jd=..)

Choose a value of eps_r=10.0..

Modify the constant value C1 accordingly

Run the program

In free space wave is propagating freely and

on the other side reflection and transmission of the wave at
the interface

52 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Wave propagates more slowly on the medium on the right
and

the pulse length is shorter

FDTD simulation of TM wave propagation due to multiple sources
(fdtd_2d_TM_multi_source.m)
3. 3-D FDTD

From two Maxwell’s curl equations,

we need to consider all 6 equations for 3-D case
r r
∂E 1 r ∂H 1 r
= ∇× H; =− ∇× E
∂t ε 0ε r ∂t µ0 µr

53 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

54 FDTD by Rakhesh Singh Kshetrimayum Fig. 4 Yee’s FDTD grid 4/8/2018
 FDTD: An Introduction
Yee’s FDTD grid

Unlike 1-D and 2-D FDTD cases,

Electric fields are calculated at “integer” time-steps

and magnetic field at “half-integer” time-steps

Electric field components are placed at mid-points of the
corresponding edges

E.g.

Ex is placed at midpoints of edges oriented along x-
direction

Ex is on half-grid in x and the integer grids in y & z

55 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

The magnetic field components are

placed at the centers of the faces of the cubes and

oriented normal to the faces

E.g.

Hx components are placed at the centers of the faces in
the yz-plane

Hence Hx is

on the integer grid in x and

on the half-grid in y & z

56 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 Electric fields are calculated at
“integer” time-steps
FDTD: An Introduction and magnetic field at “half-integer” time-steps

Ex is placed at ∂E x 1  ∂H z

∂H y 

= −
midpoints of edges ∂t ε r ε 0  ∂y ∂z 
oriented along x- Hz is placed at center
direction of faces in xy-plane
E xn +1 ( p + 1 / 2, q, r ) − E xn ( p + 1 / 2, q, r )
=
∆t
1  H zn +1/ 2 ( p + 1 / 2, q + 1 / 2, r ) − H zn +1/ 2 ( p + 1 / 2, q − 1 / 2, r )
ε r ε 0  ∆y


1  H y ( p + 1 / 2, q, r + 1 / 2 ) − H y ( p + 1 / 2, q, r − 1 / 2 ) 
n +1/ 2 n +1/ 2

−  
ε r ε 0  ∆z 
Hy is placed at center of faces in xz-plane
57 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction
∂E y 1  ∂H x ∂H z 
=  − 
∂t ε r ε 0  ∂z ∂x 

E yn +1 ( p, q + 1 / 2, r ) − E yn ( p, q + 1 / 2, r )
=
∆t
1  H xn +1/ 2 ( p, q + 1 / 2, r + 1 / 2 ) − H xn +1/ 2 ( p, q + 1 / 2, r − 1 / 2 )
ε r ε 0  ∆z


1  H zn +1/ 2 ( p + 1 / 2, q + 1 / 2, r ) − H zn +1/ 2 ( p − 1 / 2, q + 1 / 2, r )
−
ε r ε 0  ∆z



58 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
∂E z 1  ∂H y ∂H x 
=  − 
∂t ε r ε 0  ∂x ∂y 

E zn +1 ( p, q, r + 1 / 2) − E zn ( p, q, r + 1 / 2 )
=
∆t
1  H y ( p + 1 / 2, q, r + 1 / 2 ) − H y ( p − 1 / 2, q, r + 1 / 2 ) 
n +1/ 2 n +1 / 2

 
ε r ε 0  ∆x 
1  H xn +1/ 2 ( p, q + 1 / 2, r + 1 / 2 ) − H xn +1/ 2 ( p, q − 1 / 2, r + 1 / 2 )
−
ε r ε 0  ∆y



59 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
∂H x 1  ∂E z ∂E y 
= −  − 
∂t µ 0  ∂y ∂z 

H xn +1/ 2 ( p, q + 1 / 2, r + 1 / 2 ) − H xn −1/ 2 ( p, q + 1 / 2, r + 1 / 2)
=
∆t
1  E y ( p, q + 1 / 2, r + 1) − E y ( p, q + 1 / 2, r ) 
n n

 
µ 0  ∆z 
1  E zn ( p, q + 1, r + 1 / 2 ) − E zn ( p, q, r + 1 / 2 ) 
−
µ 0  ∆y



60 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

∂H y 1  ∂E x ∂E z 
=−  − 
∂t µ 0  ∂z ∂x 

H yn +1/ 2 ( p + 1 / 2, q, r + 1 / 2 ) − H yn −1/ 2 ( p + 1 / 2, q, r + 1 / 2 )
=
∆t
1  E zn ( p + 1, q, r + 1 / 2 ) − E zn ( p, q, r + 1 / 2 ) 
µ 0  ∆x


1  E xn ( p + 1 / 2, q, r + 1) − E xn ( p + 1 / 2, q, r ) 
−
µ 0  ∆z



61 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

∂H z 1  ∂E y ∂E x 
= −  − 
∂t µ 0  ∂x ∂y 
H zn +1/ 2 ( p + 1 / 2, q + 1 / 2, r ) − H zn −1/ 2 ( p + 1 / 2, q + 1 / 2, r )
=
∆t
1  E xn ( p + 1 / 2, q + 1 / 2, r ) − E xn ( p + 1 / 2, q, r ) 
µ0  ∆y


1  E n
y ( p + 1, q + 1 / 2, r ) − E y ( p, q + 1 / 2, r )
n

−  
µ 0  ∆x 
62 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

FDTD simulation of cubical cavity (fdtd_3D_demo.m)

We will use FDTD to find

the resonant frequencies of an air-filled cubical cavity
with metal walls

Peaks in the frequency response indicate

the presence of resonant modes in the cavity

An initial Hz field excites only

the TE modes

The first four eigenfrequencies

TE101, TE011/TE201, TE111 and TE102 modes
63 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

Field components placed on a grid with 3×4×2 cells

For Ex (3×5×3) and Hx (4×4×2)

64 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Field components placed on a grid with 3×4×2 cells

For Ey (4×4×3) and Hy (3×5×2)

65 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Field components placed on a grid with 3×4×2 cells

For Ez (4×5×2) and Hz (3×4×3)

66 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Artificial absorbers on the walls of anechoic chamber

67 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
4. Perfectly Matched Layers

Berenger’s PML is an artificial material

that is theoretically designed to create no reflections
regardless of the

frequency,

polarization and

angle of incidence of a plane wave upon its interface

68 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

PML could be analyzed in stretched coordinate systems

Source-free Maxwell’s equations
r r r r
∇ stretched × E = − jωµ H ; ∇ stretched × H = jωε E ;
r r
( ) ( )
∇ stretched • ε E = 0; ∇ stretched • µ H = 0;
1 ∂ 1 ∂ 1 ∂
∇ stretched = xˆ + yˆ + zˆ
( )
sx x ∂x ( )
s y y ∂y ()
sz z ∂z

Consider plane waves
r r − jkr •rr r r − jkr •rr
E = E0 e ; H = H 0e
r r
k • r = kx x + k y y + kz z

69 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction

Substituting these into Maxwell’s equations, we have,
r r r r r r
k stretched × E = ωµ H ; k stretched × H = −ε E ;
r r r r
( ) ( )
k stretched • ε E = 0; k stretched • µ H = 0;
r kx ky kz
k stretched = xˆ + yˆ + zˆ
sx sy sz
2 2 2
r r  kx   k y   kz 
k stretched • k stretched =   +   +   = k 2
   
 sx   s y   s y 

Solution to the above equation is given by
k x = ks x sin θ cos φ , k y = ks y sin θ sin φ , k z = ks z cos θ
70 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction

If sx is a complex number with a negative imaginary part,

the wave will be attenuated in the x direction

e − jk x x
=e − jks x sin θ cosφx
=e ( )
− jk s ' − js '' sin θ cosφx
, s ' ≥ 1, s '' > 0

Wave impedance is independent of coordinate stretching

r
E ωµ ωµ µ
η= r = r = r =
H kstretched k ε

71 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018

 FDTD: An Introduction
' ''
4 corners : s z = 1,
s x = s z = 1, s y = s − js
s x = s y = s ' − js ''

y
s y = s z = 1, Object
' '' x s y = s z = 1,
s x = s − js
s x = s ' − js ''

s x = s z = 1, s y = s ' − js ''
Fig. 5 Computational domain truncated using the conductor backed PMLs
72 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018
 FDTD: An Introduction
References
1. S. D. Gedney, Introduction to the Finite-Difference Time-
Domain Method for Electromagnetics, Morgan &
Claypool, 2011
2. T. Rylander, P. Ingelstrom and A. Bondeson,
Computational Electromagnetics, Springer, 2013
3. J. M. Jin, Theory and Computation of Electromagnetic
Fields, IEEE Press, 2010
4. U. S. Inan and R. A. Marshall, Numerical Electromagnetics
The FDTD Method, Cambridge University Press, 2011

73 FDTD by Rakhesh Singh Kshetrimayum 4/8/2018