COMPUTATIONAL ELECTROMAGNETICS 2017, MINIPROJECT VI 1

Finite Difference Method Applied to a Uniform

Homogeneous Guide
David Sebastian Pulido, Member, IEEE, Luis Carlos Diaz, Member, IEEE,

Abstract—through the development of this mini project we set the boundary at the open sides of the guide to be PMC
pretend to enhance our skills in the numerical solution in the walls. In the same way, the gap between the upper plate
electromagnets field. In this work we present a computational and the top of the fin is set as a PMC wall due to the symmetry.
solution for the potential Φ( x, y) of an air-filled ridged parallel-
plate wave-guide using the Method of Finite Differences (FD).
The obtained results are compared with the ANSYS HFSS (High Once the problem is simplified, the computational domain
Frequency Electromagnetic Field Simulation) software results is discretized by squared or logarithmic meshing. Then, the
for the same topology in order to determine his accuracy. A spatial distribution of the potential φ(x, y) is approximated
distributed capacitance of 67,4662pF is obtained for plates with by a finite difference approach. The solution is achieved
20 mm long spaced 3 mm between them. The fin is 2 mm long using two variations of the FD method: matrix inversion
and is attached at the mid-point of the lower plate. With respect and matrix relaxation. Finally, the distributed parameters are
to the value simulated by HFSS, a relative error of 0.341% is estimated, the electric field is sketched and the two variations
achieved. The finite difference method is implemented with a are compared in terms of simulation time and accuracy.
logarithmic meshing.
Keywords—Capacitance, Electrostatic, FD, HFSS, Ridged, Wave- The operations are performed on MatLab MathWorks and
guide. compared with a simulation in the HFSS software which
solves the problem by using Finite Element Method (FEM).
I. I NTRODUCTION Therefore, there is available for comparison a trustfully
This work is done with the intention of apprehend the approach for the numerical problem.
basic notions of the FD method. In order to do that, we
estimate the spatial distribution of the potential φ(x, y) along
an air-filled ridged parallel-plate waveguide. The structure is
presented in figure1. The structure is assumed to be infinitely
long towards the z axis, then the problem can be solved using II. M ETHODOLOGY
a 2D approach.
A. FD Method applied to an electrostatic problem

Ir order to solve the problem by FD, we must consider the

y
z
V Laplace’s equation given by equation 1. Due to the complexity
x
of this operator, we should take into account a finite difference
approach that fits the given differential equation.

∇2 φ(x, y) = 0 (1)
b
h

a The proposed approximation is based on a second order

Taylor series as follows:

Fig. 1. 3D Structure of the two parallel plates with fin in the middle

f (x + h) ≈ f (x) + hf 0 (x) + h2 f 00 (x)

As the structure is symmetric in the middle, the problem
1 (2)
can be simplified by dividing the structure by one half as it f 00 (x) ≈ 2 [f (x + h) − f (x) − hf 0 (x)]
is depicted in figure 3. Also, is important to mention that h
we are not interested in the border effects edges, then we
Department of Electrical and Electronic Engineering, Bogotá Uni- Now, suppose there is a mesh like the one is depicted in fig-
versidad de los Andes, Colombia, e-mail: ds.pulido10@uniandes.edu.co, ure 2, it can be calculate the second derivative approximation
lc.diaz12@uniandes.edu.co over a (x, y) point by considering the equation 2.
COMPUTATIONAL ELECTROMAGNETICS 2017, MINIPROJECT VI 2

Y B. Meshing
1) Square meshing: Since the mesh is constituted by squares
(see figure 3), the distance between every node is the same.
Thus, h1x = h2x = h1y = h2y = ∆.
h1y
y
h2x h1x X
(x,y)
h2y

x
PEC PMC
Fig. 2. Spatial distribution for the finite differences approximation
Fig. 3. Square meshing
The resulting expression over any axis is given by:
For a square grid, equation 5 can be simplified as:
 
00 2 h2 h2
f (x) ≈ 2 f (x − h2 ) + f (x + h1 ) − ( + 1)f (x)
h2 + h2 h1 h1 h1
(3) Φ(x − ∆, y) + Φ(x + ∆, y)
Φ(x, y) =
Thus, the laplace’s equation may be replaced by: 4 (6)
Φ(x, y − ∆) + Φ(x, y + ∆)
+
4
∇2 φ(x, y) = 0
 2) Logarithmic meshing: Due to the specific topology of
2 h2x
∇2 φ(x, y) = 2 Φ(x − h2x , y) + Φ(x + h1x , y) the problem, we propose a different type of grid mesh, instead
h2x + h2x h1x h1x a regular rectangle, there are irregular rectangles spaced in a
logarithmic distribution as is shown in figure 4.

h2x
−( + 1)Φ(x, y) +
h1x
2

h2y
y
Φ(x, y − h2y ) + Φ(x, y + h1y )
h22y + h2y h1y h1y

h2y
−( + 1)Φ(x, y) = 0
h1y
(4)
Equation 4 can be solved to find Φ(x, y), yielding:
x
ay [Φ(x − h2x , y) + byΦ(x + h1x , y)] PEC PMC
Φ(x, y) =
Dxy Fig. 4. Logarithmic meshing
(5)
ax [Φ(x, y − h2y ) + bxΦ(x, y + h1y )]
+
Dxy The separation between the grid is given by the equation:
where: log10 (b)−log10 (a)
h(i) = 10log10 (a)+ N −1 i
(7)
ax = h22x + h2x h1x Where a, b are the initial and final point respectively, and
h2x N is the total number of separations between [a, b]. As the
bx = separation between nodes h depends on i, it is possible to
h1x establish analytic expressions for ax , ay , bx , by , Dxy that are
ay = h22y + h2y h1y also dependant on i an j indexes varying along the axis.
h2y
by = C. Distributed parameters estimation
h1y
Dxy = ax (by + 1) + ay (bx + 1) Once the potential values in all the nodes is calculated by
using matrix relaxation or matrix inversion, we can use that
If there is a grid in a 2D space problem, every node can be spatial distribution to compute the electric field E and the
represented like a variable φij . Furthermore, equation 5 can be distributed parameters C 0 ,Z 0 .
used to establish a relation between all the nodes potential.
COMPUTATIONAL ELECTROMAGNETICS 2017, MINIPROJECT VI 3

1) Electric field: The electric field can be evaluated as: 1) Matrix inversion: It was performed using the MatLab
Mathworks software, the elapsed time for the matrix inversion
E = −∇Φ(x, y) (8) was measured in order to compare it with with the relaxation
As the problem has been discretized, we can approach the variation.
electrical fiels as follows:
φ(x + hx , y) − φ(x, y) φ(x, y + hy ) − φ(x, y) 2) Matriz relaxation: We assume all the free nodes to be
E(x, y) = − î− ĵ zero for the first iteration, and set the value of the nodes that
hx hy
(9) are on the plates according to their respective voltage. The
It can be done along all the computational domain where iterative process is interrupted when the relative error between
Φ(x, y) is defined. the current and the previous iteration is less than a tolerance,
we set the tolerance value to be equal to the HFFS software
2) distributed capacitance C 0 : The distributed capacitance (tol = 1e − 6).
can be computed according to the Gauss’ law by:
I Finally, with the solution of the system we can calculate the
Q0 = 0 E(x, y) · ndL (10) distributed parameters and the electric field at each node that
belongs to the computational domain.
Again, due to the discretization, Q0 can be approximated as:
X
Q0 = 0 E n (x, y)h(i) (11)
conductors B. HFSS simulation
Thus, capacitance can be expressed as:
It is necessary to compare the results of the FD method
Q0 implementations with a trustfully result, to do so, a simulation
C= (12)
V in HFSS is performed. Since the original problem requires
Where V is the difference potential applied to the parallel a electrostatic solution, Maxwell 2D Design environment
plates. is used. Figure 5 shows the geometry implemented in the
software.
3) characteristic impedance Z 0 : The characteristic
impedance may be calculated as [1]:
√
0 µ0 0
Z = (13)
C0

III. I MPLEMENTATION
A. FD method Fig. 5. 2D Structure HFSS implementation
In Order to solve the problem, firstly we discretize the
computational domain using the squared and the logarithmic
grids, we were encouraged to use the logarithmic mesh to The computational domain is limited by a perfect magnetic
improve the resolution near to the fin. wall at the sides and it is applied 1 Volt to the upper conductor.
The conductors have a thickness of 0.01 mm and the relative
Once the problem is discretized, a linear system is build, error of the setup is fixed at 0.01% with a maximum of 20
where every potential value at each node is one independent passes refining the mesh. The software allow us to calculate
variable of the problem. The equation that relates all this the distributed capacitance C 0 and sketch the electric field
variable is the Laplace’s function approximation, described in inside the computational domain. These results are presented
equation 4. There are some considerations to build this system: in section IV.

1) For the conductors, the potential must be equal to the

applied voltage at each plate.
2) For every perfect magnetic wall, we must guarantee that
IV. R ESULTS
the potential value remains constant along the normal
direction.
3) For the remaining nodes, the Laplace’s function approx- Figure 6 depicts the results for the spacial distribution
imation is applied as it is. of the potential, represented by equipotential lines that take
After that, a linear system is obtained that can be solved values from 1 to 0 [V]. Additionally it is sketched the electric
using two variations: field distribution around the fin.
COMPUTATIONAL ELECTROMAGNETICS 2017, MINIPROJECT VI 4

3 3

2.5 2.5

2 2
Y [mm]

Y [mm]
1.5 1.5

1 1

0.5 0.5

0 0
-2 -1 0 1 2 -2 -1 0 1 2
X [mm] X [mm]

Fig. 6. Sketch of the electric field around the fin with a square mesh Fig. 7. Sketch of the electric field around the fin with a logarithmic mesh

The result has the same behavior as the HFSS simulation

Given the geometry of the problem, the mesh must be made (see figure 8). However, it approximates better to reality due
according to the spatial distribution of the electrical potential to its concentrated meshing in the region of interest.
along the computing domain. The existence of the fin causes
the potential lines to be compressed upwards as shown in figure
6. This variation introduces the need of a more refined mesh,
we propose a logarithmic mesh, concentred in the center of
the guide as it is shown in figure 4.

A. Distributed parameters estimation

The calculation of the distributed capacitance and the char-

acteristic impedance distributed according to equations 12 and
13 for a logarithmic mesh are presented in table I. In addition,
the variation with respect to spatial discretization is shown.
Fig. 8. Sketch of the electric field simulated by HFSS
TABLE I. D ISTRIBUTED PARAMETERS ESTIMATION FOR M ATRIX
INVERSION AND R ELAXATION VARIATIONS COMPARED WITH N

Matrix inversion Relaxation

V. D ISCUSSION AND C ONCLUSION
N C’ [pF] Z’ [Ω] Time (sec) C’ [pF] Z’ [Ω] Time (sec) The variation of matrix relaxation corresponds to a simple
10 69,8341 47,7653 0,8588 70,6822 47,1922 2,0715
20 68,1653 48,9346 1,8996 69,1146 48,2625 16,7077
method to solve recursively the equations without arriving
30 67,6893 49,2787 8,9036 69,0552 48,3040 75,6850 to establish a complete system of equations.This is due to
40 67,4662 49,4417 58,1138 68,8840 48,4241 273,2484 the fact that one solves in a punctual form for each free
node with all of its neighbors with a known value. However,
the computational time increase exponentially when more
For all the results contained in table I, satisfactory results are precision is required. On the other hand, the matrix inversion
obtained with a relative error of less than 5.124%, knowing that method consumes only the time it takes to invert the matrix,
the value obtained in the simulation in HFSS for the distributed which makes it more efficient in terms of time, and since it
capacitance is C 0 = 67.237pF , and for the characteristic is the solution of the system, the error will be less compared
impedance is Z 0 = 49.6102Ω. However, it is observed that to the relaxation method.
when it is used matrix inversion, more accurate results are
obtained in a shorter computing time. Also, it is important to mention that when approaching the
edges as PMC walls, it is neglected the edge effects which
also introduce some discrepancies in comparison with the
physical behaviour of the guide.
B. Electric field distribution

Finally, the electric field around the fin is sketched, in figure

R EFERENCES
7, for a logarithmic mesh with N = 40 solved with matrix
inversion. [1] M. Ney, “Lecture notes in computational electromagnetism,” July 2017.