MoM Slide

10.
Method of Moments
Dr. Rakhesh Singh Kshetrimayum
1 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

10.1 Introduction
learn how to use method of moments (MoM) to solve

electrostatic problems
advanced & challenging problems in time-varying fields
brief discussion on the basic steps of MoM
solve a simple differential equation using MoM
in order to elucidate the steps involved
MoM for 1-D and 2-D electrostatic problems
MoM for electrodynamic problems

10.2 Basic Steps in Method of Moments
Method of Moments (MoM) transforms
integro-differential equations into matrix systems of linear equations
which can be solved using computers
Consider the following inhomogeneous equation
L(u ) = k
⇒ L(u ) − k = 0
where L is a linear integro-differential operator,
u is an unknown function (to be solved) and
k is a known function (excitation)

For example,
(a) consider the integral equation for a line charge density
λ ( x ')dx '
V0 = ∫
4πε 0 r ( x, x ')
Then u = λ x' ( )
k = V0
dx '
L=∫
4πε 0 r ( x, x ')

(b) consider the differential equation of the form
d 2 f ( x) 2
− = 3 + 2 x
dx 2
Then
u = f ( x)
k = 3 + 2x2
d2
L=− 2
dx

To solve u, approximate it by sum of weighted known
basis functions or 1 ∞ ∞
f (x ) = a0 + ∑ an cos(nx ) + ∑ bn sin (nx )
expansion functions 2 n =1 n =1
as given below
N N
u ≅ ∑ un =∑ I n bn , n = 1, 2,..., N
n =1 n =1
where bn is the expansion function,

I n is its unknown complex coefficients to be determined,
N is the total number of expansion functions

Since L is linear, substitution of the above equation in the
integro-differential equation,
we get,
 N 
L  ∑ I nbn  ≈ k
 n=
n =1 
where the error or residual is given by
 N 
R = k − L  ∑ I n bn 
 n =1 

Mathematicians name this method as Method of Weighted
Residuals
Why?
Next step in MoM
Enforcing the boundary condition
Make inner product of the above equation with each of the
testing or
weighting functions
should make residual or error zero

By replacing u by u n
where n=1,2,…,N
taking inner product with a set of wm
weighting or
testing functions
in the range of L, we have,
wm , ( L ( un ) − k ) = 0, m = 1, 2,..., M

Since In is a constant
we can take it outside the inner product and
write
N
∑I
n =1
n wm , L ( bn ) = wm , k , m = 1, 2,..., M
M and N should be infinite theoretically

but practically it should be a finite number

Note that a scalar product w, g is defined to be a scalar
satisfying
w, g = g , w = ∫ g ( x) w( x) dx
bf + cg , w = b f , w + c g , w
g*, g > 0 if g ≠0 g*, g = 0 if g =0
b and c are scalars and * indicates complex conjugation

In matrix form
[Z ][I ] = [V ]
with each matrix and vector defined by
T
[I ] = [I 1 I 2 ... I N ]T [V ] =  k , w1 k , w2 ... k , wM 
 w1 , L ( b1 ) w1 , L ( b2 ) K w1 , L ( bN ) 
 
 w2 , L ( b1 ) w2 , L ( b2 ) ... w2 , L ( bN ) 
 
[ Z ] =  w3 , L ( b1 ) w3 , L ( b2 ) ... w3 , L ( bN ) 
 
M M O M 
 
 wM , L ( b1 ) wM , L ( b2 ) ... wM , L ( bN ) 
For [Z] is non-singular,
Solve the unknown matrix [I] of amplitudes of basis function
as
−1
[ I ] = [ Z ] [V ] = [Y ][V ]
Galerkin’s method
bn = wn
Point matching or Collocation
The testing function is a delta function

Methods for calculating inverse of a matrix
−1
Seldom find the inverse of matrix directly [ Z ] , because,
if we have ill-conditioned matrices,
it can give highly erroneous results
MATLAB command ‘pinv’ finds pseudo inverse of a matrix
using the singular value decomposition
For a matrix equation of the form AX=B,
if small changes in B leads to large changes in the solution X,
then we call A is ill-conditioned

The condition number of a matrix is the
ratio of the largest singular value of a matrix to the smallest
singular value
Larger is this condition value
closer is the matrix to singularity
It is always
greater than or equal to 1
If it is close to one,
the matrix is well conditioned
which means its inverse can be computed with good accuracy

If the condition number is large,
then the matrix is said to be ill-conditioned
Practically,
such a matrix is almost singular, and
the computation of its inverse, or
solution of a linear system of equations is
prone to large numerical errors
A matrix that is not invertible
has the condition number equal to infinity

Sometimes pseudo inverse is also used for finding
approximate solutions to ill-conditioned matrices
Preferable to use LU decomposition
to solve linear matrix equations
LU factorization unlike Gaussian elimination,
do not make any modifications in the matrix B
in solving the matrix equation

Try to solving a matrix equation [ A][ X ] = [ B ]
using LU factorization
First express the matrix
[ A] = [ L][U ]
 l11 0 0 L 0 u11 u12 u13 L u1N 
l 0 L 0  0 u u23 L u2 N 
 21 l22  22
[ L ] =  l31 l32 l33 L 0 and [U ] =  0 0 u33 L u3 N 

   
M M M O M  M M M O M 
lN 1 lN 2 lN 3 ... lN 1   0 0 0 ... u NN 

∴[ L ][U ][ X ] = [ B ] ⇒ [ L ][Y ] = [ B ]
through the forward substitution

b1 1 i −1

y1 = ; yi = bi − ∑ lik yk  , i > 1
l11 lii  k =1 
through the backward substitution [U ][ X ] = [Y ]

yN 1  N

xN = ; xi =  yi − ∑ uik xk  , i < N
u NN uii  k =i +1 
This is more efficient than Gaussian elimination
since the RHS remain unchanged during the whole process
The main issue here is to
find the lower and upper triangular matrices.
MATLAB command for LU factorization of a matrix A is
[L U] = lu(A)

Example 10.1
Consider a 1-D differential equation
d 2 f ( x) 2
− = 3 + 2 x
dx 2
subject to the boundary condition f(0)=f(1)=0
Solve this differential equation using Galerkin’s MoM
Solution:
Note that for this case,
u = f ( x)

k = 3 + 2 x2
d2
L=− 2
dx
According to the nature of the known function k = 3 + 2 x 2 ,

it is natural to choose the basis function as bn ( x ) = x
n
However,
the boundary condition f(1)=0
can’t be satisfied with such a basis function

A suitable basis function for this differential equation
taking into account of this boundary condition is
bn ( x ) = x − x n +1 ; n = 1, 2,..., N
Assume N=2 (the total number of subsections on the

interval [0,1])
Approximation of the unknown function
f ( x ) ≅ I1b1 ( x) + I 2b2 ( x) = I1 ( x − x 2 ) + I 2 ( x − x 3 )

For Galerkin’s MoM, the weighting functions are
wm ( x ) = x − x m+1; m = 1, 2,..., M
Choosing a square [Z] matrix where M=N=2

1 1
1
Z11 = w1 , L ( b1 ) = ∫ w1 ( x) L ( b1 ( x) ) dx = ∫ ( x − x 2 ) ( 2 ) dx =
0 0
3
d2
L=− 2
dx
1 1
1
Z12 = w1 , L ( b2 ) = ∫ w1 ( x) L ( b2 ( x) ) dx = ∫ ( x − x ) ( 6 x ) dx =
2
0 0
2

1 1
1
Z 21 = w2 , L ( b1 ) = ∫ w2 ( x) L ( b1 ( x) ) dx = ∫ ( x − x 3 ) ( 2 ) dx =
0 0
2
1 1
4
Z 22 = w2 , L ( b2 ) = ∫ w2 ( x) L ( b2 ( x) ) dx = ∫ ( x − x 3 ) ( 6 x ) dx =
0 0
5
1 1
3
V1 = k , w1 = ∫ k ( x) w1 ( x) dx = ∫ ( 3 + 2 x 2 ) ( x − x 2 )dx =
0 0
5
k = 3 + 2 x2 1 1
11
V2 = k , w2 = ∫ k ( x) w2 ( x) dx = ∫ ( 3 + 2 x 2 ) ( x − x 3 ) dx =
0 0
12

Therefore, 1 1 3
3 2   I1   5 
[ Z ][ I ] = [V ] ⇒  1  = 
4   I 2   11 

 2 5  12 
 13 
 I1  10 
⇒ [I ] =   =  
I2   1 
 3 
The unknown function f(x)
13 1
f ( x) ≅ I1 ( x − x 2 ) + I 2 ( x − x 3 ) = ( x − x 2
) + ( x − x3 )
10 3

The above function satisfies the given boundary conditions
f(0)=f(1)=0
The analytical solution for this differential equation is
5 3 1
f ( x) = x − x2 − x4
3 2 6
Check whether the above solution using MoM is

different from the analytical solution obtained by direct
integration (see Fig. 10.1)

Analytical
and MoM
solution:
almost
same
coinciding
even with
two basis
functions
Fig. 10.1 Comparison of exact solution (analytical) and approximate

solution (MoM) of Example 10.1

10.3 Introductory examples from electrostatics
In electrostatics, the problem of finding the potential
due to a given charge distribution is often considered
In practical scenario, it is very difficult to
specify a charge distribution
We usually connect a conductor to a voltage source
and thus the voltage on the conductor is specified
We will consider MoM
to solve for the electric charge distribution
when an electric potential is specified
Examples 2 and 3 discuss about calculation of inverse using LU
decomposition and SVD

1-D Electrostatic case: Charge density of a straight wire
Consider a straight wire of length l and radius a (assume
a<<l),
placed along the y-axis as shown in Fig. 10.2 (a)
The wire is applied to a constant electric potential of 1V
Choosing observation along the wire axis (x=z=0) i.e.,
along the y-axis
and representing the charge density on the surface of the wire
l
1 λ ( y ' )dy '
1=
4πε 0 ∫ R( y, y ' )
0

∆

Fig. 10.2
(a) Straight wire of length l and radius a applied with a
constant potential of 1V
(b) Its segmentation: y1, y2, …, yN are observation points and
r′ shows a source point
(c) Division of the charged strip into N sections

where
r r
R( y, y ' ) = R(r , r ' ) = ( y − y ' )2 + ( x ' )2 + ( z ' ) 2 = ( y − y ' )2 + (a)2
x= z =0
It is necessary to solve the integral equation

to find the unknown function λ(y′)
The solution may be obtained numerically by
reducing the integral equation into a series of linear algebraic
equations
that may be solved by conventional matrix techniques

(a) Approximate the unknown charge density λ(y′)
by an expansion of N known basis functions with unknown
coefficients
N
λ( y' ) = ∑ I n bn ( y ' ) 1
l
λ ( y ' ) dy '
n =1 1=
4πε 0 ∫ R( y, y ' )
0
Integral equation after substituting this is
N
l ∑ I n bn ( y ' )dy ' N l
bn ( y ' )dy '
4πε 0 = ∫
n =1
R( y, y ' )
= ∑ In ∫ R ( y , y ')
0 n =1 0

Now we have divided the wire into N uniform segments each
of length Δ as shown in Fig. 10.2 (b)
We will choose our basis functions as pulse functions
1
()
bn y ' = 
0
for ( n − 1) ∆ ≤ y ' ≤ n∆
otherwise
b) Applying the testing or weighting functions

Let us apply the testing functions as delta functions  ∂ ( y − ym ) 
for point matching

Integration of any function with this delta function
will give us the function value at y = ym
Replacing observation variable y by a fixed point such as ym,
results in an integrand that is solely a function of y′
so the integral may be evaluated. N
It leads to an equation l ∑ I n bn ( y ' )dy ' N l
bn ( y ' )dy '
with N unknowns
4πε 0 = ∫
n =1
R( y, y ' )
= ∑ In ∫ R ( y , y ')
0 n =1 0
∆ 2∆ n∆ l
b1 ( y ' )dy ' b2 ( y ' ) dy ' bn ( y ' )dy ' b N ( y ' )dy '
4πε 0 = I1 ∫ R( y , y')
+ I2 ∫ R ( y , y ')
+ ... + I n ∫ R ( y , y ')
+ ... + I N ∫ R ( y , y ')
0 m ∆ m ( n −1) ∆ m ( N −1) ∆ m

Solution for these N unknown constants,

N linearly independent equations are required
N equations may be produced
by choosing an observation point ym on the wire
where m=1,2,3…, N and
at the center of each Δ length element
as shown in Fig. 10.2 (c)
Result in an equation of the form of the previous equation
corresponding to each observation point

For N such observation points we have
∆ 2∆ n∆ l
b1 ( y ' )dy ' b2 ( y ' ) dy ' bn ( y ' )dy ' b N ( y ' ) dy '
4πε 0 = I1 ∫ R( y , y ' ) + I 2 ∫ '
+ ... + I n ∫ '
+ ... + I N ∫ '
0 1 ∆ R ( y1 , y ) ( n −1) ∆ R ( y1 , y ) ( N −1) ∆ R ( y1 , y )
∆ 2∆ n∆ l
b1 ( y ' )dy ' b2 ( y ' ) dy ' bn ( y ' )dy ' b N ( y ' ) dy '
4πε 0 = I1 ∫ R( y , y')
+ I2 ∫ R ( y , y ')
+ ... + I n ∫ R ( y , y ')
+ ... + I N ∫ R ( y , y ')
0 2 ∆ 2 ( n −1) ∆ 2 ( N −1) ∆ 2
∆ 2∆ n∆ l
b1 ( y ' )dy ' b2 ( y ' )dy ' bn ( y ' ) dy ' b N ( y ' ) dy '
4πε 0 = I1 ∫ R( y , y')
+ I2 ∫ , y')
+ ... + I n ∫ , y')
+ ... + I N ∫ '
0 N ∆ R( y N ( n −1) ∆ R ( y N ( N −1) ∆ R ( y N , y )

(c) We may write the above equations in matrix form as
 Z11 Z12 ...Z1N   I1  V0 
Z Z 22 ...Z 2 N   I 2  V0 
 21
 Z 31 Z 32 K Z 3 N   I 3  = V0  ⇒ [ Z mn ][ I n ] = [Vm ]
    
M M O M   M  M 
 Z N 1 ZN 2 ... Z NN   I N  V0 
where [Vm ] = [4πε 0 ]

l yn
bn ( y ' )dy ' dy '
Z mn = ∫ = ∫
' 2 2
0 ( ym − y ) + a yn−1 ( ym − y ' ) 2 + a 2
yn yn
dy ' dy ' ∆
≅ ∫
yn−1 ( ym − y ) ' 2
= ∫
yn−1
ym − y '
≈
ym − yn
for m≠n

Special care for calculating the Zmn for m=n case
since the expression for Zmn is infinite for this case
Extraction of this singularity
Substitute ym − y ' = ξ ⇒ d ξ = − dy '
0 ∆ ∆
dξ dξ
Z mn = − ∫
∆
2
(ξ ) + a 2
=∫
0
2
(ξ ) + a 2 ( 2
= log ξ + (ξ ) + a 2
) 0
 ∆ + ∆2 + a2 
= ln  
 a 

Self or diagonal terms are the
most dominant elements in the [Z] matrix
Note that linear geometry of this problem
yields a matrix that is symmetric toeplitz, i.e.,
 Z11 Z12 ... Z1N 

Z Z11 ... Z1N −1 
[ Z mn ] =  .12 . . . 
 
 Z1N Z1N −1 ... Z11 

All other rows are a rearranged version of the first row
Required to calculate the first row of the matrix only
Remaining elements can be obtained by the rearrangement
formula:
Z mn = Z1, m − n +1 , m ≥ 2, n ≥ 1
Therefore the unknown [I] matrix could be solved as
[I n ] = [Z mn ]−1[Vm ]

Fig. 10.3 (a) Convergence plot of Z11 and Z21 (b) Plot of line charge density of
the wire (MATLAB program provided in the book)

Let us see the convergence of these two types of elements of
the Z matrix say,
Z11 and Z21
Fig. 10.3 (a) shows the convergence plot of two elements of
the Z matrix
for number of sub-sections varying from 5 to 100
The graph of Z21 (dashed line) versus number of sub-sections
is a straight line
so any number of sub-sections between 5 and 100 should give
the same result

But the graph of Z11 versus number of sub-sections is
decreasing quite fast at the initial values of number of sub-
sections and
it is decreasing more slowly for larger values of number of sub-
sections
This shows that at
higher values of number of sub-sections,
we will get a more convergent result
Choose the maximum number of sub-sections and
plot the line charge density as depicted in the Fig. 10.3 (b)

See the condition number of the [Z] matrix in order to see
whether the [Z] matrix is well-behaved or not
The condition number of [Z] matrix
(=7.1409) for maximum number of sub-sections is good
No problem in taking the inverse
Fig. 10.3 (b) line charge density is
maximum at the two end points of the wire and
minimum at the center of the wire
2-D Electrostatic case: Charge density of a square conducting
plate discussed in the book

10.4 Some commonly used basis functions
The weighted sum of basis functions is
used to represent the unknown function in MoM
Choose a basis function that reasonably approximates
the unknown function over the given interval
Basis functions commonly used in antenna or scattering
problems are of two types:
entire domain functions and
sub-domain functions

10.4.1 Entire domain basis functions
The entire domain functions exist over the full domain
-l/2<x<l/2
Some examples are:
 n − 1 2x 
Fourier (is well known) bn ( x) = cos( ) 
2 l  
2x
Chebyshev (will discuss briefly) bn ( x) = T2n − 2 (
l
)
2x
Legendre (will discuss briefly) bn ( x) = P2n − 2 ( )
l
where n=1,2,3,…,N.

Chebyshev's differential equation
(1 − x 2 )y '' − xy ' + n 2 y = 0
where n is a real number
Solutions Chebyshev functions of degree n
n is a non-negative integer, i.e., n=0,1,2,3,…,
the Chebyshev functions are called Chebyshev polynomials
denoted by Tn(x)

A Chebyshev polynomial at one point can be
expressed by neighboring Chebyshev polynomials at the same
point
T2 ( x ) = 2 xT1 (x ) − T0 (x )
Tn +1 (x ) = 2 xTn (x ) − Tn −1 (x )
whereT0(x)=1, T1(x)=x
Legendre's differential equation
(1 − x 2 )y '' − 2xy ' + n(n + 1)y = 0

where n is a real number

Solutions of this equation are called Legendre functions of
degree n
When n is a non-negative integer, i.e., n=0,1,2,3,…,
the Legendre functions are called Legendre polynomials denoted
by Pn(x)
Legendre polynomial at one point can be
expressed by neighboring Legendre polynomials at the same point
(n + 1)Pn +1 (x ) = (2n + 1)xPn (x ) − nPn −1 (x )

2 P2 (x ) = 3 xP1 ( x ) − P0 ( x )
where P (x)=1, P (x)=x
0 1

Disadvantage: entire domain basis function may not be
applicable of any general problem
Choose a particular basis function for a particular problem
Crucial and only experts in the area could do it efficiently
Developing a general purpose MoM based software,
software for analyzing almost every problem in
electromagnetics
this is not feasible
Sub-domain basis functions could achieve this purpose

10.4.2 Sub-domain basis functions
Sub-domain basis functions exist only on one of the N
overlapping segments
into which the domain is divided
Some examples are:
Piecewise constant function (pulse)
1 x[n − 1] < x < x[n]

bn ( x) = 
0 otherwise

Piecewise triangular function
∆ − x − xn
 x[n − 1] < x < x[n + 1]
bn ( x) =  ∆

 0 otherwise
 x − xn −1
x − x x[n − 1] < x < x[n]
 n n −1
 x −x
=  n +1 x[n] < x < x[n + 1]
 xn +1 − xn
 0 otherwise



Piecewise sinusoidal function {
 sin k ( ∆ − x − xn ) }
 x[n − 1] < x < x[n + 1]
bn ( x) =  sin ( k ∆ )

 0 otherwise
 sin {k ( x − xn )}
 x[n − 1] < x < x[n]
 { ( n n −1 )}
sin k x − x

 sin {k ( xn +1 − x )}
= x[n] < x < x[n + 1]
 sin {k ( xn − xn −1 )}
 0 otherwise



where Δ=l/N, assuming equal subintervals but it is not mandatory
and k is a constant

Fig. 10.5 Sub-domain basis functions (a) Piecewise constant

function (b) Piecewise triangular function (c) Piecewise
sinusoidal function

Since the derivative of the pulse function is impulsive
we cannot employ it for MoM problems

o where the linear operator L consists of derivatives
Piecewise triangular and sinusoidal functions
may be used for such kinds of problems
Piecewise sinusoidal functions are generally used
for analysis of wire antennas since
they can approximate sinusoidal currents in the wire
antennas

10.5 Wire Antennas and Scatterers
For Piece-wise triangular and sinusoidal functions
when we have N points in an interval
we will have N-1 sub-sections and
N-2 basis functions may be used

Consider application of MoM techniques
to wire antennas and scatterers

Antennas can be distinguished from scatterers
in terms of the location of the source
If the source is on the wire
it is regarded as antenna
When the wire is far from the source
it acts as scatterer
For the wire objects (antenna or scatterer)
we require to know the current distribution accurately

Integral equations are derived and
solved for this purpose
Wire antennas
Feed voltage to an antenna is known
and the current distribution could be calculated
other antenna parameters such as
impedance,
radiation pattern, etc.
can be calculated

Wire scatterers
Wave impinges upon surface of a wire scatterer
it induces current density
which in turn is used to generate the scattered fields
We will consider
how to find the current distribution on a
thin wire or
cylindrical antenna
using the MoM

10.5.1 Electric field integral equation (EFIE)
On perfect electric conductor like metal
the total tangential electric field is zero
Centrally excited cylindrical antenna (Fig. 10.6)
have two kinds of electric fields viz.,
incident and
scattered electric fields
r t ot r inc r scat r inc r scat

Etan = 0 ⇒ Etan + Etan = 0 ⇒ Etan = − Etan

2∆
Fig. 10.6 A thin wire antenna of length L, radius a (a<<L)

and feed gap 2Δ

r inc
where the is the source or impressed field and
E
r scat
E can be computed from the
current density induced on the cylindrical wire antenna due to the
incident or
impressed field
10.5.2 Hallen’s and Pocklington's Integro-differential equation
Let us consider a perfectly conducting wire of
length L and
radius a such that a<<L and λ, the wavelength corresponding to the
operating frequency

Consider the wire to be a hollow metal tube
open at both ends
r
Let us assume that an incident wave E inc ( rr )
impinges on the surface of a wire
When the wire is an antenna
the incident field is produced by the feed at the gap (see Fig.
10.6)
The impressed field Ezinc is required
to be known on the surface of the wire

Simplest excitation
delta-gap excitation
For delta gap excitation (assumption)
excitation voltage at the feed terminal is constant and
zero elsewhere
Implies incident field
constant over the feed gap and
zero elsewhere

2V0 (from + V0 to -V0) voltage source applied
across the feed gap 2Δ,
Incident field on the wire antenna can be expressed as
V0
r inc  ∆ ; z <∆
Ez =
 0; ∆ < z < L
2∆
 2
Induced current density
due to the incident or impressed electric field
r scat r
produces the scattered electric field E (r )

The total electric field is given by
r tot r r inc r r scat r
E (r ) = E (r ) + E (r )
Since the wire is assumed to be perfectly conducting,

tangential component of the total electric field on the surface of
the wire is zero
For a cylindrical wire placed along z-axis, we can write,
r tot r r inc r r scat r

Ez ( r ) = Ez ( r ) + Ez ( r ) = 0; on the wire antenna

that is,
r scat r r inc r
Ez ( r ) = − Ez ( r )
Find the electric field from the potential functions using
r r
E = − jω A − ∇ V
Lorentz Gauge condition,

r
∇ • A = − jωµ 0 ε 0V

For a thin cylinder,
current density considered to be independent of φ
1
J z (z ' ) = I (z ' )
2πa
where is J z (z ' ) the surface current density

at a point on the conductor z '
skin depth of the perfect conductor is almost zero
and therefore all the currents flow on the surface of the wire

Jz
Fig. 10.7 Cylindrical conductor of radius a with surface current

density  A
J z (z' )  
m
and its equivalence to the case of the conductor replaced by current

filament I ( z ' ) = 2π aJ z ( z ' ) ( A) at a distance a from the z-axis

The current I ( z ' ) may be assumed to be
a filamentary current located parallel to z-axis
at a distance a (a is a very small number) as shown in the Fig. 10.7
For the current flowing only in the z direction,
∂V
E z = − jωAz −
∂z
From Lorentz Gauge condition for time harmonic case,
∂Az ∂ 2 Az ∂V ∂V 1 ∂ 2 Az
= − jωµ0ε 0V ⇒ 2
= − jωµ0ε 0 ⇒− =
∂z ∂z ∂z ∂z jωµ0ε 0 ∂z 2

Therefore,
1∂ 2 Az 1  2 ∂ 2 Az  1  2 ∂ 2 Az 
Ez = − jω Az + =  ω µ0ε 0 Az + 2 =  β0 Az + 2 
jωµ0ε 0 ∂z 2 jωµ0ε 0  ∂z  jωµ ε
0 0  ∂z 
Magnetic vector potential can be expressed as

e − jβ 0 r '
Az = µ 0 ∫∫ Jz
4πr
ds
s
Putting the Jz expression from (10.20), we have,
L / 2 2π
I ( z ' ) e − jβ0 r
Az = µ0 ∫ ∫
−L/2 0
2π a 4π r
adφ ' dz '

where
r= ( ' 2
) (
' 2
x−x + y− y + z−z ' 2
) ( )
For ρ =a
L/2
 1 2π
e − j β0 r '  '
Az ( ρ = a ) = µ0 ∫ I ( z )  '
∫ dφ dz
−L/ 2  2π 0
4π r 
where
ϕ' 
( ) 2
r ρ = a = 4a sin 2   + z−z '
 2
2
( )
 

Therefore, we can write
L/2
Az ( ρ = a ) = µ 0 ∫ I ( z ' )G ( z , z ' )dz '
−L / 2
where
2π
1 e − j β0 r
G ( z, z ' ) =
2π ∫
0
4π r
dφ '
r r
G (r , r ' )
is the field at the observation point caused by a unit
point source placed at rr '

r
The field at r by a source distribution J (rr ' )
r r r r'
is the integral of J (r ' )G (r , r ' ) over the range of r occupied
by the source
The function G is called the Green's function
We have,
1 ∂2 A 
Ez =  z + β 0 Az 
2
jωε 0 µ 0  ∂z 2 
 
and
L/2
 1 2π
e − j β0 r '  '
Az ( ρ = a ) = µ0 ∫ I ( z ) 
'
∫ dφ dz
−L/2  2π 0
4π r 

From the above two equations we can write, two equations:

(a) 1  ∂2 2
L/2
 2 + β 0  ∫ I ( z )G ( z, z )dz
' ' '
Ez =
jωε 0  ∂z  −L/ 2
'
This electric field is the field due to current I ( z )
[which results because of the impressed or source field] and
this field can be written as the scattered field
Therefore,
1  ∂2  L/2
 2 + β 0  ∫ I ( z )G ( z, z )dz
2 ' ' '
Ezscat =
jωε 0  ∂z  −L/ 2

Since from the EFIE on the surface of the wire,
Ezscat ( ρ = a ) = − Ezinc ( ρ = a )
1  ∂2 2
L/2
0  ∫
' ' ' inc
 + β I ( z )G ( z , z )dz = − E z ( ρ = a)
jωε 0  ∂z 2  −L/ 2
This equation is called the Hallen's Integro-differential equation

In this case, differential is outside the integral

L/2  2  '
(b) 1  ∂
+ β 0  I ( z )G ( z , z ' )dz '
∫
Ez = 2
jωε 0  2 
− L / 2  ∂z 
This electric field is the field due to current I ( z ' )

[which results because of the impressed or source field] and
this field can be written as the scattered field
Therefore,
1
L/2
 ∂2 2
jωε 0 − L∫/ 2  ∂z 2
scat ' ' '
E z =  + β 0  I ( z )G ( z , z )dz


Since from EFIE on the surface of the wire,
Ezscat ( ρ = a ) = − Ezinc ( ρ = a )
1
L/2
 ∂2 2
∫
' ' ' inc
 2
+ β 0  I ( z )G ( z , z )dz = − E z ( ρ = a)
jωε 0 − L / 2  ∂z 
This equation is the Pocklington's Integro-differential equation

In this case, the differential has moved inside the integral
Richmond has simplified the above equation as follows:

(c) In cylindrical coordinates,
r r ' 2 r r' 2
r = r −r' = (z − z ) + ρ−ρ
r r r r
Q ρ ' = a ∴ ρ − ρ ' = ρ 2 + a 2 − 2 ρ • ρ ' = ρ 2 + a 2 − 2 ρ a cos ( φ − φ ' )
r r 2
⇒ r = r − r ' = ρ 2 + a 2 − 2 ρ a cos (φ − φ ' ) + ( z − z ' )
Problem under analysis has cylindrical symmetry and

observation for any values of φ won’t make any difference
we may assume without loss of generality φ = 0
' '
hence φ − φ = φ

2π
e − jβ0 r ' '
L/2
I (z' )
Az = µ0 ∫ ∫ dφ dz
−L/2
2π 0
4π r
where r = ρ + a − 2 ρ a cos (φ ) + ( z − z
2 2 ' ' 2
)
2π
e − j β0 r '
and the inner integration ∫
0
4π r
dφ
is also referred to as cylindrical wire kernel

2
Thin wire approximation r ≅ ρ 2 + ( z − z' )
If we assume a<<λ and is very small, we have, φ'

Inner integrand is no more dependent on the variable

Therefore
I ( z ' )e − j β 0 r '
L/2
Az = µ0 ∫ dz
−L/ 2
4π r
Also called as thin wire approximation

with the reduced kernel
For this case, we can write
e − j β0 r
G ( z, z ) ≅
'
= G (r )
4π r

Now in the light of this simplification of the magnetic vector

potential,
we can simplify equation 10.29c (see example 10.4) as
follows:
e − j β0 r 
L/2
1 dz ' = − Ezinc ( ρ = a )
jωε 0 4π − L∫/ 2
I z ( '
z )
r5 
(1 + j β 0 r ) ( 2 r 2
− 3a 2
) + ( β 0 ar )
2
This form of the Pocklington’s integro-differential is more

suitable for MoM formulation
since it does not involve any differentiation.

10.5.3 MoM Formulation of Pocklington's Integro-differential
equation
Applying MoM formulation to above integral equation
Divide the wire in to N segments
Consider pulse basis function and
express the current as a series expansion
in the form of a staircase approximation as
N where
I ( z ) = ∑ I n bn ( z ' )
'
n =1
' 1 for ∆zn'
bn ( z ) = 
0 otherwise

zn'
∆zn'
Fig. 10.8 Thin wire dipole is divided into N equal segments

∆z n' is the length of the nth segment, expressed as
L L L L
− + (n − 1) < z ≤ − + n
2 N 2 N
L/2
− Ezinc = ∫
−L / 2
I ( z ' ) F ( z , z ' ) dz '
 ∂2 
where '
F ( z, z ) =
1 
jωε  ∂z 2
+ β 0 G z, z '
2

( )
 

Substituting I(z′) and
evaluating at z=zm (middle of the mth segment) as shown in the Fig. 10.7
for point matching with weighting functions as wm ( z ) = δ ( z − zm )
where zm is the center of the segment m

L L
zm = − + (m − 0.5)
2 N
and m=1,2,3,…,M, we can write,

N
− E zi (z m ) = ∑ I n ∫ F (z m , z ' )dz '
n =1 ∆z n'

To overcome the singularity
for the self term or diagonal elements of the [Z] matrix
we have assumed that the source is on the surface of the wire
whereas the observation is the axis of the wire
Using mid-point integration, we have,
N
− E zi (z m ) = ∑ I n Fmn
n =1
∫ F (z m , z )dz ( )
' ' ≅ F z , z ' ∆z '
where Fmn = m n n
∆z n'

For m=1,2,…,M,
 F11 F12 ... F1N   I1   − Ezinc ( z1 ) 

F  
 21 F22 ... F2 N   I 2   − Ezinc ( z2 ) 
=
 : : : :   :   
    
 FM 1 FM 2 ... FMN   I N   − Ezinc ( z N ) 
In the compact form, we can write

[Fmn ][I n ] = −[E m ]
Multiplying both sides by Δz, we can write,
[ Z mn ][ I n ] = − [Vm ]

[In] can be computed by matrix inversion
Can find the approximate current distribution on the antenna
Other important antenna parameters are
input impedance of the antenna and
the total radiated fields
which can be obtained as:
2V0
Z input =
IN
2
r tot r r inc r r scat r r inc r L/2 N

1  ∂ 2
 e − j β0 r
E (r ) = E (r ) + E (r ) = E (r ) + ∫ ∑ I n bn ( z '
) 
jωε 4π  ∂z 2
+ β 2
0  dz '
− L / 2 n =1  r

Example 10.5
Consider a short dipole (thin wire antenna) of length 0.3λ
and radius 0.01λ.
Find the current distribution on the short dipole and input
impedance using MoM.
Assume frequency of operation of the antenna is at 1 MHz.
Choose the number of discretizations on the thin wire as
three segments only.

For three segments discretization ( ∆z = 0.1λ ) on the thin
wire antenna
the MoM matrix equation will be
 Z11 Z12 Z13   I1  V1 

Z Z 22 Z 23   I 2  = − V2 
 21
 Z 31 Z 32 Z 33   I 3  V3 
 F11 F12 F13   I1   E1 
⇒ ∆z  F21 F22 F23   I 2  = −∆z  E2 
 F31 F32 F33   I 3   E3 

where
e − j β 0 rmn (1 + j β 0 rmn ) ( 2rmn

2
− 3a ) + ( β 0 armn )  ∆z
2 2
Z mn = Fmn ∆z =  
jωε 0 4π rmn 5
Simplifying the above expression of the Zmn (see book)
∆z  rmn    a    a     rmn 
2 2
1 + j 2π     rmn  
− jZ 0 2 − 3   +  2π   cos 2π − j sin  2π 
λ  λ   r
 mn     λ   

 λ 
  λ  
  
Z mn = 3
2  rmn 
8π λ  
 λ 

For thin wire approximations,
2
rmn ≅ ρ 2 + ( zm − zn )
∴ r11 = a 2 = a = r22 = r33 = 0.01λ ;

r12 = a 2 + ∆z 2 = r21 = r23 = r32 = 0.1λ ;
2
r13 = a 2 + ( 2∆z ) = r31 = 0.2λ
[Z] symmetric Toeplitz matrix (need to calculate the first row of

the matrix only)  Z1 Z2 Z3 
[ Z ] =  Z 2 Z1 Z 3 
 Z 3 Z2 Z1 
About the [V] matrix, for delta gap excitation,
0 
[V ] = − 1 
0 
since a voltage 1V exists at the feed gap only

Solve the [I] matrix and get the current distribution
and the input impedance can be calculated as
2V0 1
Zinput = =
IN I2
2

10.6 Software language for implementation
of electromagnetics codes
Choice of software languages for implementing electromagnetics
code
FORTRAN language complex numbers are a built-in
datatype
Many computational electromagnetics programmer prefer
to use FORTRAN language
for implementing their algorithms
Earlier versions of FORTRAN were a functional language

New versions of FORTRAN are object-oriented languages

C++, another object-oriented language,
is also widely used for many numerical methods
FORTRAN and C++ are efficient in implementation
since the computational time is less
MATLAB is also convenient environments since
it accepts complex numbers,
graphics are very easy to create and
many in-built functions are readily available for use

MATLAB any additional “for” loop in the program,

the time it takes to run the program increases drastically
Good to consider the advantages and disadvantages
for employment of any software language
For instance,
drawing graphics in C is somewhat involved,
but MATLAB is convenient for such things.
program written in C runs faster than MATLAB and so on

Computational electromagnetics is a topic
which you can learn only by doing
Some simulation exercises are given at the end of the chapter,
you should always write down a
MATLAB or
any other software language program
in which you are comfortable and
see those results

10.7 Summary
Summarize the three steps involved in MoM:
(a) Derivation of appropriate integral equations
(b) Conversion or discretization of the integral equation into
a matrix equation using
basis or expansion functions and
weighting or testing functions
as well as evaluation of the matrix elements
(c) Solving the matrix equation and
obtaining the desired parameters

MoM Slide

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MoM Slide

Uploaded by

Copyright:

Available Formats

10.

Dr. Rakhesh Singh Kshetrimayum

1 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

learn how to use method of moments (MoM) to solve

2 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

3 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

4 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

5 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

where bn is the expansion function,

6 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

where the error or residual is given by

7 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

8 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

9 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

M and N should be infinite theoretically

10 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

g*, g > 0 if g ≠0 g*, g = 0 if g =0

b and c are scalars and * indicates complex conjugation

11 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

13 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

14 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

15 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

16 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

17 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

[ L ] =  l31 l32 l33 L 0 and [U ] =  0 0 u33 L u3 N 

18 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

through the forward substitution

through the backward substitution [U ][ X ] = [Y ]

20 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

21 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

According to the nature of the known function k = 3 + 2 x 2 ,

22 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Assume N=2 (the total number of subsections on the

23 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Choosing a square [Z] matrix where M=N=2

24 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

25 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

26 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Check whether the above solution using MoM is

27 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Fig. 10.1 Comparison of exact solution (analytical) and approximate

28 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

29 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

30 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

31 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

32 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

It is necessary to solve the integral equation

33 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

34 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

b) Applying the testing or weighting functions

35 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

36 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Solution for these N unknown constants,

37 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

38 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

where [Vm ] = [4πε 0 ]

39 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

40 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

 Z11 Z12 ... Z1N 

41 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

Therefore the unknown [I] matrix could be solved as

42 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

43 Electromagnetic Field Theory by R. S. Kshetrimayum 4/27/2016

g, g > 0 if g ≠0 g, g = 0 if g =0