35
CHAPTER 4
The Locally-Corrected Nyström
Method
In this chapter, the Nyström method is extended in order to better handle singularities in the
kernel of the integral operator. The extension, known as the locally corrected Nyström (LCN)
method, was originally proposed in [13]. In contrast to the conventional Nyström method, the
LCN enables a high accuracy discretization of integral operators whose kernels have either weak
or strong singularities. The LCN will be applied to two-dimensional electromagnetic scattering
problems using the electric-ﬁeld and magnetic-ﬁeld integral equations.
4.1 THE LOCALLY-CORRECTED NYSTRÖM METHOD
A fundamental difﬁculty with the classical Nyström method is that it fails if the kernel of the
integral operator is inﬁnite at one of the quadrature sample points. Although the MFIE considered
in Chapter 3 has a bounded kernel, most of the operators of interest in electromagnetics have kernels
with stronger singularities (in fact,most are inﬁnite when the source and observer locations coincide).
As demonstrated in Chapter 3, even if the kernel is bounded, the fact that it is not analytic degrades
the accuracy of the overall Nyström solution.
One remedy to this situation was proposed in [13], based on an approach for enhancing the
accuracy of quadrature rules [4]. The original idea was to synthesize a new quadrature rule when
the source and observer locations are in close proximity. An essentially equivalent approach is to
synthesize a new (bounded and smooth) kernel to be sampled by the original quadrature rule for
near-ﬁeld interactions. In equation form, the original approximation of the integral operator over a
cell (using the notation from Chapter 3)
cell n
J
z
(t
)K(t
mj
,t
)dt
=
q
i=1
w
ni
J
z
(t
ni
)K(t
mj
,t
ni
) (4.1)
is replaced by
cell n
J
z
(t
)K(t
mj
,t
)dt
=
q
i=1
w
ni
J
z
(t
ni
) L(t
mj
,t
ni
) (4.2)
where L is the new corrected” kernel, which must be synthesized at the necessary samples. One
way to accomplish that is to derive L so that the near ﬁelds of some set of hypothetical current
distributions are correct at the necessary sample points. The hypothetical currents are, in essence,
36 CHAPTER 4. THE LOCALLY-CORRECTED NYSTRÖM METHOD
just a set of basis functions
{
B
k
(t)
}
. By imposing the condition that
q
i=1
w
ni
B
k
(t
ni
) L(t
mj
,t
ni
)
=
q
i=1
B
k
(t
ni
)
w
ni
L(t
mj
,t
ni
)
=
cell n
B
k
(t
)K(t
mj
,t
)dt
(4.3)
for each basis function in the set, and each location t
mj
where L is needed, the near ﬁeld of the
source B
k
obtained via the summation is forced to be the same as the true near ﬁeld of B
k
, obtained
from the original integral operator. If there are q sample points in the quadrature rule, and q basis
functions in the set, this leads to the square system of equations
B
1
(t
n1
)B
1
(t
n2
) ··· B
1
(t
nq
)
B
2
(t
n1
)B
2
(t
n2
)B
2
(t
nq
)
.
.
.
.
.
.
B
q
(t
n1
)B
q
(t
n2
) ··· B
q
(t
nq
)
w
n1
L(t
mj
,t
n1
)
w
n2
L(t
mj
,t
n2
)
.
.
.
w
nq
L(t
mj
,t
nq
)
=
cell n
B
1
(t
)K(t
mj
,t
)dt
cell n
B
2
(t
)K(t
mj
,t
)dt
.
.
.
cell n
B
q
(t
)K(t
mj
,t
)dt
. (4.4)
This system can be solved for the numerical values of L(t
mj
,t
n1
) through L(t
mj
,t
nq
)
1
.A
similar system must be solved for each observer location mj in the near ﬁeld of cell n. By creating L
in this manner, the summation in (4.2) should produce accurate near ﬁelds for any current density
J
z
(t) that can be adequately represented by the basis functions. Since the synthetic kernel L is only
needed at a ﬁnite number of source/observer locations, and the system of (4.4) is only of order q, the
required set of computations is relatively inexpensive. The most expensive part of the calculation is
the accurate evaluation of the integrals appearing on the right-hand side of Equation (4.4) because
of the kernel singularity.
In the electromagnetic equations of interest, the kernel of the integral operator is complex
valued. In that situation, both the real and imaginary parts of the kernel are modiﬁed by the above
procedure.Since it is only necessary to“correct the non-analytic part of the original kernel,computa-
tional efﬁciency suggests that we separate the real and imaginary parts and apply the LCN procedure
only to the non-analytic part. Although that is common practice, in the interest of simplicity, we
present the development in the following sections as though we correct both parts.
1
In practice, we solve for
˜
L
mj ni
= w
ni
L(t
mj
,t
ni
), which are the actual matrix entries required in (4.12).

Get An Introduction to the Locally-Corrected Nystrom Method now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.