Available online at www.sciencedirect.

com

Wave Motion 45 (2008) 309324

www.elsevier.com/locate/wavemoti

On the Schur complement form of the

Dirichlet-to-Neumann operator
L. Knockaert *, D. De Zutter, G. Lippens, H. Rogier
Department of Information Technology, Ghent University, St. Pietersnieuwstraat 41, B-9000 Gent, Belgium
Received 14 December 2006; received in revised form 8 May 2007; accepted 10 July 2007
Available online 28 July 2007

Abstract
The Dirichlet-to-Neumann operator relates the values of the normal derivative of a scalar eld to the values of the eld
itself on the boundary of a simply connected domain. Although it is easy to prove analytically that the Dirichlet-to-Neumann operator is self-adjoint, the discretization of the pertinent Greens integral equation by means of the usual Galerkin
approach generally results in a non-symmetric matrix representation of that operator. In this paper we remedy this by
means of a Hamiltonian Schur complement method and compare it with other symmetrization approaches. It is also
shown that the problem of E-wave scattering by a lossy dielectric cylinder can be nicely simplied by means of the Dirichlet-to-Neumann operator. A numerical Galerkin implementation demonstrates the strength and versatility of the present
approach.
 2007 Elsevier B.V. All rights reserved.
Keywords: Dirichlet-to-Neumann operator; E-wave scattering; Reciprocity; Hamiltonian Schur complement

1. Introduction
The Dirichlet-to-Neumann [1] or PoincareSteklov [2] operator Y relates the values of the normal derivative En of a scalar electric eld E to the values of the eld itself on the boundary c of a simply connected
domain D. In other words we have En YE. Inside the domain D we suppose that the eld E satises the
Helmholtz equation for a homogeneous, non-magnetic material, i.e.,
1
r2 E k 2 E 0
p
with k jxl0 r jx. Hence Y is dependent on the wavenumber k. For any two elds u, v satisfying (1),
we readily nd by Greens theorem that
Z
Z
2
0 vun  uvn ds vYu  uYvds
c

Corresponding author.
E-mail address: luc.knockaert@intec.ugent.be (L. Knockaert).

0165-2125/$ - see front matter  2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.wavemoti.2007.07.004

310

L. Knockaert et al. / Wave Motion 45 (2008) 309324

which implies that the operator Y is self-adjoint. More precisely, we have that Y is a self-adjoint pseudo-differential operator of order 1 [1], with domain the space H1/2(c) and range the dual space H1/2(c) [3,4]. One of
the striking facts concerning the Dirichlet-to-Neumann operator is the observation [5] that the wavenumber in
the domain D may be replaced with the wavenumber k0 of the surrounding region, provided we introduce an
equivalent surface current density |s related to the value of the electric eld E on the boundary c by means of a
dierential surface admittance operator dened by:
def

|s  E

1
Y k  Y k0 E
jxl0

Note that in the sequel we will generally adopt the notation Y Y k and Y 0 Y k0 . It is shown in [5] that the
introduction of this equivalent surface current density greatly simplies skin eect calculations of good (but
not perfect) conductors in the presence of external elds. A possible description of the Dirichlet-to-Neumann
operator is given by the well-known boundary integral equation


Z 
1
o
0
0
0
0
Er
gr  r Er  gr  r En r ds0
4
0
2
on
c
2

where the Greens function is gr  r0 4j H 0 kjr  r0 j. This can be compactly written in an easily understood operator format:
1
E gn0 E  gEn
2

Hence, formally, from Eq. (5) we nd

Y g1 gn0  I=2

where I is the identity operator. It is seen that the self-adjoint character of Y is not apparent in Eq. (6) at rst
glance. Nevertheless, from the single layer potential representation
Z
gr  r0 nr0 ds0
7
Er
c

with normal derivative [6]:


Z 
1
o
0
gr  r nr0 ds0
En r  nr
2
on
c

which can be formally written as E = gn and En = n/2 + gnn, we obtain:

Y gn  I=2g1 Y 0

9
0

thereby proving the self-adjointness of Y. Here we have utilized the fact that gn gn0 and g 0 = g.
However, discretizing and solving the integral equation (4) by means of a Galerkin approach generally
results in a non-symmetric matrix representation of the operator Y, and hence the reciprocity relation (2)
at rst sight does not seem to have a discrete equivalent. This is of course counter-intuitive and un-physical.
In Section 2 we remedy this formally (in an involutive algebra setting1, see [7]) by means of a Hamiltonian
Schur complement approach. In Section 3 we successfully apply this to the problem of E-wave or transverse
magnetic (TM) scattering by a lossy dielectric cylinder: it is shown that the single integral equation treatment
of [8] can be nicely simplied by means of the Dirichlet-to-Neumann operator. In Sections 4 and 5 pertinent
Galerkin schemes are proposed, documented and implemented in order to nd the Schur and other explicitly
symmetric discretized matrix formulations of the Dirichlet-to-Neumann operator. Finally, typical low fre1

An involutive algebra is an operator algebra together with an involution such that (a 0 ) 0 = a, (a + b) 0 = a 0 + b 0 , (ab) 0 = b 0 a 0 . The
operator a 0 is called the adjoint operator of a. When a 0 = a the operator a is called self-adjoint. In a matrix context a 0 can stand for the
transpose aT (real involution) or the Hermitian transpose aH (complex involution).

L. Knockaert et al. / Wave Motion 45 (2008) 309324

311

quency wave propagation calculations are presented in order to demonstrate the versatility of the present
approach.
2. A Hamiltonian Schur complement approach
Following [2,9] we obtain an explicitly self-adjoint expression for Y by adjoining to (4) the derived integral
equation




Z  2
1
o
o
0
0
0
0
En r
gr  r En r ds0
gr  r Er 
10
2
on
onon0
c
which can formally be written as
1
En gnn0 E  gn En
2

11

By construction, it is seen that gnn0 is self-adjoint. Eqs. (5) and (11) together can be written in block format
as
  
 
 
E
E
K G
E def

H
12
En
En
J K 0
En
with G = G 0 = 2g, K 2gn0 , K 0 = 2gn and J J 0 2gnn0 . The operator H thus dened exhibits a Hamiltonian
structure [10], i.e., we have


0
I



I
J
H
0
K

K 0
G



0
I

 0
I
H
0

13

We know that En YE with Y Y 0 . From Eq. (12) it is also clear that Y satises the operator
equations
I K  GY
Y J  K 0Y

14
15

Furthermore, the square of the operator H is

 2

K  GJ KG GK 0
2
H
0
JK  K 0 J K 2  GJ

16

From Eq. (14) we obtain G KG  GYG, and from the adjoint of Eq. (14), i.e., I K 0  YG, we obtain
G GK 0  GYG, implying that KG  GK 0 = 0. Also, pre-multiplying Eq. (14) with Y we obtain
Y YK  YGY, thereby showing that K 0 Y YK. Similarly, from Eq. (15) we obtain YK JK  K 0 YK, and
from the adjoint of Eq. (15), i.e., Y J  YK, we obtain K 0 Y K 0 J  K 0 YK, implying that J K  K 0 J = 0. Hence
 2

K  GJ
0
H2
17
0
K 2  GJ 0
Eqs. (14) and (15) can be written in block format as:
 
 
I
I
H
Y
Y
Hence, applying H twice we obtain
   2
 
I
I
K  GJ

H2
Y
Y
0

0
0
K 2  GJ



18

I
Y


19

312

L. Knockaert et al. / Wave Motion 45 (2008) 309324

implying that K2  GJ = I. In other words H2 I, i.e., H is a Hamiltonian square root of the identity, implying that the so-called Calderon operators P  I  H=2 are oblique projectors in general2. The explicitly
self-adjoint form of Y can be found as the Schur complement [11,12]:

1
Y J  I  K 0 G1 I  K
20
2
For Z Y 1 we have a similar self-adjoint Schur complement:

1
Z G I KJ 1 I K 0
21
2
Note that Y is a self-adjoint solution of the algebraic operator Riccati equation
2Y YGY  J 0

22

Remark 1. Since K2  GJ = I and KG = GK 0 , implying G1K = K 0 G1, we easily obtain (see also a similar
argument in [13]):
J K 0 G1 K  G1
Inserting this in equation (20) we nd:

1
Y K 0 G1 K  G1  I  K 0 G1 I  K K 0  IG1 G1 K  I
2
2

23

24

This is exactly the symmetrized form of the solution of equation (14), which is Y G1 K  I.
~ obtained by replacing K with K, i.e.,
Remark 2. Consider the associated Hamiltonian H,


K G
~ def
H

J
K0

25

~ 2 I. The associated Hamiltonian H

~ therefore also admits an associated DirichIt is clear that we also have H
~ dened by the equations:
let-to-Neumann operator Y
~
I K  GY
~ J K 0Y
~
Y
yielding the self-adjoint Schur complement representations:


~ 1 J  I K 0 G1 I K
Y
2


~ Y
~ 1 1 G I  KJ 1 I  K 0
Z
2
It is a simple matter to prove the following important identities:
~
~
2J 1 Z Z
 2G1 Y Y;
~ Z  ZJ
~
2K GY  Y

26
27

28
29

30
31

~ with respective inverses Z and Z,

~ the identities
Note that, given two self-adjoint invertible operators Y and Y,
(30) imply (31) and are sucient to dene a Hamiltonian H such that H2 I.
Remark 3. If we have the Dirichlet eigenvalues km and eigenfunctions nm(r) for the region D at our disposal,
we can also write down the integral kernel of Y [5] as:
Yr; r0

1
X
nmn rnmn r0
k 2  km
m1

Self-adjoint idempotents are called orthogonal projectors, while non-self-adjoint idempotents are called oblique projectors.

32

L. Knockaert et al. / Wave Motion 45 (2008) 309324

313

Similarly, with the Neumann eigenvalues ~

km and eigenfunctions ~nm r for the region D, we can write down the
integral kernel of Z:
Zr; r0 

1 ~m
X
n r~
nm r0
2
~m
k k

33

m1

Interestingly enough, the representations (32) and (33) are the respective solutions to the algebraic operator
Riccati Eq. (22) in the degenerate cases G = 0 (Dirichlet Greens function) and J = 0 (Neumann Greens
function).
3. TM scattering by a dielectric cylinder
As an application, we consider the TM scattering problem of a lossy dielectric cylinder with wavenumber k
and cross-section D, in an incoming eld Ei(r). The pertinent integral equations are [8]:


Z 
1
o
0
0
0
0
Er
gr

r

Er


gr

r
E
r

ds0
34
n
2
on0
c




Z
1
o
g r  r0 Er0  g0 r  r0 En r0 ds0
35
Ei r  Er
2
on0 0
c
2

where g0 r  r0 4j H 0 k 0 jr  r0 j with k0 the wavenumber outside the cylinder. The integral equations (34)
and (35) can be written in an easily understood shorthand notation:
1
E gn0 E  gEn
2
1
Ei  E g0;n0 E  g0 En
2
1 sc
 E g0;n0 Esc  g0 Esc
n
2

36
37
38

where the scattered eld Esc = E  Ei. Note that we have used the fact:
1 i
E g0;n0 Ei  g0 Ein
2

39

which is equivalent with Ein Y 0 Ei . The scattered eld Esc can be written as a single layer potential Esc = g0 f,
yielding the single integral equation for f:


1 i
1
1
i
i
E  gn0 E gEn  g0  g gn0 g0  gg0;n f
40
2
2
2
Inside the scatterer we can write the single layer potential E = gn, yielding the single integral equation for n:


1
1
i
41
E g0 g g0;n0 g  g0 gn n
2
2
The integral equations (40) and (41), which were rst introduced in [8] and later also documented in [14], can
be further reduced to a simpler form. With the notations of Section 2, the integral equations (36) and (37) can
be written as
42

E KE  GEn
i

2E  E K 0 E  G0 En

43

The integral equation (42) can be replaced with En YE, which inserted in (43) yields

~0 Y E
Ei g0 Y

44
i

The simple integral equation (44) allows us to nd E on c directly. The relationships E = E + g0f and E = g n
(7) then yield f and n and therefore the complete solutions inside and outside the scatterer. The integral equa-

314

L. Knockaert et al. / Wave Motion 45 (2008) 309324

~ 0 g1 , Eq. (44) can

tion (44) presents an interesting twist: since the rst identity of (30) tells us that Y 0 Y
0
be written as
Ei E g0 Y 0  Y E

45

which is a Fredholm equation of the second kind. Eq. (45), which exhibits the dierential Dirichlet-to-Neumann operator [5] Y 0  Y, can also be found by another means, for instance by introducing the additional
phantom integral equation:
1
~n
E g0;n0 E  g0 E
2
~ n Y 0 E. Subtracting (46) from (37) we obtain
We can always do so provided E

~ n  En
Ei  E g0 E

46

47

which is precisely integral equation (45). Note that Eq. (45) implies that
Esc g0 Y  Y 0 E

48

resulting in f Y  Y 0 E. Finally, let us eectively prove that the formidable integral equations (40) and (41)
can be reduced to the very simple integral equation (44). Regarding integral equation (41), it is relatively easy
to show that

1
1
~0 Y g
g0 g g0;n0 g  g0 gn g0 Y
49
2
2
which is consistent with integral equation (44). We can also nd an explicit expression for n, i.e.,

~ Y 0 Ei  Y
~ Y
~ 0 Esc
n Y

50

The case of integral equation (40) is more dicult; however, it is not too hard to prove that
1 i
E  gn0 Ei gEin gY 0  Y Ei
2

51

1
1
~ 0 Y g0
 g0  g gn0 g0  gg0;n g Y
2
2

52

and

With the proviso that nullspace(g) = {0}, integral equation (40) is equivalent with

~ 0 Y Esc
Y 0  Y Ei Y

53

which is easily transformed into Eq. (48). The above derivations also show the conceptual superiority of the
Dirichlet-to-Neumann approach as compared to the usual integral equation techniques.
4. The projected Hamiltonian matrix
As was said in Section 1, Y is a self-adjoint pseudo-dierential operator of order 1 mapping H1/2(c)
to H1/2(c), where c is the boundary contour of the relevant region D. It should be noted that H1/2(c) is a
SobolevHilbert space with norm [15] p. 208:
!12
Z
Z Z
jEs  Es0 j2
2
0
kEk1=2
jEsj ds
ds ds
54
R2
c
c
c
In Eq. (54) R stands for the Euclidian distance between the two points parametrized by s and s 0 on the contour
c, and is a positive parameter proportional to the length of the contour c in order to make formula (54)
dimensionally correct. The dual norm kEk1/2 is dened via the usual procedure (see [4] p. 320), but is virtually
impossible to compute in any practical situation. Fortunately, we know [3] that

L. Knockaert et al. / Wave Motion 45 (2008) 309324

H 1=2 c  L2 c  H 1=2 c

315

55

with H1/2(c) dense in L2(c). For instance, as a consequence of the proper subset equation (55), there are functions in L2(c) that are not in H1/2(c), and functions in H1/2(c) that are not in L2(c).
In the context of a Galerkin procedure for obtaining an approximation to Y, this is important, since we
have to develop E into basis functions in H1/2(c) and En into basis functions in H1/2(c). Note that the second
integral in formula (54) is a so-called hypersingular integral, which diverges for piecewise constant functions E
but converges for piecewise linear functions E [16]. This having been specied, recall that our pertinent equations are
E KE  GEn

56

En JE  K En
En YE

57
58

The elds E and En are expanded as

E

1
X

En

ak bk s

k1
1
X

59

bk tk s

60

k1

where the set of real basis functions {bk(s)} and {tk(s)} are complete, respectively, in H1/2(c) and H1/2(c).
Inserting this in Eqs. (56) and (58) we obtain
1
X
k1
1
X
k1
1
X
k1

ak bk
bk tk
bk tk

1
X
k1
1
X
k1
1
X

ak Kbk 
ak Jbk 

1
X

bk Gtk

61

bk K 0 tk

62

k1
1
X
k1

ak Ybk

63

k1

A self-consistent Galerkin procedure is obtained by testing Eq. (61) with the set {tk}, while testing Eqs. (62)
and (63) by means of the set {bk}, resulting in
1
X
k1
1
X
k1
1
X

ak htl ; bk i
bk hbl ; tk i
bk hbl ; tk i

k1

1
X
k1
1
X
k1
1
X

ak htl ; Kbk i 
ak hbl ; Jbk i 

1
X
k1
1
X

bk htl ; Gtk i

64

bk hbl ; K 0 tk i

65

k1

ak hbl ; Ybk i

66

k1

Note that, the operator J being a hypersingular operator [16], the self-patch elements hbk, Jbki are nite if and
only if the basis functions bk belong to H1/2(c), since the hypersingular integral present in the norm (54) exhibits the same singularity as J. In the same context, it is known that operator J can alternatively be written as an
integro-dierential operator [9], i.e.,
J k 2 nr  nr0 gr  r0

ogr  r0 o
os0
os

67

resulting in the Galerkin elements:

Z Z
Z Z
hbk ; Jbl i k 2
bk sbl s0 nr  nr0 gr  r0 ds ds0 
b0k sb0l s0 gr  r0 ds ds0
c

68

316

L. Knockaert et al. / Wave Motion 45 (2008) 309324

Of course, the derivation of formula (68) requires the slightly more stringent assumption that the basis functions {bk} belong to H1(c), which is, e.g., the case when the {bk} are piecewise boundedly dierentiable, say in
practice when one uses the well-known linear rooftop functions [16].
Eqs. (64)(66) can be written in the easily understood innite matrix format
Ta Ka  Gb
T

69

T b Ja  K b

70

T b Ya

71

where
Tkl htk ; bl i

Kkl htk ; Kbl i Gkl htk ; Gtl i Jkl hbk ; Jbl i

Ykl hbk ; Ybl i

Eliminating the unknowns a and b, we obtain the projected innite block matrix equations
 
 
I
I
H
Y
Y
with projected Hamiltonian matrix
 1
 
b
T K T1 GTT def K
H
b
T T
J
K T
J

b
G
b
KT

72

73


74

The explicitly symmetric form of Y can be found as the Schur complement:

i
h
i
1h
b 1 I  K
b 1 J  T  KT G1 T  K
b TG
75
J  I  K
Y b
2
2
It is seen that, exactly in parallel with the operator formalism of Section 2, the requirement H2 = I is needed,
bb
bG
bK
b K
b Tb
b2  G
bG
b T, b
JK
J I. Also, following Remark 1 of Section 2, it follows that
J and K
implying that K
Y can similarly be written in the symmetrized form
i
h
i
1 h bT
b 1 G
b 1 K
b  I 1 K  TT G1 T TT G1 K  T
Y K
 I G
76
2
2
and hence the hypersingular Galerkin matrix J b
J can be written as
b 1 K
b 1 KT G1 K  TT G1 T
bTG
bG
JK

77

A similar (although in a dierent context) expression for J was found in [13]. This means that we only have to
calculate the entries of the matrices G, K and T. A still cheaper approach can be found in the Theorems A and
b and the
B of the Appendix, where it is shown that one only needs the eigen-decomposition of the matrix K
b
b
diagonal entries of one of the matrices G or J. Of course, it should be noted that all this will only be approximately valid when the bases {bk} and {tk} are truncated, i.e., when we have 1 6 k 6 M for some suciently
large, but nite M.
5. Verication and numerical simulations
The dierent expressions for the discretized Dirichlet-to-Neumann operator Y will be constructed for a
number of two-dimensional examples, where the cross-section D of the cylinder is a rectangle. The parameters
of the lossy cylinder are , l0, r and the working frequency is f = x/2p, resulting in the wavenumbers
p
k0 = x/c = 2p/k and k jxl0 r jx: From the derivation of formula (68) we need {bk}  H1(c),
and for all practical purposes this means that one may use the well-known linear rooftop functions [16]. In
order to avoid unnecessary complications we also consider the full Galerkin case {tk} = {bk}. The linear
rooftop basis {bk} is shown in Fig. 1. It is seen that the boundary of the cross-section of the cylinder is divided
into 32 overlapping segments of width 2d, where d = dx = dy is the discretization step. The number of discretizations in the x- and y-directions are nx = 12 and ny = 4 for the conguration shown in Fig. 1.

L. Knockaert et al. / Wave Motion 45 (2008) 309324

317

In this section, we shall rst calculate a number of projected Dirichlet-to-Neumann operators and verify the
symmetry and convergence properties. The entries of these Dirichlet-to-Neumann operators are subsequently
compared with the results obtained from the Dirichlet eigenfunction expansion method [5]. Finally, we will
discuss the performance of the dierent methods with respect to the modelling of low frequency waves.
The methods presented in the previous sections, that will be applied below, will be referred to as the Schur
complement method (see Eq. (75)), the symmetrization approach (see Eq. (76)), the asymmetric description (utilizing only the two Eqs. (69) and (71))) and the eigen-decomposition method (see Theorems A and B of the
Appendix).
5.1. Comparison of the projected operators with results from other methods
In this subsection, we compare the dierent Y matrices, obtained by applying a Galerkin projection in
the case of a rectangular cylinder with dimensions a = 6 mm and b = 3 mm. The lossless (r = 0) cylinder
has a permittivity of r = 6. The results obtained with the Schur complement approach (75) are rst
compared with those resulting from the Dirichlet method [5] and with those obtained from the asymmetric approach (76). The interactions with a rooftop in the upper right corner (row 19 of the admittance
matrix) of the cylinder are plotted as a function of frequency in Fig. 2 (imaginary part) and in Fig. 3
(real part). Figs. 2 and 3 indicate that the Schur complement method matrix elements coincide with the
matrix entries obtained with the Dirichlet method [5]. The asymmetric approach yields mostly correct
results, although the self-interaction (located at the 19th coordinate) is not truly accurate. When observing the real part (Fig. 3) of the three methods under scrutiny, which should be zero (r = 0), the result
obtained with the asymmetric approach is largely unsatisfactory. The real part is small, but non-negligible. The overall impact could be unacceptable, should an admittance model be used in a global
simulation.
The interactions of a rooftop in the middle of the upper edge of the cylinder (10th row of the Y matrix) are
shown in Fig. 4 (imaginary part) and in Fig. 5 (real part). All methods now yield correct results with respect to
the imaginary part, but the discrepancies in the real part are still important. The other non-Schur methods
result in comparable discrepancies of the real part. In Fig. 6, the real parts of the corner interactions (19th
row of the Y matrix) are presented, for three non-Schur approaches: the asymmetric method, the symmetrization approach and the eigen-decomposition algorithm. For this particular example, the symmetrization
approach performs better than the other two methods. The residual dissipation is more or less equal for
the three examples.
In Fig. 7, the real parts of the interactions with a rooftop in the middle of the upper edge (row 19) are
shown, for the three non-Schur approaches discussed above. Comparing with the corner case, we can conclude
that the behavior is similar, whereas the residual dissipation is somewhat smaller on the edges than on the
corners. In the next subsection, it will be shown that the residual dissipation resulting from the non-Schur
approaches signicantly alters the results when a conductive material is used.

y
7
1
32

13

2d

x
29

17

Fig. 1. Galerkin discretization of a rectangular cylinder. Two rooftop basis functions are shown. One of them is positioned on an edge
(number 7) and the other one is located on a corner (number 13).

318

L. Knockaert et al. / Wave Motion 45 (2008) 309324

8
x 10
7

Schur complement
Dirichlet
Asymmetric
6

0
0

10

15

20

25

30

35

40

45

50

Fig. 2. Imaginary part of the 19th row (corner interaction) of the Schur complement, Dirichlet-based and symmetrization Y matrices.

10
x 10

Schur complement
Dirichlet
Asymmetric

2
4
6
8
10

10

15

20

25

30

35

40

45

50

Fig. 3. Real part of the 19th row (corner interactions) of the Schur complement, eigenfunction-based and symmetrization Y matrices.

5.2. Verication of the DC resistance

When a low frequency transverse magnetically (TM) polarized wave impinges on a homogeneous conducting cylinder or when a low frequency signal propagates through the cylinder, a uniform current distribution
pattern develops inside the cylinder. The associated DC resistance per unit of length (p.u.l.) is given by Pouillets law:

L. Knockaert et al. / Wave Motion 45 (2008) 309324

319

7
x 10

Schur complement
Dirichlet
Asymmetric
2.5

1.5

0.5

0
0

10

15

20

25

30

35

40

45

50

Fig. 4. Imaginary part of the 10th row (middle of upper edge) of the Schur complement, Dirichlet-based and symmetrization Y matrices.

10
x 10

Schur complement
Dirichlet
Asymmetric

1.5

0.5

0.5
1
1.5
2
2.5

10

15

20

25

30

35

40

45

50

Fig. 5. Real part of the 10th row (middle of upper edge) of the Schur complement, Dirichlet-based and symmetrization Y matrices.

RDC

1 1
r ab

78

where a and b are the dimensions of the rectangular cylinder. The calculation of the DC p.u.l. resistance from
the admittance matrix is straightforward:

320

L. Knockaert et al. / Wave Motion 45 (2008) 309324

10
x 10

Asymmetric
Symmetrization
Eigendecomposition based

2
4
6
8
10

10

15

20

25

30

35

40

45

50

Fig. 6. Real part of the 19th row (corner interactions) of three non-Schur-based Y matrices.

10

x 10

Asymmetric
Symmetrization
Eigendecomposition based

1.5

0.5

0.5
1
1.5
2
2.5

10

15

20

25

30

35

40

45

50

Fig. 7. Real part of the 10th row (middle of upper edge) of three non-Schur-based Y matrices.

(
RDC R

XX
i

)1
Yij

79

It is of course very important that RDC is accurately predicted by any numerical method. We will now take a
look at the performance of the various methods discussed above with regard to this DC resistance. We con-

L. Knockaert et al. / Wave Motion 45 (2008) 309324

321

sider a copper cylinder (r = 5.72 107 S/m) with aspect ratio r = a/b = 3/2. The number of discretizations in the
x, resp. y-direction, is nx, resp. ny.
In Table 1, the theoretical DC resistance (in X/m) is shown for ve simulations with subsequent mesh
renements, together with the results obtained with the Schur complement method and with those obtained
with three non-Schur methods. These are the asymmetric approach, the symmetrized approach and the
eigen-decomposition method, which turn out to be equivalent when calculating DC resistances. This explains
why only a single value is given in the last column of Table 1. It should be noted that the DC resistances of
Table 1 are remarkably close to the theoretical values given by Eq. (78). The non-Schur results deviate only
slightly for some simulations. When simulating semiconductor type cylinders, the conclusions are very
dierent.
In Table 2, the admittances YDC = 1/RDC of a 14.4 lm (nx = 12) by 9.6 lm (ny = 8) cylinder are shown for
a total of 20 dierent r values. YDC is in units of 1012 Sm. Only the Schur results are displayed; the admittances obtained from the non-Schur methods were several orders of magnitude too large and negative, i.e.,
YDC (non-Schur) in the range [0.3283 109, 0.1442 109] Sm for all non-Schur simulations. The results
obtained with the Schur complement method are correct when a sucient number of basis functions is chosen.
The results for the lower conductivity values are slightly less accurate, probably due to the residual dissipation
which starts to play a role at these low conductivity values.
5.3. Numerical verication of symmetry
In order to verify the accuracy of the expressions (75) and (76) in a realistic context, we consider how the Y
matrices of a rectangular cylinder evolve as a function of frequency, the number of basis functions and the
material parameters. We use the same notation as in the previous example for the dimensions and discretizations. Here we consider good conductors, semiconductors and dielectric materials: copper [Cu]
(r = 5.72 107 S/m, r = 1.0), silicon [Si] (r = 1.6 103 S/m, r = 11.0) and PECVD Silicon Nitride [SiN]
(r = 1010 S/m, r = 7.0). The material name abbreviations between square brackets are used in Table 3,
where the relevant parameters for the subsequent simulations are shown.
For the simulations with Si and SiN, a frequency sweep between fl = 1 MHz and fh = 1 THz is performed.
For the copper example, the simulation bandwidth is chosen such that q = 0.5(a + b)/d 2 [102, 104], where
the skin depth d is dened as:
s
2
d
80
2pf rl0
Because of this particular choice of a set of frequency points, the transition region between the low frequency
situation (the current ows uniformly in the longitudinal direction) and the high frequency case (the current is
forced to the border of the cylinder), is completely covered. The numerical symmetry of the Schur complement
formulation is compared with the results of the asymmetric and eigen-decomposition formulations. In order to
measure the departure from symmetry, we take the relative 2-norm of the asymmetric part: e(f) = kY  YTk2/
kYk2 as a function of frequency. The maximum and minimum values over the chosen frequency range for the
Schur approach are listed in Table 3 as emax and emin for a total of twelve simulations. For the asymmetric and
eigen-decomposition formulations, e(f) as a function of frequency remains acceptable at the higher part of the
spectrum, but very rapidly increases for the lower frequencies to reach values of the order of 0.5. The observed
Table 1
DC p.u.l. resistance RDC (in X/m) of a copper cylinder (r = 5.72 107S/m) (a and b are in lm)
a

nx

ny

RDC (Pouillet)

RDC (Schur)

RDC (non-Schur)

15.0
14.4
15.0
14.7
16.2

10.0
9.6
10.0
9.8
10.8

6
12
15
21
27

4
8
10
14
18

116.6
126.5
116.6
121.4
99.9

116.6
126.5
116.6
121.4
99.9

116.7
126.6
116.7
121.4
100.0

322

L. Knockaert et al. / Wave Motion 45 (2008) 309324

Table 2
DC admittance YDC = 1/RDC of a semiconducting cylinder (in units of 1012 Sm)
r (S/m)

YDC (Pouillet)

YDC (12 8)

YDC (24 16)

YDC (36 24)

0.0008
0.0016
0.0024
0.0032
0.0040
0.0048
0.0056
0.0064
0.0072
0.0080
0.0088
0.0096
0.0104
0.0112
0.0120
0.0128
0.0136
0.0144
0.0152

0.11
0.22
0.33
0.44
0.55
0.66
0.77
0.88
0.99
1.11
1.21
1.33
1.44
1.55
1.66
1.77
1.88
1.99
2.10

0.14
0.25
0.36
0.47
0.58
0.69
0.80
0.91
1.02
1.13
1.24
1.36
1.47
1.57
1.69
1.79
1.91
2.02
2.13

0.12
0.23
0.34
0.45
0.56
0.67
0.79
0.89
1.01
1.12
1.23
1.34
1.45
1.56
1.67
1.78
1.89
2.00
2.11

0.12
0.23
0.34
0.45
0.56
0.67
0.78
0.89
1.00
1.11
1.22
1.33
1.44
1.56
1.67
1.78
1.89
2.00
2.11

Theory compared with Schur complement method (nx ny between parentheses).

Table 3
Symmetry assessment of Dirichlet-to-Neumann operators (mat.: material, a and b are in lm)
mat.

nx

ny

emin (Schur)

emax (Schur)

emin (non-Schur)

Cu
Cu
Cu
Cu
Cu

15.0
14.4
15.0
14.7
14.4

10.0
9.6
10.0
9.8
9.6

6
12
15
21
27

4
8
10
14
18

1.1 1016
4.4 1016
7.1 1016
9.0 1016
1.2 1015

8.8 108
5.3 107
9.1 107
1.6 106
9.1 107

3.8 104
7.8 104
1.0 103
9.9 104
9.5 104

Si
Si
Si
Si
Si

15.0
14.4
15.0
14.7
14.4

10.0
9.6
10.0
9.8
9.6

6
12
15
21
27

4
8
10
14
18

4.0 1014
1.8 1013
2.9 1013
6.1 1013
1.2 1012

5.8 107
3.0 106
5.9 106
1.5 105
1.9 105

5.5 103
4.0 103
3.7 103
3.6 103
3.0 103

SiN
SiN
SiN
SiN
SiN

15.0
14.4
15.0
14.7
14.4

10.0
9.6
10.0
9.8
9.6

6
12
15
21
27

4
8
10
14
18

7.4 1014
4.9 1013
5.6 1013
1.5 1012
1.5 1012

7.1 107
1.3 105
7.8 106
1.7 105
2.8 105

6.6 103
3.9 103
4.5 103
5.2 103
3.7 103

minimum values emin are also given in Table 3. These values clearly show that the Schur approach is the only
one assuring that the physically required symmetry property of the Dirichlet-to-Neumann operator is wellpreserved in its discretized version.

Acknowledgements
H. Rogier is a Postdoctoral Researcher of the FWO-V. This work was partly supported by a grant of the
Institute for the Promotion of Innovation by Science and Technology in Flanders for a joint project with Agilent Technologies.

L. Knockaert et al. / Wave Motion 45 (2008) 309324

323

Appendix A
Theorem A. Let the complex matrices G = GT, J = JT and K be such that
(1)K is diagonalizable with all its eigenvalues different.3
(2)KG = GKT, JK = KTJ and K2  GJ = I.
Then K, G and J can be written as
K P KK P 1

G P KG P T

J P T KJ P 1

A:1

where KK, KG and KJ are diagonal matrices satisfying K2K  KG KJ I.

Proof. K = PKKP1 is simply the eigen-decomposition of K where supposedly kK = diag{kk} with all kk
different. Since P is non-singular, the symmetric matrix G can be written uniquely as G = P WPT, where W
symmetric is given by W = P1GPT. Since KG = PKKWPT has to be symmetric, it is necessary that KKW is
symmetric. Hence we need
ki  kj W ij 0

8 i; j

A:2

implying that W must be a diagonal matrix. Regarding J, the same reasoning applied to the matrix V = PTJP
results in V being diagonal, and the proof is complete. h
Note that if we put KG = diag{lk}, the matrix G can be written as
X
Gij
P ik P jk lk

A:3

implying that the diagonal entries Gii are sucient to obtain the values lk by solving the linear equations
X
Gii
P 2ik lk
A:4
k

Obtaining KJ = diag{sk}, is then simply a matter of substitution (if none of the lk = 0) in the equation
k2k  lk sk 1

A:5
T

Similarly, putting Q = P
X
J ii
Q2ik sk

, the values sk can be obtained by solving the linear equations

A:6

and the lk follow from Eq. (A.5). The matrix Y = G1(K  I) is also symmetric, and can be written as
1

T
T
Y QK1
G KK  I Q QKJ KK I Q

A:7

In the case where the eigenvalues of K are algebraically degenerate, the following weaker theorem can be used.
Theorem B. Let the complex matrices G = GT, J = JT and K be such that
(1) K is diagonalizable.
(2) KG = GKT, JK = KTJ and K2  GJ = I.
Then the matrix Y = G1(K  I) is symmetric and can be constructed from the matrices K and HG alone, or,
equivalently, the matrices K and HJ alone, where
K P KK P 1
3

G P HG P T

J P T HJ P 1

No similar or other restrictions are necessary regarding the eigenvectors.

A:8

324

L. Knockaert et al. / Wave Motion 45 (2008) 309324

Proof. K = PKKP1 is again the eigen-decomposition of K, but here we do not require that all eigenvalues are
different. G can be written as G = PHGPT, where the symmetric matrix HG = P1GPT. Similarly, for J, we
obtain the symmetric matrix HJ = PTJP. Substituting this in K2  GJ = I, we obtain K2K  HG HJ I. Putting
Q = PT, the matrix Y = G1(K  I) can be written as
1

T
T
Y QH1
G KK  I Q QHJ KK I Q

A:9

From the premises KG = GK and JK = K J it is clear that the matrices KKHG and HJKK are symmetric and
hence also the matrix
1

H1
G KK KK HG KK KK
implying that Y is symmetric.

A:10
h

References
[1] L. Friedlander, Remarks on Dirichlet and Neumann eigenvalues, Am. J. Math. 117 (1995) 257262.
[2] W. Hackbusch, B.N. Khoromskij, R. Kriemann, Direct Schur complement method by domain decomposition based on H-matrix
approximation, Comput. Vis. Sci. 8 (2005) 179188.
[3] V. Isakov, Inverse problems for partial dierential equationsApplied Mathematical Sciences, vol. 127, Springer, New York, 2006.
[4] G.C. Hsiao, R.E. Kleinman, Mathematical foundations for error estimation in numerical solutions of integral equations in
electromagnetics, IEEE Trans. Antennas Propagat. 45 (1986) 316328.
[5] D. De Zutter, L. Knockaert, Skin eect modeling based on a dierential surface admittance operator, IEEE Trans. Microw. Theory
Technol. 53 (2005) 25262538.
[6] J. Van Bladel, Electromagnetic Fields, McGraw-Hill, New York, 1964 (revised printing, Hemisphere, Washington DC, (1985)).
[7] R. Dautray, J.-L. Lions, Mathematical analysis and numerical methods for science and technologySpectral Theory and Applications,
vol. 3, Springer, Berlin, 1990.
[8] L.F. Knockaert, D. De Zutter, Integral equation for the elds inside a dielectric cylinder immersed in an incident E-wave, IEEE
Trans. Antennas Propagat. 34 (1986) 10651067.
[9] S.P. Shipman, S. Venakides, Resonance and bound states in photonic crystal slabs, SIAM J. Appl. Math. 64 (2003) 322342.
[10] W.W. Lin, C.S. Wang, On computing stable Lagrangian subspaces of Hamiltonian matrices and symplectic pencils, SIAM J. Matrix
Anal. Appl. 18 (1997) 590614.
[11] G. Corach, A. Maestripieri, D. Stojano, Generalized Schur complements and oblique projections, Lin. Alg. Appl. 341 (2002) 259
272.
[12] H. Bart, I. Gohberg, M.A. Kaashoek, A.C.M. Ran, Schur complements and state space realizations, Lin. Alg. Appl. 399 (2005) 203
224.
[13] G.Y. Yu, Relationship among coecient matrices in symmetric Galerkin boundary element method for two-dimensional scalar
problems, J. Appl. Mech. 70 (2003) 479486.
[14] R.E. Kleinman, P.A. Martin, On single integral equations for the transmission problem of acoustics, SIAM J. Appl. Math. 48 (1988)
307325.
[15] M. Renardy, R.C. Rogers, An Introduction to Partial Dierential Equations, second ed., Springer, New York, 2004.
[16] L.J. Gray, Evaluation of singular and hypersingular Galerkin boundary integrals: direct limits and symbolic computation, in: V.
Sladek, J. Sladek (Eds.), Singular Integrals in the Boundary Element Method, Advances in Boundary Elements, Computational
Mechanics Publishers, Southampton, UK, 1998, pp. 3384.