Electromagnetic Scattering From An Anisotropic Impedance Half-Plane at Oblique Incidence: The Exact Solution

ELECTROMAGNETIC SCATTERING FROM AN
ANISOTROPIC IMPEDANCE HALF-PLANE AT

OBLIQUE INCIDENCE: THE EXACT SOLUTION
by Y. A. ANTIPOV
(Department of Mathematics, Louisiana State University

Baton Rouge LA 70803, USA)
and V. V. SILVESTROV
(Department of Mathematics, Chuvash State University

Cheboksary 428015, Russia)
[Received 9 March 2005. Revise 6 December 2005. Accepted 21 December 2005]
Summary
Scattering of a plane electromagnetic wave from an anisotropic impedance half-plane at skew
incidence is considered. The two matrix surface impedances involved are assumed to be com-
plex and different. The problem is solved in closed form. The boundary-value problem reduces
to a system of two rst-order difference equations with periodic coefcients subject to a sym-
metry condition. The main idea of the method developed is to convert the system of difference
equations into a scalar RiemannHilbert problem on a nite contour of a hyperelliptic surface
of genus 3. A constructive procedure for its solution and the solution of the associated Jacobi
inversion problem is proposed and described in detail. Numerical results for the edge diffraction
coefcients are reported.
1. Introduction
During the last 50 years signicant progress has been made in the mathematical theory of diffraction
which studies the inuence of material properties on edge diffraction phenomena. The achievements
have been made in large part due to the use of the Leontovich impedance boundary conditions in
modeling and the WienerHopfJones (1, 2) and Maliuzhinets (3) methods for the solution. The
former method works successfully for half-planes and right-angled wedges. Wedges of arbitrary
angles are normally treated by the Maliuzhinets technique. Recently, Daniele (4) showed that the
WienerHopfJones method is applicable for scattering problems for wedges as well. In fact, for
certain canonical problems of acoustic and electromagnetic diffraction (5 to 8), both techniques
are equivalent to the solution of a scalar RiemannHilbert problem on a Riemann surface (in some
particular cases on a sphere, that is, on a plane).
The scalar problem of electromagnetic diffraction of a plane wave from a semi-innite impedance
plane at oblique incidence was solved by Senior (9) by the WienerHopf approach. By using the
Sommerfeld integral, Maliuzhinets (3) found a closed-form solution to the problem of acoustic
diffraction by an impedance wedge with two different impedance parameters. The solution was
Corresponding author antipov@math.lsu.edu
sil@chuvsu.ru
Q. Jl Mech. Appl. Math, Vol. 59. No. 2 c The author 2006. Published by Oxford University Press;
all rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org
Advance Access publication 3 February 2006. doi:10.1093/qjmam/hbj004
212 Y. A. ANTIPOV AND V. V. SILVESTROV
derived in terms of special functions later called the Maliuzhinets functions. The homogeneous
isotropic two-face impedance half-plane problem of electromagnetic diffraction in the oblique inci-
dence case was analysed by L uneburg and Serbest (10) by using the range restrictions concept.
For the more complicated anisotropic case, two different matrix surface impedances
Z
= Z
0
_
0
1
0
_
(1.1)
are involved. Here the superscripts identify the upper and the lower faces of the plane,
1
and
2
are dimensionless parameters, Z
0
=

0
,c
0
is the intrinsic impedance, and
0
and c
0
are the
permeability and permittivity of free space. On the screen faces, the impedance boundary conditions
are E = Z
H, where E = (E
, E
z
)
and H = (H
, H
z
)
are the total electric and magnetic

elds. In the normal incidence case, the governing system of WienerHopf and rst-order difference
equations is uncoupled. The solution is obtained by quadratures.
When the incident wave is not orthogonal to the edge of the structure the equations are in general
coupled. For special cases of skew incidence the equations reduce to scalar Maliuzhinets equations.
Examples include those considered by Bernard (11) and Lyalinov and Zhu (12). For the case when
the matrices Z
+
and Z
are different, Senior (13) used the WienerHopf and Maliuzhinets ideas to
reduce the problem to functional equations. The former method gives rise to a vector WienerHopf
problem for four pairs of unknown functions. Hurd and L uneburg (14) found a closed-form solution
in the case Z
+
= Z
when the problem is reducible to the problem of factorization of a 22 matrix.

The matrix was factorized by employing the Daniele (15) method. In the more general case, when
Z
+
= Z
, the Daniele method leads to a system of highly nonlinear equations. These equations
have to be solved in order to eliminate an essential singularity of the WienerHopf factors. Apart
from the elliptic case (the number of nonlinear equations equals one) (15, 14) so far there is no
constructive (exact or approximate) technique for its solution.
Senior and Topsakal (16) employed the SommerfeldMaliuzhinets formulation and reduced the
problem to a second-order difference equation and then converted it into an integral equation. An
approximate solution was given for the case Z
+
= Z
. A method of the Riemann bilinear re-

lations for abelian integrals for a partial solution of a second-order difference was proposed by
Legault and Senior (17). The principal complexity of their method is to eliminate the polar and
cyclic periods of the solution in order to make it single-valued. The procedure requires solving a
system of nonlinear equations with respect to the unknown singularities of the abelian integrals of
the third kind. Its constructive solution is found only for those cases which are equivalent to the
elliptic case.
The main motivation behind the present work was to solve in closed form the problem on
electromagnetic scattering of a plane wave from an anisotropic impedance half-plane at oblique in-
cidence. The entries of the impedance matrices Z
+
and Z
could be complex and different. Antipov

and Silvestrov (6, 7) proposed a unied approach for a general class of problems in electromagnetic
scattering. It generalizes the WienerHopf and Maliuzhinets technique for systems of WienerHopf
and Maliuzhinets equations. The proposed method rephrases the systems of governing equations
as a scalar RiemannHilbert problem on a Riemann surface. The WienerHopf factors expressed
through the solution of the boundary-value problem on a Riemann surface are single-valued. The
essential singularity of the solution at innity is removed by xing certain free parameters which
solve the associated Jacobi inversion problem. Its solution always exists and can be constructed in
terms of the zeros of the Riemann -function. By this method, a problem on E-polarization of a
right-angled magnetically conductive wedge was solved by quadratures (8). It was pointed out (6)
THE EXACT SOLUTION FOR ELECTROMAGNETIC SCATTERING 213
that the problem of electromagnetic scattering at skew incidence from an anisotropic half-plane is
solvable by quadratures.
The main aims of this work are (i) to construct the actual solution to the problem, (ii) to develop
an efcient procedure for the associated Jacobi inversion problem, and (iii) to derive formulae for
electric and magnetic elds and to compute the diffraction coefcients.
In section 2, the governing equations are converted into two vector rst-order difference equations
(2.18) subject to a symmetry condition (2.19) and additional conditions (2.5) and (2.17). Section 3
derives a vector RiemannHilbert problem (3.14) on a system of segments. In section 4, this prob-
lem reduces to a scalar RiemannHilbert problem (4.3) on two segments of a hyperelliptic surface
of genus 3 and to a Jacobi inversion problem (4.23). The solution of the inversion problem requires
calculating the Riemann constants and nding the zeros of the associated Riemann -function of
genus 3. Formulae (5.19) and (5.33) for the Riemann constants are derived and a numerical algo-
rithm for the Jacobi problem is proposed in section 5. Section 6 presents the exact solution (6.10),
(2.10) to the physical problem and xes arbitrary constants C
j
( j = 1, 2, . . . , 15) in (6.5). The
case of normal incidence is considered in section 7. It is shown that the problem is equivalent to
two scalar RiemannHilbert problems on a segment of a complex plane. These problems are solved
in terms of one quadrature. The reected, surface and diffracted waves are recovered in section
8. Numerical results for the diffraction coefcients are reported and discussed. It is proved that
the electromagnetic eld and the diffraction coefcients are invariant with respect to the trans-
formation (,
0
, ,
1
,
2
) (,
0
, ,
1
,
2
), where is the angle of observation,
is the angle of incidence, and
0
is the angle between the screen and the incident plane
wave.
2. Formulation
PROBLEM 2.1 (Main problem) Let S be a thin semi-innite anisotropic impedance sheet {0
, = 0, z +} with surface impedances
+
1
,
+
2
on the upper side S
+
( = 0) and
1
,
2
on the lower surface S
( = + 0) (Fig. 1). The impedances may be

Fig. 1 Geometry of the problem
complex. The screen S is illuminated by a plane wave incident obliquely whose z-components are
E
i
z
= e
1
e
i k sin cos(
0
)i kz cos
, Z
0
H
i
z
= e
2
e
i k sin cos(
0
)i kz cos
, (2.1)
where k is the wave number (Imk 0), (0, ) is the angle of incidence, the angle
0

(, )\{0} denes the direction of incidence, e
1
and e
2
are prescribed parameters, and a time
factor e
i t
is suppressed. The two components of the electric and magnetic eld V
1
= E
z
and
V
2
= Z
0
H
z
solve the Helmholtz equation
(
2
+k
2
)V
j
= 0, (, , z) R
3
\S, (2.2)
and satisfy the following boundary conditions (13):
1
V
j
+(1)
j
cos
V
3j
i k
j
sin
2
V
j
= 0, = 0, j = 1, 2, (2.3)
where
1
= 1,
1
,
2
=
2
.
In this section the boundary-value problem for the Helmholtz equation will be reduced to two
vector difference equations with periodic coefcients. The total eld V = (V
1
, V
2
) is sought in the
form of the Sommerfeld integral (3)
V(, , z) =
e
i kz cos
2i
_
e
i k sin cos s
SSS(s +)ds. (2.4)
Here is the Sommerfeld contour. It consists of two loops symmetric with respect to the origin.
The asymptotes for the branches are the lines s =
3
2
and s =
1
2
for the upper loop and the lines
s =
1
2
and s =
3
2
for the lower loop. The spectral vector function SSS(s) = (S
1
(s), S
2
(s)) is
analytic everywhere in the strip Re s apart from the point s =
0
, where its components
have a simple pole with the residues dened by the incident eld (2.1)
res
s=
0
S
1
(s) = e
1
, res
s=
0
S
2
(s) = e
2
. (2.5)
At the innite points s = x i (|x| ), the functions S
1
(s) and S
2
(s) are bounded.
Because of the symmetry of the Sommerfeld contour in (2.4) the boundary conditions can be
rephrased in terms of the spectral functions as the system of the Maliuzhinets type difference equa-
tions
_
sin s
j
sin
_
S
j
(s ) +(1)
j
cos s cos S
3j
(s )
=
_
sin s
j
sin
_
S
j
(s ) +(1)
j
cos s cos S
3j
(s ), j = 1, 2. (2.6)
To satisfy the boundary conditions, we introduce the following four new functions:
+
j
(s) =
_
sin s
j
sin
_
S
j
(s) (1)
j
cos s cos S
3j
(s), j = 1, 2. (2.7)
Clearly, the transformation
T : (,
0
, ,
1
,
2
) (,
0
, ,
1
,
2
) (2.8)
does not change the electromagnetic eld. The general solution to the main problem invariant to the
transformation T is given by the Sommerfeld integral (2.4) with the spectral functions S
1
(s) and
S
2
(s) being expressed through the functions +
1
(s) and +
2
(s) as follows:
S
1
(s) =
1
2
_
S
+
1
(s) +S
1
(s)
, S
2
(s) =
1
2
_
S
+
2
(s) +S
2
(s)
, (2.9)
where
S
j
(s) =
1
I
s
(
1
,
2
)
__
sin s
3j
sin
_
+
j
(s)
+(1)
j
cos s cos +
3j
(s)
_
, j = 1, 2, (2.10)
and
I
s
(a, b) = (sin s +a sin )(sin s +b sin ) +cos
2
s cos
2
. (2.11)
The boundary conditions (2.6) for the functions S
1
(s) and S
2
(s) are equivalent to the following
conditions:
1
I
s
(1,
1
,
2
)
_
I
s
_
1
,
2
_
+
1
(s )
1
1
cos s sin 2+
2
(s )
_
=
1
I
s
(1,
1
,
2
)
_
I
s
_
1
,
2
_
+
1
(s )
1
1
cos s sin 2+
2
(s )
_
,
1
I
s
(1,
1
,
2
)
_
I
s
_
1
,
2
_
+
2
(s )
2
cos s sin 2+
1
(s )
_
=
1
I
s
(1,
1
,
2
)
_
I
s
_
1
,
2
_
+
2
(s )
2
cos s sin 2+
1
(s )
_
,
+
j
(s ) = +
j
(s ), j = 1, 2, (2.12)
Here
1
= 2
_
1
+
1
+
1
1
_
1
,
2
=

+
2
+
2
2
. (2.13)
From formulae (2.10) it follows that +
j
(s) = O(e
|s|
) ( j = 1, 2) as s (Re s is nite). Because
of the poles of the functions S
1
(s) and S
2
(s), the new functions +
1
(s) and +
2
(s) admit simple
poles at the point s =
0
. Analysis of the relations (2.10) implies that the functions S
1
(s), S
2
(s)
have inadmissible poles at the zeros of the functions I
s
(1,
1
,
2
) which lie in the strip
Re s . Let these zeros be denoted by
j
, j = 1, 2, 3, 4. The singular points
j
are the numbers
2j i log
_
i M

_
1 (M
)
2
_
, j = 0, 1, . . . , = 1, 2, (2.14)
which meet the requirement Re
j
(, ). Here
M
=
1
2
1
sin
_
1 +
2
+(1)
_
d
1
_
,
d
1
= (1
2
)
2
+4
1
(
1
) cos
2
, (2.15)
and
_
d
1
and log(i M

_
1 (M
)
2
) are xed branches of the square root and the loga-
rithmic functions, respectively. Notice that the transformation T maps the points
+
j
into
j
:
+
j
= T
j
and
j
= T
+
j
. The points
j
become removable points of the functions S
1
(s)
and S
2
(s) if the following conditions hold:
(sin s
2
sin )+
1
(s) cos s cos +
2
(s) = 0,
_
sin s
1
1
sin
_
+
2
(s) +cos s cos +
1
(s) = 0, s =
j
, j = 1, 2, 3, 4. (2.16)
Since the determinant of this system I
j
(1,
1
,
2
) is equal to 0, the above conditions are
equivalent to the following four equations:
(sin
2
sin )+
1
(
j
) cos
j
cos +
2
(
j
) = 0, j = 1, 2, 3, 4. (2.17)
Let H
+
be the innite strip H
+
= {s C| Re s 3}, and H
= {s C| 3 Re s
}. Denote the boundaries of the strips by O
= {Re = 2 } and O
1
= {Re = 2 }.
From the integral representation (2.4) and the boundary conditions (2.3) it may be deduced that the
vectors +++
(s) = (+
1
(s), +
2
(s))
solve the following problem.

PROBLEM 2.2 Find the vectors +++
(s) analytic in the strips H
apart from simple poles at the

points s =
0
and s = 2
0
, continuous up to the boundaries O
1
and satisfying the
boundary conditions
+++
() = G
()+++
( 4), O
, (2.18)
and the conditions of symmetry
+++
(s) =+++
(2 s), s H
, (2.19)
where the components G
mn
() of the matrices G
() are given by
G
+
11
() = G
11
() =
I
(1,
1
,
+
2
)I
(1,
+
1
,
2
) +
2
1
1
cos
2
sin
2
2
D()
,
G
+
22
() = G
22
() =
I
(1,
1
,
+
2
)I
(1,
+
1
,
2
) +
2
1
1
cos
2
sin
2
2
D()
,
G
12
() =
0
sin sin 2 sin 2
1
D()
, G
21
() =

2
G
12
(),
D() =
I
(1,
+
1
,
+
2
)
I
(1,
+
1
,
+
2
)
[I
(1,
1
,
+
2
)I
(1,
+
1
,
2
) +
2
1
1
cos
2
sin
2
2]
=
I
(1,
1
,
2
)
I
(1,
1
,
2
)
[I
(1,
+
1
,
2
)I
(1,
1
,
+
2
) +
2
1
1
cos
2
sin
2
2],
+
0
=
+
2

1
1
,
0
=
2

1
+
1
. (2.20)
At the ends of the strip, that is, as s (Re s is nite), +
j
(s) = O(e
|s|
) ( j = 1, 2). The
components +
j
(s) of the vectors +++
(s) satisfy the additional conditions (2.17).

REMARK 2.1 Because the entries of the matrices G
(s) are meromorphic functions, by analytical

continuation the boundary conditions (2.18) are valid not only on the contours O
but everywhere
in the complex plane apart from the poles of the solution and the functions G
i j
(s).
REMARK 2.2 In the case of normal incidence ( =
1
2
), G
() = diag{G
11
(), G
22
()}. This
case is considered in section 7.
3. Vector RiemannHilbert problem on a plane
3.1 RiemannHilbert problem for two even functions on a system of contours
Let =
1
2
. In this section Problem 2.2 will be converted into two vector RiemannHilbert prob-
lems on the segment (1, 1) and a system of branch cuts. In what follows the general method (6)
for vector difference equations with periodic meromorphic coefcients will be adjusted to Problem
2.2. The rst step of the procedure is decoupling of the vector equations (2.18). It may be achieved
by splitting the matrices G
():
G
() = T
()A
()[T
( 4)]
1
, O
, (3.1)
where A
() = diag{
1
(),
2
()}, and
1
() and
2
() are the eigenvalues of the matrices
G
(),
1
() = a
1
() +a
2
() f
1
2
(),
2
() = a
1
() a
2
() f
1
2
(). (3.2)
The matrices T
(s) meet the condition T
(s) = T
(s 4) (their elements are 2-periodic

functions), and they are given by
T
(s) =
_
1 1
f

1
(s) + f
1
2
(s) f

1
(s) f
1
2
(s)
_
. (3.3)
Here
a
1
() =
1
2
[G
11
() + G
22
()], a
2
() = G
12
(),
f

1
(s) =
G
11
(s) G
22
(s)
2G
12
(s)
=

1
(
+
0
+
0
)
2
0
sin 2 cos s
(cos
2
s cos
2
+sin
2
s g
0
sin
2
),
f

2
=
G
21
(s)
G
12
(s)
=

2
, f
(s) = [ f

1
(s)]
2
+ f

2
=
_
1
(
+
0
+
0
) tan
4
0
cos s
_
2
f

(s),
f

(s) =
_
cos
2
s + g
0

1
sin
2
_
2
+16g
1
cos
2
s cot
2
,
g
0
=
1
+
0
+
0
_
+
0
+
1
+
+
2
1
_
, g
1
=

2
+
0

1
(
+
0
+
0
)
2
. (3.4)
From the above formulae it follows that the matrices G
+
() and G
() have the same eigenvalues:
+
j
() =
j
() =
j
() and therefore A
+
() = A
() = A(). Clearly, the elements of the

matrices T
(s) are two-valued functions. They have branch points at the zeros of odd order of the
function f

(s) which are dened by the roots of the equations cos 2s = N
, s H
( = 1, 2),
where
N
= 1 +
2
sin
2
_
1 g
0
sin
2
8g
1
cos
2
+4i (1)
cos
_
d
0
_
,
d
0
= g
1
(1 g
0
sin
2
4g
1
cos
2
). (3.5)
Here

d
0
is one of the branches of the square root. In general, d
0
is complex. If d
0
= 0 then the
functions f
1,2
(s) are single-valued. This case for the strip H

+
is analysed in (6). Attention will now
be directed towards describing the details of the solution of Problem 2.2 in the case of the complex
impedance parameters (in particular, they might be real) when d
0
= 0.
Among the roots
j +
1
2
i log(N

_
N
2
1) ( = 1, 2; j = 0, 1, 2, . . .), (3.6)
only those roots s
j
should be taken which lie in the strip H
= {2 Re s 2 }. The
branch of the logarithmic functions is xed by the condition arg(N

_
N
2
1) , and
_
N
2
1 ( = 1, 2) is a xed branch. There are 16 roots in the strip H
, s
0
, s
1
, . . . , s
15
. Single
branches of the functions f
1,2
are xed by
f
1
2
(s) =
1
(
+
0
+
0
) tan
4
0
cos s
f

(s),
f

(s) cos
2
s, s x i ,
2 x 2 . (3.7)
Then, clearly, T f
1,2
+
(s) = f
1,2
(s). The branches are single-valued functions in the strips H
cut
along a system of cuts C
j
( j = 0, 1, . . . , 7) which join pairs of the branch points. The cuts are
smooth curves and they do not intersect each other.
According to the splitting (3.1) of the matrices G
() the new vector-functions
(s) = [T
(s)]
1
+
(s), s H
, (3.8)
are introduced. Their components are given by
1
(s) =
_
f

1
(s)
2 f
1,2
(s)
+
1
2
_
+
1
(s) +
+
2
(s)
2 f
1,2
(s)
,
2
(s) =
_
f

1
(s)
2 f
1,2
(s)
+
1
2
_
+
1
(s)
+
2
(s)
2 f
1,2
(s)
, s H
. (3.9)
Analysis of these formulae shows that the functions f

1
(s) f
1,2
(s) and f
1,2
(s) are free of poles

in the strips H
, respectively, and they have 16 branch points s
0
, s
1
, . . . , s
15
. Because of the poles
of the functions +
1
(s) and +
2
(s) at the points s =
0
and s = 2
0
, the functions
1
(s) and
2
(s) may have two simple poles at these points.
Clearly, the branch cuts C
j
( j = 0, 1, . . . , 7) are lines of discontinuity for the functions
1
(s)
and
2
(s). However, the vectors +
(s) = T
(s)
(s) have to be continuous through these

lines. Therefore the functions
1
(s) and
2
(s) solve the following system of two scalar difference
equations:
() =
()
( 4), = 1, 2, O
, (3.10)
subject to the RiemannHilbert boundary conditions on the cuts C
1
()|
L
=
2
()|
R
,
1
()|
R
=
2
()|
L
, C
j
( j = 0, 1, . . . , 7), (3.11)
where
()|
L
and
()|
R
are the limiting values of the functions
(s) on the contours C
j
from the left- and from the right-hand sides with respect to the positive directions of the contours
C
j
( = 1, 2; j = 1, 2, . . . , 7). It becomes evident from (3.9) that at the ends of the strip the
functions
1
(s) grow, and the functions
2
(s) are bounded:
1
(s) = O(e
|s|
),
2
(s) = O(1), s , s H
. (3.12)
By the mappings
w = i tan
s 2
4
, s = 2 +2i log
1 +w
1 w
, (3.13)
the contours O
are mapped onto the upper side of the cut [1, 1] (the left bank with respect to
the positive direction), and the left boundaries of the strips, O
1
, are mapped onto the lower side
of the cut. The images of the upper and the lower innite points of the strips H
, x i and
x +i (2 x 2 ), are the points w = 1 and w = 1, respectively. The function
log[(1 + w)(1 w)
1
] is real on the upper side of the cut. The systems of equations (3.10) and
(3.11) reduce to the following vector RiemannHilbert problem.
PROBLEM 3.1 Find the functions F
(s) analytic in the w-planes dened by (3.13), apart from the

segment [1, 1] and the poles w = w
and w = w
, where w
= i tan
1
4
(
0
), continuous
up to the segment (1, 1) and the contours C
j
( j = 0, 1, . . . , 7). The functions F
(s) are even:

F
(w) = F
(w), w C \ [1, 1], and satisfy the boundary conditions

F
+
(t ) = l
(t )F
(t ), = 1, 2, t (1, 1),
F
+
1
(t ) = F
2
(t ), F
+
2
(t ) = F
1
(t ), t C
j
, j = 0, 1, . . . , 7. (3.14)
Here the curves C
j
( j = 0, 1, . . . , 7) are the images of the branch cuts C
j
, and
F
(w) =
(s), w C, s H
,
F
+
(t ) = F
(w)|
L
, F
(t ) = F
(w)|
R
, t [1, 1],
l
(t ) =
(), t [1, 1], O
, (3.15)
where w and s are linked by (3.13) and t = i tan
1
4
( 2 ). At the ends,
F
1
(w) D
1
|w 1|
2
, F
2
(w) D
2
, w 1, (3.16)
where D
1
and D
2
are constants.
The functions F
1
(w) and F
2
(w) are even, because of the relations
+
(s) = +
(2 s), = 1, 2,
f

1
(s) = f

1
(2 s), f
1,2
(s) = f
1,2
(2 s), s H
. (3.17)
The class of solutions (3.16) for Problem 3.1 follows from the conditions (3.12).
It was proved in (6) that the even RiemannHilbert problem (3.14) has a solution if the coef-
cients l
(t ) ( = 1, 2) satisfy the condition l
(t )l
(t ) = 1. The functions l
(t ) meet the following

conditions:
l
(1) = l
(0) = l
(1) = 1, A
= 2, (3.18)
where A
are the increments of the argument of the functions l
(t ) as the point t traverses the

segment [0, 1] in the positive direction. The quantities A
are evaluated numerically. The argument

of the functions l
(t ) ( = 1, 2) is xed by the condition argl
(0) = 0.
3.2 Branch points and the function p
1
2
()
Analysis of the branch points
w
j
= i tan
s
j

4
, j = 0, 1, . . . , 15, (3.19)
shows that half of them are located in the upper half-plane. Moreover, for each branch point w
j
( j = 0, 1, . . . , 7) there is one branch point, say w
m+8
(m = 0, 1, . . . , 7), such that w
j
= w
m+8
.
Denote = w
2
= tan
2 1
4
(s ) and
j
= (w
j
)
2
( j = 0, 1, . . . , 15). Among the 16 numbers
j
there are eight distinct ones. Let they be reordered in the following manner:
Im
0
Im
1
Im
7
, (3.20)
and
Re
j
Re
j +1
if Im
j
= Im
j +1
. (3.21)
Then, clearly,
+
j
=
j
=
j
( j = 0, 1, . . . , 7). For example, if
=
1
3
,
+
1
= 1 i,
+
2
= 01 i,
1
= 2 i,
2
= 1 +i, (3.22)
then the branch points
j
are
0
= 23406 19658i,
1
= 016520 15996i,
2
= 012473 010242i,
3
= 0051898 0091307i,
4
= 025053 +021041i,
5
= 0063887 +061857i,
6
= 47887 +39321i,
7
= 47050 +82778i.
Notice that the branch points are invariant with respect to the transformation T : T
j
=
j
, j =
0, 1, . . . , 7. Since the shape of the branch cuts C
j
and the starting and the ending points of the
cuts were not chosen it may be assumed that the branch cuts are the straight lines in the -plane
which join the branch points
2 j
with
2 j +1
, j = 0, 1, 2, 3. These branch cuts are denoted by
j
.
The convention (3.20), (3.21) guarantees that the branch cuts will not cross each other (Fig. 2).
Notice that regardless of the values of the impedance parameters the rst four branch points
j
( j = 0, 1, 2, 3) are located below the real axis, and the other four branch points are in the upper
half-plane. If all the impedance parameters
j
are real, then
j
=
7j
, j = 0, 1, 2, 3.
Let p
1,2
() be the branch of the function
u
2
= p(), p() = (
0
)(
1
) . . . (
7
), (3.23)
Fig. 2 Branch cuts
0
,
1
,
2
and
3
chosen such that p
1,2
()
4
, . In what follows a relation between the single branches of
the functions f
1,2
(s) and the branch of the function p

1,2
( ) will be established. Since
= w
2
= cot
2
s
4
=
_
sin
1
2
s +1
sin
1
2
s 1
_
1
,
cos s = 1 2
_
+1
1
_
2
=
2
+6 +1
( 1)
2
, (3.24)
formula (3.7) implies
f
1
2
(s) =
1
(
+
0
+
0
) tan ( 1)
2
4
0
(
2
+6 +1)
_
f

(s). (3.25)
Use of the directly veried relations
cos 2s = 1 +
32( +1)
2
( 1)
4
(3.26)
and
f

(s) =
1
4
(cos 2s N
1
)(cos 2s N
2
) =
(1 N
1
)(1 N
2
)
4( 1)
8
p( ) =
2
12
p( )
p(1)( 1)
8
(3.27)
makes it possible to show that
_
f

(s) =
64
p
1,2
(1)( 1)
4
p
1,2
( ). (3.28)
Indeed, since cos
2
s 64( 1)
4
as 1 (s x ), the function

f

(s) meets the
condition (3.7). Therefore,
f
1,2
(s) =
16
1
(
+
0
+
0
) tan p
1,2
( )
0
p
1,2
(1)(
2
+6 +1)( 1)
2
. (3.29)
Analysis of the branches f
1,2
(s) and p
1,2
() shows that the functions f
1,2
(s) p
1,2
( ) do not have
branch points and also that at innity
f
1,2
(s)
2
1
(
+
0
+
0
) tan
0
( 1)
2
, s x i ( 1). (3.30)
4. Scalar RiemannHilbert problem on a surface of genus 3
In this section the vector problem (3.14) is formulated as a scalar RiemannHilbert problem on a
hyperelliptic surface, say R, of the algebraic function (3.23).
4.1 Formulation of the problem
The surface R is formed by gluing two copies C
1
and C
2
of the extended complex -plane C
cut along the system of the cuts
j
( j = 0, 1, 2, 3). The positive (left) sides of the cuts
j
on C
1
are glued with the negative (right) sides of the curves
j
on C
2
, and vice versa. The genus of the
surface is 3. The function u, dened by (3.23), becomes single-valued on the surface R:
u =
_
p
1,2
(), C
1
,
p
1,2
(), C
2
,
(4.1)
where a point of the surface R with afx on C
is denoted by the pair (, (1)

1
p
1,2
()),
= 1, 2. It follows from (3.14) that the functions
F
(, u) =
_
F
1
(w), (, u) C
1
,
F
2
(w), (, u) C
2
,
= w
2
= tan
2
s
4
, (4.2)
are single-valued on the Riemann surface R. They are meromorphic everywhere on R apart from
the contour L = L
1
L
2
, where L
1
= (0, 1) C
1
and L
2
= (0, 1) C
2
. Therefore, Problem 3.1
is equivalent to the following scalar RiemannHilbert problem on the surface R.
PROBLEM 4.1 Find the functions F
(, u) piecewise analytic on the surface R apart from the

geometrical optics poles at the points (
1
, p
1,2
(
1
)) and (
1
, p
1,2
(
1
)) and simple poles
at the branch points
0
,
1
, . . . ,
7
of the surface R, continuous up to the boundary L and satisfying
the boundary condition
F
+
(, ) = l(, )F
(, ), (, ) L, (4.3)
where
l(, ) =
_
l
1
(
), (, ) L
1
,
l
2
(
), (, ) L
2
,
(4.4)
= u(), and
1
= tan
2 1
4
(
0
). The functions F
(, u) are bounded at the innite points
( = 1, 2) of the surface. As 1, the functions F
(, u) are singular on the rst sheet:

F
(, u) B
( 1)
2
, and they are bounded on the second sheet. As 0 and (, u) R,
the functions F
(, u) are bounded.
Notice that a branch point
j
of a two-sheeted Riemann surface is called (18) a pole of order
j
for a function F(, u) if F(, u) Az
j
, z 0, A = const, and z = (
j
)
1,2
is a local
uniformizer of the point
j
.
4.2 Factorization of the coefcient l(t, )
The rst step of the procedure for the problem (4.3) is to factorize the function l(, ). This means
nding a meromorphic solution to the following problem:
l(, ) =
X
+
(, )
X
(, )
, (, ) L. (4.5)
Its solution is given in terms of the singular integrals with the Weierstrass kernel on the Riemann
surface
dU =
u +
2
d

, (, ) L, (, u) R, (4.6)
as follows (6):
X(, u) = e
(,u)
,
(, u) = I
1
(, u) +I
2
(, u) +I
3
(, u),
I
1
(, u) =
1
2i
_
L
logl(, )dU, I
2
(, u) =
_
p
11
p
10
dU
_
p
21
p
20
dU,
I
3
(, u) =
3
j =1
_
_
5
j
dU +m
j
_
a
j
dU +n
j
_
b
j
dU
_
. (4.7)
The function exp{I
1
(, u)} satises the boundary condition (4.3). However, its asymptotics at the
ends 1 of the contours L
1
and L
2
differs from the one required in Problem 4.1. Also, in general,
it has an essential singularity at the innite points
( = 1, 2) of the Riemann surface that is

not acceptable. The functions exp{I
2
(, u)} and exp{I
3
(, u)} are continuous through the contour
L and the contours of integration for the integrals I
2
and I
3
. The function I
2
is added to correct
the asymptotics of the solution at the ends of the contour L. The presence of the functions I
3
is
explained by the necessity to illuminate the essential singularity of the functions exp{I
1
(, u) +
I
2
(, u)} at the innite points.
The contours a
j
and b
j
( j = 1, 2, 3) are the canonical cross-sections of the surface R shown in
Fig. 3. The loops a
j
are closed contours formed by the branch cuts
1
,
2
and
3
:
a
1
= [
2
,
3
]
+
[
3
,
2
]
,
a
2
= [
4
,
5
]
+
[
5
,
4
]
,
a
3
= [
6
,
7
]
+
[
7
,
6
]
. (4.8)
The positive direction is chosen such that the rst sheet C
1
is on the left. The cross-sections b
j
are
closed contours consisting of two parts:
b
1
= [
1
,
2
]
C
2
[
2
,
1
]
C
1
,
b
2
= b
1
[
2
,
3
]
+
C
2
[
3
,
4
]
C
2
[
4
,
3
]
C
1
[
3
,
2
]
C
1
,
b
3
= b
2
[
4
,
5
]
+
C
2
[
5
,
6
]
C
2
[
6
,
5
]
C
1
[
5
,
4
]
C
1
. (4.9)
Fig. 3 Canonical cross-sections a
j
and b
j
( j = 1, 2, 3)
The rst part (the solid line in Fig. 3) lies on the sheet C
1
. The starting point of the cross-section
b
j
is
2 j
( j = 1, 2, 3), and the terminal point is
1
. The second part (the dashed line in Fig. 3)
lies on the second sheet C
2
. The contour b
j
crosses the loop a
j
from right to left and does not
cross the other loops a
k
and b
k
(k = j ) or the contour L. Notice that for all tested values of the
impedance parameters and the angle , the segment
3
4
does not cross the segment (0, 1), and the
point = 0 always lies to right of the segment
3
4
. Because of the conditions (3.18) and the choice
of the argument of the functions l
(t ), the single branch of the logarithmic function in (4.7) has the

following properties:
logl(, ) = logl
) = log |l
)| +i c
(), (, ) L
, = 1, 2,
c
() = argl
), c
() [2, 0], c
(0) = 0, c
(1) = 2. (4.10)
The contours 5
j
= r
j
q
j
( j = 1, 2, 3) are continuous curves with starting points r
j
= (
j
, n
j
)
C
1
, n
j
= p
1,2
(
j
). These points are distinct and xed arbitrarily. The ending points q
j
= (
j
, u
j
)
R, u
j
= u(
j
), are to be determined (they may lie on either sheet). The contours 5
j
cannot cross the
a- and b-loops or the contour L. It is convenient to take the contour 5
j
as a line that passes through
the branch point
0
(Fig. 4). If the terminal point q
j
C
1
, then the whole contour 5
j
C
1
. If this
point lies on the second sheet C
2
, then the contour 5
j
consists of two parts: the segment
j
0
C
1
,
and the line
0
j
C
2
. Depending on whether the point (
j
, u
j
) is to the left or to the right of the
broken line
0
1
. . .
7
, the part
0
j
of the contour 5
j
lies in the domain that is either to the left, or
to the right of the contour
0
1
. . .
7
. In Fig. 4, as an example, the following case is illustrated. The
point q
1
C
1
, and the points q
2
and q
3
lie on the sheet C
2
.
Fig. 4 Contours 5
1
, 5
2
and 5
3
The points p
1
( = 1, 2) are arbitrary, distinct and xed
p
1
= (
, (1)
1
z
) C
,
1
=
2
,
= 1, z
= p
1,2
(
), = 1, 2. (4.11)
The point p
0
coincides with the right end = 1 of the contour L
: p
0
= (1, (1)
1
p
1,2
(1)). The integers m
j
and n
j
( j = 1, 2, 3) and the points q
j
are to be determined. Formulae
(4.7) can be represented in another form more convenient for analysis
(, u) =
1
() +u()
2
(),
1
() =
1
4i
_
1
0
[logl
1
(
)+logl
2
(
)]
d

+
1
2
_

1
1
d

1
2
_

2
1
d

+
1
2
3
j =1
_
5
j
d

,
2
() =
1
4i
_
1
0
[logl
1
(
) logl
2
(
)]
d
p
1,2
()( )
+
1
2
_

1
1
d
p
1,2
()( )
+
1
2
_

2
1
d
p
1,2
()( )
+
1
2
3
j =1
_
_
5
j
+m
j
_
a
j
+n
j
_
b
j
_
d
()( )
. (4.12)
By the SokhotskiPlemelj formulae for integrals with the Weierstrass kernel the rst integrals in
the above formulae for the functions
1
( ) and
2
() are discontinuous through the contour [0, 1],
and the limiting values of the function X(, u) meet the condition (4.5). The other integrals are
continuous through the contour L. They are discontinuous through their lines of integration with
the jump multiple to 2i . The continuity of the function X(, u) = exp (, u), however, is not
affected by these jumps. Analysis of the rst, second, and third integrals in the representations
(4.12) of the functions
1
() and
2
( ) at the points = 0 and = 1 yields
(, u) const, (, u) (0, p
1,2
(0)) R,
(, u) 2 log( 1), (, u) (1, p
1,2
(1)) C
1
,
(, u) const, (, u) (1, p
1,2
(1)) C
2
. (4.13)
Thus, the function X(, u) is bounded at the points (0, p
1,2
(0)) R, and its asymptotics at the
ends (1, p
1,2
(1)) is the same as for the functions F
(, u) given by (3.16), (4.2).

4.3 Elimination of the essential singularity
The Weierstrass kernel (4.6) has a pole of order 3 at innity. Therefore, in general, the function
X(, u) has an essential singularity at innity that is not acceptable. Use of the identity
1

=
1
2

2
3
+

3
3
( )
(4.14)
gives the following asymptotic expansion of the function (, w) at innity:
(, w) =
1
2
3
=1
1
2i
_
1
0
[logl
1
(
) logl
2
(
)]
1
d
p
1,2
()
+
2
=1
_

1
d
p
1,2
()
+
3
j =1
_
_
5
j
+m
j
_
a
j
+n
j
_
b
j
_
1
d
()
u( )
+ O(1), , = 1, 2, 3.
(4.15)
Therefore the function (, w) is bounded at innity if and only if the points q
j
= (
j
, u
j
) and the
integers m
j
and n
j
solve the following system of nonlinear equations:
3
j =1
[
(q
j
) +m
j
A
j
+n
j
B
j
] = d
, = 1, 2, 3, (4.16)
where
d
=
3
j =1
(
j
, n
j
)
1
2i
_
1
0
[logl
1
(
) logl
2
(
)]
1
d
p
1,2
()

2
=1
_

1
d
p
1,2
()
. (4.17)
The differentials
d
=

1
d
()
, = 1, 2, 3, (4.18)
form a basis of abelian differentials of the rst kind on the surface R. The integrals
A
j
=
_
a
j
1
d
()
, B
j
=
_
b
j
1
d
()
(4.19)
are the A- and B-periods of the abelian integrals
(, u) =
_
(,u)
(
0
,0)
1
d
()
, = 1, 2, 3. (4.20)
The contour of integration does not intersect the canonical cross-sections. The nonlinear system
(4.16) with respect to the points q
j
= (
j
, u
j
) R and the integers m
j
and n
j
( j = 1, 2, 3) is
the Jacobi inversion problem for the surface R of genus 3. Clearly, the solution to this problem is
invariant with respect to the transformation T . To reduce the system (4.16) to standard form, the
basis of abelian integrals needs to be normalized. The A-periods of the normalized (canonical) basis
form the unit matrix. The canonical basis = {
} ( = 1, 2, 3) can be expressed through the basis

= {
} ( = 1, 2, 3) of abelian integrals (4.20) as follows:

= M, (4.21)
where M = A
1
, A = {A
j
} (, j = 1, 2, 3), and A
j
are given by (4.19). The A- and B-periods
of the canonical basis are
A
j
=
_
a
j
d
=
j
,
B
j
=
_
b
j
d
=
3
r=1
M
r
B
r j
. (4.22)
Here
j
is the Kronecker symbol. The matrix BBB = {B
j
} (, j = 1, 2, 3) is symmetric, and ImBBB is
a positive denite matrix. The Jacobi problem (4.16) can now be formulated for the canonical basis
3
j =1
[
(q
j
) +n
j
B
j
] +m
=

d
, = 1, 2, 3, (4.23)
where
=
3
j =1
M
j
d
j
, = 1, 2, 3, (4.24)
and M
j
are the elements of the matrix M.
5. Solution of the Jacobi inversion problem
Consider the Jacobi inversion problem for a Riemann surface of genus h.
PROBLEM 5.1 Given h constants

d
nd h points q
R and 2h integers n
and m
( =
1, 2, . . . , h) such that
h
j =1

(q
j
) = e
j =1
n
j
B
j
m
(modulo the periods), = 1, 2, . . . , h,

(5.1)
where e
=

d
+k
, and k
( = 1, 2, . . . , h) are the Riemann constants.

The unknown points q
j
( j = 1, 2, . . . , h) coincide with h zeros of the Riemann -function (see,
for example, (19, 5))
F(q) = (
1
(q) e
1
,
2
(q) e
2
, . . . ,
h
(q) e
h
) =
t
1
,t
2
,...,t
h
=
G
t
(q). (5.2)
where
G
t
(q) = exp
i
h
=1
h
=1
B
+2i
h
=1
t
(q) e
, t = (t
1
, t
2
, . . . , t
h
). (5.3)
Since the matrix ImB is positive denite the series (5.2) converges exponentially. The integers n
solve the system of linear algebraic equations

h
j =1
n
j
Im(B
j
) = Im(=
), = 1, 2, . . . , h, (5.4)
and the integers m
are dened explicitly through the integers n
by
m
= Re(=
)
h
j =1
n
j
Re(B
j
). (5.5)
Here
=
= e
j =1

(q
j
). (5.6)
5.1 Riemanns constants
The Riemann constants k
= k
(z)
were dened in (19, p. 144) as follows:

k
(z)
=
1
2
+
1
2
B
j =1, j =
_
a
j

(r)d
j
(r). (5.7)
Here
(r) is the limiting value of the function
(q) on the cross-section a

j
from the side of the
second sheet C
2
. It turns out that the use of these formulae gives non-integer values for the numbers
m
and n
. In what follows a correct formula for the constants k
will be derived. It will also be

shown that the zeros of the function (5.2) are the points q
j
. Following (19) consider the integral
J
=
1
2i
_
(r)d log F(r), (5.8)

where

R is the surface R cut along all the loops a
and b
( = 1, 2, . . . , h). Because the winding

number of the Riemann -function (5.2) for each contour a
equals 2, by the argument principle,

the -function has h zeros, say q
( = 1, 2 . . . , h). The residue theorem yields

J
=
h
j =1

(q
j
). (5.9)
On the other hand, the same integral may be represented as
J
=
h
j =1
(J
j
+J
j
), (5.10)
where
J
j
=
1
2i
_
a
j

+
(r)d log F
+
(r)
1
2i
_
a
j

(r)d log F
(r),
J
j
=
1
2i
_
b
j

+
(r)d log F
+
(r)
1
2i
_
b
j

(r)d log F
(r). (5.11)
Because of the properties of the abelian integrals and the -function, on the loops a
j

+
(r) =
(r) B
j
,
F
+
(r) = F
(r) exp{i B
j j
+2i [
+
j
(r) e
j
]}, (5.12)
and on the loops b
j

+
(r) =
(r) +
j
, F
+
(r) = F
(r), (5.13)
the integrals (5.11) reduce to
J
j
=
1
2i
_
a
j

(r)d log
F
+
(r)
F
(r)

B
j
2i
_
a
j
d log F
+
(r) =
_
a
j

(r)d
j
(r) n
j
B
j
,
J
j
=

j
2i
_
b
j
d log F
+
(r) =

j
2i
_
log F(r
+
j
) log F(r
++
j
)
_
=

j
2i
_
log
F(r
+
j
)
F(r
++
j
)
+2i m
j
_
, (5.14)
where n
j
and m
j
are some integers. The points r
++
j
and r
+
j
are shown in Fig. 5. Since the points
r
+
j
and r
++
j
are located on the different sides of the cross-section a
j
, from (5.12),
log
F(r
+
j
)
F(r
++
j
)
= i B
j j
2i [
j
(r
++
j
) e
j
] +2i m
j
, (5.15)
where m
j
are some integers. Therefore
J
=
h
j =1
_
_
a
j

(r)d
j
(r) n
j
B
j
_
1
2
B
(r
++
) +e
+m
+m
. (5.16)
Performing the integration and using
(r
) =
(r
+
) 1 yield
_
a
(r)d
(r) =
1
2
_
a
d[
(r)]
2
=
1
2
[
(r
+
(r
)][
(r
+
) +
(r
)] =
(r
+
)
1
2
=
(r
++
) +B

1
2
. (5.17)
Notice that this formula is not reducible to the expression
1
2
+
(r
++
) (modulo the periods) (5.18)

Fig. 5 Points r
++
j
, r
+
j
, r
+
j
and r
j
derived in (19, formula (5.15)). By substituting (5.17) into (5.16) and denoting m
= (m
+m
)+
1, n
= n
, and
k
=
1
2

1
2
B
j =1, j =
_
a
j

(r)d
j
(r), (5.19)
the integral J
is converted into
J
= e
j =1
n
j
B
j
m
. (5.20)
Comparing with (5.9) and putting q
j
= q
j
give the Jacobi inversion problem (5.1). Clearly, the
Riemann constants (5.19) and the constants (5.7) dened in (19) are not the same. Indeed, the
Jacobi problem in terms of the constants k
= k
(z)
can be written as follows:

h
j =1

(q
j
) e
k
(z)
+B
(modulo the periods), = 1, 2, . . . , h. (5.21)

Equivalently,
h
j =1

(q
j
) = e
k
(z)
j =1
(n
j

j
)B
j
, = 1, 2, . . . , h. (5.22)
Obviously, it is impossible to nd integers m
and n
such that systems (5.1) and (5.22) are equiv-

alent. This is the reason why use of the constants (5.7) for numerical computations gives fractional
values for the constants m
and n
.
To simplify the expressions for the Riemann constants, evaluate the integral in (5.19). Rewrite it
rst as
K
j
=
_
a
j

(r)d
j
(r) =
_
a
j

+
(r)d
j
(r) +B
j
. (5.23)
Let r
+
,
+
and r
be points on the left and the right sides of the cut [

2 j
,
2 j +1
] C
1
, respec-
tively. Then

+
(r
) =
_
[
0
,
2 j +1
]C
1
d
(r) +
_
[
2 j +1
,r]C
1
d
) =
+
(
2 j +1
)
+
j
(r), (5.24)
where

+
j
(r) =
_
[r,
2 j +1
]C
1
d
(
+
). (5.25)
It follows from (5.24) and
_
[
2 j
,
2 j +1
]
d
(
+
) =
1
2
_
a
j
d
(r) = 0, = j, (5.26)
that

+
(r
+
) +
+
(r
) = 2
+
(
2 j +1
) = 2
+
(
2 j
). (5.27)
Hence
K
j
= B
j
+
+
(
2 j
). (5.28)
For a hyperelliptic surface Rof genus h with the system of cross-sections as in Fig. 3, the integrals

+
(
2 j
) can be calculated. By the Cauchy theorem,
_
[
0
,
1
]
d
(r) =
h
j =1
_
a
j
d
(r) = 1. (5.29)
Therefore

+
(
2
) =
__
[
0
,
1
]
+
C
1
+
_
[
1
,
2
]C
1
_
d
(r) =
1
2

1
2
B
1
. (5.30)
Similarly, for j = 2

+
(
4
) =
+
(
2
) +
1
2
1
+
_
[
3
,
4
]C
1
d
(r) =
1
2

1
2
B
2
+
1
2
1
. (5.31)
For any j 1, it may be deduced that

+
(
2 j
) =
1
2

1
2
B
j
+
1
2
(
1
+
2
+ +
, j 1
). (5.32)
By substituting (5.32) into (5.28) and (5.19), the Riemann constants k
become
k
= 1 +
1
2

1
2
h
j =1
B
j
. (5.33)
Notice that the use of another set of the Riemann constants linked to the constants (5.33) by the
relation
= k
+ m
+ n
j =1
B
j
(5.34)
(for example,

k
=
1
2
+
1
2
(B
1
+ + B
h
)) gives also a solution to the Jacobi problem. Here
m
and n
are integers. In this case the Riemann -function should be taken in the form
F(q) = (
1
(q) e
1
,
2
(q) e
2
, . . . ,
h
(q) e
h
), e
=

d
+

k
. (5.35)
This result follows from the periodicity properties of the Riemann -function
(
1
+ j
1
,
+ j
2
, . . . ,
h
+ j
h
) = (
1
,
2
, . . . ,
h
), j
= 0, 1,
(
1
+B
1
,
2
+B
2
, . . . ,
h
+B
h
) = e
i B
2i
(
1
,
2
, . . . ,
h
), = 1, 2, . . . , h.
(5.36)
5.2 Afxes of the points q
1
, q
2
and q
3
The problem of nding the three zeros of the -function
F(q) = (
1
(q) e
1
,
2
(q) e
2
,
(
q) e
3
) =
t
1
,t
2
,t
3
=
G
t
(q) (5.37)
for h = 3 can be reduced to a cubic equation. Here
G
t
(q) = exp
i
3
=1
3
=1
B
+2i
3
=1
t
(q) e
, t = (t
1
, t
2
, t
3
),
e
=

d
+k
, k
=
1
2
+
1
2
(B
1
+B
2
+B
3
), = 1, 2, 3, (5.38)
and the coefcients

d
are given by (4.24).

Consider the integral
N
=
1
2i
_
d log F(r) (5.39)

over the boundary

Rof the surface Rcut along the cross-sections a
1
, a
2
and a
3
. By the logarithmic
residue theorem,
N
=
3
j =1
j
+
2
=1
_
res
q=0
+ res
q=
(q)
F(q)
, (5.40)
where 0
= (0, (1)
1
p
1,2
(0)) C
, and
( = 1, 2) are the two innite points of

the surface. Without loss of generality, F(0
) = 0, and F(
) = 0, = 1, 2. Because of the
asymptotics
(q)
F(q)
= O(
2
), , = 1, 2, 3, (5.41)
the residues at the innite points
vanish. The same integral N
may be represented as
N
=
3
j =1
_
a
j
t
d
+
j
(r) = 2
3
j =1
3
l=1
M
jl
_
[
2 j
2 j +1
]
l1
d
p
1,2
(
+
)
, (5.42)
where
+
[
2 j
,
2 j +1
]
+
. Comparing the two expressions for N
yields
1
1
+
1
2
+
1
3
=
, = 1, 2, 3, (5.43)
where
= 2
3
j =1
3
l=1
M
jl
_
[
2 j
2 j +1
]
l1
d
p
1,2
(
+
)

2
=1
res
q=0
(q)
F(q)
. (5.44)
Note that the integral (5.39) is different from the one used in (19, 5). The function
is replaced
by the function
. This change simplies the task of elevating the coefcients
. Instead of the
residues at the innite points
( = 1, 2), it is required to nd the residues at the two zeros of

the surface R.
The systems of algebraic equations (5.43) is converted into the cubic equation
(
3
1
3
1
2
+2
3
)
3
+3(
2

2
1
)
2
+6
1
6 = 0 (5.45)
with complex coefcients. The three roots of this equation dene the afxes of the zeros of the
-function F(q).
5.3 Numerical procedure
The rst step of the procedure is to evaluate the matrix of B-periods of the canonical basis by
formulae (4.21) and (4.22). This requires the A- and B-periods of the basis :
A
j
= 2
_
[
2 j
2 j +1
]
1
d
p
1,2
(
+
)
, j = 1, 2, 3,
B
1
= 2
_
[
1
2
]
1
d
p
1,2
()
, B
2
= B
1
2
_
[
3
4
]
1
d
p
1,2
()
,
B
3
= B
2
2
_
[
5
6
]
1
d
p
1,2
()
, = 1, 2, 3. (5.46)
These integrals are computed by the GaussChebyshev quadrature formulae. The branch of the
function p
1,2
() in the cut -plane (Fig. 2) has been xed by the condition p
1,2
( )
4
, .
For numerical purposes, this branch can algorithmically be described as follows. Dene
y
m
=
2m+1

2m
(m = 0, 1, 2, 3),
m
=
_
tan
1
(Im y
m
, Re y
m
), Re y
m
= 0,
,2, Re y
m
= 0,
j
=
j
( j = 0, 1, . . . , 7),
j
=
tan
1
(Im
j
, Re
j
), Re
j
= 0,
,2, Re
j
= 0, Im
j
> 0,
,2, Re
j
= 0, Im
j
0.
(5.47)
Then the branch p
1,2
() behaves at innity as
4
if
p
1,2
() =
_
7
j =0
|
j
| exp
i
2
7
j =0
,
j
= arg
j
. (5.48)
In the case
m
0 and ,
m
the arguments
j
are chosen to be +
m

j
+
m
(m = 0, 1, 2, 3). This means that
j
=
j
, Re
j
0,
j
+, Re
j
0,
j

m
,
j
, Re
j
0,
j
>
m
,
m = [ j,2], (5.49)
where [ j,2] is the integer part of j,2. If
m
0 and as before ,
m
, then
m

j
2 +
m
.
Therefore,
j
=
j
+, Re
j
0,
j
+2, Re
j
0,
j

m
,
j
, Re
j
0,
j

m
,
m = [ j,2]. (5.50)
The evaluation of the A-periods by formulae (5.46) requires dening p
1,2
(
+
), where
+

+
m
. In
this case for
m
0,
j
=
_
m
, j = 0, 2, 4, 6,
m
, j = 1, 3, 5, 7,
m = [ j,2], (5.51)
and if
m
0, then
j
=
_
m
, j = 0, 2, 4, 6,
m
+, j = 1, 3, 5, 7,
m = [ j,2]. (5.52)
To compute the coefcients of the cubic equation (5.45) by formulae (5.44) one needs to evaluate
the residues of the function
(q),F(q) ( = 1, 2, 3) at the points q = 0
, = 1, 2.
They are given by
res
q=0
(q)
F(q)
=1
=
F
(q)
F(q)
q=0
,
res
q=0
(q)
F(q)
=2
=
_
F
(q)
F(q)

_
F
(q)
F(q)
_
2
_
q=0
,
res
q=0
(q)
F(q)
=3
=
1
2
_
F
(q)
F(q)

3F
(q)F
(q)
F
2
(q)
+2
_
F
(q)
F(q)
_
3
_
q=0
, (5.53)
where
F
(0
) = 2i
t
1
,t
2
,t
3
=
(1)
t
G
t
(0
),
F
(0
) = 2i
t
1
,t
2
,t
3
=
_
2i
_
(1)
t
_
2
+
(2)
t
_
G
t
(0
),
F
(0
) = 2i
t
1
,t
2
,t
3
=
_
4
2
_
(1)
t
_
3
+6i
(1)
t

(2)
t
+
(3)
t
_
G
t
(0
),
(s)
t
=
3
=1
t

(s)
(0
). (5.54)
The derivatives
(s)
(0
) of the abelian integrals
at the two zeros of the surface Rare dened as

follows:

(0
) =
(1)
1
M
1
p
1,2
(0)
,

(0
) =
(1)
1
p
1,2
(0)
_
M
2

M
1
p
(0)
2p(0)
_
,

(0
) =
(1)
1
p
1,2
(0)
_
2M
3

M
2
p
(0)
p(0)

M
1
2
_
p
(0)
p(0)

3
2
_
p
(0)
p(0)
_
2
__
, (5.55)
where
p
(0)
p(0)
=
7
j =0
1
j
,
p
(0)
p(0)
=
j =0
1
j =0
1
2
j
. (5.56)
To nd a root
1
= x
1
+i x
2
of equation (5.45) the following procedure is applied. Equation (5.45)
is written as H
1
+i H
2
= 0, where
H
1
= x
3
1
3x
1
x
2
2
+a
1
(x
2
1
x
2
2
) 2a
2
x
1
x
2
+b
1
x
1
b
2
x
2
+c
1
,
H
2
= x
3
2
+3x
2
1
x
2
+a
2
(x
2
1
x
2
2
) +2a
1
x
1
x
2
+b
1
x
2
+b
2
x
1
+c
2
,
a
1
+i a
2
= 3(
2

2
1
)
1
, b
1
+i b
2
= 6
1
,
c
1
+i c
2
= 6
1
=
3
1
3
1
2
+2
3
. (5.57)
The NewtonRaphson iteration for evaluation of x
1
and x
2
x
(n)
j
= x
(n1)
j

H
1,1
H
j
+(1)
j
H
1,2
H
3j
A
,
A = H
2
1,1
+ H
2
1,2
, H
1, j
=
H
1
x
j
, j = 1, 2, (5.58)
is employed. The standard iteration formula becomes simpler because of the relations H
2,2
= H
1,1
and H
2,1
= H
1,2
. The other two roots of equation (5.45) are found from the associated quadratic
equation.
The numerical algorithm proposed can be tested by checking numerically the following
properties.
(i) The matrix B of B-periods is symmetric and ImB is a positive denite matrix.
(ii) The Riemann -function meets the conditions (5.36) (h = 3).
(iii) Among each pair of points (
j
, p
1,2
(
j
) C
1
and (
j
, p
1,2
(
j
) C
2
( j = 1, 2, 3) there is
one and only one point q
j
such that F(q
j
) = 0.
(iv) The solution n
( = 1, 2, 3) of the algebraic system (5.4) and the numbers m
dened by
(5.5) are integers.
(v) The points q
j
and the integers m
j
and n
j
satisfy the conditions (4.16), and therefore the solu-
tion X(, u) is bounded at innity.
(vi) The solution of the Jacobi problem is independent of the paths 5
j
, j = 1, 2, 3. It depends on
the location of the free points r
1
, r
2
, r
3
, p
11
and p
21
. However, the solution of Problem 2.2 is
independent of the choice of these points provided the paths 5
j
( j = 1, 2, 3) do not intersect
the a- and b-loops and the contour L.
All these properties of the solution have been successfully veried.
For the parameters
and
j
chosen to be
= +1 ( = 1, 2),
j
=
0
i j ( j = 1, 2, 3), (5.59)
and for the set of parameters (3.22) it turns out that n
= m
= 0, = 1, 2, 3,
1
= 03104 04492i,
2
= 25062 12753i,
3
= 46000 +20691i.
The points q
1
and q
3
lie on the rst sheet C
1
, and q
2
C
2
. If =
1
8
and the other parameters are
the same, then n
= m
= 0, = 1, 2, 3, q
1
C
1
, q
2
C
2
, q
3
C
2
, and their afxes are
1
= 05112 02745i,
2
= 65251 38299i,
3
= 158337 +104951i.
It is clear that if | | is large, then even a small error of computation in the afxes of the points q
j
may produce a big error for the values of the function X(, u). To overcome this difculty, in the
case | | > 1, formula (4.12) for the function
2
() is replaced by
2
() =
1
3
_
1
4i
_
1
0
[logl
1
(
) logl
2
(
)]

3
d
p
1,2
()( )
+
1
2
_

1
1
3
d
p
1,2
()( )
+
1
2
_

2
1
3
d
p
1,2
()( )
+
1
2
3
j =1
_
_
5
j
+m
j
_
a
j
+n
j
_
b
j
_

3
d
()( )
. (5.60)
Notice the above formula is not an asymptotic relation but an exact equality. The right-hand side of
the second formula in (4.12) coincides with (5.60) provided the integers n
j
and m
j
and the points
q
j
( j = 1, 2, 3) solve the Jacobi problem (4.23). Formula (5.60) guarantees the stability of the
numerical values of the function exp{u( )
2
( )} for large | | and makes the numerical algorithm
efcient for numerical purposes.
6. Solution to the physical problem and denition of the constants
To proceed now with the general solution of Problem 4.1, notice that the function X(, u) dened
by (4.7) has simple zeros at the points p
11
= (
1
, z
1
) and q
j
= (
j
, u
j
) ( j = 1, 2, 3), where
u
j
= (1)
1
p
1,2
(
j
), = 1 if q
j
C
1
and = 2 if q
j
C
2
. Therefore, the general solution
may have removable singularities at these points. Due to the presence of the prescribed poles of
the functions
(s) ( = 1, 2) at s =
0
and s = 2
0
, the functions F
(w) have simple

poles at the points w = i tan(
0
),4 and w = i tan(
0
),4. Thus, the general solution
of the problem (4.3) on the surface R has simple poles at the points of the surface with afxes
1
= tan
2 1
4
(
0
). Next, the functions
1
(s) and
2
(s) are analytic at s =
1
2
, an inner
point of the strips H
. This point is the image of the innite points
( = 1, 2) of the Riemann
surface R. Therefore, the functions F
(, u) are bounded at the innite points. By the generalized

Liouville theorem applied to the analytic continuations of the functions [X
+
(, )]
1
F
+
(, ) and
[X
(, )]
1
F
(, ) from the contour L to the surface R,

F
(, u) = X(, u)R
(, u), (6.1)
where
R
(, u) = R
1
() +u( )R
2
( ),
R
1
() =
3
j =1
C
j
u
j

j
+
C
12

1
+
C
14
z
1

1
+C
15
,
R
2
() =
3
j =1
C
j

j
+
7
j =0
C
j +4

j
+
C
13

1
+
C
14

1
, (6.2)
and C
j
( j = 1, 2, . . . , 15) are arbitrary constants to be dened. From(6.2) it is seen that the rational
functions R
(, u) have simple poles at the points q

j
= (
j
, u
j
), and they are bounded at the points
(
j
, u
j
). Because of the factor u() in the rst formula (6.2), the points
j
( j = 0, 1, . . . , 7) are
simple poles of the functions F
(, u) on the Riemann surface R. The solution has prescribed

poles at the points (
1
, p
1,2
(
1
)) and (
1
, p
1,2
(
1
)). The point (
1
, p
1,2
(
1
)) C
1
is a
removable singular point for the functions F
(, u). This is because z

1
= p
1,2
(
1
), and the function
X(, u) has a simple zero at the point (
1
, p
1,2
(
1
)). The point (
1
, p
1,2
(
1
)) C
2
is a regular
point.
The functions F
(, u) have to be bounded at the innite points
( = 1, 2). Therefore
lim

R
2
() = 0, = 1, 2, 3. (6.3)
Equivalently, these three conditions may be written as
3
j =1
C
j

1
j
+
7
j =0
C
j +4
1
j
+C
13
1
1
+C
14
1
1
= 0, = 1, 2, 3. (6.4)
It will be convenient to use the following representations for the rational functions R
1
() and
R
2
():
R
1
() =
15
j =1
C
j

j
(), R
2
() =
15
j =1
C
j

j
(), (6.5)
where
j
() =
u
j

j
,
j
() =
1

j
, j = 1, 2, 3,
j
() = 0,
j
( ) =
1

j 4
, j = 4, 5, . . . , 11,
12
( ) =
13
() =
1

1
,
12
() =
13
() = 0,
14
() =
z
1

1
,
14
() =
1

1
,
15
() = 1,
15
() = 0. (6.6)
Here z
1
= p
1,2
(
1
).
Because of the poles of the function X(, u) at the points r
= (
, n
) C
1
( = 1, 2, 3)
and p
21
= (
2
, z
2
) C
2
, the general solution (6.1) has inadmissible poles at these points. The
following conditions remove the singularities:
15
j =1
[
j
(
) +n
j
(
)]C
j
= 0, = 1, 2, 3,
15
j =1
[
j
(
2
) z
2
j
(
2
)]C
j
= 0. (6.7)
To derive the other conditions for the constants C
j
( j = 1, 2, . . . , 15), expressions for the functions
+
1
(s) and +
2
(s) are needed. Inverting the relations (3.9) and using (3.15), (4.2), (4.7), (4.12), and
(6.1) yield
+
1
(s) = F
(, u) + F
(, u),
+
2
(s) = f

1
(s)[F
(, u) + F
(, u)] + f
1,2
(s)[F
(, u) F
(, u)], (6.8)
where
F
(, u) = e
1
()+u()
2
()
[R
1
() +u()R
2
( )]. (6.9)
The above formulae may be rewritten in the form
+
1
(s) = 2
1
()[R
1
() cosh
2
() + p
1,2
()R
2
() sinh
2
( )],
+
2
(s) = f

1
(s)+
1
(s) +2 f
1,2
(s)
1
()[R
1
( ) sinh
2
() + p
1,2
( )R
2
() cosh
2
()],
(6.10)
where

1
() = e
1
()
,
2
() = p
1,2
( )
2
( ). (6.11)
By substituting (6.5) into (6.10), expressions with explicit dependence on the constants C
j
are
obtained:
+
1
(s) = 2
1
()
15
j =1
C
j
[
j
() cosh
2
() + p
1,2
()
j
() sinh
2
()],
+
2
(s) = 2
1
()
15
j =1
C
j
_
[f

1
(s)
j
( ) + f
1,2
(s) p
1,2
( )
j
()] cosh
2
( )
+[f

1
(s) p
1,2
()
j
() + f
1,2
(s)
j
()] sinh
2
()
_
. (6.12)
The functions +
1
(s) and +
2
(s) have to meet the four conditions (2.17) which give the next four
equations for the constants C
j
( j = 1, 2, . . . , 15):
15
j =1
_
j
(
p
1,2
(
)
j
(
)
+[
p
1,2
(
)
j
(
j
(
)] tanh
2
(
)
_
C
j
= 0, = 1, 2, 3, 4, (6.13)
where
= cot
2 1
4
(
),
= sin
2
sin +cos
cos f

1
(
),
= cos
cos f
1,2
), = 1, 2, 3, 4. (6.14)
The function +
2
(s) dened by (6.12) has two simple poles at the points
1
2
and
1
2
. To remove
these singularities the following conditions have to be imposed:
res
s=
1
2
2
(s) = 0, res
s=
1
2
2
(s) = 0. (6.15)
Because of the symmetry property +
2
(s) = +
2
(2 s), the function +
2
(s) is regular at the
points s =
3
2
and s =
5
2
provided it meets the conditions (6.15). Equations (6.15) can be
rewritten in the form
15
j =1
_
j
(
) +
sin p
1,2
(
)
j
(
)
+
_
p
1,2
(
)
j
(
) +
sin
j
(
)
_
tanh
2
(
)
_
C
j
= 0, = 1, 2, (6.16)
where
1
= cot
2 1
8
,

2
= tan
2 1
8
,
=
1 g
0
sin
2
sin
,
=
_
f

_
1
2
_
=
_
f

_
1
2
_
=
64p
1,2
(
1
)
p
1,2
(1)(
1
1)
4
. (6.17)
The last two conditions for the constants can be obtained by xing the residues of the functions
+
1
(s) and +
2
(s) at the geometric optics pole s =
0
:
res
s=
0
+
1
(s) = e
1
_
sin
0

sin
1
_
+e
2
cos
0
cos ,
res
s=
0
+
2
(s) = e
2
_
sin
0

2
sin
_
e
1
cos
0
cos . (6.18)
Evaluating the residue of the function (
1
)
1
at the point s =
0
yields
= res
s=
0
1

1
=
2 sin
3 1
4
(
0
)
cos
1
4
(
0
)
. (6.19)
The conditions (6.18) become therefore
C
12
+ p
1,2
(
1
) tanh
2
(
1
)C
13
=
1
2

1
(
1
) cosh
2
(
1
)
_
e
1
_
sin
0

1
1
sin
_
e
2
cos
0
cos
_
,
[ f
1,2
(
0
) tanh
2
(
1
) f

1
(
0
)]C
12
+[ f
1,2
(
0
) f

1
(
0
) tanh
2
(
1
)] p
1,2
(
1
)C
13
=
1
2

1
(
1
) cosh
2
(
1
)
[e
2
_
sin
0

2
sin
_
+e
1
cos
0
cos ]. (6.20)
The solution of the problem is dened by quadratures (2.4) in terms of the spectral functions S
1
(s)
and S
2
(s) expressed through the functions +
1
(s) and +
2
(s) and the integrals (4.12) by formulae
(2.9), (2.10), (6.12) (6.11) and (6.5). The constants C
j
( j = 0, 1, . . . , 15) solve the two separate
systems of linear algebraic equations (6.4), (6.7), (6.13), (6.16) and (6.20) which can be written in
the form
15
j =1
a
j
C
j
= h
, = 1, 2, . . . , 15. (6.21)
It turns out, for some physical parameters, the coefcients a
j
might be badly scaled. For numerics,
the system (6.21) is rescaled
15
j =1
a
j
C
j
=

h
, = 1, 2, . . . , 15, (6.22)
where
a
j
= 10
s
j
,

h
= 10
s
, s
= [lg a
], = 1, 2, . . . , 15. (6.23)
For the parameters (3.22) and (5.59), 4 s
+
0, 5 s
0.
Because of the relations

= T

( = 1, 2, 3, 4),
1+
= T
1
, and
+
= T
, the
transformation T maps the constants C
+
j
into the constants C
j
: T C
+
j
= C
j
( j = 1, . . . , 15).
7. Normal incidence
In the case =
1
2
, the representations (2.10) become
S
(s) =
+
(s)
sin s
, = 1, 2. (7.1)
The matrices G
() are diagonal, and Problem 2.2 is uncoupled

+
() = G
()+
( 4), O
,
+
(s) = +
(2 s), s H
, = 1, 2, (7.2)
where
G
() =
(sin +
+
)(sin +
)
(sin
+
)(sin
)
. (7.3)
As s (Re s is nite), +
(s) = O(e
|s|
). The functions +
(s) have to satisfy the conditions

(2.17) which read
+
) = 0, , = 1, 2, (7.4)
where sin
and Re
(, ). If the impedances
1
and
2
are real and positive,
then
1
= sin
1

2
=
1
. Otherwise
are those numbers
= i log
_
i
(1)
_
1 (
)
2
_
+2m, m = 0, 1, . . . , (7.5)
which lie in the strip Re s .
By the mapping (3.13), the difference equations (7.2) reduce to the scalar RiemannHilbert prob-
lems on the segment (1, 1)
F
+
(t ) = l
(t )F
(t ), t (1, 1),
F
(w) = F
(w), w = w
= C \ [1, 1], = 1, 2. (7.6)

where
l
(t ) = G
(), +
(s) = F
(w), w = i tan
1
4
(s ). (7.7)
The functions F
(w) grow at the ends

F
(w) = O(|w 1|
2
), w 1. (7.8)
The coefcients l
(t ) have the following properties:

l
(t )l
(t ) = 1, l
(0) = l
(1) = 1,
[argl
(t )]
[1,0]
= [argl
(t )]
[0,1]
= 2. (7.9)
Here [g(t )]
[a,b]
is the increment of a function g(t ) as t traverses the contour [a, b] in the positive
direction. Choose argl
(0) = 0. Then argl
(1) = 2. Factorize next the coefcients l
(t ):
l
(t ) =
X
+
(t )
X
(t )
, t (1, 1), (7.10)
where X
(t ) = X
(t i 0), and
X
(w) = (w
2
1)
1
exp
_
1
2i
_
1
1
logl
(t )
t w
dt
_
= (w
2
1)
1
exp
_
1
i
_
1
0
logl
(t )
t
2
w
2
t dt
_
. (7.11)
Analysis of the Cauchy integrals (7.11) shows that
X
(w) A
(w 1)
2
, w 1, A
= const. (7.12)
Thus, the functions X
(w) are even and have the asymptotics at the ends w = 1 required by the
class of solutions. The generalized Liouville theorem yields the solution
F
(w) = C
(w)
_
1
w
2
w
2
+d
1
+d
2
w
2
_
, w C, = 1, 2, (7.13)
where C
and d
( = 1, 2, = 0, 1, 2) are arbitrary constants. The functions F
(w) are even,

bounded at innity, have the asymptotics (7.8) at the ends w = 1 and possess the geometrical op-
tics poles at the points w = w
, w
= i tan
1
4
(
0
). The conditions (7.4) give the expressions
for the constants d
1
and d
2
d
1
=
(
1
)
2
2
(
2
)
2
1
(
2
)
2
(
1
)
2
, d
2
=
1

d
2
(
2
)
2
(
1
)
2
, (7.14)
where
= i tan

4
,

d
=
1
(
)
2
w
2
. (7.15)
The physical conditions (2.5) and the invariance of the solution with respect to the transformation
,
0

0
, and
( = 1, 2) dene the constants C
C
+
=
e
(sin
0

+
)
X
( w
+
)
+
, C
=
e
(sin
0
+
)
X
( w
, = 1, 2, (7.16)
where
are given by (6.19). Formula (7.16) indicates that T C

+
= C
, = 1, 2.
8. High frequency asymptotics. Numerical results
High frequency asymptotics of the electrical and magnetic eld can be represented in the form
E
z
E
i
z
+ E
r
z
+ E
s
z
+ E
d
z
,
H
z
H
i
z
+ H
r
z
+ H
s
z
+ H
d
z
, k , (8.1)
where E
i
z
and H
i
z
are the incident waves, E
r
z
and H
r
z
are the reected waves, E
s
z
and H
s
z
are the
surface waves, and E
d
z
and H
d
z
are the diffracted waves. These waves may be recovered by applying
the steepest descent method to the Sommerfeld integrals (2.4). Analysis of the spectral functions
S
1
(s) and S
2
(s) shows that the geometrical optics poles should be found among the roots of the
equation
tan
2
s
4
= tan
2

0

4
, (8.2)
which lie in the strip + Re s + . The pole s =
0
gives rise to the incident waves
(2.1). The next two poles s = 2
0
and s = 2
0
reproduce the reected waves provided
that the angle of incidence
0
and the angle of observation satisfy the conditions
0
0
and
0
in the case s = 2
0
, and 0
0
and
0
in the
case s = 2
0
. By the Cauchy theorem the reected waves become
_
E
r
z
H
r
z
_
= e
i kz cos +i k sin cos(+
0
)
__
R
e
R
h
_
(,
0
) +
_
R
+
e
R
+
h
_
(
0
, )
_
,
(8.3)
where
(a, b) =
_
1, [a, b],
0, , [a, b].
(8.4)
The reection coefcients R
e
and R
h
are dened by
_
R
e
R
h
_
=
1
2I
0
(1,
+
1
,
+
2
)
_
sin
0
+
+
2
sin cos
0
cos
cos
0
cos sin
0
+1,
+
1
sin
_
res
s=
0
2
_
+
+
1
(s)
+
+
2
(s)
_
+
1
2I
0
(1,
1
,
2
)
_
sin
0

2
sin cos
0
cos
cos
0
cos sin
0
1,
1
sin
_
res
s=
0
2
_
+
1
(s)
+
2
(s)
_
. (8.5)
To evaluate the residues
res
s=
0
2
_
+
+
1
(s)
+
+
2
(s)
_
and res
s=
0
+2
_
+
1
(s)
+
2
(s)
_
, (8.6)
the vector functions +++
+
(s) and +++
(s) should be continued analytically into the strips 5

Re s and Re s 5, respectively. This can be done by the relations
+++
+
(s) = [G
+
(s)]
1
+++
+
(s +4), 5 Re s .
+++
(s) = G
(s)+++
(s 4), Re s 5. (8.7)
Clearly, the residues in (8.5) can be expressed explicitly through the angles
0
and and the param-
eters
, e
1
, and e
2
. To nd them, there is no need to solve Problem 2.2 since the residues (2.5) are
prescribed.
Because of the conditions (2.17) the points s = +
j
( j = 1, 2, 3, 4) are removable points of
the functions S
1
(s +) and S
2
(s +). However, outside the strip Re(s +) , the points
s = +
j
+2m (m = 1, 2, . . .) are simple poles of the functions S
1
(s) and S
2
(s). The poles
s = +
j
+2 and +
j
2

H may give rise to the surface waves. The domain

H is a
curved strip. The right-hand boundary is described by the equation Re s = +gd(Ims) sgn(Ims),
where gd x is the Gudermann function gd x = cos
1
(1, cosh x). This curve is symmetric with
respect to the point s = . Its starting and ending points are s =
1
2
i and s =
3
2
+ i ,
respectively. The lines Re s =
1
2
and Re s =
3
2
are the asymptotes of the lower and upper parts
of the path. The left-hand boundary of the domain

H is symmetric to the the right-hand side with
respect to the point s = 0. Since (, ) and Re
j
(0, ) ( j = 1, 2, 3, 4) for Re
> 0,
the surface waves are given by
_
E
s
z
H
s
z
_
=
4
j =1
__
W
+
ej
W
+
hj
_
(, + g
+
j
) +
_
W
++
ej
W
++
hj
_
( + g
+
j
, )
_
e
i kz cos +i k sin cos(
+
j
)
+
4
j =1
__
W
ej
W
hj
_
(, + g
j
)
+
_
W
+
ej
W
+
hj
_
( + g
j
, )
_
e
i kz cos +i k sin cos(
j
)
, (8.8)
where
W
+
ej
=
1
2
I
+
j
[(sin
+
j
+
+
2
sin )+
+
1
(
+
j
2) cos
+
j
cos +
+
2
(
+
j
2)],
W
+
hj
=
1
2
I
+
j
__
sin
+
j
+
1
+
1
sin
_
+
+
2
(
+
j
2) +cos
+
j
cos +
+
1
(
+
j
2)
_
,
W
ej
=
1
2
j
[(sin
2
sin )+
1
(
j
2) cos
j
cos +
2
(
j
2)],
W
hj
=
1
2
j
__
sin
j

1
1
sin
_
+
2
(
j
2) +cos
j
cos +
1
(
j
2)
_
,
g
j
= Re
j
gd(Im
j
) sgn(Im
j
),
j
=
d
ds
I
s
_
1
,
2
_
s=
j
. (8.9)
The steepest descent method applied to (2.4) for k yields the diffracted eld
_
E
d
z
H
d
z
_
=
e
i k(z cos + sin )
k
_
D
e
()
D
h
()
_
, (8.10)
where the diffraction coefcients D
e
() and D
h
() are expressed through the spectral functions
S
1
(s) and S
2
(s) as follows:
D
() =
e
i ,4
2 sin e
1
[S
( ) S
( +)], = 1, 2, (8.11)
and D
1
= e
1
1
D
e
and D
2
= Z
0
e
1
1
D
h
. Since the functions S
1
(s) and S
2
(s) have a simple pole at
the point s =
0
, the diffraction coefcients are innite at the points = |
0
| and = +|
0
|.
To evaluate the diffraction coefcients by formulae (8.11), one needs to use formulae (2.9), (2.10),
and the analytical continuation (8.7). If s =
j
( j = 1, 2, 3, 4) is a real zero of the function
I
s
(1,
1
,
2
), and +
j
(, 0) ( +
j
(0, )), then the diffraction coefcients are
innite at the point = +
j
( = +
j
). Since T C
+
j
= C
j
( j = 1, . . . , 15), analysis
of formulae (2.9), (2.10), (2.12) and (8.11) shows that the diffraction coefcients are invariant with
respect to the transformation T dened by (2.8).
When the upper and lower sides of the screen have the same impedances
, and =
1
2
,
formula (8.11) becomes simpler:
D
() =
e
i ,4
2
2e
1
_
+
+
( )
sin
+
+
( )
sin +
+
+
+
( +)
sin +
+
+
( +)
sin
_
, = 1, 2.
(8.12)
Here +
( +) = G
( +)+
( 3) if 0 , and +
+
( ) = +
+
( +3),G
(
) if 0. Since C
+
(
0
) = C
(
0
) and
+
1
( ) = +
1
( +), +
1
( +) = +
1
( ), (8.13)
Fig. 6 Normal incidence. The real and imaginary parts of the diffraction coefcient D
1
() for
0
=
1
3
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
Fig. 7 Normal incidence. The real and imaginary parts of the diffraction coefcient D
2
() for
0
=
1
3
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
the diffraction coefcients are invariant with respect to the transformation and
0

0
:
D
(,
0
) = D
(,
0
). For the case of normal incidence ( =
1
2
) numerical calculations
for the diffraction coefcients D
1
() and D
2
() are illustrated in Figs 6 and 7 for the impedance
parameters (3.22),
0
=
1
3
, and e
= e
1
,e
2
= 1. Numerical values are the same if ,
0
, and
j
( j = 1, 2) are replaced by ,
0
, and
j
. As = , the diffraction coefcients D
( = 1, 2)
vanish. As approaches the angles
2
3
and
2
3
, that is, the angles
0
, the real and imaginary
parts of the diffraction coefcients grow to innity.
The skew incidence case is presented in Figs 8 to 10 for the same impedance parameters when
0
=
1
3
, e
= 1, and =
1
4
. The numerical values are symmetric with respect to the
transformation T . The surface waves poles are not close to the real axis, and the graphs are smooth
curves. However, for some sets of parameters these poles could be real or close to the real axis. For
example, for the impedance parameters (3.22) and =
1
3
, the zeros
j
of the of the functions
I
s
(1,
1
,
2
) are
+
1
= 27363 06508i,
+
2
= 01342, 10348i,
+
3
= 04053 +06508i,
+
4
= 3007 +10348i,
1
= 25114 +13157i,
2
= 04547 +00129i,
3
= 06302 13157i,
4
= 26869 00129i.
Because the imaginary part of the zeros
2
and
4
is small, the graphs of the real and imaginary
parts of the diffraction coefcients change sharply (Fig. 11) in small neighbourhoods of the points
Re
2
and Re
4
, namely as = 154
and = 26
. It is known (21) that (8.10) is valid

for large k and in those regions where the reected and surface waves are not present. For angles
close to these critical values, the total eld should be evaluated separately by using methods of the
uniform geometrical theory of diffraction (20, 21).
In Fig. 12, the real and imaginary parts of the diffraction coefcient D
1
are given for = 049.
It is seen that as
1
2
, the skew incidence coefcient D
1
approaches the values of the normal
incidence diffraction coefcient D
1
. The same tendency is observed for the second coefcient D
2
.
Fig. 8 Skew incidence. The real and imaginary parts of the diffraction coefcients D
1
() and D
2
() for
+
0
,
0
=
1
3
, =
1
4
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
1
() and D
2
() for
+
0

0
,
0
=
1
3
, =
1
4
,
+
1
= 1 i ,
+
2
= 0.1 i ,
1
= 2 i , and
2
= 1 +i .
1
() and D
2
() for

0
,
0
=
1
3
, =
1
4
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
1
() and D
2
() for
+
0
,
0
=
1
3
, =
1
3
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
Fig. 12 Skew incidence. The real and imaginary parts of the diffraction coefcient D
1
() for
0
=
1
3
, = 049,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
Fig. 13 Skew incidence. The real and imaginary parts of the diffraction coefcient D
1
() and D
2
() for
0
=
1
3
, =
5
6
,
+
1
= 1 i ,
+
2
= 01 i ,
1
= 2 i , and
2
= 1 +i
For
1
2
, the branch points
j
approach the images of the second-order zeros of the function
f

(s) = (cos
2
s + g
0
1)
2
( =
1
2
). For example, for = 0499, the branch points are
0
= 11414 20527i,
1
= 10124, 20090i,
2
= 00166 01168i,
3
= 00271 01167i,
4
= 02069 +03721i,
5
= 02000 +03970i,
6
= 18877 +81293i,
7
= 11927 +83925i.
Notice that the point = 0 is to the right of the segment
3
4
. As
1
2
, the accuracy of the
numerical solution to the Jacobi problem increases. For 0 or , the distance between
the branch points
0
and
7
grows. For =
1
10
and the impedance parameters (3.22), for example,
|
7

0
| = 39654. The values of the function p
1,2
() become large even for small values of .
Therefore, the solution of the problem requires higher accuracy for small or close to incidence
angles. The dependence of the coefcients D
1
and D
2
on the angle of incidence is shown in
Fig. 13.
9. Conclusion
Scattering of a plane wave at skew incidence from an anisotropic impedance half-plane for two
different matrix impedances Z
+
and Z
is a long-standing problem of diffraction theory (13). Only

some particular cases were studied in the literature. It turns out, using a new technique by Antipov
and Silvestrov (6), the problem in its general formulation with arbitrary complex entries of the
impedance matrices is exactly solvable. The key step of the method is to formulate the govern-
ing system of difference equations as a scalar RiemannHilbert problem on two unit segments of
a hyperelliptic surface of genus 3. Its solution has been found in terms of singular integrals with
the Weierstrass kernel on a Riemann surface. Initially, the solution had an essential singularity at
innity and possessed some free parameters. The singularity has been removed by xing these pa-
rameters which solve the associated Jacobi inversion problem for the surface. Regardless of which
analogue of the Cauchy kernel on the Riemann surface is chosen, in general, the solution of the Ja-
cobi problem cannot be bypassed. In this paper, a constructive numerical procedure for the inversion
problem for a surface of any nite genus has been proposed and described in detail. This technique
made it possible to recover the total electric and magnetic elds and compute the edge diffraction
coefcients. The solution has been found in closed form in terms of a nite number of quadratures.
It possesses three transcendents which are the roots of a certain cubic equation with known complex
coefcients and also 15 constants which solve a non-homogeneous system of 15 linear equations.
It has been proved that the solution is invariant with respect to the symmetric transformation of
the angles ,
0
and and the impedance parameters
j
( j = 1, 2). As the angle of incidence
approaches
1
2
, the diffraction coefcients approach the limiting values for the case of normal
incidence.
The technique employed here can be applied for more complicated boundary-value problems for-
mulated either as a system of the Maliuzhinets difference equations, or a system of the WienerHopf
functional equations. If, ultimately, the problem reduces to a scalar RiemannHilbert problem on a
hyperelliptic surface of high genera, then the procedure proposed should be modied accordingly.
The method of the paper has the potential to be extended to those systems of the Maliuzhinets and
the WienerHopf equations which are equivalent to a scalar RiemannHilbert problem on n-sheeted
(n 3) Riemann surfaces.
Acknowledgments
The support of the rst author by LSU Faculty Research grant 107-60-9102 and Louisiana Board
of Regents grant LEQSF(2005-07)-ENH-TR-09 and the second author by Russian Foundation for
Basic Research through contract 04-01-00160 is acknowledged. The authors are grateful to the
referees for their comments.
References
1. N. Wiener and E. Hopf,

Uber eine klasse singul arer intergleichungen, Sitz. Berliner Akad. Wiss.
(1931) 696706.
2. D. S. Jones, A symplifying technique in the solution of a class of diffraction problems, Quart.
J. Math. 3 (1952) 189196.
3. G. D. Maliuzhinets, Excitation, reection and emission of surface waves from a wedge with
given face impedances, Soviet Phys. Dokl. 3 (1958) 752755.
4. V. G. Daniele, The WienerHopf technique for impenetrable wedges having arbitrary aperture
angle, SIAM J. Appl. Math. 63 (2003) 14421460.
5. Y. A. Antipov and V. V. Silvestrov, Factorization on a Riemann surface in scattering theory,
Q. Jl Mech. Appl. Math. 55 (2002) 607654.
6. Y. A. Antipov and V. V. Silvestrov, Vector functional-difference equation in electromagnetic
scattering, IMA J. Appl. Math. 69 (2004) 2769.
7. Y. A. Antipov and V. V. Silvestrov, Second-order functional-difference equations. I: Method
of the RiemannHilbert problem on Riemann surfaces, Q. Jl Mech. Appl. Math. 57 (2004)
245265.
8. Y. A. Antipov and V. V. Silvestrov, Second-order functional-difference equations. II: Scattering
from a right-angled conductive wedge for E-polarization, ibid. 57 (2004) 267313.
9. T. B. A. Senior, Diffraction by an imperfectly conducting half-plane at oblique incidence, Appl.
Sci. Res. B 8 (1960) 3561.
10. E. L uneburg and A. H. Serbest, Diffraction of an obliquely incident plane wave by a two-face
impedance half plane: WienerHopf approach, Radio Sci. 35 (2000) 13611374.
11. J. M. L. Bernard, Diffraction at skew incidence by an anisotropic impedance wedge in elec-
tromagnetism theory: a new class of canonical cases, J. Physics A: Math. Gen. 31 (1998)
595613.
12. M. A. Lyalinov and N. Y. Zhu, Exact solution to diffraction problem by wedges with a class
of anisotropic impedance faces: Oblique incidence of a plane electromagnetic wave, IEEE
Trans. Antennas and Propagation 51 (2003) 12161220.
13. T. B. A. Senior, Some problems involving imperfect half planes, Electromagnetic Scattering
(ed. P.L.E.Uslenghi; Academic Press, New York 1978) 185219.
14. R. A. Hurd and E. L uneburg, Diffraction by an anisotropic impedance half plane, Canad. J.
Phys. 63 (1985) 11351140.
15. V. G. Daniele, On the solution of two coupled WienerHopf equations, SIAM J. Appl. Math. 44
(1984) 667680.
16. T. B. A. Senior and E. Topsakal, Diffraction by an anisotropic impedance half-plane-revised
solution, Progress in Electromagnetics Research 53 (2005) 119.
17. S. R. Legault and T. B. A. Senior, Solution of a second-order difference equation using the
bilinear relations of Riemann, J. Math. Phys. 43 (2002) 15981621.
18. G. Springer, Introduction to Riemann Surfaces (AddisonWesley, Reading, Mass. 1956).
19. E. I. Zverovich, Boundary value problems in the theory of analytic functions in H older classes
on Riemann surfaces, Russian Math. Surveys 26 (1971) 117192.
20. R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical theory of diffraction for an edge
in a perfectly conducting surface, Proc. IEEE 62 (1974) 14481461.
21. T. B. A. Senior and J. L. Volakis, Approximate Boundary Conditions in Electromagnetics (The
Institution of Electrical Engineers, London 1995).
22. V. G. Daniele, An Introduction to the WienerHopf technique for the Solution of the Electro-
magnetic Problems, Rapporto Interno ELT 2004-1 (Dipartimento di Elettronica, Politecnico
di Torino 2004).

Electromagnetic Scattering From An Anisotropic Impedance Half-Plane at Oblique Incidence: The Exact Solution

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Electromagnetic Scattering From An Anisotropic Impedance Half-Plane at Oblique Incidence: The Exact Solution

Загружено:

Авторское право:

Доступные форматы

ELECTROMAGNETIC SCATTERING FROM AN

ANISOTROPIC IMPEDANCE HALF-PLANE AT

(Department of Mathematics, Louisiana State University

(Department of Mathematics, Chuvash State University

Corresponding author antipov@math.lsu.edu

are the total electric and magnetic

when the problem is reducible to the problem of factorization of a 22 matrix.

. A method of the Riemann bilinear re-

could be complex and different. Antipov

( = + 0) (Fig. 1). The impedances may be

solve the following problem.

(s) analytic in the strips H

apart from simple poles at the

(s) satisfy the additional conditions (2.17).

(s) are meromorphic functions, by analytical

(s) meet the condition T

(s 4) (their elements are 2-periodic

() have the same eigenvalues:

() = A(). Clearly, the elements of the

(s) are single-valued. This case for the strip H

1 ( = 1, 2) is a xed branch. There are 16 roots in the strip H

(s). The branches are single-valued functions in the strips H

() the new vector-functions

(s) are free of poles

, respectively, and they have 16 branch points s

(s) have to be continuous through these

(s) on the contours C

(s) analytic in the w-planes dened by (3.13), apart from the

(s) are even:

(w), w C \ [1, 1], and satisfy the boundary conditions

(), t [1, 1], O

(t ) ( = 1, 2) satisfy the condition l

(t ) meet the following

are the increments of the argument of the functions l

(t ) as the point t traverses the

are evaluated numerically. The argument

(t ) ( = 1, 2) is xed by the condition argl

(s) and the branch of the function p

is denoted by the pair (, (1)

(, u) piecewise analytic on the surface R apart from the

(, u) are bounded at the innite points

( = 1, 2) of the surface. As 1, the functions F

(, u) are singular on the rst sheet:

( = 1, 2) of the Riemann surface that is

(t ), the single branch of the logarithmic function in (4.7) has the

(, u) given by (3.16), (4.2).

} ( = 1, 2, 3) can be expressed through the basis

} ( = 1, 2, 3) of abelian integrals (4.20) as follows:

(modulo the periods), = 1, 2, . . . , h,

( = 1, 2, . . . , h) are the Riemann constants.

solve the system of linear algebraic equations

are dened explicitly through the integers n

were dened in (19, p. 144) as follows:

(r) is the limiting value of the function

(q) on the cross-section a

. In what follows a correct formula for the constants k

will be derived. It will also be

(r)d log F(r), (5.8)

( = 1, 2, . . . , h). Because the winding

equals 2, by the argument principle,

( = 1, 2 . . . , h). The residue theorem yields

) (modulo the periods) (5.18)