Article

JOURNAL OF PARTIAL DIFFERENTIAL EQUATIONS
J. Part. Diff. Eq., Vol. 24, No. 2, pp. 114-139

doi: 10.4208/jpde.v24.n2.2
May 2011
Construction of Greens Functions for the Two-Dimensional
Static Klein-Gordon Equation
MELNIKOV Yu. A.
Department of Mathematical, Sciences Middle Tennessee State University,

Murfreesboro, Tennessee 37132, USA.
Received 29 December 2009; Accepted 10 March 2011
Abstract. In contrast to the cognate Laplace equation, for which a vast number of
Greens functions is available, the eld is not that developed for the static Klein-Gordon
equation. The latter represents, nonetheless, a natural area for application of some of
the methods that are proven productive for the Laplace equation. The perspective
looks especially attractive for the methods of images and eigenfunction expansion.
This study is based on our experience recently gained on the construction of Greens
functions for elliptic partial differential equations. An extensive list of boundary-value
problems formulated for the static Klein-Gordon equation is considered. Computer-
friendly representations of their Greens functions are obtained, most of which have
never been published before.
AMS Subject Classications: 35J08, 65N80
Chinese Library Classications: O175.2, O241
Key Words: Static Klein-Gordon equation; Greens function.
1 Introduction
To prevent a possible confusion as to the subject of the present study, note that our fo-
cus is not on the hyperbolic Klein-Gordon equation representing a classical mathematical
model in quantum eld theory and for which the Greens function formalism is well de-
veloped. This project deals instead with the elliptic two-dimensional static Klein-Gordon
equation (SKGE)
2
u(P)k
2
u(P) =0 (1.1)
with
2
representing the Laplace operator written in the coordinates of point P and the
parameter k is a real constant.
Corresponding author. Email address: ymelniko@mtsu.edu (Y. A. Melnikov)

http://www.global-sci.org/jpde/ 114
Construction of Greens Functions for Static Klein-Gordon Equation 115
To our best knowledge, [1] and [2] represent the only book-format publications cover-
ing, to a certain extend, the Greens function topic for the SKGE. Just a limited number of
boundary-value problems has been reviewed in those books, with the method of eigen-
function expansion being used. The present study aims at obtaining computer-friendly
forms of Greens functions for a signicant list of problems stated for the SKGE.
The Greens function G(P, Q) for the SKGE will be introduced, in this study, by setting
up the homogeneous boundary-value problem
M
i
u(P)
i
(P)
u(P)
n
i
+
i
(P)u(P) =0, P
i
, (1.2)
for the nonhomogeneous equation
2
u(P)k
2
u(P)=f (P), P, (1.3)
where: represents a simply connected region in two-dimensional Euclidean space with
=
m
i=1
i
denoting a piecewise smooth contour of ;
i
(P) and
i
(P) represent given
functions dened on in such a way that at least one of them is nonzero for every piece
i
of ; and n
i
represents the normal direction to
i
at the point P. The right-hand side
function f (P) in (1.3) is supposed to be integrable on .
Assume that the boundary-value problem in (1.2) and (1.3) is well-posed. This im-
plies, in fact, that it has a unique solution or, in other words, the corresponding homoge-
neous problem, with f (P)0, has only the trivial u(P) 0 solution. With this, we dene
the Greens function for the SKGE. That is, if, for any integrable on right-hand side
term f (P) in (1.3), the solution to the boundary-value problem in (1.2) and (1.3) is found
in the form
u(P) =

G(P, Q) f (Q)d(Q), (1.4)

then the kernel G(P, Q) of the above representation is said to be the Greens function for
the homogeneous problem corresponding to that of (1.2) and (1.3).
A standard terminology will apply, according to which P and Q in (1.4) are referred
to as the eld (observation) point and the source point, respectively.
For any location of the source point Q, the Greens function, as a function of the
coordinates of the observation point P, holds the following properties (being referred to
herein as the dening properties):
1. at any point P, except at P=Q, G(P, Q) satises the homogeneous equation in
(1.1), that is
(
2
k
2
)G(P, Q)=0, P=Q.
2. For PQ, G(P, Q) approaches innity like the modied cylindrical Bessel (or Mac-
donald) function K
0
(k|PQ|) of the second kind of order zero.
3. G(P, Q) satises the boundary conditions in (1.2), that is
M
i
G(P, Q)=0, P
i
, i =1,m.
116 Y. A. Melnikov / J. Partial Diff. Eq., 24 (2011), pp. 114-139
Due to the properties of the Macdonald function (see, for example, [35], the Greens
function G(P, Q) of the SKGE possesses the type of logarithmic singularity of Greens
function for the Laplace equation. To support this claim, explore the nature of the Mac-
donald function K
0
(x). For this we present its standard series expansion
K
0
(x)=
C+ln
x
2

j=0
x
2j
2
2j
(j!)
2
+
m=1
x
2m
2
2m
(m!)
2
n=1
1
n
, (1.5)
where C0.5772157 is referred to as the Eulers constant.
The representation in (1.5) appears, in fact, computer-friendly. This is so because both
of its innite series components uniformly converge for any value of x(0,); and their
convergence rate is fairly high.
From the denition just introduced, the Greens function of the homogeneous boun-
dary-value problem corresponding to (1.2) and (1.3) can be expressed as
G(P, Q)=
1
2
K
0
(k|PQ|)+R(P, Q), (1.6)
where R(P, Q), as a function of P, satises the homogeneous SKGE everywhere on ,
regardless of a mutual location of P and Q.
A special comment is appropriate as to the use of the terms regular component and
singular component of the Greens function. Upon explicitly expressing the logarithmic
term in (1.5) as
ln
x
2
j=0
x
2j
2
2j
(j!)
2
=ln
x
2
1
x
2
4
+
x
4
64
,
one realizes that the singularity in (1.6) is only associated with the j =0 term of the se-
ries. With this in mind, we are going to refer to the additive terms K
0
(k|PQ|)/2 and
R(P, Q) in (1.6) as the singular component and the regular component of the Greens func-
tion, respectively.
2 Method of images
We turn nowto the construction of Greens functions for the SKGE. The method of images
scheme [3, 69] will be applied to a variety of boundary-value problems.
The singular component [K
0
(k|PQ|)]/2 of G(P, Q) in (1.6) is interpreted, similarly
to the term[ln|PQ|]/2 in the case of Laplace equation, as the response at a eld point P
to a unit source placed at an arbitrary point Q. With this, the regular component R(P, Q)
of G(P, Q) is intended, in the method of images, to be expressed as a sum of a number of
unit sources and sinks placed at points Q
1
, Q
2
, , Q
m
outside the region . This makes
the regular component
R(P, Q) =
m
j=1
1
2
K
0
(k|PQ
j
|)
of G(P, Q) a function satisfying the homogeneous SKGE at any point P in (since all the
source points Q
j
are outside ).
The algorithm of the method of images appears instrumental in obtaining a number
of Greens functions for the SKGE.
Consider a trivial case of the Dirichlet problem stated for the SKGE on the upper
half-plane
={(x,y) | <x<, y>0}.
Inuence of the unit source at a point Q(,) represents the singular component
1
2
K
0
(x)
2
+(y)
2
of the Greens function. The above can be compensated with a single unit sink placed at
the point Q
(,) located at the lower half-plane and symmetric to Q(,) about the
boundary y=0. Having the inuence of this sink given as
1
2
K
0
(x)
2
+(y+)
2
,
the Greens function for the upper half-plane is ultimately found as
G(x,y;,) =
1
2
K
0
(k|z|) K
0
k|z|
, (2.1)
where the complex variable notation
z=x+iy and =+i
is used for the observation and the source point, respectively.
The method of images allows us to prove the following three statements.
Theorem 2.1. Greens function of the Dirichlet problem stated on the innite circular sector
=
(r, )
0<r <,0<
with the opening angle of /2, which represents actually the rst quadrant quater-plane, is given
as
G(r, ;,)=
1
2
2
n=1
K
0
r
2
2rcos(((n1)+))+
2
K
0
r
2
2rcos((n))+
2
. (2.2)
Proof. Since the distance between two points (r
1
,
1
) and (r
2
,
2
) is dened in polar coor-
dinates as
r
2
1
2r
1
r
2
cos(
1
2
)+r
2
2
,
the singular component of the Greens function G(r, ;,) reads as
1
2
K
0
r
2
2rcos()+
2
, (2.3)
which represents response of the eld at an observation point (r, )to the unit source
acting at (,).
In order to compensate the trace of the function in (2.3) (or, in other words, to support
the Dirichlet condition) on the boundary segment y=0, we place the unit sink at the point
(,2). The inuence of this sink is given by
1
2
K
0
r
2
2rcos((2))+
2
. (2.4)
Similarly, with the unit sink at (,), whose inuence is dened as
1
2
K
0
r
2
2rcos(())+
2
, (2.5)
we compensate the trace of (2.3) on the boundary segment x=0, while to compensate the
traces of the functions in (2.4) and (2.5) on x=0 and y=0, respectively, the unit source is
required at (,+), with the inuence
1
2
K
0
r
2
2rcos((+))+
2
. (2.6)
Hence, the Greens function of the Dirichlet problem stated on the quarter-plane rep-
resents the sum of the components shown in (2.3)(2.6), which converts to the form
shown in (2.2).
Theorem2.2. Greens function of a mixed boundary-value problem for the innite circular sector
with the opening angle of /2, with Dirichlet and Neumann boundary conditions imposed on the
boundary segments y=0 and x=0, respectively, is given as
G(r, ;,)=
1
2
2
n=1
(1)
n
K
0
r
2
2rcos((n))+
2
+(1)
n
K
0
r
2
2rcos(((n1)+))+
2
. (2.7)
Proof. To support the Dirichlet condition on the boundary segment y=0, we compensate
the trace of the function
1
2
K
0
r
2
2rcos()+
2
(2.8)
with the unit sink at (,2), the inuence of which is given by
1
2
K
0
r
2
2rcos((2))+
2
. (2.9)
The Neumann condition on the boundary segment x=0 will be supported if the unit
source is put at (,), with the inuence dened as
1
2
K
0
r
2
2rcos(())+
2
, (2.10)
while to support both the boundary conditions imposed on x=0 and y=0, the unit sink
is required at (,+), with the inuence
1
2
K
0
r
2
2rcos((+))+
2
. (2.11)
Summing up the components in (2.8) through (2.11), one obtains the Greens function
of the mixed problem under consideration in the form shown in (2.7).
Theorem 2.3. Greens function of the Dirichlet problem set up on the innite circular sector
=
(r, )
0<r <, 0<<
with the opening angle of /3 is given as

G(r, ;,)=
1
2
3
n=1
K
0
r
2
2rcos
(2(n1)
3
+)
+
2
K
0
r
2
2rcos
(2n
3
)
+
2
. (2.12)
Proof. Compensate the inuence of the singular component of the Greens function on
the boundary fragment =0 with a compensatory unit sink at (,2), while another
unit sink is required at (,2/3) to support the Dirichlet condition on =/3. To
compensate the trace of the latter sink on the boundary fragment =0, a unit source is
required at (,4/3+). The trace of the latter source is compensated on =/3 with
the unit sink at (,4/3), while the trace of this sink is compensated on =0 with the
unit source placed at (,2/3+).
Thus, the regular component R(r, ;,) of the Greens function can be formed as the
aggregate of the ve compensatory sources and sinks just located outside . This implies
that the Greens function G(r, ;,) itself is obtained by adding the singular component
to R(r, ;,), which ultimately yields the expression in (2.12).
Recall the Greens function exhibited in (2.7) for the mixed (Dirichlet-Neumann) prob-
lem stated on the innite circular sector with the opening angle of /2. We can readily
present, however, an example of the same type of mixed problem, for which the method
of images procedure fails. This claim will be justied by considering the Dirichlet-Neum-
ann problem stated for the innite circular sector with the opening angle of /3.
Clearly, the Dirichlet condition on =0 is supported with the unit sink placed at
(,2). To allow this sink to support the Neumann condition on =/3, the unit
sink is also required at (,2/3+). As to the Neumann condition on =/3, the unit
source at (,) ought to be supportedwith the unit source at (,2/3), which, in turn,
ought to be accompanied with the unit sink placed at (,4/3+). The latter sink has to
be accompanied with the unit sink at (,4/3) for the Neumann condition supported
on =/3. If we now take a look at the two sinks at (,2/3+) and (,4/3), they
do not, unfortunately, support the Dirichlet condition on the boundary fragment =0.
And this is what indicates, in fact, the methods failure for the mixed problem under
consideration.
Although the method of images might fail for some problems stated on an innite
circular sector, the range of problems, for which it appears efcient, is not limited to
those considered so far. Indeed, considering the case of Dirichlet problem stated on the
innite circular sector ={(r, ) | 0<r<, 0<</4} with the opening angle of /4.,
one arrives at its Greens function
G(r, ;,)=
1
2
4
n=1
K
0
r
2
2rcos
((n1)
2
+)
+
2
K
0
r
2
2rcos
(n
2
)
+
2
. (2.13)
In the case of Dirichlet-Neumann problem stated for the innite circular sector with
the opening angle of /4, with Dirichlet and Neumann conditions imposedon the bound-
ary segments =0 and =/4, respectively, we come up with the Greens function
expressed as
G(r, ;,) =
1
2
2
n=1
K
0
r
2
2rcos(((n1)+))+
2
+K
0
r
2
2rcos
((2n1)
2
)
+
2
K
0
r
2
2rcos
((2n1)
2
+)
+
2
K
0
r
2
2rcos((n))+
2
. (2.14)
Greens function to the Dirichlet problem stated on the innite circular sector with the
angle of /6, is obtained as an aggregate of the total number of twelve appropriately
placed unit sources and sinks. Omitting details, we present just its ultimate expression
G(r, ;,)=
1
2
6
n=1
K
0
r
2
2rcos
((n1)
3
+)
+
2
K
0
r
2
2rcos
(n
3
)
+
2
. (2.15)
Certain generalizations can be drawn from the analysis of the forms of Greens functions
derived thus far for the boundary-value problems stated on an innite circular sector.
Observing the expressions derived earlier for the Greens function of the Dirichlet
problem stated on the half-plane, on the circular sector with the opening angles of /4
and /2, one arrives at the following generalization
G(r, ;,)=
1
2
2
m
n=1
K
0
r
2
2rcos
(2(n1)

2
m
+)
+
2
K
0
r
2
2rcos
(2n

2
m
)
+
2
, (2.16)
which represents the Greens function of the Dirichlet problemstated on the set of innite
circular sectors with the opening angle of /2
m
, where m=0,1,2, .
Note that the case of m=0, which corresponds to the circular sector with the opening
angle of or, in other words, to the upper half-plane y>0, reads from (2.16) as
G(r, ;,)=
1
2
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
, (2.17)
representing the Greens function derived earlier in (2.1) and expressed here in polar
coordinates.
Another signicant generalization can be drawn out upon analyzing the expressions
in (2.12) and (2.15), from which one derives a Greens function of the Dirichlet problem
stated for the set of circular sectors with the angle of /(32
m
) where m=0,1,2, . It is
found in the form
G(r, ;,)=
1
2
32
m
n=1
K
0
r
2
2rcos
(2(n1)

32
m
+)
+
2
K
0
r
2
2rcos
(
2n
32
m
)
+
2
, (2.18)
with the case of m=0 staying for the sector of /3, while the case of m=1 stays for the
sector of /6.
Speaking of the success or failure of the method of images in constructing Greens
functions for innite circular sectors, remind that earlier we have shown that the method
fails for a mixed (Dirichlet-Neumann) problem stated on the sector with the opening
angle of /3. But it is worth noting that the method is not necessarily successful for
Dirichlet problems either. The following problem is presented to illustrate the point.
Consider the Dirichlet problemon the innite sector ={(r, ) | 0<r<, 0<<2/3}
and try to construct its Greens function.
Analogously to this same problem stated for the Laplace equation, the method of
images also fails for the SKGE. To observe the failure, let the unit source (which produces
the singular component of the Greens function) be located at (,). To compensate its
trace on the fragment =0 of the boundary of , place the compensatory sink at (,2
)/ . The trace of the latter on the boundary fragment =2/3 is compensated, in turn,
with a unit source at (,4/3+)/ , whose trace on =0 ought to be compensated with
a unit sink at (,2/3), which is, unfortunately, located inside . And this is what
justies the failure of the method.
We have just illustrated that it could potentially be the case that the method of images
fails for a Dirichlet problem stated on an innite circular sector which allows a cyclic
symmetry.
Note that all the Greens functions, constructedso far, are expressedin terms of a nite
sum of Macdonald functions. In examples that follow, we focus on a few other settings
for which the method of images is applicable but allows us to obtain Greens functions in
terms of an innite series of Macdonald functions.
Theorem2.4. Greens function of the Dirichlet problem set up on the innite strip ={(x,y) |
<x<, 0<y<b} reads as
G(x,y;,) =
1
2
n=
K
0
(x)
2
+(y+2nb)
2
K
0
(x)
2
+(y+2nb)
2
. (2.19)
Proof. Place a unit source S
+
0
at an arbitrary point (,) in . The response to S
+
0
at a
eld point (x,y) is the fundamental solution
G
+
0
(x,y;,) =
1
2
K
0
(x)
2
+(y)
2
of the SKGE.
It is evident that the function G
+
0
(x,y;,) conicts with the Dirichlet conditions on
the boundary fragments y=0 and y=b (it does not vanish on these lines). To compensate
the traces of G
+
0
(x,y;,) on y =0 and y =b, we place two unit sinks S
1,0
and S
1,b
at the
points (,) and (,2b), which represent the images of (,) about the lines y =0
and y=b, respectively. The responses to these sinks at (x,y) evidently are
G
1,0
(x,y;,) =
1
2
K
0
(x)
2
+(y+)
2
and
G
1,b
(x,y;,2b) =
1
2
K
0
(x)
2
+(y(2b))
2
.
The functions G
1,0
(x,y;,) and G
1,b
(x,y;,2b) do not vanish on the boundary
lines y=0 and y=b; and their traces can, in turn, be compensated with the unit sources
S
+
2,0
and S
+
2,b
which are located at (,2b+) and (,2b+). The responses to these at
(x,y) are given as
G
+
2,0
(x,y;,2b+) =
1
2
K
0
(x)
2
+(y(2b+))
2
,
and
G
+
2,b
(x,y;,2b+) =
1
2
K
0
(x)
2
+(y(2b+))
2
.
To properly compensate the traces of the functions G
+
2,0
(x,y;,2b+) and G
+
2,b
(x,y;
,2b+) on y=0 and y=b, we place unit sinks S
3,0
and S
3,b
at (,2b) and (,4b),
respectively.
Upon following the described procedure of properly placing compensatory unit sou-
rces that alternate with compensatory unit sinks, one expresses the Greens function G=
G(x,y;,) that we are looking for in the form
G=G
+
0
+
i=1
2i1,0
+G
2i1,b
i=1
G
+
2i,0
+G
+
2i,b
,
which, after a trivial algebra, transforms ultimately into the single innite series exhibited
in (2.19).
Since the Greens function in (2.19) is expressed in a series form, convergence of the
latter ought to be specically addressed to ensure its computability. Note rst that the
singularity in (2.19) is provided by the component which is a part of a single term (n=0),
and cannot, in any way, affect the series convergence.
A close analysis reveals convergence of the series in (2.19) at a high rate. This makes
it suitable for immediate computer implementations. This assertion is supported by the
observation
lim
n
(x)
2
+(y2nb)
2
2nb
=1
which implies that the arguments of the Macdonal functions in (2.19) are asymptotically
close to 2nb and converge to zero, when n increases, at the same rate as terms of the
sequence {K
0
(2nb)}. But the Macdonald function K
0
(x) is [10, 11] continuous, positive,
decreasing (at a very high rate) and bounded below by zero. For the strip of a unit width
(b=1), for example, the rst few terms of the sequence {K
0
(2nb)} are
K
0
(2).11389, K
0
(4).01116, K
0
(6).00124, K
0
(8).00015
approaching zero at a rate close to that of a geometric sequence whose ratio is of the order
of 10
1
and which therefore rapidly converges.
The brief analysis just completed brings a condence in a practicality of the represen-
tation in (2.19). One might, however, call in question the rigor of the analysis. Indeed, it
in no ways proves the convergence.
Theorem2.5. Greens function of a mixed problem stated on the innite strip ={(x,y) | <
x<, 0<y<b}, with Dirichlet condition imposed on y=0 and Neumann condition imposed on
y=b, is expressed as
G(x,y;,) =
1
2
n=
K
0
(x)
2
+(y+4nb)
2
K
0
(x)
2
+(y++4nb)
2
+K
0
(x)
2
+(y++2(2n+1)b)
2
K
0
(x)
2
+(y+2(2n+1)b)
2
. (2.20)
Proof. The trace of the fundamental solution G
+
0
(x,y;,) on the boundary line y=0, can
be compensated with a unit sink S
1,0
placed at the point (,), with the response at
(x,y) given by
G
1,0
(x,y;,) =
1
2
K
0
(x)
2
+(y+)
2
.
As to the Neumann condition imposed on y=b, it can be supported by placing a unit
source S
+
1,b
at the point (,2b). This yields
G
+
1,b
(x,y;,2b) =
1
2
K
0
(x)
2
+(y(2b))
2
.
To compensate the trace of the function G
+
1,b
(x,y;,2b) on the boundary line y=0
a unit sink S
2,0
is placed at (,2b+), with the response at (x,y)
G
2,0
(x,y;,2b+) =
1
2
K
0
(x)
2
+(y+(2b))
2
,
while the Neumann condition on y =b can be supported with a unit sink S
2,b
located at
(,2b+), whose response at (x,y) reads as
G
2,b
(x,y;,2b+) =
1
2
K
0
(x)
2
+(y(2b+))
2
.
The trace of the function G
2,b
(x,y;,2b+) on y=0 can, in turn, be compensated with
a unit source S
+
3,0
placed at (,2b), with the response
G
+
3,0
(x,y;,2b) =
1
2
K
0
(x)
2
+(y+(2b+))
2
,
while the Neumann condition on y=b can be supported with a unit sink S
3,b
at (,4b),
whose response at (x,y) is given as
G
3,b
(x,y;,4b) =
1
2
K
0
(x)
2
+(y(4b))
2
.
Proceeding with this pattern in compliance with the approach described in the proof
of Theorem 2.4, the Greens function that we are looking for is ultimately obtained in
the form of (2.20) which convergence at a high rate comparable to that of the series in
(2.19).
The pattern used in the proof of Theorems 2.4 and 2.5 appears also workable for the
Dirichlet problem set up on the semi-innite strip ={(x,y) | 0<x<,0<y<b}.
The fundamental solution of the SKGE, which represents the eld generated by the
unit source acting at an arbitrary point (,) in , can be compensated on the edges
y=0 and y= b with a set of unit sources and sinks placed at the regular set of points
(,), (,2b), (,2b+), (,2b+), (,2b), (,4b), located out of . In
other words, these sources and sinks allowus to satisfy the homogeneous Dirichlet bound-
ary condition imposed on the edges y=0 and y=b of .
To satisfy the boundary condition imposed on the edge x =0, the inuence of the
sources and sinks acting at the set of points just specied can, in turn, be compensated on
that boundary line with unit sources and sinks if we place them at another regular set of
points (,), (,), (,2b), (,2b+), (,2b+), (,2b), (,4b
), located also out of . It is evident that the latter sources and sinks are in agreement
with the boundary conditions imposed on y=0 and y=b.
Upon combining the inuence of all the compensatory sources and sinks placed by
the described scheme, one arrives at the series form
G(x, y; , ) =
1
2
n=
K
0
(x)
2
+(y+2nb)
2
K
0
(x)
2
+(y+2nb)
2
+K
0
(x+)
2
+(y+2nb)
2
K
0
(x+)
2
+(y+2nb)
2
(2.21)
for the Greens function of the Dirichlet problem stated on the semi-innite strip =
{(x,y) | 0<x<,0<y<b}. It is evident that the convergence rate of the above represen-
tation is comparable to that of the forms in (2.19) and (2.20).
As another example for the semi-innite strip ={(x,y) | 0<x <,0<y<b}, we
consider a mixed boundary-value problem. That is the one with Dirichlet conditions
imposed on the boundary fragments y=0 and y=b, and Neumann condition imposed on
x=0
Similarly to the Dirichlet problem considered above, the traces of the fundamental
solution of the SKGE (the eld generated by the unit source acting at an arbitrary point
(,) in ) on the edges y=0 and y=b are compensated with unit sources and sinks
placed at the set of exterior to points (,), (,2b), (,2b+), (,2b+), (,2b
), (,4b), .
As to the Neumann condition imposed on the boundary fragment x =0, it can be
supported by the aggregate inuence of the sources and sinks acting at the set of points
just specied. The latter can, similarly to the Dirichlet problem, be compensated with
unit sources and sinks placed at the set of exterior to points (,), (,), (,2b
), (,2b+), (,2b+), (,2b), (,4b), . The order of sources and sinks
is, however, different, in this case, of that suggested earlier for the Dirichlet problem.
Combining inuence of all the sources and sinks in compliance with our procedure,
one arrives at the rapidly convergent series representation
G(x,y;,) =
1
2
n=
K
0
(x)
2
+(y+2nb)
2
K
0
(x)
2
+(y+2nb)
2
+K
0
(x+)
2
+(y+2nb)
2
K
0
(x+)
2
+(y+2nb)
2
(2.22)
of the Greens function for the mixed problem under consideration.
3 Method of eigenfunction expansion
As it was illustrated in Section 2, the method of images appears successful for a number
of problemsettings with Dirichlet and Neumann conditions imposed. We turn nowto the
methodof eigenfunction expansion, which is also applicable to all the problems workable
with the method of images, but whose application range is notably wider.
We start with a trivial problems already reviewed in the previous section. An alter-
native to (2.19) representation will be obtained, in what follows, for the Greens function
G(x,y;,) to the homogeneous Dirichlet problem corresponding to
(
2
k
2
)u(x,y)=f (x,y), (x,y), (3.1)
u(x,0) =u(x,b) =0, (3.2)
stated on the innite strip ={(x,y)|<x<, 0<y<b} of width b. To ensure the well-
posedness to the setting in (3.1) and (3.2), we also require, for u(x,y) to remain bounded
as x approaches innity. In addition, assume that the right-hand side term f (x,y) in (3.1)
is integrable on
f (x,y)d(x,y)
<.
Express the solution u(x,y) to the problem in (3.1) and (3.2), and the right-hand side
function f (x,y) of (3.1) by the following Fourier sine-series
u(x,y) =
n=1
u
n
(x)siny, =
n
b
, (3.3)
and
f (x,y) =
n=1
f
n
(x)siny. (3.4)
This yields the set of self-adjoint boundary-value problems
d
2
u
n
(x)
dx
2
(
2
+k
2
)u
n
(x)=f
n
(x), n=1,2,3, , (3.5)
|u
n
()|<, |u
n
()|<, (3.6)
for the coefcients u
n
(x) of the series in (3.3).
Clearly, a fundamental set of solutions of the homogeneous equation corresponding
to (3.5) can be represented by the functions
exp(
2
+k
2
x) and exp(
2
+k
2
x)
and the Greens function g
n
(x,) of the homogeneous problem corresponding to (3.5)
and (34), which can be constructed by using either the method based on the dening
properties or by the method of variation of parameters, appears as
g
n
(x,)=
1
2
2
+k
2
exp(
2
+k
2
(x)), for x,
exp(
2
+k
2
(x)), for x.
(3.7)
This yields [2] the Greens function G(x,y;,) that we are looking for in the form
G(x,y;,)=
2
b
n=1
g
n
(x,)sinysin. (3.8)
It is evident that the non-uniform convergence of the series in (3.8) is preconditioned
by the logarithmic singularity of the Greens function. This signicantly limits the effec-
tiveness of direct computer implementations of the form in (3.8). To improve the con-
vergence, we rewrite the branch of the function g
n
(x,), which is valid for x, in the
form
g
n
(x,) =
exp(h(x))
2h

exp((x))
2
+
exp((x))
2
,
where the parameter h=
2
+k
2
is introduced for compactness. With this, the expression
in (3.8) transforms into
G(x,y;,)=
1
b
n=1
exp(h(x))
h

exp((x))
sinysin
+
1
b
n=1
exp((x))
sinysin, (3.9)
where the rst series uniformly converges and its numerical implementations bring no
problem. The second series in (3.9) does not converge uniformly but allows a complete
summation. The summation is conducted with the aid of the classical [11] summation
formula
n=1
p
n
n
cosn=
1
2
ln(12pcos+p
2
), (3.10)
that holds for p<1 and 0<2.
After an elementary algebra, the expression in (3.9) reads as
G(x,y;,) =
1
2
ln
E(z)
E(z)
1
b
n=1
H
n
(x,)sinysin, (3.11)
where z and stay for the observation and the source point, respectively; the real-valued
function E(w) of a complex variables w is introduced as
E(w) =
1exp(
b
w)
and the coefcient H

n
(x,) in (39) is given by
H
n
(x,) =
1
h
hexp((x))exp(h(x))
, for x.
Notice that the expression of H
n
(x,) for x can be obtained from that above by the
interchanging x with . Notice also that H
n
(x,) vanishes when k=0 and the representa-
tion in (39) reduces to the well-known closed form
G(x,y;,) =
1
2
ln
E(z)
E(z)
of the Greens function of the Dirichlet problem for Laplace equation (see, for example,
[3, 8, 9, 12]).
To review the computability of the representation in (39), we turn now to its trigono-
metric series. To explore its convergence, analyze the modulus of its remainder
|R
N
(x,y;,)| =
n=N+1
1
h
hexp((x))exp(h(x))
sinysin
.
Upon ignoring the trigonometric factors, one obtains
|R
N
(x,y;,)|
n=N+1
1
h
hexp((x))exp(h(x))
.
Clearly, the parameter h exceeds the value of and, consequently, the rst termin the
brackets always exceeds the second. This yields
|R
N
(x,y;,)|
n=N+1
1
h
hexp((x))exp(h(x))
.
Accounting for the fact that <h, the above estimate can be enforced as
|R
N
(x,y;,)| <
n=N+1
1
hexp((x))exp(h(x))
. (3.12)
This can be further developed by using the obvious relation
h=
2
+k
2
<+k,
that might be thought of as the triangle inequality for the values of h, , and k. With this,
one enforces the inequality in (3.12)
|R
N
(x,y;,)|<
n=N+1
1
(+k)exp((x))exp((+k)(x))
.
An elementary algebra yields
|R
N
(x,y;,)| <
n=N+1
1
[1exp(k(x))]exp((x))+kexp((x))
1exp(k(x))
n=N+1
1
exp((x))+k
n=N+1
1
2
exp((x))
1exp(k(x))
n=N+1
1
exp((x))+k
n=N+1
1
2
, (3.13)
or
|R
N
(x,y;,)| <
b
1exp(k(x))
n=1
1
n
exp
n
b
(x)
n=1
1
n
exp
n
b
(x)
+k
b
2
n=1
1
n
2
n=1
1
n
2
. (3.14)
The second series in (3.14) represents the p-series whose sum [11] is
n=1
1
n
2
=

2
6
,
while the sum of the rst series in (3.14) can be found if we take a convergent geometric
series and integrate it in the term-by-term manner. Indeed, consider the series
n=1
expnp=
expp
1expp
, p<0,
and integrate it, along with its sum. This yields
n=1
1
n
expnp=ln(1expp),
which translates, in our terms, into
n=1
1
n
expn
b
(x)
=ln
1exp(
b
(x))
.
This ultimately yields the following form
|R
N
(x,y;,)| <
b
1exp(k(x))
ln
1exp
b
(x)
n=1
1
n
exp
n
b
(x)
+k
b
2
6

N
n=1
1
n
2
(3.15)
for the estimate in (3.14).
Hence, the series in (3.11) converges uniformly and can be accurately computed by
a direct truncation, and the estimate in (3.15) helps appropriately nd a value of the
truncating parameter N required for attaining a desired accuracy.
For the next example on the use of the method of eigenfunction expansion, we set up
the mixed boundary-value problem
u(x,0) =
u(x,b)
y
=0 (3.16)
posedfor the nonhomogeneous SKGE on the innite strip ={(x,y)| <x<, 0<y<b}.
In this case, the solution u(x,y) and the right-hand side function f (x,y) of the problem
in (3.1) and (3.16) ought to be expressed by the following Fourier sine-series
u(x,y) =
n=1
u
n
(x)siny, =
(2n1)
2b
, (3.17)
and
f (x,y) =
n=1
f
n
(x)siny.
Proceeding then in compliance with the algorithm used in the previous example, we
arrive, for the coefcient u
n
(x) of the series in (3.17), at the boundary-value problem
in (3.5) and (34) whose Greens function was already presented in (3.7). This leads to the
Greens function for the setting in (3.1) and (3.16) obtained also as the series in (3.8) which
ultimately converts, in the current setting, into a quite compact and computer-friendly
form as
G(x,y;,) =
1
2
ln
E
1
(z)E
2
(z)
E
2
(z)E
1
(z)
1
b
n=1
exp((x))

exp(h(x))
h
sinysin, (3.18)
where the real-valued functions E
1
(w) and E
2
(w) of a complex variable w are dened as
E
1
(w)=
exp
w
2b
+1
, E
2
(w)=
exp
w
2b
.
Notice that the coefcient
exp((x))

exp(h(x))
h
of the series in (3.18) is valid for x, whereas its expression for x can be obtained
from the above by interchanging x with .
The series in (3.18) is uniformly convergent in . Its analysis can be accomplished in
exactly same fashion as for the Dirichlet problem considered earlier.
Omitting details, we present an ultimate expression for the Greens function of the
Dirichlet problem
u(0,y) =u(x,0) =u(x,b) =0
stated for the SKGE on the semi-innite strip ={(x,y) | 0<x<, 0<y<b}. It is found
as
G(x,y;,) =
1
2
ln
E(z)E(z+)
E(z)E(z+)
2
b
n=1
sinhx
exp
sinhhx
hexph
sinysin. (3.19)
The coefcient of the series in (3.19) is valid for x whereas its expression for x
can be obtained from the above by interchanging of x with . The series converges
uniformly at a high rate, and its remainder can readily be estimated by using the scheme
described earlier. As another example for the semi-innite strip ={(x,y) | 0<x<, 0<
y<b}, we set up a mixed boundary-value problem
u(x,0) =
u(x,b)
y
=
u(0,y)
x
u(0,y) =0, 0, (3.20)
whose remarkable feature is that three different kinds of boundary conditions are im-
posed on different segments of the contour of .
Proceeding in compliance with our approach, the Greens function for the formulation
in (3.20) is ultimately obtained in the form
G(x,y;,) =
1
2
ln
E
1
(z)E
2
(z+)E
1
(z+)E
2
(z)
E
1
(z+)E
2
(z)E
1
(z)E
2
(z+)
2
b
n=1
coshx
exp
coshhx
hexph
sinysin
2
b
n=1
exp(h(x+))
h(h+)
sinysint, x, (3.21)
where
h=
2
+k
2
, =(2n1)

2b
. (3.22)
The coefcient of the rst series in (3.21) is valid for x, while its expression for x
can be obtained by interchanging x with . Clearly, the coefcient of the second series in
(3.21) remains invariant to the interchange.
The series in (3.21) are uniformly convergent, and one can readily estimate the re-
mainder of the rst of them by using the approach that has already been implemented to
the series of (3.11). To estimate the remainder R
N
(x,y;,) of the second series in (3.21),
we proceed as follows
|R
N
(x,y;,)| =
n=N+1
exp(h(x+))
h(h+)
sinysin
n=N+1
exp(h(x+))
h(h+)
n=N+1
exp(h(x+))
h(h+)
<
n=N+1
exp((x+))
(+)
. (3.23)
Recalling from (3.22) the expression for in terms of n, we rewrite the estimate in (3.23)
in the explicit form
|R
N
(x,y;,)|<
4b
2
n=N+1
exp
(2n1)
2b
(x+)
(2n1)[(2n1)+
0
]
,
where
0
=2b/. The above can be rewritten as
|R
N
(x,y;,)|<
4b
2
n=N+1
exp
(x+)
2b
2n1
(2n1)[(2n1)+
0
]
=
4b
2
2
p
n=N+1
p
2n
(2n1)2n
, (3.24)
where we introduce for compactness
p=exp
2b
(x+)
.
To further proceed with the estimate obtained in (3.24), we will derive the summation
formula for the series
n=1
p
2n
(2n1)2n
,
whose N-th remainder appears in (3.24). In doing so, let us take advantage of the stan-
dard [11] relation
n=1
p
2n1
2n1
=
1
2
ln
1+p
1p
(3.25)
valid for p
2
<1. Since the above series uniformly converges for p (0,1), one is allowed
to integrate it in the term-by-term manner. This yields
p
0
n=1
t
2n1
2n1
dt =
n=1
1
2n1
p
0
t
2n1
dt =
n=1
p
2n
(2n1)2n
.
On the other hand, upon integrating the right-hand side of the relation in (3.25), one
obtains
p
0
1
2
ln
1+t
1t
dt =
1
2
[(1+p)ln(1+p)+(1p)ln(1p)].
Hence, we have nally derived the summation formula
n=1
p
2n
(2n1)2n
=
1
2
[(1+p)ln(1+p)+(1p)ln(1p)], (3.26)
which is important in the fulllment of the estimation which is under way.
We now return to the inequality in (3.24) and rewrite it as
|R
N
(x,y;,)| <
4b
2
2
p
n=N+1
p
2n
(2n1)2n
=
4b
2
2
p
n=1
p
2n
(2n1)2n
n=1
p
2n
(2n1)2n
.
This can be completed in view of the relation that has just been derived in (3.26), i.e.,
|R
N
(x,y;,)| <
4b
2
2
p
1
2
[(1+p)ln(1+p)+(1p)ln(1p)]
N
n=1
p
2n
(2n1)2n
, (3.27)
where the parameter p was introduced right after the relation in (3.24).
Thus, from the estimate just obtained, it follows that the representation of the Greens
function presentedin (3.21) can be accurately computed by a direct truncation of its series
components.
We focus now on the Dirichlet problem set up on the rectangle ={(x,y) | 0<x<
a, 0<y<b}. Tracing out the technique productively used before, one obtains the series
representation
G(x,y;,) =
2
b
n=1
g
n
(x,)sinysin, =
n
b
(3.28)
of its Greens function. The coefcient g
n
(x,) appears, for x, in the form
g
n
(x,) =
1
4hsinh(ha)
exp(h(xa))exp(h(x+a))
+exp(h(xa))exp(h(x+a))
, (3.29)
where the parameter h is again introduced as h=
2
+k
2
.
By a trivial algebra, the above expression for g
n
(x,) reduces to
g
n
(x,) =
sinhhxsinhh
hexphasinhha
+
1
2h
exp(h(x))exp(h(x+))
.
Upon adding and subtracting the function
1
2
exp((x+))exp((x))
to the bracket term of g

n
(x,), conducting a partial summation of the series in (3.28), and
performing then a trivial algebra, one obtains the following representation
G(x,y;s,) =
1
2
ln
E(z)E(z+)
E(z)E(z+)
2
b
n=1
sinhx
exp
sinhhx
hexph
sinysin
2
b
n=1
sinhhxsinhh
hexphasinhha
sinysin, x (3.30)
of the Greens function for the Dirichlet problem set up for the SKGE on the rectangle .
As to the representation of G(x,y;,) valid for x, note that it follows from (3.30) if
we interchange the variables x with only in its rst of the two series, because the second
of the series is invariant to the interchange.
Notice that the representation of the Greens function, which was earlier derived in
(3.19), immediately follows from that of (3.30) when a is taken to innity. This claim is
supported by the following features: (i) the logarithmic as well as the rst series term in
(3.30) do not depend on a, and (ii) the limit of the second series term, as a goes to innity,
is equal to zero
lim
a
n=1
sinhhxsinhh
hexphasinhha
sinysin =0.
Speaking of computational implementations of the representation in (3.30), note that,
as we already mentioned, the rst series in it is computer-friendly and can be accurately
valuated by an appropriate truncation. The second series in (3.30) is not, however, uni-
formly convergent. It possesses the logarithmic singularity which shows up if both the
eld and the source point go to the edge x=a.
For another example on the success of the eigenfunction expansion method in a case
when the method of images fails, consider the boundary-value problem
u(r,0) =u(r,) =0 (3.31)
set up for the nonhomogeneous SKGE
1
r
r
u(r, )
r
+
1
r
2
2
u(r, )
2
k
2
u(r, ) = f (r, ) (3.32)
on the innite wedge (r, )={(r, )| 0<r<,0<<}, with 0<<2.
Theorem 3.1. Greens function to the homogeneous problem corresponding to (3.31) and (3.32)
can be expressed as
G(r, ;,) =
2
n=1
g
n
(r,)sinsin, (3.33)
where =n/ and
g
n
(r,)=
(kr)K
(k), for r ,
I
(k)K
(kr), for r,
(3.34)
with I
(kr) and K
(kr) representing the modied Bessel functions (see [3, 6, 7]) of order of the
rst and the second kind, respectively.
Proof. To make the statement in (3.31) and (3.32) well posed, we require that its solution
u(r, ) remains bounded whenr approaches zero and innity. Assume also that the right-
hand side term f (r, ) in (3.32) is integrable on .
To obtain the solution u(r, ) of the problem in (3.31) and (3.32) in the body integral
form
u(r, ) =

0
G(r, ;,) f (,)dd
delivering the Greens function G(r, ;,) that we are looking for, we express u(r, ) and
f (r, ) by the following Fourier sine-series
u(r, ) =
n=1
u
n
(r)sin, =
n
, (3.35)
and
f (r, ) =
n=1
f
n
(r)sin. (3.36)
This yields the set of boundary-value problems
d
2
u
n
(r)
dr
2
+
1
r
du
n
(r)
dr

k
2
+
2
r
2
u
n
(r)=f
n
(r), n=1,2,3, , (3.37)
lim
r0
|u
n
(r)| <, lim
r
|u
n
(r)|<, (3.38)
in the coefcients u
n
(r) of the series in (3.35).
The fundamental set of solutions of the homogeneous equation in (3.37) can be rep-
resented by the modied Bessel functions I
(kr) and K
(kr). By using the latter set, the

Greens function g
n
(r,) to the homogeneous problem corresponding to (3.37) and (3.38)
appears in the form of (3.34). This yields then the Greens function G(r, ;,) as shown
in (3.33) .
A particular case of the expansion in (3.33) is of a specic interest. Assuming therein
=, one comes up with the series expansion
G(r, ;,) =
2
n=1
I
n
(kr)K
n
(k)sinnsinn, for r , (3.39)
of the Greens function for the Dirichlet problem set up for the SKGE on the half-plane
={(r, ) |0<r<,0<<}. In compliance with (3.34), the variables r and in (3.39)
ought to be interchanged for r.
It is worth while to remind that another, alternative to (3.39), form of the Greens
function for the Dirichlet problem stated on the half-plane has earlier been obtained by
the method of images. Indeed, that is the one shown in (2.1) as
G(r, ; , ) =
1
2
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
, (3.40)
where the moduli of z and z are expressed, for convenience, in polar coordinates.
Upon equating the two equivalent representations of (3.39) and (3.40), one arrives at
the following relation
4
n=1
I
n
(kr)K
n
(k)sinnsinn
=
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
, for r , (3.41)
representing a nontrivial, in fact, multi-variable identity for the modied Bessel func-
tions.
It is evident that some other particular cases of the expansion in (3.39) could also
generate interesting identities for the modied Bessel functions. Assuming, for example,
=/2 in (3.33), one transforms it into
G(r, ;,)=
4
n=1
I
2n
(kr)K
2n
(k)sin2nsin2n, for r , (3.42)
representing the Greens function for the Dirichlet problem stated on the quarter-plane
={(r, )|0<r <,0< </2}. Upon equating the form in (3.42) with its equivalent
obtained earlier in Section 2 by the method of images and presented in (2.2), we obtain
another nontrivial identity
8
n=1
I
2n
(kr)K
2n
(k)sin2nsin2n
=
2
n=1
K
0
r
2
2rcos(((n1)+))+
2
K
0
r
2
2rcos((n))+
2
, for r , (3.43)
where for r the variables r and ought to be interchanged.
Consider the Dirichlet problem
u(R, ) =u(r,0) =u(r,) =0
set up on the half-disk ={(r, ) | 0<r<R, 0<<} of radius R.
We present just an ultimate expression of the Greens function to this problem. It is
obtained by the method of eigenfunction expansion in the form
G(r, ;,) =
1
2
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
n=1
I
n
(kr)I
n
(k)K
n
(kR)
I
n
(kR)
sinnsinn. (3.44)
Clearly, the Greens function of the Dirichlet problem for the half-plane, presented in
(3.40), follows from the above as R approaches innity. Indeed, the series term in (3.44)
vanishes as R, because of the evident [11] observation
lim
R
K
n
(kR)
I
n
(kR)
=0,
which directly follows from the asymptotic (as R) behavior
lim
R
K
n
(kR) =0 and lim
R
I
n
(kR) =
of the modied Bessel functions.
To conclude this presentation, we consider the following mixed boundary-value prob-
lem
u(R, )
r
+u(R, ) =u(r,0) =u(r,) =0, 0, (3.45)
set up on the half-disk ={(r, ) | 0<r<R, 0<<}.
Greens function for this problem is found by means of our approach as
G(r, ;,) =
1
2
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
n=1
I
n
(kr)I
n
(k)[K
n
(kR)+K
n
(kR)]
I
n
(kR)+I
n
(kR)
sinnsinn. (3.46)
Greens functions for two other boundary-value problems can be obtained from the
above representation. First, the formin (3.46) reduces to the expression of (3.44) (Dirichlet
problem for the half-disk), when the parameter in the boundary condition of (3.45)
approaches innity. Second, as goes to zero, the representation in (3.46) reduces to the
Greens function
G(r, ;,) =
1
2
K
0
r
2
2rcos()+
2
K
0
r
2
2rcos(+)+
2
n=1
I
n
(kr)I
n
(k)K
n
(kR)
I
n
(kR)
sinnsinn (3.47)
for the mixed boundary-value problem
u(R, )
r
=u(r,0) =u(r,) =0
stated on the half-disk.
4 Concluding remarks
Summarizing the results of the present study, point out that a variety of boundary-value
problems for the static Klein-Gordon equations, for which Greens functions can poten-
tially be constructed by using the techniques developed herein, is not limited to the cases
just considered. Many other problems can also be successfully tackled. The presented
work especially aimed at those researchers who might be interested in a numerical im-
plementation of Greens functions, in which case the practical computability is a critical
issue.
References
[1] Melnikov Yu. A., Greens Functions in Applied Mechanics. Computational Mechanics Pub-
lications, Southampton-Boston, 1995.
[2] Melnikov Yu. A., Inuence Function and Matrices, Marcel Dekker. New York-Basel, 1998.
[3] Smirnov V. I., A Course of Higher Mathematics. Pergamon Press, Oxford-New York, 1964.
[4] Watson G. N., A Treatise on the Theory of Bessel Functions. Cambridge University Press,
Cambridge, 1966.
[5] Tranter C. J., Bessel Functions with Some Physical Applications. Hard Publishing Company,
New York, 1969.
[6] Morse P. M. and Feshbach H., Methods of Theoretical Physics. McGraw Hill, New York-
Toronto-London, 1953.
[7] Courant R. and Hilbert D., Methods of Mathematical Physics. Interscience, New York, 1953.
[8] Stakgold I., Greens Functions and Boundary Value Problems. John Wiley, New York, 1980.
[9] Duffy D., Greens Functions with Applications. CRC Press, Boca Raton, 2001.
[10] Abramovitz M. and Stegun I., Handbook of Mathematical Functions. National Bureau of
Standards, Washington D.C., 1972.
[11] Gradstein I. S. and Ryzhik I. M., Tables of Integrals, Series and Products. Academic Press,
New York, 1980.
[12] Economou E. N., Greens Functions in Quantum Physics. Springer-Verlag, Berlin, 1983.

Article

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Article

Enviado por

Direitos autorais:

Formatos disponíveis

JOURNAL OF PARTIAL DIFFERENTIAL EQUATIONS

J. Part. Diff. Eq., Vol. 24, No. 2, pp. 114-139

Department of Mathematical, Sciences Middle Tennessee State University,

Corresponding author. Email address: ymelniko@mtsu.edu (Y. A. Melnikov)

G(P, Q) f (Q)d(Q), (1.4)

0<r <, 0<<

with the opening angle of /3 is given as

and the coefcient H

to the bracket term of g

(kr). By using the latter set, the

Você também pode gostar