Beruflich Dokumente
Kultur Dokumente
Yuri A. Melnikov
Volodymyr N. Borodin
Green's
Functions
Potential Fields on Surfaces
Developments in Mathematics
Volume 48
Series editors
Krishnaswami Alladi, Gainesville, USA
Hershel M. Farkas, Jerusalem, Israel
More information about this series at http://www.springer.com/series/5834
Yuri A. Melnikov Volodymyr N. Borodin
•
Green’s Functions
Potential Fields on Surfaces
123
Yuri A. Melnikov Volodymyr N. Borodin
Department of Mathematical Sciences Department of Mechanical Engineering
Middle Tennessee State University Tennessee Technological University
Murfreesboro, TN Cookeville, TN
USA USA
vii
viii Preface
ix
x Contents
xi
xii Introduction
Fig. 1 Geometry of a
spherical surface
that is used to model potential phenomena on the spherical surface of radius a, can
directly be obtained from the standard three-dimensional Laplacian
1 @ 2 @ 1 @ @ 1 @2
r2 ðr; #; uÞ r þ sin # þ 2 2 ð3Þ
r @r
2 dr r sin # @#
2 @# r sin # @u2
written in spherical coordinates. At the first step of the transformation of (3) onto
(2), we drop the first of the three additive components in (3). This implies in fact
that the derivatives with respect to the variable r are just neglected. At the second
step, we let the variable r, in the remainder of (3), be fixed at a.
A circular toroidal surface represents another quite frequently encountered in this
study element. In Chaps. 3–5, we turn repeatedly to thin toroidal shells. To derive
the two-dimensional operator analogous to (2) in a toroidal surface case, we let R
represent the distance between the center of the meridian cross-section of radius a
and the axis of revolution (see Fig. 2).
To derive a two-dimensional differential operator that simulates potential phe-
nomena on the toroidal surface, a certain coordinate system should be accepted.
Locating a point on the surface, a frame of reference can be introduced in the way
similar to that used earlier for a spherical surface. That is, upon accepting the
geographical coordinates # (the latitude) and u (the longitude), we come up with
the following parameterization
@2 @2 @2
r2 ðx; y; zÞ þ 2þ 2
@x 2 @y @z
It is worth noting that the above operator might reduce to that of (2). Tracking
out the reduction process, one might assume that the parameter R in Dð#Þ of (5) is
set to equal zero. If so, then the circular toroidal surface of radii R and a degenerates
onto the spherical surface of radius a, in which case the form in (6) reduces directly
to that of (2).
Each of the operators shown in (2) or (6) comes into the play in this volume as
soon as potential fields are analyzed as induced in fragments of either spherical or
toroidal shells (see Chaps. 2 and 3). Both of these operators are also dealt with in
Chaps. 4 and 5, where we are involved with fields induced in joint thin-wall
constructions for which either a spherical or a toroidal shell fragment is a com-
ponent of an assembly.
Note that potential fields generated in thin-wall elements different of spherical or
toroidal are also of interest in our study. Some flat and cylindrical fragments, in
particular, represent elements of assemblies of shells studied in Chaps. 4 and 5. The
forms of the Laplace operator for such trivial cases can be found in every standard
text, and we believe that there is no need to specially recollect them at this moment.
Speaking about the practical solution of boundary value problems that simulate
potential fields induced in thin shell structures, it is important to pay attention, in
Introduction xv
particular, to the configuration of the region hosting the problem. The region’s
shape is always a decisive factor in choosing a solution method for any partial
differential equation. In this regard, we would like to advise the reader to consider
the Green’s function approach to a problem as one of prospective options in
resolving the region’s shape factor. In Chap. 5, one can find a number of illus-
trations of the flexibility of this approach.
The material presented in this volume requires no special preparation from the
reader. A standard knowledge, as assumed by the undergraduate applied mathe-
matics curriculum, seems to be sufficiently enough. Some superficial preparedness
in undergraduate physics can nevertheless also help.
The general organization of the material in this presentation is as follows.
Chapter 1 is intended as a preparatory segment helping the reader to be fluent in
dealing with the remainder of the whole volume. It introduces the complementary
concepts of the Green’s function formalism applied to various types of boundary
value problems for ordinary high-order linear differential equations and their sys-
tems. The emphasis is on issues important for a special approach to the practical
construction of PDE Green’s functions, which we advocate further in the volume.
In addition to the standard type of two-point-posed boundary value problems,
the Green's function topic for which is traditionally covered in relevant texts, we
discuss also in Chap. 1 an extension of the Green’s function formalism to
multi-point-posed problems set up for specific sets of ordinary differential equa-
tions. The notion of matrix of Green’s type, which naturally arises in the extension,
was originally introduced in [18] and formalized then in [31]. Perhaps this is the
only subject touched upon in the entire manual that does not belong to the under-
graduate mathematics curriculum. That is why our intention was to make the dis-
cussion as transparent and self-explained as it could be.
Chapter 2 covers a lion’s share of boundary value problems that simulate
potential fields in regions of various configuration representing segments of a
spherical surface. Regions that represent a spherical belt, a triangle, a quadrilateral,
and a sector are considered. Note that at this stage of our presentation, all the
regions are regular in shape and bounded, in other words, with segments of coor-
dinate lines (either meridians or parallels). All the three classical types (Dirichlet,
Neumann, and Robin) of boundary conditions appear in well-posed problem set-
tings. A special version of the eigenfunction expansion method [16, 21, 38, 42],
which we proposed back in [14, 15], created a workable background for our
approach to the construction of required Green’s functions.
With the methodological background worked out in necessary details in Chap. 2
for spherical surfaces, we appeal in Chap. 3 to the development of the Green’s
function approach to boundary value problems that simulate potential phenomena
occurring in a variety of regions belonging to a circular toroidal shell. A specific
form of the operator in (6) raises some extra complications compared to the case
of the spherical surface, and the reader is instructed on a strategy to overcome the
complications. Similar to the preceding chapter, regions of regular configurations
are encountered, with all the three classical types of boundary conditions involved
in well-posed problem settings.
xvi Introduction
Green’s functions for a specific type of elliptic second-order partial differential equa-
tions with variable coefficients is what we will be involved with in this volume. In
this regard, one might assume that the present chapter stays aside of the principal
track of the whole work, since it turns to ordinary differential equations and their
systems. But such an assumption is pointless because our version [8, 14, 15, 31–36]
of the classical eigenfunction expansion method [16, 21, 38, 42], offered for the
construction of Green’s functions to PDEs, requires this turn.
To construct Green’s functions for ODE, most of the traditional texts in the field
(see, for example, [16, 21, 37]) offer a standard procedure that flows down from the
proof of the existence and uniqueness theorem for the Green’s function. The proof
is constructive in nature allowing the reader to actually obtain the required Green’s
function by simply following in footsteps of the procedure. In addition to that, we
invite the reader to take a look at another construction procedure [16, 21] in which
the Lagrange’s method of variation of parameters is a key. It is based on the fact
that solution of a nonhomogeneous equation can be written in terms of the Green’s
function for the corresponding homogeneous equation.
Along with classical two-point-posed boundary value problems conventionally
considered in all standard texts in the field, some multiple-point-posed problems
will be discussed later in this chapter for specific systems of ODE. Earlier (see, for
example, [31, 33, 35]), we proposed to treat such problems with a special extension
of the Green’s function formalism. This gives birth to the notion of matrix of Green’s
type.
The Green’s function formalism is recalled in this section as applied to a linear two-
point boundary value problem posed for an ordinary nth order differential equation.
We provide a detailed description of the classical method for the construction of
Green’s functions, which is based on their defining properties. With a number of
© Springer International Publishing AG 2017 1
Y.A. Melnikov and V.N. Borodin, Green’s Functions,
Developments in Mathematics 48, DOI 10.1007/978-3-319-57243-7_1
2 1 Green’s Functions for ODE
dn y d n−1y
L[y(x)] ≡ p0 (x) + p 1 (x) + · · · + pn (x)y = 0 (1.1)
dxn d x n−1
be subject to the boundary conditions
n−1
k k
i d y(a) i d y(b)
Mi [y(a), y(b)] ≡ αk + βk = 0, (i = 1, n) (1.2)
k=0
dxk dxk
Let the above problem setting be well-posed on the interval (a, b) implying that it
has only the trivial solution. The equation coefficients pi (x), (i = 0, n) are supposed
to be continuous functions on (a, b), where the leading coefficient p0 (x) is not
supposed to equal zero at a single point in (a, b).
The abbreviated appearance of the boundary conditions in (1.2) needs a clarifi-
cation. The forms Mi [y(a), y(b)], (i = 1, n), with constant coefficients αik and βki ,
are supposed to be linearly independent. This holds the total number of boundary
conditions in (1.2) at n, ensuring the fact that the trivial solution y(x) ≡ 0 represents
the only solution for the problem shown in (1.1) and (1.2).
Another important feature of the relations in (1.2) is that they are written in a two-
point form. This implies that some of them may involve both the end points a and b
of the interval. If, however, a certain of these conditions is written in a single-point
form or, in other words, is imposed, say, at x = a only, then all the coefficients βki in
(1.2) are zero, while at least one of the coefficients αik is not. Same comment applies
to the case in which some of the boundary conditions are imposed at the right-end
point x = b of [a, b].
We recall now the classical [16, 21] definition of the Green’s function for the
homogeneous boundary value problem stated in (1.1) and (1.2). The theorem of
existence and uniqueness will later be formulated in the way that its proof is instruc-
tive providing in fact a constructing method for the Green’s function.
Definition The function g(x, s) is said to be the Green’s function for the boundary
value problem in (1.1) and (1.2), if as a function of its first variable x, it satisfies the
following defining properties, for any s ∈ (a, b):
1. On both of the intervals [a, s) and (s, b], g(x, s) is a continuous function having
continuous derivatives of up to the nth order included, and satisfies the governing
equation in (1.1) on (a, s) and (s, b), i. e.:
∂ k g(x, s) ∂ k g(x, s)
lim+ − lim = 0, (k = 0, n − 2);
x→s ∂x k x→s − ∂x k
The following theorem specifies conditions of existence and uniqueness for the
Green’s function.
Theorem 1.1 (existence and uniqueness) If the homogeneous boundary value prob-
lem in (1.1) and (1.2) has only the trivial solution, then there exists its unique Green’s
function g(x, s).
The reader is suggested to carefully read the proof of this theorem because of its
constructive nature. By saying this, we mean that the Theorem delivers a straight-
forward algorithm for the actual construction of Green’s functions. Throughout the
present manual, we will frequently use this algorithm for a variety of problem set-
tings.
In numerous practical situations, one can find the fundamental set of solutions
for (1.1) analytically. This can, in particular, be done for most of the equations that
we arrive at in the present study. If, however, the governing differential equation
does not allow an analytical solution, then appropriate numerical procedures may be
employed. Later, in this book, we will discuss this point in more detail.
In compliance with property 1 of the definition, for any arbitrarily fixed position
of s ∈ (a, b), the Green’s function g(x, s) has to be a solution of (1.1) in (a, s) (on
the left of s), as well as in (s, b) (on the right of s). As soon as y j (x), ( j = 1, n) is
a fundamental set of solutions for (1.1), any its solution can be expressed as a linear
4 1 Green’s Functions for ODE
combination of the components y j (x). One may consequently write g(x, s) in the
following form
n
y j (x)A j (s), for a ≤ x ≤ s
g(x, s) = (1.3)
y j (x)B j (s), for s ≤ x ≤ b
j=1
where A j (s) and B j (s) represent functions to be yet determined. Clearly, the number
of these functions is 2n and the number of linear relations, which can be derived for
g(x, s) from properties 2, 3, and 4 of the definition, is also 2n. Thus, we have to
derive a system of 2n linear equations in 2n unknowns A j (s) and B j (s). It is evident
that (n − 1) of those equations can be obtained from property 2, a single equation
comes out from property 3, and the remaining n equations follow from property 4.
Hence, the key issue to be clarified at the current stage of the proof is whether
the described system of 2n equations in A j (s) and B j (s), ( j = 1, n) is well posed.
The well-posedness implies the consistence of the system and the uniqueness of its
solution.
By virtue of property 2, which stipulates the continuity of g(x, s) itself and its
partial derivatives with respect to x of up to the (n − 2) order included, as x = s,
one arrives at the following system of (n − 1) linear algebraic equations
n
d k y j (s)
C j (s) = 0, (k = 0, n − 2) (1.4)
j=1
dxk
n
d n−1 y j (s) 1
C j (s) =− (1.6)
j=1
dx n−1 p0 (s)
in the same set {C j (s)| j = 1, n} of unknowns. Hence, the relations in (1.4) along
with that of (1.6) constitute a system of n simultaneous linear algebraic equations
in n unknowns. The determinant of the coefficient matrix in this system is not zero,
because it represents the Wronskian for the fundamental set of solutions {y j (x),
j = 1, n}. Thus, the system has a unique solution. In other words, one can readily
obtain the explicit expressions for C j (s) by solving the system.
Once the functions C j (s) are at hand, the relations in (1.5) represent another
underdetermined system of n linear algebraic equations in the 2n functions A j (s)
and B j (s). To eliminate this occurrence of the system, we take advantage of the
1.1 Two-Point-Posed Boundary Value Problems 5
defining property 4. In doing so, let us first breakdown the forms Mi [y(a), y(b)] in
(1.2) into two additive parts as
n−1
n−1
Pi [y(a)] = αik y (k) (a), Q i [y(b)] = βki y (k) (b).
k=0 k=0
In compliance with property 4, we now substitute the expression for g(x, s) from
(1.3) into (1.2)
Mi [g(a, s), g(b, s)] ≡ Pi [g(a, s)] + Q i [g(b, s)] = 0, (i = 1, n). (1.7)
Since the operator Pi in (1.7) governs the values of g(a, s) and its derivatives at
the left-end point x = a of the interval [a, b], while Q i governs those at the right-end
point x = b, the branch of g(x, s) valid for a ≤ x ≤ s from (1.3) goes to Pi [g(a, s)],
while the branch valid for s ≤ x ≤ b must be substituted into Q i [g(b, s)]. This
yields
Mi [g(a, s), g(b, s)]
n
≡ Pi [g(a, s)]A j (s) + Q i [g(b, s)]B j (s) = 0, (i = 1, n).
j=1
Replacing the values of A j (s) in accordance with (1.5), one rewrites the above
system in the form
n
Pi [g(a, s)](B j (s) − C j (s)) + Q j [g(b, s)]B j (s) = 0, (i = 1, n).
j=1
Combining then the terms with B j (s) and taking the term with C j (s) to the right-
hand side, one obtains
n
n
{Pi [g(a, s)] + Q i [g(b, s)]} B j (s) = Pi [g(a, s)]C j (s), (i = 1, n).
j=1 j=1
Upon recalling the relations from (1.7), the above equations can finally be rewritten
in the form
n
n
Mi [g(a, s), g(b, s)]B j (s) = Pi [g(a, s)]C j (s), (i = 1, n). (1.8)
j=1 j=1
6 1 Green’s Functions for ODE
d 2 y(x)
= 0, x ∈ (0, a) (1.9)
dx2
be subject to the boundary conditions
dy(0) dy(a)
= 0, + hy(a) = 0, (1.10)
dx dx
where h represents a constant.
To ensure the existence of the Green’s function for the above problem, we check
out if this setting has only the trivial solution.
The most elementary set of functions constituting a fundamental set of solutions
for (1.9) is represented by
y1 (x) ≡ 1, y2 (x) ≡ x
This yields the general solution yg (x) for the equation in (1.9) in the form
yg (x) = D1 + D2 x
where D1 and D2 are arbitrary constants. A substitution of the above into the boundary
conditions of (1.10) yields the homogeneous system of linear algebraic equations in
D1 and D2 , with a well-posed coefficient matrix
0 1
h 1 + ah
Hence, the problem in (1.9) and (1.10) has only the trivial solution. Thus, there exists
its unique Green’s function g(x, s), and in compliance with the procedure developed
in the proof of Theorem 1.1, we express g(x, s) as
1.1 Two-Point-Posed Boundary Value Problems 7
A1 (s) + x A2 (s), for 0 ≤ x ≤ s
g(x, s) = (1.11)
B1 (s) + x B2 (s), for s ≤ x ≤ a
we form a system of two linear algebraic equations in these functions (see the system
in (1.4) and (1.6)) written as
C1 (s) + sC2 (s) = 0
(1.13)
C2 (s) = −1
in B1 (s) and B2 (s), from which it follows that B1 (s) = (1 + ha)/ h. This in turn
yields A1 (s) = [1 + h(a − s)]/ h.
Substituting the values of A j (s) and B j (s) just found into (1.11), we ultimately
obtain the Green’s function to the boundary value problem posed by (1.9) and (1.10)
in the form
(a − s) + h −1 , for 0 ≤ x ≤ s
g(x, s) = (1.14)
(a − x) + h −1 , for s ≤ x ≤ a
It is interesting to note that if the parameter h is taken to infinity, the second term
h −1 in (1.14) vanishes yielding the Green’s function
a − s, for 0 ≤ x ≤ s
g(x, s) = (1.15)
a − x, for s ≤ x ≤ a
dy(0)
= 0, y(a) = 0 . (1.16)
dx
Another interesting observation follows from (1.14), if the parameter h in it is set
to equal zero. It suggests that a Green’s function to the boundary value problem
dy(0) dy(a)
= 0, =0 (1.17)
dx dx
stated for the equation in (1.9) does not exist.
8 1 Green’s Functions for ODE
It is quite frequently the case in applied sciences that research projects are con-
sidered for phenomena occurring in infinite media. The Green’s function formalism
can successfully be extended to associated boundary value problems formulated over
infinite intervals. In our next example, we explain how this feature of a problem can
be taken care of.
Example 1.1.2 Consider the following differential equation
d 2 y(x)
− k 2 y(x) = 0, x ∈ (0, ∞) (1.18)
dx2
subject to boundary conditions imposed as
It can be shown that the conditions of existence and uniqueness for Green’s func-
tion are met in this case assuring a unique Green’s function of the above formulation.
Since roots of the characteristic (auxiliary) equation for (1.18) are k and −k, the
two exponential functions
represent a fundamental set of solutions for (1.18). Hence, one can express the Green’s
function for the boundary value problem in (1.18) and (1.19) in the form
A1 (s) exp (kx) + A2 (s) exp (−kx), for x ≤ s
g(x, s) = (1.20)
B1 (s) exp (kx) + B2 (s) exp (−kx), for s ≤ x
Denoting Ci (s) = Bi (s) − Ai (s), (i = 1, 2), one obtains the following system of
linear algebraic equations
exp (ks)C1 (s) + exp (−ks)C2 (s) = 0
k exp (ks)C1 (s) − k exp (−ks)C2 (s) = −1
1 1
C1 (s) = − exp(−ks), C2 (s) = exp(ks) (1.21)
2k 2k
The first condition in (1.19) implies
while the second condition results in B1 (s) = 0. This is so because the exponential
function exp (kx) is unbounded as x approaches infinity, and the only way to satisfy
the second condition in (1.19) is to set B1 (s) equals zero. This immediately yields
1.1 Two-Point-Posed Boundary Value Problems 9
1
A1 (s) = exp (−ks)
2k
and the relation in (1.22) consequently provides
1
A2 (s) = − exp (−ks)
2k
Hence, based on the known functions C2 (s) and A2 (s), one obtains
1
B2 (s) = [exp (ks) − exp (−ks)]
2k
Upon substituting the values of the coefficients A j (s) and B j (s) just found into
(1.20), one ultimately obtains the Green’s function to the problem posed by (1.18)
and (1.19) in the form
1 exp(k(x − s)) − exp(−k(x + s)), for x ≤ s
g(x, s) =
2k exp(k(s − x)) − exp(−k(s + x)), for s ≤ x
1
g(x, s) = (exp(−k|x − s|) − exp(−k(x + s))) (1.23)
2k
in terms of the absolute value function.
Example 1.1.3 Another boundary value problem is targeted as stated for the same
governing ODE as in Example 1.1.2
d 2 y(x)
− k 2 y(x) = 0, x ∈ (0, a), (1.24)
dx2
but formulated over a finite domain. Let it be subject to the following specific set of
boundary conditions
dy(0) dy(a)
y(0) = y(a), = (1.25)
dx dx
This problem setting is a sample of an important type of formulations in applied
sciences. The relations in (1.25) specify conditions of the a -periodicity of the solu-
tion.
The reader is recommended to confirm that the above boundary value problem
has only the trivial solution, providing existence of its unique Green’s function.
Since the formulation in (1.24) and (1.25) entails the same differential equation
as considered in Example 1.1.2, the beginning stage of the construction procedure
for the Green’s function resembles that from the previous problem. We again express
the Green’s function by (1.20), and the coefficients C1 (s) and C2 (s) are again found
as in (1.21).
10 1 Green’s Functions for ODE
implying that the relations in (1.26) and (1.27) along with those in (1.21) form a well-
posed system of four linear algebraic equations in the functions A1 (s), A2 (s), B1 (s),
and B2 (s). From the equations in (1.26) and (1.27), it follows that
1
− A1 (s) + B1 (s) = − exp(−ks) (1.29)
2k
we solve then the equations in (1.28) and (1.29) simultaneously, and obtain
To find the functions A2 (s) and B2 (s), we subtract (1.27) from (1.26). This results
in
A2 (s) − B2 (s) exp(−ka) = 0 (1.30)
1
− A2 (s) + B2 (s) = exp(ks) (1.31)
2k
we solve then (1.30) and (1.31) simultaneously. This yields
Substituting the values of A1 (s), A2 (s), B1 (s), and B2 (s) just found into (1.20),
we ultimately obtain the Green’s function to the boundary value problem in (1.24)
and (1.25) as
1.1 Two-Point-Posed Boundary Value Problems 11
exp(k(x − s + a)) + exp(k(s − x)), for x ≤ s
g(x, s) = K 0 (1.32)
exp(k(s − x + a)) + exp(k(x − s)), for s ≤ x
In order to make sure that the above equation does not degenerate a singular point
on (0, a), we impose certain limitations on its constant parameters m and b. Namely,
we assume that m > 0 and b > 0, which clearly implies that the function mx + b
does not take on zero value anywhere on the interval [0, a].
If boundary conditions are imposed for (1.33) as
dy(0)
= 0, y(a) = 0 (1.34)
dx
then the problem in (1.33) and (1.34) is well-posed allowing only the trivial solution.
The fundamental set of solutions
can, for example, be obtained for (1.33) by two successive integrations. Indeed, the
first integration yields
dy
(mx + b) = C1
dx
while separating variables in the above, we have
dx
dy = C1
mx + b
C1
y(x) = ln(mx + b) + C2
m
Since the boundary value problem in (1.33) and (1.34) has only the trivial solution,
there exists its unique Green’s function which can be presented in the form
12 1 Green’s Functions for ODE
A1 (s) + ln (mx + b)A2 (s), for 0 ≤ x ≤ s
g(x, s) = (1.35)
B1 (s) + ln (mx + b)B2 (s), for s ≤ x ≤ a
Tracing out then our customary procedure, one obtains the system of linear alge-
braic equations
C1 (s) + ln (ms + b)C2 (s) = 0
m(ms + b) C2 (s) = −(ms + b)−1
−1
1 1
C1 (s) = ln (ms + b), C2 (s) = −
m m
1 ma + b
A1 (s) = ln
m ms + b
Substituting the values of A j (s) and B j (s) just found into (1.35), one obtains the
Green’s function that we are looking for in the form
1 ln[(ma + b)(ms + b)−1 ], for 0 ≤ x ≤ s
g(x, s) = (1.36)
m ln[(ma + b)(mx + b)−1 ], for s ≤ x ≤ a
Example 1.1.5 Consider a boundary value problem in which the governing equation
d dy(x)
x = 0, x ∈ (0, a) (1.37)
dx dx
dy(a)
lim |y(x)| < ∞, + hy(a) = 0 (1.38)
x→0 dx
Note that the form, the first boundary condition of (1.38) presented in, is supported
by the singularity of the governing equation at the left-end point x = 0 of the domain.
1.1 Two-Point-Posed Boundary Value Problems 13
Integrating the equation in (1.37) successively two times, one obtains its funda-
mental set of solutions as
y1 (x) ≡ 1, y2 (x) ≡ ln x
The problem in (1.37) and (1.38) has only the trivial solution, allowing us to look
for its unique Green’s function in the form
A1 (s) + ln x A2 (s), for 0 ≤ x ≤ s
g(x, s) =
B1 (s) + ln x B2 (s), for s ≤ x ≤ a
In compliance with our customary procedure, we form the system of linear alge-
braic equations
C1 (s) + ln s C2 (s) = 0
s −1 C2 (s) = −s −1
in C1 (s) and C2 (s), whose solution is: C1 (s) = ln s and C2 (s) = −1.
The boundness of the Green’s function at x = 0 implies A2 (s) = 0. Consequently,
we find B2 (s) = −1. The second condition in (1.38) yields
Notice that as the value of h is taken to infinity, the first term (ah)−1 in (1.39)
vanishes, yielding the Green’s function
− ln[(a)−1 s], for 0 ≤ x ≤ s
g(x, s) = (1.40)
− ln[(a)−1 x], for s ≤ x ≤ a
Observing appearance of the Green’s functions obtained thus far, one may notice
their common property. Indeed, they are symmetric in a certain sense. That is, the
interchange of x with s in their expressions valid for x ≤ s yields that one valid for
x ≥ s and vice versa. We will discuss this issue in more detail later. In the mean
time, the example that follows focuses on a problem whose Green’s function appears
to be nonsymmetric.
14 1 Green’s Functions for ODE
Example 1.1.6 A boundary value problem is set up for the second-order linear
equation
d 2 y(x) dy(x)
+ − 2y(x) = 0, x ∈ (0, ∞) (1.42)
dx2 dx
subject to the boundary conditions
Direct analysis shows that the above problem has only the trivial solution, allow-
ing, subsequently, a unique Green’s function. Since
represent a fundamental set of solutions to (1.42), one can express the Green’s func-
tion to the problem of (1.42) and (1.43) in the form
A1 (s) exp(x) + A2 (s) exp(−2x), for x ≤ s
g(x, s) = (1.44)
B1 (s) exp(x) + B2 (s) exp(−2x), for s ≤ x
Upon comparison with the forms which arrived in (1.14), (1.15), (1.23), (1.32),
(1.36), (1.39), and (1.40), the above expression fails to be symmetric. The question
is, what makes the statement in (1.42) and (1.43) different of all the others considered
earlier in this section. The reasoning for this occurrence can be found in the next
section.
1.1 Two-Point-Posed Boundary Value Problems 15
In order to address the symmetry for Green’s functions with respect to the observation
and the source point, a certain preparatory work has to be carried out. In doing so,
we consider the linear nth-order homogeneous differential equation
dn y d n−1y
L[y(x)] ≡ p0 + p 1 + · · · + pn y = 0
dxn d x n−1
dp0 (x)
2 − p1 (x) = p1 (x)
dx
which must hold for the self-adjointness of (1.46). This implies
dp0 (x)
p1 (x) = (1.49)
dx
Differentiating the above relation, we realize that the sum of the first two terms
in the coefficient
d 2 p0 (x) dp1 (x)
− + p2 (x)
dx2 dx
of y(x) in (1.48) is equal to zero. This means that the self-adjointness of (1.46)
implies the relation between the coefficients p0 (x) and p1 (x) and puts no constraint
on the coefficient p2 (x). In other words, if (1.46) is self-adjoint, then it can be written
as
d 2 y(x) dp0 (x) dy(x)
p0 (x) + + p2 (x)y(x) = 0
dx2 dx dx
which reads in a short-handed form as
d dy(x)
p0 (x) + p2 (x)y(x) = 0 (1.50)
dx dx
The above is usually referred to as the standard form of the second-order self-
adjoint equation.
Thus, if the coefficients p0 (x) and p1 (x) in (1.46) satisfy the relation in (1.49),
then it is in a self-adjoint form. The fact that (1.49) does not involve the coefficient
p2 (x) prompts a simple idea of how a linear second-order differential equation can
reduce to a self-adjoint form. Indeed, multiplying (1.46) through by a certain nonzero
function (we call it the integrating factor) and applying then the relation in (1.49)
to the coefficients of d 2 y/d x 2 and dy/d x of the resultant equation, one can readily
formulate a relation from which the integrating factor can afterwards be found. The
procedure of finding the integrating factor is quite straightforward. In the example
that follows, we discuss it in detail.
Example 1.1.7 Clearly, the condition in (1.49) is not met for the equation
d 2 y(x) dy(x)
+ 4x − 2y(x) = 0 (1.51)
dx2 dx
1.1 Two-Point-Posed Boundary Value Problems 17
d 2 y(x) dy(x)
μ(x) 2
+ 4xμ(x) − 2μ(x)y(x) = 0 (1.52)
dx dx
The leading coefficient p0 (x) of the above equation is μ(x), while the coefficient
p1 (x) is 4xμ(x). Thus, in compliance with (1.49), the equation in (1.52) would be
in a self-adjoint form if
dμ(x)
= 4xμ(x) (1.53)
dx
dμ(x)
= 4xd x
μ(x)
ln |μ(x)| = 2x 2 + C
By solving the above equation for μ(x), we obtain the general solution to (1.53)
as
2
+C
μ(x) = e2x
Any function from this family can be considered as the integrating factor for the
equation in (1.52). In other words, constant C can be arbitrarily fixed and we assume,
say, C = 0, which yields
2
μ(x) = e2x
Substituting now the found expression into (1.52), we reduce the latter to the form
2 d 2 y(x) 2 dy(x) 2
e2x + 4xe2x − 2e2x y(x) = 0,
dx2 dx
which is self-adjoint.
At this point in our presentation we assume that L represents a self-adjoint second-
order operator. That is
d d
L≡ p0 (x) + p2 (x)
dx dx
Consider two functions u(x) and v(x), and assume that each of them is two times
continuously differentiable on (a, b), and form the following bilinear combination
of them
18 1 Green’s Functions for ODE
Upon removing the outer parentheses in both the additive components, the above
combination transforms into
d dv d du
u L(v) − vL(u) = u p0 (x) −v p0 (x)
dx dx dx dx
When the product rule is applied and some regrouping accomplished, the above
expression simplifies as
d dv d du
u p0 (x) −v p0 (x)
dx dx dx dx
dp0 (x) dv d 2v dp0 (x) du d 2u
=u + p0 (x) 2 −v + p0 (x) 2
dx dx dx dx dx dx
Integrating both sides of (1.55) from a to b, one obtains the following relation
b
dv du b
[u L(v) − v L(u)] d x = p0 (x) u −v (1.56)
a dx d x a
which is usually [2, 12, 16, 21, 24, 41, 42] referred to as the Green’s formula for
a self-adjoint operator. From the recent development, it follows that the Green’s
formula holds for a self-adjoint operator L and two continuously differentiable on
(a, b) functions u(x) and v(x).
1.1 Two-Point-Posed Boundary Value Problems 19
If in addition to being two times continuously differentiable on (a, b), the func-
tions u(x) and v(x) make zero the right-hand side in (1.56), then the Green’s formula
reduces to a compact form. That is, if
dv du b
p0 (x) u −v =0 (1.57)
dx d x a
then we have b
[u L(v) − v L(u)] d x = 0 (1.58)
a
So, the Green’s formula in (1.58) is valid for a self-adjoint operator L, with u(x)
and v(x) being two times continuously differentiable on (a, b) and satisfying the
relation in (1.57). This relation is, however, implicit in nature, which makes it too
cumbersome to deal with over and over again in actual computations. Therefore, it
is important to find some of its explicit equivalents which are more convenient for
practical use.
Aiming at some explicit equivalents of (1.57), we rewrite it in the extended form
dv(b) du(b) dv(a) du(a)
p0 (b) u(b) − v(b) − p0 (a) u(a) − v(a) = 0 (1.59)
dx dx dx dx
Notice that this relation contains the values of u(x) and v(x), as well as of their
derivatives, at the end points of [a, b]. This observation makes it obvious that the
relation in (1.59) holds, if both u(x) and v(x) satisfy one of the following types of
boundary conditions at x = a and x = b:
(1) y(a) = 0, y(b) = 0
(2) y(a) = 0, y (b) = 0
(3) y (a) = 0, y (b) = 0
It is also directly seen that the condition in (1.59) is valid in the so-called singular
case, when the leading coefficient p0 (x) in (1.57) is equal to zero at one of the
end points of [a, b]. In such a case we usually require y(x) to be bounded at the
corresponding end point, with a value of either y(x) or y (x) being zero at the other
end point, that is:
(4) lim x→a |y(x)| < ∞, y(b) = 0
(5) lim x→a |y(x)| < ∞, y (b) = 0
In addition, from a close analysis, it follows that the condition in (1.59) holds also
for both u(x) and v(x) satisfying one of the following sets of boundary conditions:
(6) y(a) = 0, y (b) + hy(b) = 0
(7) y (a) = 0, y (b) + hy(b) = 0
(8) y (a) + h 1 y(a) = 0, y (b) + h 2 y(b) = 0
(9) y(a) = y(b), p0 (a) y (a) = p0 (b) y (b)
(10) lim x→a |y(x)| < ∞, y (b) + hy(b) = 0
The last set of conditions presumes (similarly to cases in (4) and (5)) that the
leading coefficient p0 (x) of (1.54) equals zero at x = a.
20 1 Green’s Functions for ODE
Note that the end points a and b, in all the types of boundary conditions (1)–(10),
are interchangeable. Namely, the set of conditions
for example, falls into the type in (6). This is also true for the boundary conditions
of the types in (4), (5), (7) and (10).
The discussion just completed brings another important terminological issue. A
boundary value problems set up for the equation in (1.50) subject to either one of
the types of boundary conditions listed above, belongs to the class of the so-called
self-adjoint boundary value problems.
We are in a position now to redirect the reader’s attention to one of the basic
questions in this section. That is, what makes a Green’s function symmetric in the
sense mentioned in Sect. 1.1.3? The theorem below does, in fact, specify conditions
that a boundary value problem should meet for its Green’s function to be of that type
symmetric.
is self-adjoint, then its Green’s function g(x, s) is symmetric, provided that its expres-
sion valid for x ≥ s can be obtained from that valid for x ≤ s by interchanging of x
with s.
Proof The following proof is based on a slight modification of that procedure which
has been used in the proof of Theorem 1.1. Herein, we also choose two linearly
independent particular solutions y1 (x) and y2 (x) of the governing equation in (1.60).
But contrary to Theorem 1.1, some additional constraints will be put on y1 (x) and
y2 (x). We are going to choose them in a specific manner.
Namely, let y1 (x) and y2 (x) be two linearly independent particular solutions to
(1.60). Let y1 (x) satisfy, in addition, the first boundary condition in (1.61), while
y2 (x) satisfies the second condition in (1.61). Clearly, neither y1 (x) nor y2 (x) can
satisfy both boundary conditions in (1.61). Indeed, by making such an assumption
we come to a conflict with the well posedness of the statement in (1.60) and (1.61),
which implies that the trivial solution is the only solution to this problem.
Let us form the bilinear combination based on y1 (x) and y2 (x)
which identically equals zero on (a, b), since L[y1 (x)] ≡ 0 and L[y2 (x)] ≡ 0 for
x ∈ (a, b).
1.1 Two-Point-Posed Boundary Value Problems 21
Recalling the form from (1.54) and rewriting it in terms of the functions y1 (x)
and y2 (x) yields
d dy2 dy1
y1 L(y2 ) − y2 L(y1 ) = p0 (x) y1 − y2
dx dx dx
Since the left-hand side of the above relation is identically zero, so is its right-hand
side as well. That is
d dy2 dy1
p0 (x) y1 − y2 =0
dx dx dx
which implies
dy2 dy1
p0 (x) y1 − y2 =C (1.62)
dx dx
So, we can assume that, if the particular solutions y1 (x) and y2 (x) are chosen
for (1.60) in the manner specified earlier, then these solutions meet the condition in
(1.63) throughout (a, b). This implies that for any location of a point s ∈ (a, b), we
can express the Green’s function g(x, s) to the problem in (1.60) and (1.61) in the
form
c (s) y1 (x), for a ≤ x ≤ s
g(x, s) = 1 (1.64)
c2 (s) y2 (x), for s ≤ x ≤ b
in c1 (s) and c2 (s). Clearly, the coefficient matrix of this system is not singular,
because its determinant
is the Wronskian for the two linearly independent functions y1 (s) and y2 (s). Hence,
the system in (1.65) has a unique solution which appears in the form
y2 (s) y1 (s)
c1 (s) = − , c2 (s) = −
p0 (s) W (s) p0 (s) W (s)
Upon substituting these expressions for c1 (s) and c2 (s) into (1.64), the branch of
the Green’s function valid for x ≤ s is
y1 (x) y2 (s)
g(x, s) = − , x ≤s (1.66)
p0 (s) W (s)
y2 (x) y1 (s)
g(x, s) = − , s≤x (1.67)
p0 (s) W (s)
According to the relation in (1.63), the denominator in (1.66) and (1.67) meets
the condition
p0 (s) W (s) ≡ p0 (s) y1 (s)y2 (s) − y2 (s)y1 (s) ≡ −1
This allows us to finally obtain the Green’s function g(x, s) for the boundary value
problem posed by (1.60) and (1.61) in the following ‘symmetric’ form
y2 (s) y1 (x), for a ≤ x ≤ s
g(x, s) = (1.68)
y1 (s) y2 (x), for s ≤ x ≤ b
Thus, the Theorem has been proven. Indeed, from the above representation, it
follows that the Green’s function g(x, s) of a self-adjoint boundary value problem
is invariant to the interchange of the observation point x with the source point s.
In other words, the Green’s function is symmetric in the sense that whenever the x
variable is interchanged with the s variable in one of the branches of g(x, s), we
obtain the other branch.
The symmetry of Green’s functions, whose analysis is just completed, has impor-
tant implementations in various applied sciences. It is related to the so-called
Maxwell’s reciprocity [16, 21] asserting that the response of a field at an obser-
vation point x due to a source at s is the same as the response at s due to a point
source at x.
1.1 Two-Point-Posed Boundary Value Problems 23
In the next section, we will revisit the basic issue of this chapter, which is the
construction of Green’s functions for ODE. An alternative construction procedure
will be presented in detail.
Clearly, the notion of Green’s function is introduced for a boundary value prob-
lem where both the governing differential equation and the boundary conditions are
homogeneous. Such settings are referred to as homogeneous boundary value prob-
lems. In this section, we will switch over the reader’s attention to nonhomogeneous
linear differential equations subject to homogeneous boundary conditions.
An important theorem will be formulated and proved herein. It builds up a theoret-
ical background for utilization of Green’s function instrument for solving boundary
value problems for nonhomogeneous linear equations. Then, we will review the clas-
sical [16, 21, 41] procedure for the construction of Green’s functions, which is based
on that theorem and the Lagrange method of variation of parameters that is tradition-
ally used in ODE to analytically solve nonhomogeneous linear differential equations
if the fundamental set of solutions is available for the corresponding homogeneous
equation.
Consider, the linear nonhomogeneous equation
dn y d n−1y
L[y(x)] ≡ p0 (x) + p 1 (x) + · · · + pn (x)y = − f (x) (1.69)
dxn d x n−1
n−1
d k y(a) k
i d y(b)
Mi [y(a), y(b)] ≡ [αik + βk ] = 0, (i = 1, n) (1.70)
k=0
dxk dxk
where the coefficients p j (x) are continuous while the right-hand side term f (x) of
the governing equation is integrable on [a, b], with p0 (x) = 0, and Mi [y(a), y(b)]
represent linearly independent forms with constant coefficients.
Connection will be established between the uniqueness of the solution to (1.69)
and (1.70), and the corresponding homogeneous problem. For this reason, we focus
on a theorem that makes things ready for the use of a Green’s function, constructed
for a homogeneous problem, in solving corresponding nonhomogeneous equations.
Theorem 1.3 The boundary value problem in (1.69) and (1.70) is well posed (has,
in other words, a unique solution), if the corresponding homogeneous setting has
only the trivial solution.
Proof The statement of this theorem follows from the linearity of the setting. Indeed,
let Y1 (x) and Y2 (x) represent two distinct solutions to (1.69) and (1.70). This implies
24 1 Green’s Functions for ODE
that each of these solutions is supposed to make the equation in (1.69) true. That is,
d n Y1 d n−1Y1
p0 (x) + p 1 (x) + · · · + pn (x)Y1 = − f (x)
dxn d x n−1
and
d n Y2 d n−1Y2
p0 (x) + p 1 (x) + · · · + pn (x)Y2 = − f (x)
dxn d x n−1
Subtracting these in the term-by-term manner, we have
d n (Y1 − Y2 ) d n−1(Y1 − Y2 )
p0 (x) + p 1 (x) + · · · + pn (x)(Y1 − Y2 ) = 0
dxn d x n−1
Thus, if Y1 (x) and Y2 (x) represent two distinct solutions to (1.69), then their
difference Y12 (x) = Y1 (x) − Y2 (x) is a solution to the corresponding homogeneous
equation. In the same fashion, taking advantage of the linearity of the forms Mi ,
we can show that Y12 (x) should satisfy the homogeneous boundary conditions in
(1.70). In other words, Y12 (x) represents a solution to the homogeneous boundary
value problem corresponding to (1.69) and (1.70). But, according to the statement of
this theorem, the corresponding homogeneous problem has only the trivial solution,
which implies that the difference Y1 (x) − Y2 (x) should be identical zero.
So, our assumption about existence of two distinct solutions of the original setting
in (1.69) and (1.70) is wrong and there exists, therefore, a unique solution of that
problem, if the corresponding homogeneous problem has only the trivial solution.
The theorem that follows establishes a direct way for expressing the solution to
the problem in (1.69) and (1.70) in terms of the Green’s function constructed for the
corresponding homogeneous problem.
Theorem 1.4 If the boundary value problem posed by the nonhomogeneous equa-
tion of (1.69) subject to the homogeneous conditions of (1.70) is well posed, then
the unique solution for (1.69) and (1.70) can be expressed [12, 16, 21, 41] by the
integral
b
y(x) = g(x, s) f (s) ds (1.71)
a
whose kernel g(x, s) represents the Green’s function of the corresponding homoge-
neous problem.
Proof The theorem statement implies that a proof is required for two independent
statements. First, that the integral in (1.71) satisfies the equation in (1.69), and second,
that it complies with the boundary conditions of (1.70).
Since the Green’s function g(x, s) is defined in two pieces, we break down the
integral in (1.71) into two integrals as shown
1.1 Two-Point-Posed Boundary Value Problems 25
x b
−
y(x) = g (x, s) f (s) ds + g + (x, s) f (s) ds (1.72)
a x
where by g + (x, s) and g − (x, s) we denote the branches of g(x, s) valid for x ≤ s
and for x ≥ s, respectively.
As to the satisfaction of the governing equation of (1.69), we take into account a
specific occurrence of y(x) in (1.72). The point is that it is defined in terms of the
two definite integrals (with s representing the variable of integration), which contain
a parameter x and have the variable limits depending on x. Therefore, one has to
recall (from the fundamental theorem of integral calculus [41]) that, if a function
(x) is defined in the integral form
β(x)
(x) = F(x, s)ds
α(x)
This implies that, since both the integrals in (1.72) contain x as a parameter and
their limits depend upon x, the derivative of y(x) can formally be written as
dy(x) x
∂g − (x, s)
= f (s)ds + g − (x, x) f (x)
dx a ∂x
b
∂g + (x, s)
+ f (s)ds − g + (x, x) f (x)
x ∂x
Combining the integrals in the above relation and realizing that the nonintegral
terms are eliminated due to the continuity of the Green’s function as x = s, one
obtains b
dy(x) ∂g(x, s)
= f (s)ds
dx a ∂x
Thus, the derivative of the form in (1.71) can be taken by the direct differentiation
of its kernel function. Recalling the continuity of the derivatives of the Green’s
function, of up to the (n − 2)nd order included, as x = s (see property 2 of the
Definition), the higher order derivatives of the integrals in (1.72), of up to the (n −1)st
order included, can be computed, analogously to the first derivative, as
d k y(x) b
∂ k g(x, s)
= f (s)ds, (k = 1, n − 1) (1.74)
dxk a ∂x k
26 1 Green’s Functions for ODE
Since the boundary conditions in (1.70) involve only the derivatives of y(x) of
up to the order of (n − 1), all the derivatives in Mi [y(a), y(b)] can formally be
computed under the integral sign. This yields
n−1
b
∂ k g(a, s) b
∂ k g(b, s)
Mi [y(a), y(b)] ≡ αk
i
f (s)ds + βki f (s)ds
k=0 a ∂x k a ∂x k
n−1
b
∂ k g(a, s) i ∂ g(b, s)
k
= αik + β f (s)ds = 0, i = 1, n
a k=0
∂x k k
∂x k
because the expressions in the square brackets equal zero due to property 4 in the
Definition of the Green’s function. Thus, the boundary conditions in (1.70) are indeed
satisfied by the form in (1.71).
To complete the part of the theorem related to the satisfaction of the governing
equation in (1.69), we compute the nth derivative of y(x) by differentiating the
relation in (1.74), where k is fixed as n − 1. This yields
n−1 −
d n y(x) b
∂ n g(x, s) ∂ g (x, x) ∂ n−1 g + (x, x)
= f (s)ds + − f (x)
dxn a ∂x n ∂x n−1 ∂x n−1
Upon substituting y(x) and its derivatives found earlier into (1.69) and combining
all the integral terms into a single term form, one finally obtains
b
L[g(x, s)] f (s) ds − f (x) = − f (x)
a
Clearly, the above equality is an identity, since L[g(x, s)] = 0 on (a, b). This
completes the proof of the theorem.
Earlier in Sect. 1.1.3, we obtained a series of Green’s functions by means of the
approach based on their defining properties. Utilizing the theorem just proven, we
will explore, in what follows, an alternative approach which employs the Lagrange’s
method of variation of parameters [16, 21, 41] that is traditionally used in solving
nonhomogeneous linear differential equations.
We will restrict our discussion in this section to the second-order equation
d 2 y(x) dy(x)
p0 (x) 2
+ p1 (x) + p2 (x)y(x) = − f (x) (1.75)
dx dx
1.1 Two-Point-Posed Boundary Value Problems 27
because the scope of the present volume is limited to the analysis of second-order
PDEs.
Let the above equation be set up on the interval (a, b) and subject to the simplest
set of boundary conditions
y(a) = 0, y(b) = 0 (1.76)
Assume that the above boundary value problem has a unique solution, which
implies (in view of Theorem 1.3) that the corresponding homogeneous problem has
only the trivial solution. Let y1 (x) and y2 (x) be a fundamental set of solutions to the
homogeneous equation associated with (1.75). Express then the general solution of
the above in compliance with the Lagrangian method of variation of parameters, in
the form
y(x) = C1 (x)y1 (x) + C2 (x)y2 (x) (1.77)
where C1 (x) and C2 (x) are differentiable on (a, b) functions to be found in what
follows.
The first impression is that the idea of expressing a solution for (1.75) in the form
of (1.77), which contains two unknown functions C1 (x) and C2 (x), is unproductive,
unless a second (in addition to the equation in (1.75) itself) relation is established.
Lagrange’s method does provide an effective and elegant choice of such a relation.
The direct substitution of y(x) from (1.77) into (1.75) would result in a cum-
bersome single second-order differential equation in two unknown functions C1 (x)
and C2 (x). In order to avoid such an unfortunate complication, the procedure in the
Lagrange’s method suggests to differentiate y(x) from (1.77)
y (x) = C1 (x)y1 (x) + C1 (x)y1 (x) + C2 (x)y2 (x) + C2 (x)y2 (x)
resulting in
y (x) = C1 (x)y1 (x) + C1 (x)y1 (x) + C2 (x)y2 (x) + C2 (x)y2 (x) (1.80)
To go further, we substitute the expressions for the functions y(x), y (x), and
y (x) just found into (1.75)
+ p2 (C1 y1 + C2 y2 ) = − f (x)
28 1 Green’s Functions for ODE
Since y1 (x) and y2 (x) represent particular solutions of the homogeneous equation
associated with (1.75), the coefficients of C1 (x) and C2 (x) in the above equation are
zero. This yields
C1 (x)y1 (x) + C2 (x)y2 (x) = − f (x) p0−1 (x) (1.81)
The relations in (1.78) and (1.81) constitute a well-posed linear system in C1 (x)
and C2 (x). This assertion is based of the fact that determinant of the coefficient
matrix of the system represents Wronskian
to the two linearly independent functions y1 (x) and y2 (x), and is therefore nonzero.
Upon solving the system, we obtain
and x
y1 (s) f (x)
C2 (x) = ds + H2
a p0 (s)W (s)
move the factors y1 (x) and y2 (x) under the integral signs (this is a formal operation
since the variable of integration is s but not x), and combine then the two integrals
in one. This yields
x
y1 (s)y2 (x) − y1 (x)y2 (s)
y(x) = ds + H1 y1 (x) + H2 y2 (x) (1.82)
a p0 (s)W (s)
1.1 Two-Point-Posed Boundary Value Problems 29
Upon satisfying the boundary conditions in (1.76) with y(x) as expressed above,
we obtain the following linear system
y1 (a) y2 (a) H1 0
× = (1.83)
y1 (b) y2 (b) H2 P(a, b)
where
R(b, s) = y1 (b)y2 (s) − y1 (s)y2 (b)
With this, we arrive at the solution to the system in (1.83) in the form
b
y2 (a)R(b, s) f (s)
H1 = − ds
a p0 (s)R(a, b)W (s)
and b
y1 (a)R(b, s) f (s)
H2 = ds
a p0 (s)R(a, b)W (s)
Upon substituting the above into (1.82), we express the solution of the boundary
value problem in (1.75) and (1.76) as
x
R(x, s) f (s) b
R(a, x)R(b, s) f (s)
y(x) = − ds + ds
a p0 (s)W (s) a p0 (s)R(a, b)W (s)
with the kernel function g(x, s) expressed in two pieces, one of which appears as
R(a, x)R(b, s)
g(x, s) = , x ≤s (1.85)
p0 (s)R(a, b)W (s)
After a trivial but quite cumbersome transformation, the above expression sim-
plifies to the form
R(a, s)R(b, x)
g(x, s) = , x ≥s (1.86)
p0 (s)R(a, b)W (s)
Notice that, since the solution to the problem posed by (1.75) and (1.76) is found as
the integral in (1.84), we conclude (by virtue of Theorem 1.4) that the kernel function
g(x, s) in (1.84) does in fact represent the Green’s function to the homogeneous
boundary value problem corresponding to that of (1.75) and (1.76).
It is clear that if the setting in (1.75) and (1.76) is self-adjoint, then the product
p0 (s)W (s) is equal to a constant. This obviously makes the expressions in (1.85)
and (1.86) symmetric in the sense discussed earlier.
Thus, the approach based on the method of variation of parameters allows us to
construct Green’s functions. Indeed, once the solution to a nonhomogeneous linear
differential equation subject to homogeneous boundary conditions is expressed in
the integral form of the type in (1.84), the kernel of the latter represents the Green’s
function to the corresponding homogeneous boundary value problem.
The described approach can be considered as an alternative to the one based on
the defining properties of Green’s function and which was discussed in some detail
earlier. We present below a couple of examples illustrating some peculiarities of its
application.
Example 1.1.8 Construct the Green’s function to the problem set up for the non-
homogeneous equation
d 2 y(x)
+ k 2 y(x) = − f (x), x ∈ (0, a) (1.87)
dx2
subject to homogeneous boundary conditions
The right-hand side function f (x) in the governing equation of (1.87) is supposed
to be integrable on [0, a].
One can readily realize that the corresponding to (1.87) and (1.88) homogeneous
problem has only the trivial solution. This implies that the conditions of existence
and uniqueness of its Green’s function are met.
The general solution to (1.87) can be written in the form
because y1 (x) ≡ sin kx and y2 (x) ≡ cos kx might constitute a fundamental set of
solutions for the corresponding homogeneous equation.
Following in footsteps of the procedure described earlier for the general case (see
the development for the problem in (1.75) and (1.76)), we arrive, in the current case,
1.1 Two-Point-Posed Boundary Value Problems 31
at the well-posed system of linear algebraic equations in C1 (x) and C2 (x)
sin kx cos kx C1 (x) 0
× =
k cos kx −k sin kx C2 (x) − f (x)
resulting in
1 1
C1 (x) = cos kx f (x), C2 (x) = − sin kx f (x)
k k
This yields x
1
C1 (x) = cos ks f (s)ds + H1
0 k
and x
1
C2 (x) = − sin ks f (s)ds + H2
0 k
Substituting these into (1.89) and combining then the two integral terms into one,
we express the general solution to (1.87) in the form
x
1
y(x) = sin k(x − s) f (s)ds + H1 sin kx + H2 cos kx (1.90)
0 k
The first condition in (1.88) yields H1 = 0, while the second condition results in
a
cos k(a − s) f (s)ds − H2 k sin ka = 0
0
Plugging the values of H1 and H2 just found in (1.90), one obtains the solution to
the problem in (1.87) and (1.88) as
x
sin k(x − s) a
cos k(a − s)
y(x) = f (s)ds + cos(kx) f (s)ds
0 k 0 k sin ka
Hence, in compliance with Theorem 1.4, the above represents the Green’s function
to the homogeneous boundary value problem corresponding to (1.87) and (1.88).
Example 1.1.9 Consider a boundary value problem posed for the equation with
variable coefficients
d dy(x)
(x + 1)
2
= f (x), x ∈ (0, a) (1.93)
dx dx
and describe in brief the construction procedure for the Green’s function to the
corresponding homogeneous problem.
y(x) = C arctan x + D
Upon satisfying the boundary conditions in (1.94), the values of D1 and D2 are
found as a
ω − arctan s
D1 = 0, D2 = f (s)ds
0 ω
where ω = arctan a.
1.1 Two-Point-Posed Boundary Value Problems 33
Substituting these into the above expression for the general solution and rearrang-
ing the integral terms, we obtain the solution to the original boundary value problem
as the single integral a
y(x) = g(x, s) f (s)ds
0
whose kernel
1 arctan x(ω − arctan s), for 0 ≤ x ≤ s
g(x, s) = (1.95)
ω arctan s(ω − arctan x), for x ≤ s ≤ a
The opening segment of the current chapter reviewed the classical Green’s function
approach. We tried to familiarize the reader with some of its applications. The review
discloses importance of this approach which represents a powerful instrument in the
qualitative as well as quantitative analysis of boundary value problems posed for
linear ODEs and PDEs.
Coefficients of the governing differential equation in a boundary value problem
are assumed differentiable, to a certain order, functions of the independent variable.
Note, however that coefficients of differential equations that simulate many physical
phenomena are not necessarily smooth functions. They might, for example, be dis-
continuous, in which case the classical Green’s function formalism does not directly
apply.
This creates a situation to accordingly adjust the Green’s function formalism and
to make it workable for such an irregularity of differential equations. Earlier in [8,
31], we had reported on our work on such an adjustment and on a certain progress
made. In this section, a discussion is launched on the indicated adjustment. A novel
notion of the matrix of Green’s type will be introduced for specific sets of ordinary
differential equations, and a number of its applications will be presented. Later in
34 1 Green’s Functions for ODE
our volume, the matrix of Green’s type formalism will be further extended to make
it applicable to PDEs as well.
Specific sets of linear ordinary differential equations are considered as posed on finite
weighted graphs, where each of the equations governs a single unknown function
and is posed on a single edge of a graph. The individual equations are put in a system
format by subjecting contact and boundary conditions at the vertices and endpoints
of the graph. This creates a multiple-point-posed boundary value problem, and our
objective is to determine its matrix of Green’s type. The issue of existence and
uniqueness of such a matrix is addressed and an analytical method is proposed for
its construction.
To state a boundary value problem of the discussed kind, we consider a finite
weighted graph R (see Fig. 1.1). For terminological purposes, vertices of degree
one will be referred to as the endpoints. Let the graph have n edges denoted with
ei , (i = 1, n), m endpoints E h , (h = 1, m), and r vertices Vk , (k = 1, r ). Let dk
represent the vertex Vk degree, and let also positive real numbers li , (i = 1, n), each
representing the length of the edge ei , be regarded as its weight.
Let u i (x) represent unknown functions, each to be defined on the corresponding
edge ei of R. We will determine each of these functions by the following set of linear
second-order differential equations
d du i (x)
pi (x) + qi (x)u i (x) = − f i (x), x ∈ (0, li ), (i = 1, n) (1.96)
dx dx
The above individual equations are arranged into a system format by imposing a
set of uniqueness conditions. That is, the contact conditions
dk
du j (Vk )
u 1 (Vk ) = · · · = u dk (Vk ), p j (Vk ) = 0, (k = 1, r ) (1.97)
j=1
dx
are imposed at each vertex Vk , of degree dk . Formulating the above conditions, we use
a ‘local’ numbering for the edges incident to the vertex Vk . It can easily be seen that
the number of contact conditions imposed at each vertex equals the vertex degree.
In addition, the end (boundary) conditions
du i (E h )
αh + βh u i (E h ) = 0, (h = 1, m) (1.98)
dx
are imposed at each endpoint E h of R. This implies that the functions u i (x) in (1.98)
are defined on the edges ei incident to E h .
Note that the number of contact conditions imposed at a vertex Vk is equal to the
vertex degree, while a single boundary condition is imposed at each endpoint E h .
This implies that the total number N of uniqueness conditions imposed in (1.97) and
(1.98) is defined as
r
N =m+ dk
k=1
which is, according to graph theory, two times the number n of the edges in R, that
is N = 2n.
As an example of possible physical interpretation of the problem stated in (1.96)–
(1.98), one might think of it as the simulation of a steady-state either heat-transfer
or mass-transfer process taking place in an assembly of one-dimensional conduc-
tive elements. Clearly, the contact conditions in (1.97) simulate, in this case, the
conservation of energy law at every vertex Vk of R.
and
∂gii (x, ξ) ∂gii (x, ξ) 1
lim+ − lim− =
ξ→x ∂x ξ→x ∂x pi (ξ)
Before going to the existence and uniqueness of the matrix of Green’s type, it is
appropriate to make a note. If the problem in (1.96)–(1.98) is well posed having a
unique solution, then the trivial solution
u i (x) ≡ 0, x ∈ (0, li ), (i = 1, n)
The functions ai1 (ξ), ai2 (ξ), bi1 (ξ), bi2 (ξ), ci j (ξ), and di j (ξ) in the above
representations are to be determined upon applying the remaining defining properties
of the matrix of Green’s type. Notice that the total number of these functions equals
2n(n + 1) while the total number of the relations provided by properties 2 and 4 is
also 2n(n + 1).
By virtue of property 2, one obtains n well-posed systems of linear algebraic
equations
u i1 (ξ) u i2 (ξ) Ci1 (ξ) 0
× = , (i = 1, n) (1.101)
u i1
(ξ) u i2 (ξ) Ci2 (ξ) pi−1 (ξ)
in two unknowns each, of the total amount of 2n equations in 2n unknowns Ci1 (ξ)
and Ci2 (ξ), (i = 1, n). These unknowns are expressed in terms of the ξ depending
functions of gii (x, ξ) in (1.99) as
The well posedness of the system in (1.101) follows from the fact that the deter-
minant of its coefficient matrix represents Wronskian for the linearly independent
functions u i1 (x) and u i2 (x). Hence, the unique expressions for Ci1 (ξ) and Ci2 (ξ) can
readily be obtained. Subsequently, in compliance with (1.102), the functions ai1 (ξ)
and ai2 (ξ) can uniquely be expressed in terms of bi1 (ξ) and bi2 (ξ) and vice versa.
Thus, the number of undetermined functions in (1.99) and (1.100) reduces to 2n 2 .
And they can ultimately be found by applying the defining property 4. Indeed, by
satisfying the entire set of boundary and contact conditions posed in (1.97) and (1.98)
n times (once for each location of the source point ξ ∈ e j , j = 1, n), we finally
obtain a nonhomogeneous system of 2n 2 linear algebraic equations in 2n 2 unknowns.
The coefficient matrix of this system reduces to the following partitioned diagonal
form ⎛ ⎞
A11 0 . . . 0
⎜ 0 A22 . . . 0 ⎟
⎜ ⎟
⎜ ⎟
M =⎜⎜ ⎟
. . . . . . ⎟
⎜ ⎟
⎝ ⎠
0 0 . . . Ann
Notice that the proof, just completed, is constructive. That is, it offers a straightfor-
ward procedure for the actual construction of matrices of Green’s type for multiple-
point-posed boundary value problems posed on graphs.
Another alternative procedure can also be used for obtaining matrices of Green’s
type for homogeneous boundary value problems of the type posed by (1.96)–(1.98).
It is based on the method of variation of parameters. To describe the procedure, a
vector function U(x) is introduced, whose components Ui (x), (i = 1, n) are defined
in terms of the solutions u i (x) of the governing equation in (1.96) as
u (x), for x ∈ ei
Ui (x) = i (1.103)
0, for x ∈ R \ ei
We also introduce a vector function F(x) whose components Fi (x) are defined in
terms of the right-hand side functions f i (x) of (1.96) in the form
f i (x), for x ∈ ei
Fi (x) = (1.104)
0, for x ∈ R \ ei
Theorem 1.6 If G(x, ξ) represents the matrix of Green’s type of the homogeneous
multiple-point-posed boundary value problem corresponding to (1.96)–(1.98), then
the solution of the latter posed on R can be written in the integral form
U(x) = G(x, ξ)F(ξ)d R(ξ), x∈R (1.105)
R
where the integration is carried out over the entire graph R. The converse is also true.
That is, if the solution of the problem in (1.96)–(1.98) is obtained in the integral form
of (1.105), then the kernel G(x, ξ) of the integral represents the matrix of Green’s
type for the homogeneous problem corresponding to (1.96)–(1.98).
Proof Upon using the components Ui (x) and Fi (x) of the vector functions U(x) and
F(x) as shown in (1.103) and (1.104), the integral of (1.105) spells out in the scalar
form
n
u i (x) = gi j (x, ξ) f j (ξ)de j (ξ), i = 1, n
j=1 ej
Since the diagonal gii (x, ξ) and the peripheral gi j (x, ξ) elements of the matrix of
Green’s type are defined in a different manner (see (1.99) and (1.100)), we isolate
the ith term of the finite sum in (1.106)
i−1
lj li
u i (x) = gi j (x, ξ) f j (ξ)dξ + gii (x, ξ) f i (ξ)dξ
j=1 0 0
n
lj
+ gi j (x, ξ) f j (ξ)dξ, x ∈ [0, li ], i = 1, n
j=i+1 0
As soon as the diagonal elements of G(x, ξ) are defined in pieces, we break down
the integral containing gii (x, ξ) in the above representation of u i (x) into two additive
terms as shown
i−1 l j
u i (x) = gi j (x, ξ) f j (ξ)dξ
j=1 0
x li
+ gii− (x, ξ) f i (ξ)dξ + gii+ (x, ξ) f i (ξ)dξ
0 x
n
lj
+ gi j (x, ξ) f j (ξ)dξ, x ∈ [0, li ], i = 1, n
j=i+1 0
where gii− (x, ξ) and gii+ (x, ξ) represent the lower and the upper branches of the
diagonal elements of G(x, ξ), which are valid for x ≥ ξ and x ≤ ξ, respectively (see
(1.99)).
To properly differentiate the functions u i (x), we recall the defining properties of
the elements of G(x, ξ) and notice also that the above expression for u i (x) contains
integrals involving parameter and having variable limits. With this in mind, one
obtains
du i (x) ∂gii− (x, ξ)
i−1 lj
∂gi j (x, ξ) x
= f j (ξ)dξ + f i (ξ)dξ
dx j=1 0 ∂x 0 ∂x
−
li
∂gii+ (x, ξ)
+ gii (x, x ) f i (x) + f i (ξ)dξ − gii (x, x+ ) f i (x)
x ∂x
n
lj
∂gi j (x, ξ)
+ f j (ξ)dξ, x ∈ [0, li ], i = 1, n
j=i+1 0
∂x
40 1 Green’s Functions for ODE
x
∂gii− (x, ξ) li
∂gii+ (x, ξ)
+ f i (ξ)dξ + f i (ξ)dξ
0 ∂x x ∂x
n
lj
∂gi j (x, ξ)
+ f j (ξ)dξ, x ∈ [0, li ], i = 1, n
j=i+1 0
∂x
since, in compliance with property 1 of the definition of G(x, ξ), the sum
of the nonintegral terms equals zero. So, the derivative of u i (x) reads as
du i (x)
n lj
∂gi j (x, ξ)
= f j (ξ)dξ, x ∈ [0, li ], i = 1, n (1.107)
dx j=1 0 ∂x
implying that the first derivatives of the integral representations of u i (x) in (1.106)
can be obtained by a straightforward differentiation of the integrands. Consequently,
with the representations for u i (x) and du i (x)/d x shown in (1.106) and (1.107), the
boundary conditions of (1.97) and (1.98) are satisfied.
To find out whether the integral representations of u i (x) shown in (1.106) satisfy
the governing differential equations, we obtain the second derivatives of u i (x)
d 2 u i (x)
i−1 lj
∂ 2 gi j (x, ξ)
= f j (ξ)dξ
dx2 j=1 0 ∂x 2
x
∂ 2 gii− (x, ξ) ∂gii (x, x− )
+ f i (ξ)dξ + f i (x)
0 ∂x 2 ∂x
li
∂ 2 gii+ (x, ξ) ∂gii (x, x+ )
+ f i (ξ)dξ − f i (x)
x ∂x 2 ∂x
n
lj
∂ 2 gi j (x, ξ)
+ f j (ξ)dξ, x ∈ [0, li ], i = 1, n
j=i+1 0
∂x 2
1.2 Multiple-Point-Posed Problems 41
And for the second derivative of u i (x), we finally obtain its compact representation
as
d 2 u i (x)
n lj
∂ 2 gi j (x, ξ) f i (x)
= f j (ξ)dξ − , x ∈ [0, li ], i = 1, n (1.108)
dx2 j=1 0 ∂x 2 pi (x)
Upon substituting the values of u i (x) and their derivatives from (1.106)–(1.108)
into (1.96), we ultimately obtain
n
lj
L[gi j (x, ξ)] f j (ξ)dξ − f i (x) = − f i (x), x ∈ (0, li )
j=1 0
Based on this and following the standard routine of the method of variation of
parameters, one obtains the well-posed systems of linear algebraic equations
u i1 (x) u i2 (x) Di1 (x) 0
× = , i = 1, n
u i1 (x) u i2 (x) Di2 (x) − f i (x)/ pi (x)
42 1 Green’s Functions for ODE
in the derivatives of the functions Di1 (x) and Di2 (x) in (1.109). This yields
and x
u i1 (ξ) f i (ξ)
Di2 (x) = − dξ + E i2 , i = 1, n
0 pi (ξ)Wi (ξ)
Substituting the expressions for Di1 (x) and Di2 (x) just found into (1.109), the
latter can be rewritten as
x x
u i2 (ξ) f i (ξ) u i1 (ξ) f i (ξ)
u i (x) = u i1 (x) dξ − u i2 (x) dξ
0 p i (ξ)W i (ξ) 0 pi (ξ)Wi (ξ)
+E i1 u i1 (x) + E i2 u i2 (x), i = 1, n
By combining the integral terms in the above, the general solutions of the gov-
erning equations in (1.96) are finally obtained in the form
x
u i1 (x)u i2 (ξ) − u i2 (x)u i1 (ξ)
u i (x) = f i (ξ)dξ
0 pi (ξ)Wi (ξ)
Example 1.2.1 Consider a three-point-posed boundary value problem set up for two
Cauchy–Euler equations
1.2 Multiple-Point-Posed Problems 43
d dy1 (x) 1
x − y1 (x) = f 1 (x), x ∈ (0, a) (1.111)
dx dx x
d dy2 (x) 1
x − y2 (x) = f 2 (x), x ∈ (a, ∞) (1.112)
dx dx x
each stated in an individual domain, with boundary and contact conditions imposed
as
lim |y1 (x)| < ∞, lim |y2 (x)| < ∞ (1.113)
x→0 x→∞
in C1 (x) and C2 (x). The solution to this system is found as
f 1 (x) x f 1 (x)
C1 (x) = , C2 (x) = −
2x 2
Integrating and substituting then the expressions for C1 (x) and C2 (x) into
(1.115), we have
x
x 2 − ξ2
y1 (x) = f 1 (ξ)dξ + D11 x + D12 x −1 . (1.116)
0 2xξ
It is not hard to realize that for the second condition in (1.113) to hold, the integral-
containing factor of x in (1.118) must be zero, resulting in
∞
1
D21 = − f 2 (ξ)dξ .
a 2ξ
With the values of D12 and D21 already at hand, the first condition in (1.114)
yields
∞
−1
a 2
a − ξ2 a
D11 a − D22 a =− f 1 (ξ)dξ − f 2 (ξ)dξ (1.119)
0 2aξ a 2ξ
To account for the second condition in (1.114), we first differentiate the expres-
sions in (1.116) and (1.117)
x
x 2 + ξ2
y1 (x) = f 1 (ξ)dξ + D11
0 2ξx 2
and x
x 2 + ξ2
y2 (x) = f 2 (ξ)dξ + D21 − D22 x −2
a 2ξx 2
This yields
∞
a
a2 + ξ2 λ
D11 + λD22 a −2 = − f 1 (ξ)dξ − f 2 (ξ)dξ (1.120)
0 2ξa 2 a 2ξ
The relations in (1.119) and (1.120) form a well-posed system of linear algebraic
equations in D11 and D22 , whose solution is found as
∞
λ a
(a 2 + ξ 2 ) + λ(a 2 − ξ 2 )
D11 = − f 2 (ξ)dξ − f 1 (ξ)dξ
a (1 + λ)ξ 0 2(1 + λ)a 2 ξ
and a
(a 2 + ξ 2 ) − λ(a 2 − ξ 2 )
D22 = − f 1 (ξ)dξ .
0 4λξ
1.2 Multiple-Point-Posed Problems 45
Once all the four just found values of Di j , (i, j = 1, 2) are substituted into (1.116)
and (1.117), we obtain the solution to the problem posed in (1.111)–(1.114) in the
form a
x[(a 2 + ξ 2 ) + λ(a 2 − ξ 2 )]
y1 (x) = − f 1 (ξ)dξ
0 2(1 + λ)a 2 ξ
∞
x − ξ2
x 2
λx
+ f 1 (ξ)dξ − f 2 (ξ)dξ
0 2xξ a (1 + λ)ξ
and a
(a 2 + ξ 2 ) − λ(a 2 − ξ 2 )
y2 (x) = − f 1 (ξ)dξ
0 4λxξ
∞
x
x 2 − ξ2 x
+ f 2 (ξ)dξ − f 2 (ξ)dξ
a 2xξ a 2ξ
Expressing the first two integrals in y1 (x), as well as the last two integrals in y2 (x)
in a single-integral form each, we are able to visualize ultimately the elements
−1
x[(a 2 + ξ 2 ) + λ(a 2 − ξ 2 )][2(1 + λ)a 2 ξ] , for 0 ≤ x ≤ s < a
g11 (x, ξ) = −1
ξ[(a 2 + x 2 ) + λ(a 2 − x 2 )][2(1 + λ)a 2 x] , for 0 < s ≤ x ≤ a,
Example 1.2.2 Construct matrix of Green’s type for a problem that simulates the
steady-state heat-transfer process in the assembly of rods depicted in Fig. 1.2. Each
of the rods is made of a homogeneous material with thermal conductivity h i .
Notice that a local coordinate system is introduced for each edge of the graph R.
This yields the following five-point-posed boundary value problem
d 2 u i (x)
hi = − f i (x), x ∈ (0, li ), i = 1, 4 (1.121)
dx2
46 1 Green’s Functions for ODE
du 3 (0) du 4 (l4 )
u 3 (0) = u 4 (l4 ), h 3 − h4 =0 (1.124)
dx dx
then the above transforms, in compliance with the procedure of the method of vari-
ation of parameters, to
x
ξ−x
u i (x) = f i (ξ)dξ + E i1 + E i2 x, x ∈ (0, li ), i = 1, 4 (1.127)
0 hi
1.2 Multiple-Point-Posed Problems 47
E 11 = E 21 = E 41 = 0.
For the rest of the constants in (1.127), one obtains a well-posed system of linear
algebraic equations which appears as
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
l1 −l2 0 0 0 E 12 A2 − A1
⎜ l1 −1 −l3 0 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 0 ⎟ ⎜ E 22 ⎟ ⎜ A3 − A1 ⎟
⎜ h1 h2 0 h3 ⎜
0 ⎟ × ⎜ E 31 ⎟ = ⎜ B1 + B2 + B3 ⎟
⎟ ⎜ ⎟ (1.128)
⎜ ⎟
⎝0 0 1 0 −l4 ⎠ ⎝ E 32 ⎠ ⎝ A4 ⎠
0 0 0 h3 −h 4 E 42 −B4
To make the remainder of our development more compact, we assume that the
rods in the assembly (the edges of the graph) have equal lengths, that is l1 = l2 =
l3 = l4 = l. This converts the determinant of the coefficient matrix in (1.128) into
= l 2 [(h 1 + h 2 )(h 3 + h 4 ) + h 3 h 4 ] .
After solving the system in (1.128) and substituting thereupon values of the found
constants E i1 and E i2 into (1.127), we have
l
x
u 1 (x) = {∗ − ξ[h 2 (h 3 + h 4 ) + h 3 h 4 ]} f 1 (ξ)dξ
0 ∗ h 1
x
ξ−x l
xξ
+ f 1 (ξ)dξ + (h 3 + h 4 ) f 2 (ξ)dξ
0 h1 0 ∗
l l
x xξ
+ (lh 3 + ξh 4 ) f 3 (ξ)dξ + h 3 f 4 (ξ)dξ, (1.129)
0 ∗ 0 ∗
l
xξ
u 2 (x) = (h 3 + h 4 ) f 1 (ξ)dξ
0 ∗
x
ξ−x l
x
+ f 2 (ξ)dξ + {∗ − ξ[h 1 (h 3 + h 4 ) + h 3 h 4 ]} f 2 (ξ)dξ
0 h2 0 ∗ h 2
48 1 Green’s Functions for ODE
l l
x xξ
+ (lh 3 + ξh 4 ) f 3 (ξ)dξ + h 3 f 4 (ξ)dξ, (1.130)
0 ∗ 0 ∗
l
ξ l
ξ
u 3 (x) = (lh 3 + xh 4 ) f 1 (ξ)dξ + (lh 3 + xh 4 ) f 2 (ξ)dξ
0 ∗ 0 ∗
l
1
+ [l(h 1 + h 2 + h 3 ) − ξ(h 1 + h 2 )](lh 3 + xh 4 ) f 3 (ξ)dξ
0 ∗ h 3
x
ξ−x l
ξ
+ f 3 (ξ)dξ + [l( p1 + p2 + p3 ) − x( p1 + p2 )] f 4 (ξ)dξ, (1.131)
0 p3 0 ∗
and
l l
xξ xξ
u 4 (x) = h 3 f 1 (ξ)dξ + h 3 f 2 (ξ)dξ
0 ∗ 0 ∗
l
x
+ [l(h 1 + h 2 + h 3 ) − ξ(h 1 + h 2 )] f 3 (ξ)dξ
0 ∗
x
ξ−x l
x
+ f 4 (ξ)dξ + [∗ − ξh 3 (h 1 + h 2 )] f 4 (ξ)dξ (1.132)
0 h4 0 ∗ h 4
where ∗ = /l.
At this stage of the derivation procedure, we group the f 1 (ξ) containing integrals
of (1.129), the f 2 (ξ) containing integrals of (1.130), the f 3 (ξ) containing integrals of
(1.131), and the f 4 (ξ) containing integrals of (1.132), in a single-integral form each.
This yields the solution to the five-point-posed boundary value problem of (1.121)–
(1.125) in a single-integral form of (1.126). The elements gi j (x, ξ) of the matrix of
Green’s type G(x, ξ) of the corresponding homogeneous problem can explicitly be
read off from the integral representations in (1.129) through (1.132).
For the sake of compactness, we present only the elements gi1 (x, ξ) of the first
column of G(x, ξ). They are found as
⎧
1 ⎨x{∗ − ξ[h 2 (h 3 + h 4 ) + h 3 h 4 ]}, for x ≤ ξ
g11 (x, ξ) =
∗ h 1 ⎩ξ{∗ − x[h (h + h ) + h h ]}, for x ≥ ξ
2 3 4 3 4
xξ ξ
g21 (x, ξ) = ∗
(h 3 + h 4 ), g31 (x, ξ) = ∗ (lh 3 + xh 4 )
xξ
g41 (x, ξ) = h3
∗
1.2 Multiple-Point-Posed Problems 49
The above stay, in physical terms, for responses of the assembly depicted in
Fig. 1.2 to a unit point source released at a source point ξ arbitrarily located in
the first rod. The rest of the elements of the matrix of Green’s type G(x, ξ),
which represent responses of the assembly to a unit source released at other rods,
could also be directly read off from the integral representations of (1.129) through
(1.132).
Exercising with the ODE problems offered below is important in developing
adequate skills to successfully comprehend the material that the reader will face in
the following chapters.
Exercise 1.1 Determine the well posedness of the boundary value problem:
(a) y (x) = 0, with y (0) = 0 and y (a) + my(a) = 0;
(b) y (x) − k 2 y(x) = 0, with y(0) = 0 and lim x→∞ |y(x)| < ∞;
(c) ((mx + p)y (x)) = 0, with y (0) = 0 and y(a) = 0.
Exercise 1.2 Construct the Green’s function using the defining properties:
(a) y (x) = 0, with y(0) = 0 and y (a) = 0;
(b) y (x) = 0, with y(0) = 0 and y (a) + hy(a) = 0, h ≥ 0;
(c) y (x) = 0, with y (0) − h 1 y(0) = 0 and y (a) + h 2 y(a) = 0;
(d) ((mx + p)y (x)) = 0, with y(0) = y(a) = 0, m > 0 and p > 0.
(e) (x − a)y (x) + y (x) − y(x) = 0, with lim x→a |y(x)| < ∞, and y (b) +
hy(b) = 0, h > 0;
Exercise 1.6 Use the method of variation of parameters to construct the Green’s
function to the boundary value problem:
Exercise 1.7 Based on Theorem 1.4, solve the boundary value problem:
As it follows from the preceding segments in this volume, the lion share of our
work will be devoted to a specific class of applied problems that have never been
touched upon before in standard texts on differential equations. Boundary value prob-
lems will be analyzed as setup for second-order elliptic partial differential equations
that simulate potential fields induced in thin-wall structures. The construction and
implementation of Green’s functions for such problems will be in our focus. While
investigating this topic, one of the primary objectives will be to meet the needs
of practitioners who might be potentially interested in the employment of Green’s
functions in their numerical work.
Chapter 1 was devoted to Green’s functions for ordinary differential equations.
The intention was to maintain a basis for a productive work with partial differential
equations later on. The objective in the current chapter is to describe a workable algo-
rithm developed in [31, 32] for the construction of computer friendly representations
of Green’s functions for problems formulated in regions that represent fragments of
a spherical surface.
The presentation is organized, herein, in such a way that each section deals with
a problem set up for a specific fragment of a spherical surface. The idea behind of
this kind of organization is that different regions on a sphere reveal different peculiar
features of our algorithm. Section 2.1 overviews the algorithm, while each of the
following sections focuses on region’s shape specificities which require individual
considerations. Namely, it is important to know, for example, whether or not the
poles of the sphere represent parts of the region’s boundary. It is also an issue if
the solution of the encountered problem is periodic with respect to one or both the
independent variables.
= {ϑ, ϕ| ϑ0 ≤ ϑ ≤ ϑ1 , 0 ≤ ϕ ≤ ϕ1 }
and
B3 [u (ϑ0 , ϕ)] = 0 and B4 [u (ϑ1 , ϕ)] = 0, (2.3)
and
∞
f (ϑ, ϕ) = f n (ϑ) sin νϕ, (2.6)
n=1
With the above problem formulation, the reader arrives at the familiar territory
of Chap. 1, which has provided us with a necessary experience, and equips with a
workable instrument for dealing with ordinary differential equations. Our next move
will be to properly implement the gained experience while analyzing the problem in
(2.7)–(2.8).
In order to solve the problem in (2.7)–(2.8), we recall the method of variation of
parameters which just represents one of the potential options for that. As the reader
has learned from Chap. 1, this method requires a fundamental set of solutions to the
homogeneous equation corresponding to (2.7). But it is not quite trivial to find two
linearly independent particular solutions representing that set. The variable coeffi-
cients of (2.7) are evidently an issue. However, upon introducing a new independent
variable
ϑ
ω = ln tan ,
2
the governing differential equation in (2.7) reduces to a trivial form, converting the
whole problem in (2.7)–(2.8) to
d 2 u n (ω)
− ν 2 u n (ω) = −
f n (ω) (2.9)
dω 2
B3 [u n (ω1 )] = 0 and
B4 [u n (ω2 )] = 0. (2.10)
2.1 Basics of the Resolving Algorithm 55
In view of the recent change of the independent variable, the backward substitution
reveals the set of functions
ϑ ϑ
tanν and tan−ν
2 2
that represent a fundamental set of solutions to the homogeneous equation corre-
sponding to (2.7). Using the above set of functions and following the standard vari-
ation of parameters procedure, we look for the general solution to the problem in
(2.7) and (2.8) in the form
ϑ ϑ
u n (ϑ) = C1 (ϑ) tanν + C2 (ϑ) tan−ν . (2.11)
2 2
In compliance with the method routine, the form (2.11) yields the well-posed
system of linear algebraic equations
⎛ ϑ ϑ
⎞⎛ ⎞ ⎛ ⎞
tanν 2
tan−ν 2
C1 (ϑ) 0
⎝ ⎠⎝ ⎠=⎝ ⎠
ν tanν ϑ
2
−ν tan−ν ϑ
2
C2 (ϑ) −
f n (ϑ)
in the derivatives of the functions C1 (ϑ) and C2 (ϑ) from (2.11). The solution for this
system appears as
1 ϑ 1 ϑ
C1 (ϑ) = − tan−ν f n (ϑ) and C2 (ϑ) = tanν f n (ϑ)
2ν 2 2ν 2
Upon integrating the above, the functions C1 (ϑ) and C2 (ϑ) themselves are found
as ϑ
1 τ
C1 (ϑ) = − tan−ν f n (τ )dτ + D1
2ν ϑ0 2
and ϑ
1 τ
C2 (ϑ) = tanν f n (τ )dτ + D2
2ν ϑ0 2
Substituting the found expressions for C1 (ϑ) and C2 (ϑ) into (2.11), we come up
with the integral-containing form
ϑ
1 tan2ν (τ /2) − tan2ν (ϑ/2)
u n (ϑ) = f n (τ ) dτ
2ν ϑ0 tanν (ϑ/2) tanν (τ /2)
56 2 Spherical Surface
ϑ ϑ
+ D1 tanν + D2 tan−ν (2.12)
2 2
for the general solution to the equation in (2.7). The constants of integration D1
and D2 can be found upon satisfying the boundary conditions of (2.8). As it follows
from the standard variation of parameters procedure, these constants are expressed
as definite integrals from ϑ0 to ϑ1 of a product of two functions one of which is
f n (τ ) = a 2 sin τ f n (τ ). This allows us to transform the expression for u n (ϑ) from
(2.12) onto the integral-only representation
ϑ1
u n (ϑ) = gn (ϑ, τ ) a 2 sin τ f n (τ ) dτ , (2.13)
ϑ0
where the kernel function gn (ϑ, τ ) represents the Green’s function to the homoge-
neous ODE boundary value problem corresponding to (2.7) and (2.8). It is expressed
in two pieces whose explicit expressions depend on specifics flowing out from the
boundary conditions operators of (2.8). While considering particular problems later
on, we will discuss this issue in more detail, which at this moment look unnecessary.
Note that the functions f n (τ ) in (2.13) represent Fourier coefficients of the series
in (2.6). At this point in our development, they ought to be expressed in terms of the
right-hand side function f (ϑ, ϕ) of the governing equation (2.1). Applying hence,
the Fourier–Euler formula to f n (τ )
ϕ1
2
f n (τ ) = f (τ , ψ) sin νψdψ, n = 1, 2, 3, . . .
ϕ1 0
Substituting now the above form of u n (ϑ) into (2.5), the solution u (ϑ, ϕ) to the
problem of (2.1)–(2.3) is ultimately found as
∞
ϕ1 ϑ1
2
u (ϑ, ϕ) = gn (ϑ, τ ) sin νϕ sin νψ f (τ , ψ) a 2 sin τ dτ dψ
0 ϑ0 ϕ1 n=1
At this point, it is not hard to realize that the integral form we just came up with
in an extended version of (2.4), and its kernel function
∞
2
G (ϑ, ϕ; τ , ψ) = gn (ϑ, τ ) sin νϕ sin νψ. (2.14)
ϕ1 n=1
2.1 Basics of the Resolving Algorithm 57
We proceed now with the implementation of our algorithm (which was sketched in
the preceding section) for the construction of Green’s functions to boundary value
problems that simulate potential fields induced in various fragments of a thin spherical
shell. In doing so, we consider the simply connected region
= {ϑ, ϕ | 0 ≤ ϑ ≤ ϑ1 ; 0 ≤ ϕ ≤ ϕ1 }
a part of the “boundary” for . It will be seen soon that this specific circumstance
notably affects the algorithm.
In the described spherical triangle, we consider a boundary value problem, where
the governing equation
1 ∂ ∂u 1 ∂2u
sin ϑ + = − f (ϑ, ϕ) in (2.15)
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
and
∂u (ϑ1 , ϕ)
lim |u (ϑ, ϕ)| < ∞ and = 0. (2.17)
ϑ→0 ∂ϑ
Note that there is a reason behind expressing the first condition of (2.17) in its
current nontrivial form. The thing is that any standard boundary condition is mean-
ingless at ϑ = 0. This is so because ϑ = 0 represents a point of singularity to the
governing equation in (2.15).
Taking into account the boundary conditions of (2.16), we express the solution
function u (ϑ, ϕ) and the right-hand side function f (ϑ, ϕ) of the governing equation
in the Fourier sine series form
∞
nπϕ
u (ϑ, ϕ) = u n (ϑ) sin (2.18)
n=1
ϕ1
and
∞
nπϕ
f (ϑ, ϕ) = f n (ϑ) sin (2.19)
n=1
ϕ1
where
f n (ϑ) = a 2 sin ϑ f n (ϑ), and n = 1, 2, 3, . . ., subject to the boundary condi-
tions
du n (ϑ1 )
lim |u n (ϑ)| < ∞ and = 0, (2.21)
ϑ→0 dϑ
Clearly, the first condition in (2.21) symbolizes the boundness of u n (ϑ) at the
point of singularity ϑ = 0 of the governing equation.
Following in the footsteps of the procedure described in Sect. 2.1, the general
solution to the equation in (2.20) appears as
ϑ
1 tan2ν (τ /2) − tan2ν (ϑ/2)
u n (ϑ) = f n (τ ) dτ
2ν 0 tanν (ϑ/2) tanν (τ /2)
ϑ ϑ
+ D1 tanν + D2 tan−ν , (2.22)
2 2
The uniqueness conditions from (2.21) allow us to specify the constants of integra-
tion D1 and D2 . Observing the above form, one realizes that the function tan−ν (ϑ/2)
is undefined at ϑ = 0. Thus, the only way to satisfy the first boundary condition in
(2.21), which requires for the solution u n (ϑ) to be bounded at ϑ = 0, is to let D2 = 0.
Upon applying the second condition of (2.21), the constant D1 can also be found.
Omitting a quite trivial algebra, we present just its ultimate expression
ϑ1
1 2n (ϑ1 ) + 2n (τ )
D1 = f n (τ ) dτ
2ν 0 n (τ ) 2n (ϑ1 )
whose kernel function gn (ϑ, τ ), representing, by the way, the Green’s function to
the homogeneous ODE boundary value problem corresponding to (2.20) and (2.21),
is found in two pieces
60 2 Spherical Surface
⎧ −
⎨ gn (ϑ, τ ) , 0 ≤ τ ≤ ϑ
gn (ϑ, τ ) = (2.24)
⎩
gn+ (ϑ, τ ) , ϑ ≤ τ ≤ ϑ1
Turning back to the expression for u n (ϑ) in (2.23), notice that the functions
f n (τ ) , as factors of
f n (τ ) , represent Fourier coefficients of the series in (2.19).
Expressing them in terms of the right-hand side function f (ϑ, ϕ) of the governing
equation in (2.15), we have
ϕ1
2
f n (τ ) = f (τ , ψ) sin νψdψ
ϕ1 0
Substituting the above into (2.18), one arrives at an ultimate representation for
the solution to the problem of (2.15)–(2.17), which reads as
u (ϑ, ϕ) = G (ϑ, ϕ; τ , ψ) f (τ , ψ) dτ ,ψ ,
at all) to make the representation in (2.25) practical and, as a result, more attractive
to potential users.
In what follows, we will show that the series representation of the Green’s function
to the problem in (2.15)–(2.17) which is presented in (2.25) appears completely
summable. It reduces to a computer friendly series-free form expressed in terms of
elementary functions. This summability ensures high potential of our algorithm and
plays an important role in attracting possible Green’s functions users.
To begin with the summation procedure, we refer the reader to the standard
[1, 19] summation formula
∞
pn 1
cos nx = − ln 1 − 2 p cos x + p 2 , (2.26)
n=1
n 2
which is valid for p 2 < 1 and 0 ≤ x < 2π. It is worth noting that this series expansion
will play a substantial role in the development that follows.
Proceeding with the summation itself, let us convert the expansion for the Green’s
function from (2.25) to the equivalent form
∞
1
G (ϑ, ϕ; τ , ψ) = gn (ϑ, τ ) [cos ν(ϕ − ψ) − cos ν(ϕ + ψ)] (2.27)
ϕ1 n=1
Note that the expression for gn (ϑ, τ ) presented in (2.24) is given in two pieces.
Note also that each of the pieces (either gn− (ϑ, τ ) or gn+ (ϑ, τ )) is equivalently feasible
for employment in the coming summation procedure. We take gn− (ϑ, τ ), but before
going any further, transform it to the equivalent form
n n
1 (τ ) (τ ) (ϑ)
gn− (ϑ, τ ) = +
2ν (ϑ) 2 (ϑ1 )
∞
1 1 (τ ) n (τ ) (ϑ) n
− + cos nβ, (2.28)
2π n=1 n (ϑ) 2 (ϑ1 )
where
π(ϕ − ψ) π(ϕ + ψ)
α= and β =
ϕ1 ϕ1
62 2 Spherical Surface
The single-variable function (x) = tanπ/ϕ1 (x/2), just recently introduced in this
section, is increasing in its domain, and the two-variable function gn− (ϑ, τ ) is defined
for τ ≤ ϑ. With this in mind, one realizes that the expansion in (2.28) represents the
sum of four similar summable series of the type in (2.26). The summation, hence,
yields the following closed form
⎧
1 ⎨ (τ ) (τ ) 2
G (ϑ, ϕ; τ , ψ) = ln 1 − 2 cos β +
2π ⎩ (ϑ) (ϑ)
2
(τ ) (ϑ) (τ ) (ϑ)
+ ln 1−2 cos β +
2 (ϑ1 ) 2 (ϑ1 )
2
(τ ) (τ )
− ln 1−2 cos α +
(ϑ) (ϑ)
⎫
2 ⎬
(τ ) (ϑ) (τ ) (ϑ)
− ln 1−2 cos α +
2 (ϑ1 ) 2 (ϑ1 ) ⎭
for the expansion that appeared in (2.28). After an elementary transformation, it can
be rewritten in a more compact form as
⎧
1 ⎨ 2 (ϑ) − 2 (ϑ) (τ ) cos β + 2 (τ )
G (ϑ, ϕ; τ , ψ) = ln
2π ⎩ 2 (ϑ) − 2 (ϑ) (τ ) cos α + 2 (τ )
⎫
4 (ϑ1 ) − 22 (ϑ1 ) (ϑ) (τ ) cos β + 2 (τ ) 2 (ϑ) ⎬
+ ln (2.29)
4 (ϑ1 ) − 22 (ϑ1 ) (ϑ) (τ ) cos α + 2 (τ ) 2 (ϑ) ⎭
representing the Green’s function to the homogeneous boundary value problem cor-
responding to (2.15)–(2.17). The appearance of the form in (2.29) speaks for itself.
It is compact and closed in the conventional sense, implying that it is expressed in
terms of elementary functions. That is why we can call this form computer friendly.
While observing the above expression, the reader is advised to turn back to that
point in our recent development at which we had chosen the piece gn− (ϑ, τ ) of
the two-piece-defined function gn (ϑ, τ ). We claimed over there that each piece of
gn (ϑ, τ ) is equivalently eligible for use. Indeed, in view of the symmetry of (2.29),
the interchange of ϑ with τ does not affect it at all. If so then it becomes absolutely
clear that giving preference to the piece gn+ (ϑ, τ ) instead, would never affect the
ultimate form of (2.29).
2.2 Triangular Shaped Region 63
3
K j G ϑ, ϕ; τ j , ψ j
j=1
of three profiles of the Green’s function shown in (2.29) represents the field.
Appendix segment of this volume provides the reader with an extensive list of
computer friendly representations of Green’s functions constructed for a number of
boundary value problems that simulate potential fields induced in elements of thin
shells. Among of many others, Green’s functions for some different of (2.15)–(2.17)
problems posed in the spherical triangle are also available over there.
The work is continued in this section on the construction of Green’s functions for
problems stated in regions belonging to a spherical surface. Another practically
important shape of a spherical region is brought to the reader’s attention. That is the
double-connected region
64 2 Spherical Surface
belonging to a spherical surface of radius a and depicted in Fig. 2.4. The region
is bounded with two parallels ϑ = ϑ0 and ϑ = ϑ1 , and will be referred to as the
spherical belt.
An important feature of any boundary value problem posed in is that its solution
has to be 2π-periodic with respect to the longitudinal coordinate ϕ. This brings
new nuances to our approach, which must, of course, be somewhat different of that
developed in Sect. 2.2.
For an illustrative example, let the equation
1 ∂ ∂u 1 ∂2u
sin ϑ + = − f (ϑ, ϕ) in , (2.30)
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
with an integrable in right-hand side function f (ϑ, ϕ), be subject to the Neumann
and the Dirichlet conditions
∂u (ϑ0 , ϕ)
= 0 and u (ϑ1 , ϕ) = 0 (2.31)
∂ϑ
imposed on the boundary fragments ϑ = ϑ0 and ϑ = ϑ1 of , respectively. The
conditions
u (ϑ, 0) − u (ϑ, 2π) = 0 (2.32)
and
∂u (ϑ, 0) ∂u (ϑ, 2π)
− =0 (2.33)
∂ϕ ∂ϕ
have also to be imposed to reflect the 2π-periodicity of the solution we are looking
for.
2.3 Belt-Shaped Region 65
∞ ∞
1
u (ϑ, ϕ) = u 0 (ϑ) + u (c)
n (ϑ) cos nϕ + u (s)
n (ϑ) sin nϕ (2.34)
2 n=1 n=1
and
∞ ∞
1
f (ϑ, ϕ) = f 0 (ϑ) + f n(c) (ϑ) cos nϕ + f n(s) (ϑ) sin nϕ. (2.35)
2 n=1 n=1
Evidently, the expansion in (2.34) complies with the conditions in (2.32) and
(2.33). Substituting now the above expansions into (2.30) and (2.31), one arrives at
the ODE boundary value problem
d du n (ϑ) n 2 u n (ϑ)
sin ϑ − =−
f n (ϑ) , n = 0, 1, 2, . . . (2.36)
dϑ dϑ sin ϑ
du n (ϑ0 )
= 0 and u n (ϑ1 ) = 0 (2.37)
dϑ
in the coefficients u n (ϑ) of the series in (2.34). The reader, who delves into every
detail, notices perhaps that the functions u n (ϑ) and f n (ϑ) in (2.36) and (2.37) are
not decorated with the superscripts (c) and (s) . This is so because for both the cosine
and sine coefficients of the Fourier series of (2.34) and (2.35), we arrive at the same
problem in u n (ϑ), making actually needless the use of these subscripts.
Before we proceed to the solution of the problem in (2.36) and (2.37), it is worth
noting that its treatment for the cases of n = 0 and of n ≤ 1 ought to be different.
Why so? To answer this question, the reader is advised to recall that the standard
method of variation of parameters is in use at this very stage of our procedure (see
Sects. 2.1 and 2.2). And a substantial element of this method is a fundamental set
of solutions to the homogeneous equation corresponding to (2.36). But fundamental
sets of solutions for the cases of n = 0 and n ≤ 1 are obviously different.
Let us focus on the case of n = 0 first. The setting in (2.36) and (2.37) reduces,
in this case, to
d du 0 (ϑ)
sin ϑ =− f 0 (ϑ) (2.38)
dϑ dϑ
du 0 (ϑ0 )
= 0 and u 0 (ϑ1 ) = 0 (2.39)
dϑ
66 2 Spherical Surface
and in compliance with the method of variation of parameters, we thus arrive at the
general solution to (2.38) in the form
ϑ
0 (τ )
u 0 (ϑ) = ln f 0 (τ ) dτ + D1 ln 0 (ϑ) + D2 ,
ϑ0 0 (ϑ)
where ⎧
⎪ 0 (ϑ1 )
⎨ ln 0 (τ )
, if ϑ ≤ τ
g0 (ϑ, τ ) = (2.41)
⎪
⎩ ln 0 (ϑ1 )
0 (ϑ)
, if τ ≤ ϑ
represents the Green’s function to the homogeneous ODE boundary value problem
corresponding to (2.36) and (2.37) in the case of n = 0.
The case of n ≥ 1 turns us back to the problem in (2.36) and (2.37) in its current
form, with n representing an integer parameter. Analogously to the case of n = 0 just
completed, the solution u n (ϑ) of the setting in (2.36) and (2.37) can also be obtained
with the aid of the method of variation of parameters whose procedure was already
explained for this type of problems in our preceding sections. That is why we omit
a lengthy but very straightforward routine and present just the ultimate form
2.3 Belt-Shaped Region 67
ϑ1
u n (ϑ) = gn (ϑ, τ )
f n (τ ) dτ (2.42)
ϑ0
for u n (ϑ) expressed in terms of the Green’s function gn (ϑ, τ ) of the homogeneous
ODE problem corresponding to (2.36) and (2.37). It is customarily found in two
pieces. Its expression valid for ϑ0 ≤ τ ≤ ϑ ≤ ϑ1 reads as
n0 (τ ) n0 (ϑ0 ) n0 (ϑ1 ) n0 (ϑ)
× + − n , (2.43)
n0 (ϑ0 ) n0 (τ ) n0 (ϑ) 0 (ϑ1 )
∞
1
G (ϑ, ϕ; τ , ψ) = g0 (ϑ, τ ) + gn (ϑ, τ ) cos nϕ cos nψ
2 n=1
∞
+ gn (ϑ, τ ) sin nϕ sin nψ
n=1
∞
1
= g0 (ϑ, τ ) + gn (ϑ, τ ) cos n (ϕ − ψ) (2.44)
2 n=1
So, at this stage in our procedure, similarly to the situation in Sect. 2.2, the Green’s
function that we are looking for is obtained in a series form. Note that both the
series in (2.28) and (2.44) are nonuniformly convergent in and both have the
same convergence rate. But in contrast to the form in (2.28), which was completely
summed up in Sect. 2.2, the form in (2.44) does not allow a complete summation. A
significant increase of its convergence rate is nevertheless possible notably enhancing
its computational potential.
Convergence of series representing Green’s functions for partial differential equa-
tions was always an issue for researchers (see, for example, [21, 29]). In [32, 34],
we proposed a special technique aimed at the convergence improvement for series of
68 2 Spherical Surface
the type in (2.44). We are not going to describe this technique in full detail in relation
to the current situation, but its brief sketch would, in our opinion, help the reader.
To figure out a feature which makes the series in (2.44) hard to sum up, let us take
a close look at both (2.28) and (2.44), and compare. The form in (2.28) contains four
series of the type
∞
1 P1 n
cos nx,
n=1
n P2
where ||P1 || < ||P2 || and 0 ≤ x < 2π. This makes the series in (2.28) completely
summable with the aid of the standard summation formula of (2.26) which was
already implemented in Sect. 2.2. The series in (2.44) is, however, different. Its type
can be expressed as
∞
1 P1n P2n
cos nx, (2.45)
n=1
n P22n + P12n
where ||P1 || < ||P2 || and 0 ≤ x < 2π. It is evident that the formula in (2.26) appears
useless in the case of (2.45). But upon transforming it in the way shown below
∞
1 P1n P2n
cos nx
n=1
n P22n + P12n
∞
1 P1n P2n P1n P1n
= − + cos nx
n=1
n P22n + P12n P2n P2n
∞ ∞
1 P1n P2n P1n 1 P1 n
= − n cos nx + cos nx
n=1
n P22n + P12n P2 n=1
n P2
∞ ∞
1 P13n 1 P1 n
=−
cos nx + cos nx , (2.46)
n=1
n P2n P22n + P12n n=1
n P2
we managed to decompose the series in (2.45) onto two other series, one of which is
of the familiar type allowing the complete summation, whereas the convergence rate
of the other series in (2.46) is evidently higher then that of (2.45) making it uniformly
convergent in .
Applying the technique, a brief sketch of which was just presented, to the series
in (2.44), after a tedious but quite straightforward algebra, we obtain the ultimate
form ∞
1
G (ϑ, ϕ; τ , ψ) = Rn (ϑ, τ ) cos n (ϕ − ψ)
2π n=1
2.3 Belt-Shaped Region 69
20 (ϑ)40 (ϑ0 ) − 220 (ϑ0 )20 (ϑ1 )0 (ϑ)0 (τ ) cos(ϕ − ψ) + 20 (τ )40 (ϑ1 )
+ ln
40 (ϑ1 ) 20 (ϑ) − 20 (ϑ)0 (τ ) cos(ϕ − ψ) + 20 (τ )
20 (ϑ)20 (τ ) − 220 (ϑ1 )0 (ϑ)0 (τ ) cos(ϕ − ψ) + 40 (ϑ1 )
+ ln (2.47)
20 (ϑ)20 (τ ) − 220 (ϑ0 )0 (ϑ)0 (τ ) cos(ϕ − ψ) + 40 (ϑ0 )
of the Green’s function to the boundary value problem in (2.30)–(2.33). The coeffi-
cient Rn (ϑ, τ ) of the series component in (2.47) is defined in two pieces. The piece
valid for ϑ0 ≤ τ ≤ ϑ ≤ ϑ1 reads as
2n 2n
0 (ϑ0 ) 0 (ϑ) − 0 (ϑ1 ) 0 (τ ) + 0 (ϑ0 )
2n 2n 2n
Rn (ϑ, τ ) = 2n
n 2n0 (ϑ1 ) 0 (ϑ) 0 (τ ) 0 (ϑ0 ) + 0 (ϑ1 )
n n 2n
while the expression for Rn (ϑ, τ ) valid for ϑ0 ≤ ϑ ≤ τ ≤ ϑ1 can be obtained from
the above by interchanging ϑ and τ in its numerator. The denominator is evidently
indifferent to this interchange.
Observing the form in (2.47), one might get an impression that it is too heavy
loaded for computer implementations, but we can refute such an impression. One of
the arguments for this is that the logarithmic components, looking a sort of heavy
loaded, are expressed in fact in terms of elementary functions, making them easy
to compute. As to the series component, we have already highlighted its uniform
convergence in .
Hence, the form in (2.47) is indeed computer friendly, allowing direct implemen-
tations in a numerical work.
Illustrating the computability of (2.47), we present, in Fig. 2.5, a potential field
induced by a single unit point source in a thin-shell element whose middle surface
Fig. 2.5 The field induced by a point source in the spherical belt
70 2 Spherical Surface
is a spherical belt. Its shape is defined by ϑ0 = 0.2π and ϑ1 = 0.5π, and the source
is positioned at (0.35π, 0.55π).
= {ϑ, ϕ | ϑ0 ≤ ϑ ≤ ϑ1 ; 0 ≤ ϕ ≤ ϕ1 }
∂u (ϑ, ϕ1 )
u (ϑ, 0) = 0, = 0, (2.49)
∂ϕ
and
∂u (ϑ1 , ϕ)
u (ϑ0 , ϕ) = 0, = 0. (2.50)
∂ϑ
Aiming at a separation of variables for the above problem and taking into account
the Dirichlet–Neumann combination of boundary conditions in (2.49), we expand the
solution function u (ϑ, ϕ) and the right-hand side f (ϑ, ϕ) of the governing equation
in the Fourier sine series form
∞
u (ϑ, ϕ) = u n (ϑ) sin νϕ (2.51)
n=1
and
∞
f (ϑ, ϕ) = f n (ϑ) sin νϕ, (2.52)
n=1
where the factor ν of the argument of the sine function is specifically defined in terms
of the summation index n of the above series as
ν = (2n − 1) π/(2ϕ1 )
2.4 Quadrilateral-Shaped Region 71
This makes u (ϑ, ϕ) in (2.51) complying with the conditions in (2.49) and yields
the boundary value problem
d du n (ϑ) ν 2 u n (ϑ)
sin ϑ − =−
f n (ϑ) in (ϑ0 , ϑ1 )
dϑ dϑ sin ϑ
du n (ϑ1 )
u n (ϑ0 ) = 0 and =0
dϑ
for the sought-after Green’s function of the homogeneous boundary value problem
corresponding to (2.48)–(2.50).
Thus, analogously to the situation that took place in Sect. 2.3 with the series of
(2.44), we are once again at the familiar point in the development where the Green’s
function of our interest is expressed in the form of a nonuniformly convergent series.
A certain effort is required therefore towards an improvement of its convergence.
72 2 Spherical Surface
In doing so, we will focus on a partial summation of the series in (2.55). It looks
like it might potentially be accomplished in the way proposed earlier in Sect. 2.3.
Indeed, the transformation sketched over there (see (2.46)) appears also workable
for the series in (2.55) splitting it onto two other series. One of them converges
uniformly and can therefore be considered as computer friendly. But, as to the second
of those series, the summation formula from (2.26) is not unfortunately immediately
applicable.
With all the foregoing comments in mind, we recall another standard summation
formula [1, 19]
∞
p 2n−1 1 1 + 2 p cos x + p 2
cos(2n − 1)x = ln , (2.56)
n=1
2n − 1 4 1 − 2 p cos x + p 2
which presumes that its parameters p and x satisfy the conditions: p 2 < 1 and 0 ≤
x < 2π. It perfectly fits the situation with the second of the two series resulting from
the just mentioned splitting of the expansion in (2.55).
Omitting a tedious algebra which resembles, in most details, the work done on
the series from (2.44) in Sect. 2.3, we reveal just an ultimate representation for the
Green’s function to the problem in (2.48)–(2.50). It was found in the form
∞
1
G (ϑ, ϕ; τ , ψ) = Rn (ϑ, τ ) sin νϕ sin νψ
2π n=1
√
(ϑ)(τ ) α β (ϑ) α β
+ ln H , , H , ,
(ϑ1 ) 2 2 (τ ) 2 2
(ϑ0 ) α β (ϑ0 ) (ϑ) α β
− ln H √ , , H , , , (2.57)
(ϑ)(τ ) 2 2 (ϑ1 ) (τ ) 2 2
where the three-variable function H (x, ξ, η), in terms of which the arguments of the
logarithmic functions are expressed, is defined as
while its expression valid for τ ≤ ϑ can customarily be obtained by the interchange
of the variables ϑ and τ .
In Fig. 2.6, the reader finds a profile of the just obtained Green’s function, where the
parameters defining the region are given as: ϑ0 = 0.2π, ϑ1 = 0.5π, and ϕ1 = 0.5π,
with a unit source released at the point (0.35π, 0.35π).
Note that the boundary condition operators are of either Dirichlet or Neumann kind
in all the problems encountered so far in this chapter. But, as we have asserted earlier
in Sect. 2.1, the technique, we propose for the construction of Green’s functions, is
potentially applicable to problems with Robin condition imposed as well. And it is
about right time to confirm this assertion. For illustration, we consider a problem
stated in the region
on a sphere of radius a, where 0 < ϑ1 < π. The region is simply connected and
bounded with the parallel ϑ = ϑ1 in spherical coordinates. This shape will be referred
to as the spherical cap.
The shape of combines two specific features each of which has been touched
upon earlier in this chapter. Namely, analogously to the spherical belt considered in
Sect. 2.3, the spherical cap is closed in the longitudinal direction, and as well as in
74 2 Spherical Surface
the case of a spherical triangle (see Sect. 2.2), the north pole ϑ = 0 represents a part
of the “boundary” for .
Our objective is the Green’s function to the homogeneous boundary value problem
corresponding to the well-posed setting
1 ∂ ∂u 1 ∂2u
sin ϑ + = − f (ϑ, ϕ) , (ϑ, ϕ) ∈ (2.58)
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
∂u (ϑ1 , ϕ)
lim |u (ϑ, ϕ) | < ∞ and u (ϑ1 , ϕ) + λ =0 (2.59)
ϑ→0 ∂ϑ
and
∂u (ϑ, 0) ∂u (ϑ, 2π)
u (ϑ, 0) = u (ϑ, 2π) and = , (2.60)
∂ϕ ∂ϕ
∞ ∞
1 (c)
u (ϑ, ϕ) = u 0 (ϑ) + u n (ϑ) cos nϕ + u (s)
n (ϑ) sin nϕ (2.61)
2 n=1 n=1
and
∞ ∞
1
f (ϑ, ϕ) = f 0 (ϑ) + f n(c) (ϑ) cos nϕ + f n(s) (ϑ) sin nϕ (2.62)
2 n=1 n=1
are required, in this case, for the solution function u (ϑ, ϕ) and the right-hand side
f (ϑ, ϕ) of (2.58).
This yields the ODE boundary value problem
d du n (ϑ) n 2 u n (ϑ)
sin ϑ − =−
f n (ϑ) , n = 0, 1, 2, . . . (2.63)
dϑ dϑ sin ϑ
du n (ϑ1 )
lim |u n (ϑ) | < ∞ and u n (ϑ1 ) + λ =0 (2.64)
ϑ→0 dϑ
∞
1
G (ϑ, ϕ; τ , ψ) = g0 (ϑ, τ ) + gn (ϑ, τ ) cos n (ϕ − ψ) , (2.65)
2 n=1
2.5 Robin Problem for Spherical Cap 75
where g0 (ϑ, τ ) and gn (ϑ, τ ) represent Green’s functions to the homogeneous prob-
lem settings corresponding to (2.63) and (2.64), for the cases n = 0 and n ≥ 1,
respectively.
Constructing the Green’s function gn (ϑ, τ ) to the actual setting in (2.63) and
(2.64) by the method variation of parameters, we customarily obtain the general
solution to (2.63) as
ϑ
1 2n
0 (τ ) − 0 (ϑ)
2n
u n (ϑ) = f n (τ ) dτ + D1 n0 (ϑ) + D2 −n
0 (ϑ), (2.66)
2n 0 n0 (τ )n0 (ϑ)
Upon substituting the just presented values of the constants of integration D1 and
D2 into (2.66), the latter reduces to
ϑ
1 sin ϑ1 − nλ n0 (ϑ)n0 (τ ) n0 (τ )
u n (ϑ) = − n f n (τ ) dτ
2n 0 sin ϑ1 + nλ 2n0 (ϑ1 ) 0 (ϑ)
ϑ1
1 sin ϑ1 − nλ n0 (ϑ)n0 (τ ) n0 (ϑ)
+ − n f n (τ ) dτ (2.67)
2n ϑ sin ϑ1 + nλ 2n0 (ϑ1 ) 0 (τ )
This reveals the Green’s function gn (ϑ, τ ) to the homogeneous ODE problem
corresponding to (2.63) and (2.64). Its representation valid for ϑ ≤ τ appears as
1 sin ϑ1 − nλ n0 (ϑ)n0 (τ ) n0 (ϑ)
gn (ϑ, τ ) = − n , (2.68)
2n sin ϑ1 + nλ 2n0 (ϑ1 ) 0 (τ )
whereas the expression of gn (ϑ, τ ) valid for τ ≤ ϑ can be obtained from the above
by interchanging ϑ with τ .
We turn now to the component g0 (ϑ, τ ) of the series expansion in (2.65). It
represents the Green’s function to the homogeneous boundary value problem corre-
sponding to
d du 0 (ϑ)
sin ϑ =− f 0 (ϑ) (2.69)
dϑ dϑ
du 0 (ϑ1 )
lim |u 0 (ϑ) | < ∞ and u 0 (ϑ1 ) + λ =0 (2.70)
ϑ→0 dϑ
76 2 Spherical Surface
The above problem is what the setting in (2.63) and (2.64) reduces to in the case of
n = 0. Via the method of variation of parameters, the general solution of the equation
in (2.69) appears as
ϑ
0 (τ )
u 0 (ϑ) = ln f 0 (τ ) dτ + D1,0 ln 0 (ϑ) + D2,0
0 0 (ϑ)
The boundness condition in (2.70) implies D1,0 = 0, while the Robin condition
at ϑ = ϑ1 leads to
ϑ1
0 (ϑ1 ) λ
D2,0 = ln + f 0 (τ ) dτ
0 0 (τ ) sin ϑ1
resulting in
ϑ
0 (ϑ1 ) λ
u 0 (ϑ) = ln + f 0 (τ ) dτ
0 0 (ϑ) sin ϑ1
ϑ1
0 (ϑ1 ) λ
+ ln + f 0 (τ ) dτ
0 0 (τ ) sin ϑ1
for the Green’s function to the homogeneous ODE problem corresponding to (2.69)
and (2.70) in the case of n = 0.
So, with both g0 (ϑ, τ ) and gn (ϑ, τ ) at hand, the form in (2.65) gives us an
explicit nonuniformly convergent series representation for the Green’s function
G (ϑ, ϕ; τ , ψ) to the homogeneous problem corresponding to (2.58)–(2.60). To
enhance the computational potential of the series in (2.65), we transform the expres-
sion from (2.68) for gn (ϑ, τ ), valid for ϑ ≤ τ , to the equivalent one
1 2nλ n0 (ϑ)n0 (τ ) n0 (ϑ)
gn (ϑ, τ ) = 1− − n
2n sin ϑ1 + nλ 2n0 (ϑ1 ) 0 (τ )
This yields
1 0 (ϑ1 ) λ
G (ϑ, ϕ; τ , ψ) = ln +
2 0 (τ ) sin ϑ1
∞
1 1 n0 (ϑ)n0 (τ ) n0 (ϑ)
+ − cos n (ϕ − ψ)
2π n=1 n 2n
0 (ϑ1 ) n0 (τ )
2.5 Robin Problem for Spherical Cap 77
∞
λ n0 (ϑ)n0 (τ )
− cos n (ϕ − ψ) ,
2π n=1 (sin ϑ1 + nλ) 2n 0 (ϑ1 )
where the first of the two series is completely summable, while the second is uni-
formly convergent for any mutual location of the observation and the source points
inside of . The partial summation ultimately yields
∞
λ λ n0 (ϑ)n0 (τ )
G (ϑ, ϕ; τ , ψ) = + cos n (ϕ − ψ)
sin ϑ1 2π n=1 (sin ϑ1 + nλ) 2n 0 (ϑ1 )
1 40 (ϑ1 ) − 220 (ϑ1 )0 (ϑ)0 (τ ) cos (ϕ − ψ) + 20 (ϑ)20 (τ )
+ ln (2.72)
2π 20 (ϑ1 ) 20 (τ ) − 20 (ϑ)0 (τ ) cos (ϕ − ψ) + 20 (ϑ)
∂u (ϑ1 , ϕ)
lim |u (ϑ, ϕ) | < ∞ and =0
ϑ→0 ∂ϑ
which the statement in (2.58)–(2.60) reduces to in this case, is ill-posed, and its
classical Green’s function does not therefore exist.
78 2 Spherical Surface
The term spherical sector will be used in this section in reference to the simply
connected region
= {ϑ, ϕ | 0 < ϑ ≤ π; 0 ≤ ϕ < ϕ1 }
and
∞
nπϕ
f (ϑ, ϕ) = f n (ϑ) sin
n=1
ϕ1
where gn (ϑ, τ ) represents the Green’s functions to the homogeneous problem set-
tings corresponding to (2.77) and (2.78).
Our customary routine, based on the method of variation of parameters, provides
us with the general solution to (2.77) in the form
ϑ
1 2n (τ ) − 2n (ϑ)
u n (ϑ) = f n (τ ) dτ + D1 n (ϑ) + D2 −n (ϑ),
2ν 0 n (τ )n (ϑ)
which implies
1 ϑ
n (ϑ)
lim − f n (τ ) dτ + D 1 n
(ϑ) <∞
ϑ→π 2ν 0 (τ )
n
we have π
1 1 (τ ) dτ
D1 = fn
2ν 0 n (τ )
This ultimately transforms the solution of the problem in (2.77) and (2.78) to the
integral form
ϑ π
1 n (τ ) 1 n (ϑ)
u n (ϑ) = f n (τ ) dτ + f n (τ ) dτ
2ν 0 (ϑ)
n 2ν ϑ n (τ )
80 2 Spherical Surface
So, substituting the above into (2.79), one arrives at the series representation of
the Green’s function G (ϑ, ϕ; τ , ψ) to the homogeneous boundary value problem
corresponding to (2.73)–(2.75), which appears summable. The summation can read-
ily be accomplished with the aid of the standard summation formula (2.26) which
was presented earlier in Sect. 2.2 and has repeatedly been used in this chapter. The
summation yields
1 2 (ϑ) − 2(ϑ)(τ ) cos β + 2 (τ )
G (ϑ, ϕ; τ , ψ) = ln , (2.80)
2π 2 (ϑ) − 2(ϑ)(τ ) cos α + 2 (τ )
where π π
α= (ϕ − ψ) and β = (ϕ + ψ) .
ϕ1 ϕ1
It is interesting to note that the development of this section appears workable for
another boundary value problem stated in . Namely, it works smoothly if in (2.75)
the condition on the meridian ϕ = 0 stays unchanged, but instead of the Dirichlet
condition on the meridian ϕ = ϕ1 , the Neumann condition
∂u(ϑ, ϕ1 )
u(ϑ, 0) = 0 and =0 (2.81)
∂ϕ
is imposed.
Omitting details, we present just the ultimate closed form
√
1 (ϑ) + 2 (ϑ)(τ ) cos (α/2) + (τ )
G (ϑ, ϕ; τ , ψ) = ln √
2π (ϑ) − 2 (ϑ)(τ ) cos (α/2) + (τ )
√
1 (ϑ) + 2 (ϑ)(τ ) cos (β/2) + (τ )
− ln √ (2.82)
2π (ϑ) − 2 (ϑ)(τ ) cos (β/2) + (τ )
For the homogeneous equation corresponding to (2.1), construct the Green’s func-
tions to the boundary value problem stated in the indicated region:
Exercise 2.1
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
= {0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 2.2
∂u (ϑ, ϕ1 )
u (ϑ, 0) = 0 and =0
∂ϕ
∂u (ϑ1 , ϕ)
lim |u (ϑ, ϕ) | < ∞ and =0
ϑ→0 ∂ϑ
= {0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 2.3
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ1 , ϕ)
lim |u (ϑ, ϕ) | < ∞ and + βu(ϑ1 , ϕ) = 0
ϑ→0 ∂ϑ
= {0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 2.4
∂u (ϑ, 0) ∂u (ϑ, 2π)
u (ϑ, 0) = u (ϑ, 2π) and =
∂ϕ ∂ϕ
82 2 Spherical Surface
Exercise 2.5
∂u (ϑ, 0) ∂u (ϑ, 2π)
u (ϑ, 0) = u (ϑ, 2π) and =
∂ϕ ∂ϕ
∂u (ϑ1 , ϕ)
u (ϑ0 , 0) = 0 and u (ϑ1 , ϕ) + β =0
∂ϑ
in the spherical belt
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ < 2π} .
Exercise 2.6
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ1 , ϕ)
u (ϑ0 , ϕ) = 0 and =0
∂ϑ
in the spherical quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 2.7
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ0 , ϕ) ∂u (ϑ1 , ϕ)
u (ϑ0 , ϕ) + β = 0 and =0
∂ϑ ∂ϑ
in the spherical quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 2.8
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ0 , ϕ)
u (ϑ0 , ϕ) + β = 0 and u (ϑ1 , ϕ) = 0
∂ϑ
in the spherical quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
2.7 Chapter Exercises 83
Exercise 2.9
∂u (ϑ, ϕ1 )
u (ϑ, 0) = 0 and =0
∂ϕ
∂u (ϑ0 , ϕ)
u (ϑ0 , ϕ) + β = 0 and u (ϑ1 , ϕ) = 0
∂ϑ
in the spherical quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 }.
Exercise 2.10
u (ϑ, ϕ1 )
u (ϑ, 0) = 0 and =0
∂ϕ
= {0 < ϑ < π, 0 ≤ ϕ ≤ ϕ1 } .
Chapter 3
Toroidal Surface
The emphasis in the previous chapter was on the investigation of potential phe-
nomena taken place in a variety of fragments of a spherical surface. We concen-
trated in Chap. 2 on the development of a workable algorithm for the construction of
computer-friendly representations of Green’s functions for boundary value problems
that simulate relevant potential fields. This gives the reader a valuable knowledge
and accumulates necessary skills for implementing our experience to problems of
potential formulated on surfaces of some different geometries and configurations.
It is worth noting that a toroidal surface is mathematically the knottiest of all
standard surfaces of revolution, making intricate differential equations that simulate
physical phenomena occurring in relevant shell structures. This is so, in particular,
because of a specificity of the toroidal surface which combines regions of both
positive and negative Gaussian curvature.
We believe that having studied a variety of situations with a spherical surface,
the reader has been efficiently prepared for dealing with potential fields induced in
fragments of a circular toroidal surface as well. To make the transaction between these
two surfaces smoother, and to achieve a more notable progress in comprehending the
content of the present chapter, it is recommended to review the material of Chap. 2
prior to jumping into details of our upcoming developments.
toroidal surface which represents the shell’s middle surface. Geographical coordinate
system is introduced on the surface in the way shown in Fig. 3.1.
Location of a point on a circular toroidal surface will be determined by the geo-
graphical coordinates ϑ (the latitude) and ϕ (the longitude).
On the described toroidal surface, we consider a fragment occupying the simply
connected region
= {(ϑ, ϕ)|ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 }
and
u (ϑ0 , ϕ) = 0 and u (ϑ1 , ϕ) = 0 (3.3)
where the function D(ϑ) in (3.1) is introduced in terms of the radii a and R of the
encountered surface and the latitudinal variable ϑ as
D(ϑ) = R + a sin ϑ
3.1 Quadrilateral-Shaped Region 87
and
∞
f (ϑ, ϕ) = f n (ϑ) sin νϕ (3.5)
n=1
where the parameter ν is directly proportional to the series summation index n, that
is, ν = nπ/ϕ1 . This gives birth to the following self-adjoint ODE boundary value
problem
1 d du n (ϑ) ν2a
D (ϑ) − u n (ϑ) = −
f n (ϑ) , n = 1, 2, 3, . . . (3.6)
a dϑ dϑ D (ϑ)
where
2a a + R tan ϑ/2
ω(ϑ) = √ arctan √ (3.9)
R2 − a2 R2 − a2
where the constants C1 and C2 can be determined upon satisfying the boundary
conditions in (3.7). This yields the well-posed system of linear algebraic equations
⎛ ⎞⎛ ⎞ ⎛ ⎞
eνω(ϑ0 ) e−νω(ϑ0 ) C1 0
⎝ ⎠⎝ ⎠=⎝ ⎠
eνω(ϑ1 ) e−νω(ϑ1 ) C2 I (ϑ0 , ϑ1 )
in C1 and C2 , where
ϑ1
1
I (ϑ0 , ϑ1 ) = sinh ν (ω (ϑ1 ) − ω (τ ))
f n (τ ) dτ
ν ϑ0
and ϑ1
eνω(ϑ0 )
C2 = sinh ν (ω (ϑ1 ) − ω (τ ))
f n (τ ) dτ
2ν ϑ0
where
= sinh ν[ω (ϑ1 ) − ω (ϑ0 )]
By substituting the just found expressions for C1 and C2 into (3.10), the solution
of the boundary value problem in (3.6) and (3.7) reduces to
ϑ
1
u n (ϑ) = sinh ν[ω (ϑ0 ) − ω (τ )] sinh ν[ω (ϑ1 ) − ω (ϑ)]
f n (τ ) dτ
ν ϑ0
ϑ1
+ sinh ν[ω (ϑ0 ) − ω (ϑ)] sinh ν[ω (ϑ1 ) − ω (τ )]
f n (τ ) dτ
ϑ
by introducing the kernel function gn (ϑ, τ ) defined in two pieces. Its expression
valid for ϑ0 ≤ τ ≤ ϑ is found as
1
gn (ϑ, τ ) = {sinh ν[ω (ϑ0 ) − ω (τ )] sinh ν[ω (ϑ1 ) − ω (ϑ)]} (3.12)
ν
while the one valid for ϑ0 ≤ τ ≤ ϑ can be obtained from the above by interchanging
ϑ with τ . This symmetry feature of gn (ϑ, τ ) agrees with the self-adjointness of
the ODE setup in (3.6) and (3.7) for which it serves as the Green’s function to the
corresponding homogeneous problem.
Expressing now the Fourier coefficients f n (τ ) of the series from (3.5) in terms
of the series-generating function f (τ , ψ)
ϕ1
2
f n (τ ) = f (τ , ψ) sin νψdψ, n = 1, 2, 3, . . .
ϕ1
0
we recall the introduced earlier in this section relation f n (ϑ) = [D(ϑ)/a] f n (ϑ), and
substitute u n (ϑ) from (3.11) into (3.4). This yields the following compact represen-
tation
ϕ1 ϑ1
∞
2
u (ϑ, ϕ) = gn (ϑ, τ ) sin νϕ sin νψ f (τ , ψ)[D(τ )/a]dτ dψ
ϕ1 n=1
0 ϑ0
cannot uniformly converge, implying that the closer the source point (τ , ψ) is to the
observation point (ϑ, ϕ), the lower is the series convergence rate.
Hence, a potential Green’s function user might be hesitant as to practical applica-
tions of the form in (3.14) to a numerical work. In other words, an improvement of
the series convergence is required to make it more attractive. This can be done by a
partial summation whose procedure had been used for a number of other problems
in Chap. 2.
As we mentioned in Chap. 2, a special technique was proposed in [32, 34] for
improvement of the convergence for series of the type in (3.14). In what follows,
we will provide a brief sketch of that technique at the current conjuncture. In our
opinion, this might help the reader to develop necessary skills required for practical
implementation of our technique when needed.
A partial summation of the series in (3.14) could be accomplished with the aid of
the standard [1, 19, 25] summation formula
∞
pn 1
cos nα = − ln 1 − 2 p cos α + p 2 (3.15)
n=1
n 2
valid for p 2 < 1 and 0 ≤ α < 2π, which had repeatedly been used in Chap. 2. But
the direct use of (3.15) in the case of (3.14) is problematic due to the form of the
series coefficients gn (ϑ, τ ) displaced in (3.12). This becomes evident if the hyper-
bolic sine functions in gn (ϑ, τ ) are expressed in terms of exponential functions,
while the product of sine functions sin νϕ sin νψ is replaced with the difference of
corresponding cosine functions. With the foregoing transformations accomplished,
the form in (3.14) decomposes onto a few series of the type
∞
P1n
cos nα (3.16)
n=1
n(P2n − 1)
where||P1 || < ||P2 || and 0 ≤ α < 2π. It is evident that the formula in (3.15) appears
useless in the case of (3.16). But upon transforming the latter in the way shown below
∞
1 ∞
P1n P1n P1n P1n
cos nα = − n + n cos nα
n=1
n(P2 − 1)
n
n=1
n P2n − 1 P2 P2
∞ ∞
1 P1n P1n 1 P1 n
= − cos nα + cos nα
n=1
n P2n − 1 P2n n=1
n P2
∞
1 ∞ n
P1n P1
= cos nx + cos nx , (3.17)
n=1
n P2 P2 − 1
n n
n=1
n P2
the series in (3.16) splits onto two other series, the second of which, namely
3.1 Quadrilateral-Shaped Region 91
∞
1 P1 n
cos nx
n=1
n P2
is directly summable with the aid of (3.15). As to the first series in (3.17), namely
∞
P1n
cos nx
n=1
n P2n P2n − 1
⎞
(1 − 2Y cos α + Y 2 )(1
− 2Z cos α + Z 2)
⎠
+ ln
(1 − 2Y cos β + Y 2 )(1 − 2Z cos β + Z 2 )
∞
2 gn (ϑ, τ ) sin νϕ sin νψ
− (3.18)
a 2 ϕ1 n=1 exp (2ν[ω(ϑ1 ) − ω(ϑ0 )])
and π π
α= (ϕ − ψ) and β = (ϕ + ψ) .
ϕ1 ϕ1
92 3 Toroidal Surface
Note that the representation in (3.18) is valid for τ ≤ ϑ and to get its form valid
for ϑ ≤ τ , the variables ϑ and τ in it have to be interchanged.
Although the form in (3.18) might look heavy loaded, it is, nonetheless, computer
friendly. As to its logarithmic terms, they are, of course, easily computable, while the
series term is uniformly convergent, with a very high convergence rate. Clearly, all
this together makes (3.18) well prepared for immediate computer implementations.
This assertion is illustrated by Fig. 3.2, where a potential field is depicted as
induced by two point sources released at (0.05π, 0.15π) and (−0.15π, 0.65π) in the
quadrilateral determined by ϕ1 = π, ϑ0 = −0.5π, and ϑ1 = 0.5π.
on a toroidal surface of radii a and R. The region is closed in the latitudinal direction
and bounded by the two meridians ϕ = 0 and ϕ = ϕ1 (see Fig. 3.3). We will refer
to this region, in what follows, as the toroidal sector. The term sector is chosen
on purpose due to the fact that the region is bounded by two meridians. In this
regard, the reader might recall a similar situation from the previous chapter, where a
spherical sector was considered.
We target a potential phenomenon as simulated in by the mixed boundary value
problem
1 ∂ ∂u(ϑ, ϕ) a 2 ∂ 2 u(ϑ, ϕ)
D(ϑ) + 2 = − f (ϑ, ϕ) in (3.19)
D(ϑ) ∂ϑ ∂ϑ D (ϑ) ∂ϕ2
3.2 Toroidal Sector 93
∂u (ϑ, ϕ1 )
u (ϑ, 0) = 0 and =0 (3.20)
∂ϕ
and
∂u (−π, ϕ) ∂u (π, ϕ)
u (−π, ϕ) = u (π, ϕ) and = (3.21)
∂ϑ ∂ϑ
The objective is to develop an efficient algorithm for the construction of the
Green’s function G(ϑ, ϕ; τ , ψ) to the homogeneous problem ( f (ϑ, ϕ) ≡ 0) corre-
sponding to (3.19)–(3.21).
The form of boundary conditions imposed in (3.20) suggests for the solution-
function u (ϑ, ϕ) and the right-hand side f (ϑ, ϕ) of the governing equation of (3.19)
the Fourier-sine series expansions
∞
u (ϑ, ϕ) = u n (ϑ) sin νϕ (3.22)
n=1
and
∞
f (ϑ, ϕ) = f n (ϑ) sin νϕ (3.23)
n=1
du n (−π) du n (π)
u n (−π) = u n (π) and = (3.25)
dϑ dϑ
for the homogeneous equation corresponding to (3.24) was introduced in Sect. 3.1
(see Eqs. (3.8) and (3.9)). Implementing the standard variation of parameters routine,
with the above fundamental set of solutions at hand, the general solution to the
equation in (3.24) is obtained in the form
ϑ
1
− sinh ν[ω(ϑ) − ω(τ )]
f n (τ ) dτ (3.26)
ν
−π
Upon satisfying then the boundary conditions of (3.25), we arrive at the well-posed
system ⎛ νω(−π) ⎞⎛ ⎞ ⎛ ⎞
e − eνω(π) e−νω(−π) − e−νω(π) C1 I1
⎝ ⎠⎝ ⎠ = ⎝ ⎠ (3.27)
eνω(−π) − eνω(π) e−νω(π) − e−νω(−π) C2 I2
and π
1
I2 = − cosh ν[ω(π) − ω(τ )]
f n (τ ) dτ
ν −π
Since the introduced in (3.9) function ω(ϑ) is undefined at the points ϑ = ±π,
the values ω(π) and ω(−π), which form arguments of the exponential functions in
(3.27), are understood as the corresponding limits of this function. That is
2a a + R tan ϑ/2 aπ
ω (±π) = lim √ arctan √ = ±√ (3.28)
ϑ→±π R2 − a2 R −a
2 2 R2 − a2
and π
1 e−ν[ω(π)−ω(τ )]
C2 = f n (τ ) dτ
4ν −π sinh νω(π)
Upon substitution of the above into (3.26), the solution to the boundary value
problem in (3.24) and (3.25) is obtained as
ϑ
1
u n (θ) = − sinh ν[ω(ϑ) − ω(τ )]
f n (τ ) dτ
ν
−π
π
1 cosh ν [ω(π) + ω(ϑ) − ω(τ )]
+ f n (τ ) dτ
2ν −π sinh νω(π)
Combining the two integrals, we rewrite the above expression in the single-integral
form
π
1
u n (θ) = gn (ϑ, τ )
f n (τ ) dτ (3.29)
2ν
−π
whose kernel-function gn (ϑ, τ ) represents, in fact, the Green’s function to the homo-
geneous ODE boundary value problem corresponding to (3.24) and (3.25). It is
expressed in two pieces, with the piece valid for ϑ ≤ τ found as
2aπ
σ=√
R2 − a2
while the expression of gn (ϑ, τ ) valid for τ ≤ ϑ can be obtained from the above
by interchanging ϑ with τ . This is so because of the self-adjointness of the problem
setting in (3.24) and (3.25).
96 3 Toroidal Surface
If we recall now the relevant stage of our development in Sect. 3.1, and conduct
the corresponding transformations, then the solution of the problem in (3.19)–(3.21)
arrives in the form
∞
2
u (ϑ, ϕ) = gn (ϑ, τ ) sin νϕ sin νψ f (τ , ψ)dτ ,ψ
a 2 n=1
for the sought-after Green’s function to the homogeneous boundary value problem
corresponding to (3.19)–(3.21).
As it always occurs with series representations of Green’s functions, the conver-
gence issue comes into the play. That is why we have, at this point of our develop-
ment, to take care of the series in (3.30) convergence rate, since it is not uniformly
convergent in . The standard, repeatedly used in our volume before, summation
formula
∞
p 2n−1 1 1 + 2 p cos α + p 2
cos(2n − 1)α = ln , (3.31)
n=1
2n − 1 4 1 − 2 p cos α + p 2
valid for p < 1 and 0 ≤ α < 2π, can help us with this intention.
A complete summation of the series in (3.30) would totally eliminate all doubts
as to its practicality. But it is clear, however, that it is impossible, and the factor
(1 − cosh νσ) in the denominator of gn (ϑ, τ ) is, probably, the most critical hurdle
for a complete summation. This turn us, therefore, to a partial summation option,
which can significantly improve the convergence rate of the series. As a preparatory
step to a partial summation, we transform the reciprocal of (1 − cosh νσ) as
1 2 − e−νσ 2
= νσ − νσ
1 − cosh νσ e (1 − cosh νσ) e
Based on that, we rewrite the expression for the coefficient gn (ϑ, τ ) of the series
in (3.30), which is valid for τ ≤ ϑ, in the expanded form
1 ν[ω(τ )−ω(ϑ)−σ]
gn (ϑ, τ ) = − e − e−ν[ω(τ )−ω(ϑ)+σ]
ν
−eν(ω(τ )−ω(ϑ)) + e−ν[ω(τ )−ω(ϑ)+2σ] + gn (ϑ, τ ) , (3.32)
gn (ϑ, τ ) reads as
where the last additive component
3.2 Toroidal Sector 97
This decomposes the series in (3.30) onto the sum of five other series. The first four
of them (those whose coefficients are the pure exponential components of gn (ϑ, τ )
in (3.32)) are just summable. The summation can be readily accomplished with the
aid of the standard formula in (3.31). As to the fifth of those series, the one related
to the component gn (ϑ, τ ), the reader can easily realize that its convergence is much
higher than of (3.30), and it is uniformly convergent in .
Hence, substituting the expression of gn (ϑ, τ ) from (3.32) into (3.30) and accom-
plishing the prepared partial summation, we arrive at the sought-after Green’s func-
tion
π π
[ω(τ )−ω(ϑ)] − [ω(τ )−ω(ϑ)+σ]
G (ϑ, ϕ; τ , ψ) = H1 e 2ϕ1 , α, β + H1 e 2ϕ1 , α, β
π π
[ω(τ )−ω(ϑ)−σ] − [ω(τ )−ω(ϑ)+2σ]
−H1 e 2ϕ1 , α, β − H1 e 2ϕ1 , α, β
∞
+
gn (ϑ, τ ) sin νϕ sin νψ (3.33)
n=1
where the three-variable function H1 , which defines the non-series additive compo-
nents in (3.33), reads as
1 1 + 2x cos α + x 2 1 1 + 2x cos β + x 2
H1 (x, α, β) = ln − ln
4π 1 − 2x cos α + x 2 4π 1 − 2x cos β + x 2
while the second and the third arguments α and β of H1 (x, α, β) are introduced as
π (ϕ − ψ) π (ϕ + ψ)
α= and β =
2ϕ1 2ϕ1
on the surface of radii R and a. Since the region is closed in the longitudinal direction
and bounded by two parallels ϑ = ϑ0 and ϑ = ϑ1 , it will be referred to as the toroidal
belt. Let the mixed boundary value problem
1 ∂ ∂u (ϑ, ϕ) a 2 ∂ 2 u (ϑ, ϕ)
D (ϑ) + = − f (ϑ, ϕ) (3.34)
D (ϑ) ∂ϑ ∂ϑ D 2 (ϑ) ∂ϕ2
∂u (ϑ0 , ϕ)
= 0 and u (ϑ1 , ϕ) = 0 (3.35)
∂ϑ
and
∂u (ϑ, 0) ∂u (ϑ, 2π)
u (ϑ, 0) = u (ϑ, 2π) and = (3.36)
∂ϕ ∂ϕ
reflect the periodicity which prompts the general Fourier series expansions
∞ ∞
1
u (ϑ, ϕ) = u 0 (ϑ) + u (c)
n (ϑ) cos nϕ + u (s)
n (ϑ) sin nϕ (3.37)
2 n=1 n=1
and
∞ ∞
1
f (ϑ, ϕ) = f 0 (ϑ) + f n(c) (ϑ) cos nϕ + f n(s) (ϑ) sin nϕ (3.38)
2 n=1 n=1
for the solution-function u (ϑ, ϕ) and the right-hand side f (ϑ, ϕ) of the governing
equation in (3.34). This yields the self-adjoint ODE boundary value problem
1 d du n (ϑ) n2a
D(ϑ) − u n (ϑ) = −
f n (ϑ) , n = 0, 1, 2, . . . (3.39)
a dϑ dϑ D(ϑ)
du n (ϑ0 )
= 0 and u n (ϑ1 ) = 0 (3.40)
dϑ
in the coefficients u n (ϑ) of the series in (3.37), where the right-hand side is
f n (ϑ) =
[D(ϑ)/a] f n (ϑ).
For the thoughtful reader, who was able to review the details of our developments
in Sects. 2.3 and 2.5, it becomes clear that to adequately treat the problem in (3.39)
and (3.40), the cases of n = 0 and of n ≥ 1 have to be considered separately.
Let us turn first to the case of n ≥ 1, for which the general solution
ϑ
1
u n (ϑ) = − sinh n[ω(ϑ) − ω(τ )]
f n (τ ) dτ + C1 enω(ϑ) + C2 e−nω(ϑ) (3.41)
n ϑ0
to the equation in (3.39) can readily be obtained from that derived earlier in Sect. 3.1.
The point is that the form in (3.41) directly follows from that of the equation in (3.10)
by just replacing in the latter the parameter ν with n.
Proceeding further with the method of variation of parameters routine, we satisfy
the boundary conditions in (3.40). This yields the following well-posed system
⎛ ⎞⎛ ⎞ ⎛ ⎞
enω(ϑ0 ) −e−nω(ϑ0 ) C1 0
⎝ ⎠⎝ ⎠=⎝ ⎠
nω(ϑ1 ) −nω(ϑ1 )
e e C2 I (ϑ0 , ϑ1 )
of linear algebraic equations in the constants C1 and C2 , where the second element
in the right-hand side vector is found in the integral form
ϑ1
1
I (ϑ0 , ϑ1 ) = sinh n[ω (ϑ1 ) − ω (τ )]
f n (τ ) dτ
n ϑ0
100 3 Toroidal Surface
and ϑ1
enω(ϑ0 )
C2 = sinh n[ω (ϑ1 ) − ω (τ )]
f n (τ ) dτ
2n ϑ0
where
= cosh n[ω (ϑ1 ) − ω (ϑ0 )]
Upon substituting the just determined values of C1 and C2 into (3.41) and per-
forming some algebra, the solution to the problem in (3.39) and (3.40) is found
as
ϑ
1
u n (ϑ) = cosh n[ω (τ ) − ω (ϑ0 )] sinh n[ω (ϑ1 ) − ω (ϑ)]
f n (τ ) dτ
n ϑ0
ϑ1
+ cosh n[ω (ϑ) − ω (ϑ0 )] sinh n[ω (ϑ1 ) − ω (τ )]
f n (τ ) dτ
ϑ
1
gn (ϑ, τ ) = {cosh n[ω (τ ) − ω (ϑ0 )] sinh n[ω (ϑ1 ) − ω (ϑ)]} (3.43)
n
while, upon interchanging ϑ with τ in the above, one obtains the expression for
gn (ϑ, τ ) valid for ϑ ≤ τ ≤ ϑ1 . This is supported by the self-adjointness of the prob-
lem in (3.39) and (3.40).
At this point in our development, we turn back to the problem in (3.39) and (3.40),
and consider its case of n = 0 that reduces the latter to
1 d du 0 (ϑ)
(R + a sin ϑ) =− f 0 (ϑ) (3.44)
a dϑ dϑ
du 0 (ϑ0 )
= 0 and u 0 (ϑ1 ) = 0 (3.45)
dϑ
3.3 Belt-Shaped Region 101
In helping the reader to tackle the above boundary value problem with the aid of
the method of variation of parameters, we consider the homogeneous equation
1 d du 0 (ϑ)
(R + a sin ϑ) =0 (3.46)
a dϑ dϑ
du 0 (ϑ)
(R + a sin ϑ) = D1
dϑ
where D1 is an arbitrary constant. Separating variables, one arrives at the general
solution to (3.46) in the form
dϑ
u 0 (ϑ) = D1 (3.47)
R + a sin ϑ
From the course of calculus, the reader learns that the substitution method might
be recommended for the integration in this case, and the standard for the above type
of integrals substitution t = tan ϑ/2 reduces the integral in (3.47) to
dt
Rt 2 + 2at + R
whose further treatment looks more or less trivial with completing a perfect square of
the quadratic trinomial. Leaving details aside, we present just the ultimate expression
2a a + R tan ϑ/2
u 0 (ϑ) = D1 √ arctan √ + D2
R2 − a2 R2 − a2
for the general solution to the homogeneous equation in (3.46). Retrieving from
Sect. 3.1 the notation
2a a + R tan ϑ/2
ω(ϑ) = √ arctan √
R2 − a2 R2 − a2
1 and ω(ϑ)
to the boundary value problem in (3.44) and (3.45). In a single-integral form, the
above representation reads
ϑ1
u 0 (ϑ) = g0 (ϑ, τ )
f n (τ ) dτ
ϑ0
where ⎧
⎨ ω (ϑ1 ) − ω (ϑ) , ϑ0 ≤ τ ≤ ϑ
g0 (ϑ, τ ) = (3.48)
⎩
ω (ϑ1 ) − ω (τ ) , ϑ ≤ τ ≤ ϑ1
∞
1
G (ϑ, ϕ; τ , ψ) = g0 (ϑ, τ ) + gn (ϑ, τ ) cos n (ϕ − ψ) (3.49)
2 n=1
with compact representations for its coefficients g0 (ϑ, τ ) and gn (ϑ, τ ) shown earlier
in (3.48) and (3.43).
From the experience gained earlier in this volume, the reader is supposed to learn
that the series in (3.49) cannot uniformly converge in . This is so because Green’s
functions for problems of the type in (3.34)–(3.36) possess logarithmic singularity.
3.3 Belt-Shaped Region 103
Upon completion of a partial summation of the series in (3.49), the Green’s func-
tion of the homogeneous boundary value problem corresponding to (3.34)–(3.36) set
up in the toroidal belt is ultimately derived as
1
G (ϑ, ϕ; τ , ψ) = 1 (ϑ1 , τ )
2a 2
1 1 − 2e2−1 (ϑ,τ ) cos (ϕ − ψ)+ e2[2−1 (ϑ,τ )]
+ ln
2πa 2 1 − 2e1 (ϑ,τ ) cos (ϕ − ψ) + e21 (ϑ,τ )
1−2e2 (ϑ,τ )−2ω(ϑ1 ) cos (ϕ − ψ) + e2[2 (ϑ,τ )−2ω(ϑ1 )]
+ ln
1−2e2ω(ϑ0 )−2 (ϑ,τ ) cos (ϕ − ψ) + e2[2ω(ϑ0 )−2 (ϑ,τ )]
∞
1 gn (ϑ, τ )
− cos n (ϕ − ψ) (3.50)
4a 2 n=1 ne2n
Note that the above form is valid for τ ≥ ϑ, and to obtain the one valid for ϑ ≥ τ ,
the variables ϑ and τ ought to be interchanged.
The form in (3.50) is quite compact and well-prepared for an immediate computer
implementation, because its series component converges in at a high rate. To
illustrate this point, we depict in Fig. 3.5 the potential field induced in the toroidal
belt by a point source released at (−0.25π, 0.9π). The belt is bounded by the two
parallels: ϑ0 = −π and ϑ1 = 0.0.
104 3 Toroidal Surface
In order to develop the reader’s skills in implementing out technique, a few exer-
cises are offered in what follows. Their answers are available in the appendix to this
volume.
Exercise 3.1
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ1 , ϕ)
u (ϑ0 , ϕ) = 0 and =0
∂ϑ
in the toroidal quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 3.2
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
3.4 Chapter Exercises 105
∂u (ϑ0 , ϕ) ∂u (ϑ1 , ϕ)
+ βu (ϑ0 , ϕ) = 0 and =0
∂ϑ ∂ϑ
in the toroidal quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 3.3
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (ϑ0 , ϕ)
+ βu (ϑ0 , ϕ) = 0 and u (ϑ1 , ϕ) = 0
∂ϑ
in the toroidal quadrilateral
= {ϑ0 ≤ ϑ < ϑ1 , 0 ≤ ϕ ≤ ϕ1 } .
Exercise 3.4
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
∂u (−π, ϕ) ∂u (π, ϕ)
u (−π, ϕ) = u (π, ϕ) and =
∂ϑ ∂ϑ
in the toroidal sector
= {−π ≤ ϑ < π, 0 ≤ ϕ ≤ ϕ1 }.
Exercise 3.5
u (ϑ, 0) = 0 and u (ϑ, ϕ1 ) = 0
The goal of this chapter is to provide a retrospective analysis of the work started in
[18] which we had later adjusted in [31]. It aimed at the extension of the Green’s
function formalism to specific systems of elliptic partial differential equations. Those
are not in fact systems in a conventional sense, according to which all the governing
equations in a system share the same region for the independent variables. Each of
the governing equations in systems to be encountered herein is, however, hosted by
an individual region, which might share parts of its boundary with other hosting
regions.
In order to consistently treat such specific boundary value problems, we introduced
the notion of matrix of Green’s type for partial differential equations. The presentation
in this chapter touches upon problems that simulate potential phenomena taking place
in joint thin-wall structures. Some selective problems of this type can be found in
some of our earlier publications [8, 32, 33, 35].
In this section, we will lay down a foundation of the Green’s function formalism
targeted at boundary value problems that simulate potential fields induced in com-
positions of thin-wall elements.
To facilitate the comprehension of the material and to make it an easy read, the
presentation is limited to a two-fragment-containing region = 1 ∪ 2 shown in
Fig. 4.1. We believe that any further extension to cases with multiple fragments could
be relatively trivial, given an experience gained from this section.
Let the components 1 and 2 of be fragments of different surfaces. Let also
1 and 2 host the functions u 1 (P) and u 2 (P), respectively. In , we formulate the
non-conventional boundary value problem
B [u i (P)] = 0, P ∈ 0 , i = 1, 2 (4.2)
∂u 1 (P) ∂u 2 (P)
u 1 (P) = u 2 (P), =λ , P ∈ 12 (4.3)
∂n 12 ∂n 12
with their components Ui (P) and Fi (P) defined in terms of u i (P) and f i (P) in a
piecewise fashion as
4.1 Matrices of Green’s Type for PDE 109
u i (P), P ∈ i
Ui (P) =
0, P ∈ /i
and
f i (P), P ∈ i
Fi (P) = .
0, P ∈ /i
The problem setting in (4.1)–(4.3), with the introduced vector functions U(P)
and F(P), creates a perfect environment for introduction of a key concept for the
upcoming presentation.
Definition: If for any integrable in vector function F(P), the vector function
U(P) is expressed in the form
U(P) = G(P, Q)F(Q)d(Q) (4.5)
then the kernel-matrix G(P, Q) of the above integral representation stays for the
matrix of Green’s type to the homogeneous problem corresponding to (4.1)–(4.3).
The term ‘integrable’ with respect to F(Q) implies
F(Q)d(Q) < ∞.
1. The peripheral elements G 12 (P, Q) and G 21 (P, Q), being considered as func-
tions of the coordinates of the field point P in i , satisfy the homogeneous
equation corresponding to (4.1), that is
2. The diagonal elements G 11 (P, Q) and G 22 (P, Q), being considered as functions
of the coordinates of the field point P in i , satisfy the homogeneous equation
corresponding to (4.1) everywhere in i except for P = Q, that is
3. For P = Q, the diagonal elements G 11 (P, Q) and G 22 (P, Q) possess the loga-
rithmic singularity
1 1
G ii (P, Q) = ln ;
2π |P − Q|
4. The elements G i j (P, Q) satisfy all the boundary and contact conditions imposed
by (4.2) and (4.3), which they are involved in.
Matrix of Green’s type will be constructed, in this section, for a boundary value
problem of the type in (4.1)–(4.3). To be specific, we target a problem that simulates
potential fields induced in a thin hemispherical shell of radius a composed of two
congruent segments made of materials whose conductivities are specified by the
constants λ1 and λ2 . The shell’s middle surface occupies the region
and
2 = {π/2 < ϑ < π, 0 < ϕ < π}.
lim |u 1 (ϑ, ϕ)| < ∞ and lim |u 2 (ϑ, ϕ)| < ∞ (4.8)
ϑ→0 ϑ→π
∂u 1 (π/2, ϕ) ∂u 2 (π/2, ϕ)
u 1 (π/2, ϕ) = u 2 (π/2, ϕ) and =λ (4.9)
∂ϑ ∂ϑ
in u i = u i (ϑ, ϕ), where λ = λ2 /λ1 .
Note that the hemisphere is actually a particular case of a spherical sector (see
our work in Sect. 2.6), with the opening angle ϕ = π. Thus, both poles ϑ = 0 and
ϑ = π of the sphere are parts of the outer “boundary” of representing at the same
time the points of singularity for the governing equations in (4.6). This explains the
reasoning behind inclusion of the boundness conditions of (4.8) in the above problem
formulation.
The goal in the upcoming development is an efficient procedure for obtaining
computer-friendly representations for the elements of the matrix of Green’s type
G(ϑ, ϕ; τ , ψ) for the homogeneous ( f i (ϑ, ϕ) ≡ 0) problem corresponding to (4.6)–
(4.9). Note that the variables τ and ψ in G(ϑ, ϕ; τ , ψ) stay for the latitude and the
longitude, respectively, of the source point.
Following the pattern worked out in Chaps. 2 and 3, we aim, in the current case,
at expressing the vector function U(ϑ, ϕ) in terms of F(ϑ, ϕ) in the form of (4.5),
because the latter delivers an explicit form for the looked-for matrix G(ϑ, ϕ; τ , ψ).
The goal will again be achieved by means of a separation of variables combined with
the method of variation of parameters. Proceeding with this strategy, we express the
functions u i (ϑ, ϕ) and f i (ϑ, ϕ) in the Fourier sine-series form
∞
∞
u i (ϑ, ϕ) = u i,n (ϑ) sin nϕ and f i (ϑ, ϕ) = f i,n (ϑ) sin nϕ (4.10)
n=1 n=1
complying with the boundary conditions of (4.7). This yields, for the coefficients
u i,n (ϑ) of the first of the series from (4.10), the following three-point-posed ODE
boundary value problem
d du 1,n (ϑ) n 2 u 1,n (ϑ)
sin ϑ − =−
f 1,n (ϑ) , ϑ ∈ (0, π/2) (4.11)
dϑ dϑ sin ϑ
112 4 Compound Structures
d du 2,n (ϑ) n 2 u 2,n (ϑ)
sin ϑ − =−
f 2,n (ϑ) , ϑ ∈ (π/2, π) (4.12)
dϑ dϑ sin ϑ
lim |u 1,n (ϑ)| < ∞ and lim |u 2,n (ϑ)| < ∞ (4.13)
ϑ→0 ϑ→π
and
du 1,n (π/2) du 2,n (π/2)
u 1,n (π/2) = u 2 (π/2) and =λ (4.14)
dϑ dϑ
ϑ ϑ
u 1,n (ϑ) = C1 tann + D1 tan−n
2 2
ϑ
1 ϑ τ ϑ τ
+ tan−n tann − tann tan−n f 1,n (τ )dτ (4.15)
2n 2 2 2 2
0
and
ϑ ϑ
u 2,n (ϑ) = C2 tann + D2 tan−n
2 2
ϑ
1 ϑ τ ϑ τ
+ tan−n tann − tann tan−n f 2,n (τ )dτ . (4.16)
2n 2 2 2 2
π/2
The four constants of integration in (4.15) and (4.16) can uniquely be determined
by means of the boundness conditions of (4.13) and the contact conditions of (4.14).
The total number of these conditions is four ensuring the well-posedness of this
operation.
4.2 Compound Hemisphere 113
As to the first condition in (4.13), the only way to make it true is to set up D1 = 0.
The second condition in (4.13), being applied to the form from (4.16), allows us to
directly determine C2
π
1 τ
C2 = tan−n f 2,n (τ ) dτ .
2n 2
π/2
π/2
1 τ 1−λ τ
C1 = tan−n + tann f 2,n (τ ) dτ
2n 2 1+ λ 2
0
π
λ τ
+ tan−n f 2,n (τ ) dτ ,
n(1+ λ) 2
π/2
⎛ ⎞
π/2 π
1 ⎜ n τ 1−λ τ ⎟
D2 = ⎝ tan f 1,n (τ )dτ − tan−n f 2,n (τ ) dτ ⎠ .
n(1+ λ) 2 2 2
0 π/2
This completes that stage of our solution procedure which deals with the problem in
(4.11)–(4.14). With the just found values of the constants of integration C1 , D1 , C2 ,
and D2 , the expressions for u 1,n (ϑ)and u 2,n (ϑ)from (4.15) and (4.16) transform to
ϑ
1 τ ϑ ϑ τ
u 1,n (ϑ) = tann tan−n −tann tan−n f 1,n (τ ) dτ
2n 2 2 2 2
0
π/2
1 −n τ 1−λ n τ ϑ
+ tan + tan tann f 1,n (τ ) dτ
2n 2 1+λ 2 2
0
π
λ τ ϑ π
+ tan−n tann f 2,n (τ ) dτ , ϑ ∈ [0, ] (4.17)
n(1+λ) 2 2 2
π/2
and
114 4 Compound Structures
π/2
λ τ ϑ
u 2,n (ϑ) = tann tan−n f 1,n (τ ) dτ
n(1+λ) 2 2
0
ϑ
1 nτ −n ϑ nϑ −n τ
+ tan tan −tan tan f 2,n (τ ) dτ
2n 2 2 2 2
π/2
π
1 ϑ 1−λ ϑ τ π
+ tann − tan−n tan−n f 2,n (τ ) dτ , ϑ ∈ [ , π]. (4.18)
2n 2 1+λ 2 2 2
π/2
One might think that the forms in (4.17) and (4.18) are too heavy-loaded, but this
situation can however be improved. As to (4.17), we can make it more compact by
combining the first two integrals into a single definite integral, from 0 to π/2, of a
function defined in two pieces. At the same time, we leave the last integral of (4.17)
in its current form. This yields
π/2 π
u 1,n (ϑ) = (n)
g11 (ϑ, τ ) f 1,n (τ ) dτ + (n)
g12 (ϑ, τ )
f 2,n (τ ) dτ , (4.19)
0 π/2
(n) (n)
where the kernel functions g11 (ϑ, τ ) and g12 (ϑ, τ ) represent in fact the elements of
the first row of the matrix gn (ϑ, τ ) of Green’s type to the homogeneous problem cor-
(n)
responding to (4.11)–(4.14). The element g11 (ϑ, τ ) is defined, as we just mentioned,
in two pieces, with its expression valid for 0 ≤ ϑ ≤ τ ≤ π/2 found as
(n) 1 τ 1−λ τ ϑ
g11 (ϑ, τ ) = tan−n + tann tann ,
2n 2 1+λ 2 2
(n)
while the expression of g11 (ϑ, τ )valid for 0 ≤ τ ≤ ϑ ≤ π/2 can be obtained from the
(n)
above by interchanging ϑ and τ . The element g12 (ϑ, τ ) is defined in a single piece
form, and appears as
(n) λ τ ϑ π
g12 (ϑ, τ ) = tan−n tann , 0 ≤ ϑ ≤ ≤ τ ≤ π.
n(1+λ) 2 2 2
π/2 π
u 2,n (ϑ) = g21 (ϑ, τ )
(n)
f 1,n (τ ) dτ + (n)
g22 (ϑ, τ )
f 2,n (τ ) dτ (4.20)
0 π/2
4.2 Compound Hemisphere 115
expressed in terms of the elements of the second row of gn (ϑ, τ ) where the element
(n)
g21 (ϑ, τ ) is defined in the single piece form
(n) 1 ϑ τ π
g21 (ϑ, τ ) = tan−n tann , 0 ≤ τ ≤ ≤ ϑ ≤ π.
n(1+λ) 2 2 2
(n) (n)
The element g22 (ϑ, τ ) is, analogously to g11 (ϑ, τ ), defined in two pieces. Its
expression valid for π/2 ≤ ϑ ≤ τ ≤ π is found as
(n) 1 n ϑ 1−λ −n ϑ τ
g22 (ϑ, τ ) = tan − tan tan−n ,
2n 2 1+λ 2 2
while the expression valid for π/2 ≤ τ ≤ ϑ ≤ π can be obtained from the above by
interchanging ϑ and τ .
To proceed further, we remind that the weight functions f i,n (τ ) in (4.19) and
(4.20) are expressed in terms of the Fourier coefficients f i,n (τ ) of the second series
in (4.10) as
f i,n (τ ) = a 2 sin τ f i,n (τ ). Expressing now the functions f i,n (τ ) in terms
of their generating functions f i (τ , ψ)
π
2
f i,n (τ ) = f i (τ , ψ) sin nψdψ, i = 1, 2
π
0
π π/2
2 (n)
u 1,n (ϑ) = g (ϑ, τ ) sin nψ f 1 (τ , ψ) a 2 sin τ dτ dψ
π 11
0 0
π π
2 (n)
+ g (ϑ, τ ) sin nψ f 2 (τ , ψ) a 2 sin τ dτ dψ
π 12
0 π/2
and
π π/2
2 (n)
u 2,n (ϑ) = g (ϑ, τ ) sin nψ f 1 (τ , ψ) a 2 sin τ dτ dψ
π 21
0 0
π π
2 (n)
+ g (ϑ, τ ) sin nψ f 2 (τ , ψ) a 2 sin τ dτ dψ.
π 22
0 π/2
Upon substituting the above into the first series from (4.10), we arrive eventually
at the solution to the boundary value problem of (4.6)–(4.9) which appears in the
116 4 Compound Structures
form
u 1 (ϑ, ϕ) = G 11 (ϑ, ϕ; τ , ψ) f 1 (τ , ψ) d1 (τ , ψ)
+ G 12 (ϑ, ϕ; τ , ψ) f 2 (τ , ψ) d2 (τ , ψ), (ϑ, ϕ) ∈ 1 (4.21)
and
u 2 (ϑ, ϕ) = G 21 (ϑ, ϕ; τ , ψ) f 1 (τ , ψ) d1 (τ , ψ)
+ G 22 (ϑ, ϕ; τ , ψ) f 2 (τ , ψ) d2 (τ , ψ), (ϑ, ϕ) ∈ 2 . (4.22)
Hence, in light of (4.5), the kernel functions G i j (ϑ, ϕ; τ , ψ) in (4.21) and (4.22)
represent the elements of the sought-after matrix G(ϑ, ϕ; τ , ψ) of Green’s type to
the homogeneous boundary value problem corresponding to (4.6)–(4.9). For these
elements, we subsequently have
∞
2 (n)
G i j (ϑ, ϑ; τ , τ ) = g (ϑ, τ ) sin nϕ sin nψ
π n=1 i j
∞
1 (n)
= g (ϑ, τ )[cos n(ϕ−ψ)− cos n(ϕ+ ψ)], (i, j = 1, 2). (4.23)
π n=1 i j
∞
pn
cos nx, where p 2 < 1 and 0 ≤ x < 2π.
n=1
n
This makes the series totally summable with the aid of the standard [1, 19, 25]
summation formula
∞
pn
cos nx = − ln 1 − 2 p cos x + p 2 ,
n=1
n
which had been referred to for the first time in this volume in Chap. 2, and was
repeatedly used afterward.
Upon accomplishing the summation, we managed to obtain the following closed
computer-friendly forms:
4.2 Compound Hemisphere 117
⎛
1 ⎝ 2(ϑ) −2(ϑ)(τ ) cos(ϕ +ψ)+ 2 (τ )
G 11 (ϑ, ϕ; τ , ψ) = ln
2π 2(ϑ) −2(ϑ)(τ ) cos(ϕ − ψ)+ 2 (τ )
⎞
1−λ 1 −2(ϑ)(τ ) cos(ϕ+ψ)+ 2 (ϑ)2 (τ ) ⎠
+ ln (4.24)
1+λ 1 −2(ϑ)(τ ) cos(ϕ− ψ)+ 2 (ϑ)2 (τ )
λ 2(ϑ) −2(ϑ)(τ ) cos(ϕ +ψ)+ 2 (τ )
G 12 (ϑ, ϕ; τ , ψ) = ln
π(1+λ) 2(ϑ) −2(ϑ)(τ ) cos(ϕ − ψ)+ 2 (τ )
(4.25)
1 2(ϑ) −2(ϑ)(τ ) cos(ϕ +ψ)+ 2 (τ )
G 21 (ϑ, ϕ; τ , ψ) = ln
π(1+λ) 2(ϑ) −2(ϑ)(τ ) cos(ϕ − ψ)+ 2 (τ )
(4.26)
and
⎛
1 ⎝ 2(ϑ) −2(ϑ)(τ ) cos(ϕ +ψ)+ 2 (τ )
G 22 (ϑ, ϕ; τ , ψ) = ln
2π 2(ϑ) −2(ϑ)(τ ) cos(ϕ − ψ)+ 2 (τ )
⎞
1−λ 1 −2(ϑ)(τ ) cos(ϕ+ψ)+ 2 (ϑ)2 (τ )
⎠
− ln (4.27)
1+λ 1 −2(ϑ)(τ ) cos(ϕ− ψ)+ 2 (ϑ)2 (τ )
that represents, in fact, the Green’s function for the Dirichlet problem stated in the
homogeneous hemispherical surface
And this is what the statement in (4.6)–(4.9) reduces actually to, if λ = 1 in (4.9),
or, in other words, if both regions 1 and 2 are filled in with the same material.
Figure 4.3 depicts the potential field generated by a point source located at
(0.45π, 0.35π) for four different values of parameter λ.
118 4 Compound Structures
Fig. 4.3 Potential field induced by a point source in the compound hemisphere with
(a) λ = 0.1, (b) λ = 1.0, (c) λ = 5.0, (d) λ = 10.0
Upon comparison the fragments (a), (b), (c), and (d) of Fig. 4.3, one comes to
a conclusion that the magnitude of the coefficient λ significantly affects the field
induced by a point source.
Conductivities of the materials, of which the shells are made, are specified with
constants λ1 and λ2 .
In the assembly = 1 ∪ 2 , we consider the boundary value problem
4.3 Hemisphere Joint to Cylinder 119
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + 2 2 = − f 1 (ϑ, ϕ) in 1 (4.28)
a sin ϑ ∂ϑ
2 ∂ϑ a sin ϑ ∂ϕ2
∂ 2 u 2 (z, ϕ) 1 ∂ 2 u 2 (z, ϕ)
+ = − f 2 (z, ϕ) in 2 (4.29)
∂z 2 a 2 ∂ϕ2
∂u 2 (l, ϕ)
lim |u 1 (ϑ, ϕ)| < ∞, u 2 (l, ϕ) + β =0 (4.30)
ϑ→0 ∂z
∂u 1 (π/2, ϕ) ∂u 2 (0, ϕ)
u 1 (π/2, ϕ) = u 2 (0, ϕ), =λ (4.31)
∂ϑ ∂z
where λ = λ2 /λ1 .
The above problem simulates a field of potential induced in the considered assem-
bly of shells. And we are looking for the matrix G(P, Q) of Green’s type of the
corresponding to (4.28)–(4.33) homogeneous problem. From our experience, it fol-
lows that the elements G i j of G could be explicitly revealed if the solution functions
u 1 (ϑ, ϕ) and u 2 (z, ϕ) of the above problem are expressed as
u 1 (ϑ, ϕ) = G 11 (ϑ, ϕ; τ , ψ) f 1 (τ , ψ)d1 (τ , ψ)
+ G 12 (ϑ, ϕ; ζ, ψ) f 2 (ζ, ψ)d2 (ζ, ψ)
and
120 4 Compound Structures
u 2 (z, ϕ) = G 21 (z, ϕ; τ , ψ) f 1 (τ , ψ)d1 (τ , ψ)
+ G 22 (z, ϕ; ζ, ψ) f 2 (ζ, ψ)d2 (ζ, ψ).
∞
1
u 1 (ϑ, ϕ) = u 10 (ϑ) + u (c) (s)
1n (ϑ) cos nϕ + u 1n (ϑ) sin nϕ (4.34)
2 n=1
∞
1 (c) (s)
f 1 (ϑ, ϕ) = f 10 (ϑ) + f 1n (ϑ) cos nϕ + f 1n (ϑ) sin nϕ (4.35)
2 n=1
and
∞
1
u 2 (z, ϕ) = u 20 (z) + u (c) (s)
2n (z) cos nϕ + u 2n (z) sin nϕ (4.36)
2 n=1
∞
1 (c) (s)
f 2 (z, ϕ) = f 20 (z) + f 2n (z) cos nϕ + f 2n (z) sin nϕ. (4.37)
2 n=1
d 2 u 2n (z) n 2
− 2 u 2n (z) = − f 2n (z), 0 < z < l (4.39)
dz 2 a
du 2n (l)
lim |u 1n (ϑ)| < ∞, u 2n (l) + β =0 (4.40)
ϑ→0 dz
and
du 1n (π/2) du 2n (0)
u 1n (π/2) = u 2n (0), =λ (4.41)
dϑ dz
in the coefficients u 1n (ϑ) and u 2n (z) of the series in (4.34) and (4.36). The right-hand
side of (4.38) is defined in terms of the Fourier coefficients of (4.35) as f 1n (ϑ) =
a 2 sin ϑ f 1n (ϑ).
4.3 Hemisphere Joint to Cylinder 121
Remind that the superscripts ‘c’ and ‘s’ on u 1n (ϑ) and u 2n (z) are omitted in (4.38)
through (4.41), because the sine and the cosine coefficients of the series in (4.34)
and (4.36) will be identically treated in what follows until a certain moment in our
work.
Another important detail in the upcoming development is related to the summation
index n of the foregoing Fourier series. The point is that n represents a parameter
in the problem of (4.38)–(4.41), significantly influencing the fundamental sets of
solutions of the governing differential equations. That is why the cases n = 0 and
n ≥ 1 in (4.38)–(4.41) will be treated separately.
We begin with the case of n = 0, which transforms the governing differential
equations of the setting in (4.38)–(4.41) into
1 d du 10 (ϑ)
sin ϑ =−
f 10 (ϑ), 0 < ϑ < π/2 (4.42)
sin ϑ dϑ dϑ
d 2 u 20 (z)
= − f 20 (z), 0 < z < a. (4.43)
dz 2
can be recalled from our work in Chap. 2, whereas for the trivial case of (4.43), we
have
u (1) (2)
20 (z) = 1 and u 20 (z) = z.
With the aid of the boundary and contact conditions, the constants of integration
for (4.44) and (4.45) are obtained as
π τ l
2 β +l
C1 = −ln tan f 10 (τ )dτ + [β +(l −ζ)] f 20 (ζ)dζ
0 λ 2 0
π
2 β +l l
C2 = f 10 (τ )dτ + [β +(l −ζ)] f 20 (ζ)dζ
0 λ 0
122 4 Compound Structures
and π
2 1
D2 = − f 10 (τ )dτ and D1 = 0.
0 λ
Substituting the found values into (4.44) and (4.45), the solution functions u 10 (ϑ)
and u 20 (z) to the setting in (4.38)–(4.41) are ultimately expressed, for the case of
n = 0, in the form
π l
2
u 10 (ϑ) = 0
g11 (ϑ, τ )
f 10 (τ )dτ + 0
g12 (ϑ, ζ) f 20 (ζ)dζ (4.46)
0 0
and π
2 l
u 20 (z) = 0
g21 (z, τ )
f 10 (τ )dτ + 0
g22 (z, ζ) f 20 (ζ)dζ. (4.47)
0 0
0
Note that the kernel functions g11 (ϑ, τ ), g12
0
(ϑ, ζ), g21
0
(z, τ ), and g22
0
(z, ζ) repre-
sent the elements of the matrix of Green’s type for the homogeneous ODE boundary
value problem corresponding to (4.38)–(4.41) in the case of n = 0. They are defined
as
1 (β +l)−λ ln(tan τ /2), 0 ≤ ϑ ≤ τ ≤ π/2
g11 (ϑ, τ ) =
0
λ (β +l)− λ ln(tan ϑ/2), 0 ≤ τ ≤ ϑ ≤ π/2
0
g12 (ϑ, ζ) = [β +(l −ζ)], 0 ≤ ϑ ≤ π/2, 0 ≤ ζ ≤ l
β +(l −z)
0
g21 (z, τ ) = , 0 ≤ z ≤ l, 0 ≤ τ ≤ π/2
λ
and
β +(l −ζ), 0 ≤ z ≤ ζ ≤ l
0
g22 (z, ζ) = .
β +(l −z), 0 ≤ ζ ≤ z ≤ l
This completes our analysis for the boundary value problem of (4.38)–(4.41) in
the case of n = 0, and we are in a position now to turn to the case of n ≥ 1. As to the
required fundamental sets of solutions to the homogeneous equations corresponding
to (4.38) and (4.39), we have
u (1) (2)
1n (ϑ) = tan ϑ/2 and u 1n (ϑ) = cot ϑ/2
n n
and
u (1)
2n (z) = e
nz/a
and u (2)
2n (z) = e
−nz/a
,
respectively.
Proceeding in compliance with the Lagrange’s method of variation of parameters,
the general solutions to the governing equations of (4.38) and (4.39) are found as
4.3 Hemisphere Joint to Cylinder 123
ϑ
1 nτ nϑ nϑ nτ
u 1n (ϑ)= tan cot − tan cot f 1n (τ )dτ
0 2n 2 2 2 2
ϑ ϑ
+ C1 tann + D1 cot n (4.48)
2 2
and z
1 n
u 2n (z) = sinh (ζ −z) f 2n (ζ)dζ + C2 enz/a + D2 e−nz/a , (4.49)
0 n a
respectively.
The boundness condition of (4.40) yields D1 = 0, and the rest of the constants
of integration in (4.48) and (4.49), after a tedious but quite straightforward algebra,
are found as π
λ(βn−1)e−nl/a 2 τ
C2 = ∗
tann f 1n (τ )dτ
n 0 2
l
1+λ n
+ βn cosh n(ζ −l)−sinh (ζ −l) f 2n (ζ)dζ
n∗ 0 a
π
λ(βn+1)enl/a 2 τ
D2 = tann f 1n (τ )dτ
n∗ 0 2
l
1−λ n n
+ βn cosh (ζ −l)−sinh (ζ −l) f 2n (ζ)dζ
n∗ 0 a a
and π
2 λ (1+βn)enl/a −(1 −βn)e−nl/a τ
C1 = tann
0 n∗ 2
1 nτ τ
− tan −cot n f 1n (τ )dτ
2n 2 2
l
2 n n
+ ∗ βn cosh (ζ −l)−sinh (ζ −l) f 2n (ζ)dζ,
n 0 a a
where
∗ = (1−λ)(1−βn)e−nl/a +(1+λ)(1+βn)enl/a .
Upon substituting the just found values of the constants of integration into (4.48)
and (4.49), we arrive at
π l
2
u 1n (ϑ) = n
g11 (ϑ, τ )
f 1n (τ )dτ + n
g12 (ϑ, ζ) f 2n (ζ)dζ (4.50)
0 0
124 4 Compound Structures
and π
2 l
u 2n (z) = n
g21 (z, τ )
f 1n (τ )dτ + n
g22 (z, ζ) f 2n (ζ)dζ (4.51)
0 0
and ⎧
⎪
⎪ βn cosh an (z − l)− β sinh an (z −l)
⎪
⎪×[(1+λ)enζ/a + (1−λ)e−nζ/a ], ζ ≤ z
1 ⎨
g22 (z, ζ) =
n
.
n∗ ⎪
⎪ βn (ζ −l)−sinh (ζ −l)
⎪
⎪ cosh n n
⎩ a a
× (1+λ)enz/a + (1−λ)e−nz/a , z ≤ ζ
Summarizing the intermediate progress yet achieved in our development, note that
working on the matrix G(P, Q) of Green’s type to the problem setting in (4.28)–
(4.33), we obtained explicit expressions for the elements ginj of the matrices of Green’s
type to the ODE homogeneous problem corresponding to (4.38)–(4.41) for the cases
of n = 0 and n ≥ 1. These elements will play a decisive role in the completion of
our development.
To continue, we recall the Fourier series expansions of (4.34) through (4.37).
Notice that their coefficients are decorated with the superscripts ‘c’ and ‘s’ to dis-
tinguish the sine components from the cosine components of the series. The point is
that we need, at this stage of our development, to take that difference into account.
Namely, the integrals representing u 1n (ϑ) and u 2n (z) in (4.46), (4.47), (4.50), and
(4.51) contain the functions f 1n (τ ), f 1n (τ ), f 2n (ζ), and f 2n (ζ), which are the Fourier
coefficients of the series in (4.35) and (4.37). And this is just that moment in our
development when the superscripts ‘c’ and ‘s’ come into the play.
Expressing the coefficients of the series (4.35) and (4.37) in terms of their gener-
ating functions f 1 (τ , ψ) and f 2 (ζ, ψ) as
4.3 Hemisphere Joint to Cylinder 125
2π
(c) 1
f 1n (τ ) = f 1 (τ , ψ) cos nψdψ, n = 0, 1, 2, . . .
π 0
2π
(s) 1
f 1n (τ ) = f 1 (τ , ψ) sin nψdψ, n = 1, 2, 3, . . .
π 0
2π
(c) 1
f 2n (ζ) = f 2 (ζ, ψ) cos nψdψ, n = 0, 1, 2, . . .
π 0
and 2π
(s) 1
f 2n (ζ) = f 2 (ζ, ψ) sin nψdψ, n = 1, 2, 3, . . .
π 0
(c) (c)
we substitute f 1n (τ ) and f 2n (ζ) into (4.46), (4.47), (4.50) and (4.51)
π 2π
2 1 n
u (c)
1n (ϑ) = g (ϑ, τ ) cos nψ f 1 (τ , ψ)a 2 sin τ dψdτ
0 0 π 11
l 2π
1 n
+ g (ϑ, ζ) cos nψ f 2 (ζ, ψ)dψdζ, n = 0, 1, 2, . . .
0 0 π 12
and π 2π
2 1 n
u (c)
2n (z) = g (z, τ ) cos nψ f 1 (τ , ψ)a 2 sin τ dψdτ
0 0 π 21
l 2π
1 n
+ g (z, ζ) cos nψ f 2 (ζ, ψ)dψdζ, n = 0, 1, 2, . . .
0 0 π 22
(s) (s)
Substituting then f 1n (τ ) and f 2n (ζ) into (4.50) and (4.51), we have
π 2π
2 1 n
u (s)
1n (ϑ) = g (ϑ, τ ) sin nψ f 1 (τ , ψ)a 2 sin τ dψdτ
0 0 π 11
l 2π
1 n
+ g (ϑ, ζ) sin nψ f 2 (ζ, ψ)dψdζ, n = 1, 2, 3, . . .
0 0 π 12
and π 2π
2 1 n
u (s)
2n (z) = g (z, τ ) sin nψ f 1 (τ , ψ)a 2 sin τ dψdτ
0 0 π 21
l 2π
1 n
+ g (z, ζ) sin nψ f 2 (ζ, ψ)dψdζ, n = 1, 2, 3, . . .
0 0 π 22
126 4 Compound Structures
l 2π
+ G 12 (ϑ, ϕ; ζ, ψ) f 2 (ζ, ψ)dψdζ, (ϑ, ϕ) ∈ 1
0 0
revealing the first two entries G 11 (ϑ, ϕ; τ , ψ) and G 12 (ϑ, ϕ; ζ, ψ) of the matrix
G(P, Q) of Green’s type to the homogeneous boundary value problem corresponding
to (4.28)–(4.33). For G 11 (ϑ, ϕ; τ , ψ) we came up with the series expansion
∞
1 0 1 n
G 11 (ϑ, ϕ; τ , ψ) = g11 (ϑ, τ )+ g (ϑ, τ ) cos n(ϕ−ψ)
2π π n=1 11
0
whose coefficient functions g11 (ϑ, τ ) and g11
n
(ϑ, τ ) have been derived earlier in this
section. Substituting them into the above series, we transform it, for ϑ ≤ τ , to
1 τ
G 11 (ϑ, ϕ; τ , ψ) = λ(β +l) − ln tan
2π 2
ϑ
∞
1 1 τ
+ ∗
(βn+1)enl/a tann (1+λ) cot n
2π n=1 n 2 2
τ ϑ
− (1−λ) tann −(βn−1)e−nl/a tann
2 2
τ τ
× (1−λ) cot n −(1+λ) tann cos n(ϕ−ψ). (4.52)
2 2
The variables ϑ and τ should be interchanged in the above in order to get the
expression of G 11 (ϑ, ϕ; τ , ψ) valid for τ ≤ ϑ.
For the entry G 12 (ϑ, ϕ; ζ, ψ) of G(P, Q), we accordingly have
1
G 12 (ϑ, ϕ; ζ, ψ) = [β + (l − ζ)]
2π
∞
1 1 ϑ
+ ∗
(1+βn)en(l−ζ)/a−(1−βn)en(ζ−l)/a tann cos n(ϕ−ψ). (4.53)
2π n=1 n 2
Analogously, upon substituting the expressions for u c2n (z) and u s2n (z) into the
expansion of u 2 (z, ϕ) from (4.36), one arrives at
4.3 Hemisphere Joint to Cylinder 127
π 2π
2
u 2 (z, ϕ) = G 21 (z, ϕ; τ , ψ) f 1 (τ , ψ)a 2 sin τ dψdτ
0 0
l 2π
+ G 22 (z, ϕ; ζ, ψ) f 2 (ζ, ψ)dψdζ, (z, ϕ) ∈ 2
0 0
disclosing the other two elements of the matrix G(P, Q) of Green’s type which we
are looking for. The element G 21 (z, ϕ; τ , ψ) is obtained as
λ
G 21 (z, ϕ; τ , ψ) = [β +(l −z)]
2π
∞
λ 1 τ
+ (1+βn)en(l−z)/a −(1−βn)en(z−l)/a tann cos n(ϕ−ψ) (4.54)
π n=1 n∗ 2
and, at last, for the expression of G 22 (z, ϕ; ζ, ψ), valid for z ≤ ζ, we have
∞
β + (l −ζ) 1 (1−λ)e−nz/a +(1+λ)enz/a
G 22 (z, ϕ; ζ, ψ) = +
2π 2π n=1 n∗
× (1 +βn)en(l−ζ)/a −(1 −βn)en(ζ−l)/a cos n(ϕ−ψ) (4.55)
and the variables z and ζ should be interchanged in the above, to get an expression
for G 22 (z, ϕ; ζ, ψ) valid for ζ ≤ z.
Evidently, the convergence of the series in the just found elements G i j (P, Q) of
G(P, Q) can be improved in the way recommended earlier in this volume. But it
is worth noting that there exists an important particular case of the problem setting
in (4.28)–(4.33) for which all the series allow a complete summation. Indeed, if (i)
both the fragments in the assembly of shells are made of the same material, in which
case λ1 = λ2 and, subsequently, λ = 1; and (ii) the Dirichlet boundary condition is
imposed at z = l in (4.30) (implying that β = 0), then the entries of G(P, Q) reduce
to
1 0
G 11 (ϑ, ϕ; τ , ψ) = g (ϑ, τ )
2π 11
1 e2l/a −2(ϑ)(τ ) cos(ϕ−ψ)+e−2l/a 2 (ϑ)2 (τ )
+ ln
2π 2 (ϑ) −2(ϑ)(τ ) cos(ϕ−ψ)+2 (τ )
1 0
G 12 (ϑ, ϕ; ζ, ψ) = g (ϑ, ζ)
2π 12
1 e2l/a −2eζ/a (ϑ) cos(ϕ− ψ)+e2(ζ−l)/a 2 (ϑ)
+ ln
2π e2ζ/a −2eζ/a (ϑ) cos(ϕ− ψ)+2 (ϑ)
128 4 Compound Structures
1 0
G 21 (z, ϕ; τ , ψ) = g (z, τ )
2π 21
1 e2l/a −2ez/a (τ ) cos(ϕ− ψ)+e2(z−l)/a 2 (τ )
+ ln
2π e2z/a −2ez/a (τ ) cos(ϕ− ψ)+2 (τ )
and
1 0
G 22 (z, ϕ; ζ, ψ) = g (z, ζ)
2π 22
1 e2(l−z)/a −2e(ζ−z)/a cos(ϕ−ψ)+e2(ζ−l)/a
+ ln
2π 1−2e(ζ−z)/a cos(ϕ−ψ)+e2(ζ−z)/a
of two fragments. But it is worth noting that the number of fragments in an assembly
does not represent an issue for our approach. Justifying this assertion in what follows,
we are going to present a quite convincing example. 3
To introduce a three-fragment thin-wall assembly = i=1 i , the axial cross
section of which is depicted in Fig. 4.6, let a thin spherical shell of radius a be
composed of two fragments whose middle surfaces occupy the regions
and
2 = {(ϑ, ϕ)| ϑ1 < ϑ < π, 0 < ϕ < 2π}.
Let, in addition, a thin annular plate of radii b and c, with the middle plane occupying
the region
3 = {(r, ϕ)| b <r < c, 0 < ϕ < 2π}
attached to the sphere in the way shown. Let also each fragment i in be made of
an isotropic homogeneous conductive material whose conductivity is specified by a
constant λi .
Under the assumption of insulated facial surfaces of the fragments i , the two-
dimensional field of potential in the encountered assembly can be simulated by
the following set of equations:
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + = − f 1 (ϑ, ϕ) (4.56)
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
1 ∂ ∂u 2 (ϑ, ϕ) 1 ∂ 2 u 2 (ϑ, ϕ)
sin ϑ + 2 2 = − f 2 (ϑ, ϕ) (4.57)
a sin ϑ ∂ϑ
2 ∂ϑ a sin ϑ ∂ϕ2
130 4 Compound Structures
and
1 ∂ ∂u 3 (r, ϕ) 1 ∂ 2 u 3 (r, ϕ)
r + 2 = − f 3 (r, ϕ) (4.58)
r ∂r ∂r r ∂ϕ2
lim |u 1 (ϑ, ϕ)| < ∞, lim |u 2 (ϑ, ϕ)| < ∞, u 3 (c, ϕ) = 0 (4.59)
ϑ→0 ϑ→π
and
∂u 1 (ϑ1 , ϕ) ∂u 2 (ϑ1 , ϕ) ∂u 3 (b, ϕ)
λ1 − λ2 − λ3 = 0. (4.61)
∂ϑ ∂ϑ ∂r
Clearly, the radius a of the spherical shell, the inner radius b of the plate, and
the parameter ϑ1 are not independent in the above problem statement. Indeed, from
Fig. 4.6, it follows that
b = a sin ϑ1 .
We believe that it is worth reminding that the first two conditions in (4.59) reflect
the fact that ϑ = 0 and ϑ = π represent points of singularity for equations (4.56) and
(4.57).
Thus, our goal, in what follows, is the three-by-three matrix of Green’s type for the
well-posed homogeneous boundary value problem corresponding to (4.56)–(4.61).
The problem setting in (4.56)–(4.61) is 2π-periodic with respect to the longitudinal
variable ϕ. This type of problem statements was encountered numerous times before
in our volume. It allows expansion of the unknown functions u i ( p, ϕ) in the standard
trigonometric Fourier series
∞
1 (c) (s)
u i ( p, ϕ) = u i,0 ( p)+ u i,n ( p) cos nϕ+u i,n ( p) sin nϕ, i = 1, 2, 3 (4.62)
2 n=1
where the conditional p notation is introduced for the first variable in u i ( p, ϕ). This
is done for generality since the latitudinal coordinates in u 1 (ϑ, ϕ) and u 2 (ϑ, ϕ), on
one hand, and u 3 (r, ϕ), on the other hand, are different.
The expansions in (4.62) routinely yield an ODE boundary value problem in the
coefficients of the Fourier series. Namely, we arrive at the following set of ordinary
differential equations:
4.4 Three-Fragment Assembly 131
d du 1,n (ϑ) n 2 u 1,n (ϑ)
sin ϑ − = −
f 1,n (ϑ) (4.63)
dϑ dϑ sin ϑ
d du 2,n (ϑ) n 2 u 2,n (ϑ)
sin ϑ − = −
f 2,n (ϑ) (4.64)
dϑ dϑ sin ϑ
and
d du 3,n (r ) n 2 u 3,n (r )
r − = −
f 3,n (r ) (4.65)
dr dr r
and
du 1,n (ϑ1 ) du 2,n (ϑ1 ) du 3,n (b)
λ1 − λ2 − λ3 = 0. (4.68)
dϑ dϑ dr
The right-hand sides of equations in (4.63)–(4.65) are expressed in terms of the
coefficients f i,n ( p) of the Fourier series
∞
1 (c) (s)
f i ( p, ϕ) = f i,0 ( p)+ f i,n ( p) cos nϕ+ f i,n ( p) sin nϕ, i = 1, 2, 3
2 n=1
f 1,n (ϑ) = a 2 sin ϑ f 1,n (ϑ) ,
f 2,n (ϑ) = a 2 sin ϑ f 2,n (ϑ)
and
f 3,n (r ) = r f 3,n (r ) .
and
1 r
ρ2n − r 2n
u 3,n (r ) = f 3,n (ρ) dρ + C3,n r n + D3,n r −n , (4.71)
2n b r n ρn
λ1 = λ2 = λ3 .
b
a≡ = 1.
sin ϑ1
of linear algebraic equations in the remaining constants C1,n , D2,n , C3,n , and D3,n ,
where I1 , I2 , and I3 are found as
ϑ1
1 n (τ ) n (ϑ1 )
I1 = − f 1,n (τ ) dτ ,
2n 0 n (ϑ1 ) n (τ )
ϑ1
1 n (τ ) n (ϑ1 )
I2 = + f 1,n (τ ) dτ ,
2n 0 n (ϑ1 ) n (τ )
4.4 Three-Fragment Assembly 133
and c
1 ρn cn
I3 = − f 3,n (ρ) dρ.
2n b cn ρn
ϑ1 n
n (ϑ1 ) b cn n (τ )
D2,n = 2 − f 1,n (τ ) dτ
2n 0 cn bn n (ϑ1 )
π n
b cn n (ϑ1 )
+ + f 2,n (τ ) dτ
ϑ1 cn bn n (τ )
c n
ρ cn
2 − f 3,n (ρ) dρ ,
b cn ρn
ϑ1 n π n
1 (τ ) (ϑ1 )
C3,n = f 1,n (τ ) dτ + f 2,n (τ ) dτ
ncn 0 n (ϑ1 ) ϑ1 (τ )
n
c n
3 ρ cn
+ − f 3,n (ρ) dρ,
2nbn b cn ρn
and
ϑ1 n π n
cn (τ ) (ϑ1 )
D3,n = − f 1,n (τ ) dτ + f 2,n (τ ) dτ
n 0 n (ϑ1 ) ϑ1 (τ )
n
bn c
ρn cn
− − n f 3,n (ρ) dρ,
2n b c n ρ
where
b2n − 3c2n
= .
bn cn
134 4 Compound Structures
Substituting the just presented expressions for the constants Ci and Di into (4.69)–
(4.71), after some transformations, we have
ϑ
1 n (ϑ) n (τ ) bn cn n (τ )
u 1,n (ϑ) = + + f 1,n (τ ) dτ
2n 0 2n (ϑ1 ) cn bn n (ϑ)
ϑ1 n
1 (ϑ) n (τ ) bn cn n (ϑ)
+ + n + n f 1,n (τ ) dτ
2n ϑ 2n (ϑ1 ) cn b (τ )
π n n
1 (ϑ) b cn
+ − f 2,n (τ ) dτ
n ϑ1 n (τ ) cn bn
c n n
1 (ϑ) ρ cn
+ − f 3,n (ρ) dρ,
n b n (ϑ1 ) cn ρn
ϑ1
1 n (τ ) bn cn
u 2,n (ϑ) = − n f 1,n (τ ) dτ
n 0 n (ϑ) cn b
ϑ n
1 2n (ϑ1 ) b cn n (τ )
+ + + f 2,n (τ ) dτ
2n ϑ1 n (ϑ) n (τ ) cn bn n (ϑ)
π n
1 2n (ϑ1 ) b cn n (ϑ)
+ + + f 2,n (τ ) dτ
2n ϑ n (ϑ) n (τ ) cn bn n (τ )
c n
1 (ϑ1 ) ρn cn
+ − f 3,n (ρ) dρ,
n b n (ϑ) cn ρn
and
ϑ1
1 n (τ ) r n cn
u 3,n (r ) = − n f 1,n (τ ) dτ
n 0 n (ϑ1 ) cn r
π n n
1 (ϑ1 ) r cn
+ − n f 2,n (ϑ) dϑ
n ϑ1 n (ϑ) cn r
r
1 rn cn ρn bn
+ − n 3 n − n f 3,n (ρ) dρ
2n b cn r b ρ
c n
1 ρ cn rn bn
+ − n 3 n − n f 3,n (ρ) dρ,
2n r cn ρ b r
from which one directly derives the elements gi j,n ( p, q) of the three-by-three matrix
gn ( p, q) of Green’s type to the homogeneous ODE boundary value problem corre-
sponding to (4.63)–(4.68). As to the conditional notations p and q, introduced here
for the arguments of gn ( p, q), the reader is advised to revisit our comment provided
earlier (see the text following Eq. (4.62)) regarding the first variables in the functions
u i ( p, ϕ).
4.4 Three-Fragment Assembly 135
and
1 n (ϑ) ρn cn
g13,n (ϑ, ρ) = − . (4.75)
n n (ϑ1 ) cn ρn
and
1 n (ϑ1 ) ρn cn
g23,n (ϑ, ρ) = − . (4.78)
n n (ϑ) cn ρn
and n
1 ρn cn r bn
g33,n (r, ρ) = − 3 n − n . (4.81)
2n cn ρn b r
and r ρ
u 3,0 (r ) = ln
f 3,0 (ρ) dρ + C3,0 ln r + D3,0 . (4.84)
b r
ϑ1 π
C3,0 = −
f 1,0 (τ ) dτ −
f 2,0 (τ ) dτ ,
0 ϑ1
c ϑ1 π
D1,0 = ln
f 1,0 (τ ) dτ +
f 2,0 (τ ) dτ
b 0 ϑ1
c ϑ1
ρ (τ )
− ln f 3,0 (ρ) dρ − ln f 1,0 (τ ) dτ ,
b c 0 (ϑ1 )
c ϑ1 π
D2,0 = ln
f 1,0 (τ ) dτ +
f 2,0 (τ ) dτ
b 0 ϑ
c 1π
ρ
− ln f 3,0 (ρ) dρ − ln (ϑ1 ) f 2,0 (τ ) dτ ,
b c ϑ1
and
ϑ1 π
D3,0 = ln c
f 1,0 (τ ) dτ +
f 2,0 (τ ) dτ
0 ϑ1
c ρ
− ln
f 3,0 (ρ) dρ.
b c
If the found above expressions for Ci,0 and Di,0 , (i = 1, 2, 3) are substituted into
(4.82)–(4.84), the latter reduce to
4.4 Three-Fragment Assembly 137
ϑ c
(ϑ1 )
u 1,0 (ϑ) = ln + ln f 1,0 (τ ) dτ
0 (ϑ) b
ϑ1 c
(ϑ1 )
+ ln + ln f 1,0 (τ ) dτ
ϑ (τ ) b
π c
c c
+ ln f 2,0 (τ ) dτ + ln f 3,0 (ρ) dρ,
ϑ1 b b ρ
ϑ c
(τ )
u 2,0 (ϑ) = ln + ln f 2,0 (τ ) dτ
ϑ1 (ϑ1 ) b
π c
(ϑ)
+ ln + ln f 2,0 (τ ) dτ
ϑ (ϑ1 ) b
ϑ1 c
c c
+ ln f 1,0 (τ ) dτ + ln f 3,0 (ρ) dρ,
0 b b ρ
and
ϑ1 c π
c
u 3,0 (r ) = ln f 1,0 (τ ) dτ + ln f 2,0 (τ ) dτ
0 r ϑ r
r c1
c c
+ ln f 3,0 (ρ) dρ + ln f 3,0 (ρ) dρ,
b r r ρ
∞
1 1
G i j ( p, ϕ; q, ψ) = gi j,0 ( p, q) + gi j,n ( p, q) cos n (ϕ − ψ) , (4.92)
2 2π n=1
appears also workable in the current case. That is, we transform the expression in
(4.93) as shown
1 bn cn bn bn
= 2n + n − n
b − 3c 2n 3c 3c
1 b3n bn
= n 2n ! − n. (4.94)
3c b − 3c2n 3c
This splits the factor 1/ into two additive components, and makes completely
summable the series in (4.92) associated with the second component of (4.94). The
summation can be accomplished with the aid of the standard summation formula
∞
pn
cos nx = − ln 1 − 2 p cos x + p 2 .
n=1
n
The series in (4.92) associated with the first additive component of (4.94) con-
verges at a very high rate, and could therefore be immediately involved in computa-
tional routines without affecting their efficiency.
Leaving the detailed algebra for the reader, we present just ultimate expressions
for the elements of G( p, ϕ; q, ψ). The elements G 1 j ( p, ϕ; q, ψ) of the first row are
found as
2
1 c 1 b (ϑ) (τ )
G 11 (ϑ, ϕ; τ , ψ) = ln + H , ϕ−ψ
2 b 3 c2 2 (ϑ1 )
1 (ϑ) (τ ) 1 (ϑ1 ) (ϑ)
+ H , ϕ−ψ + ln −H , ϕ−ψ
3 2 (ϑ1 ) 2 (τ ) (τ )
∞
1 (b2n +c2n )n (ϑ) n (τ )
+ cos n (ϕ−ψ) ,
4π n=1 n∗ bn cn 2n (ϑ1 )
2
1 c 2 b (ϑ)
G 12 (ϑ, ϕ; τ , ψ) = ln + H 2 ,ϕ − ψ
2 b 3 c (τ )
∞
2 (ϑ) 1 (b2n −c2n )n (ϑ)
− H , ϕ−ψ + cos nϕ−ψ,
3 (τ ) 2π n=1 n∗ bn cn n (τ )
140 4 Compound Structures
and
1 c 2 bρ (ϑ)
G 13 (ϑ, ϕ; ρ, ψ) = ln + H 2 ,ϕ − ψ
2 ρ 3 c (ϑ1 )
∞
2 b (ϑ) 1 (ρ2n −c2n )n (ϑ)
− H , ϕ−ψ + cos n (ϕ−ψ) .
3 ρ (ϑ1 ) 2π n=1 n∗ cn ρn n (ϑ1 )
2 2
1 c 1 b (ϑ1 )
G 22 (ϑ, ϕ; τ , ψ) = ln + H 2 , ϕ−ψ
2 b 3 c (ϑ) (τ )
1 2 (ϑ1 ) 1 (ϑ) (ϑ)
+ H , ϕ−ψ + ln −H , ϕ−ψ
3 (ϑ) (τ ) 2 (ϑ1 ) (τ )
∞
1 (b2n +c2n )n (ϑ1 )
+ cos n (ϕ − ψ) ,
4π n=1 n∗ bn cn n (ϑ) (τ )
and
1 c 2 bρ (ϑ1 )
G 23 (ϑ, ϕ; ρ, ψ) = ln + H , ϕ − ψ
2 ρ 3 c2 (ϑ)
∞
2 b (ϑ1 ) 1 (ρ2n −c2n )n (ϑ1 )
− H , ϕ−ψ + cos n (ϕ−ψ) .
3 ρ (ϑ) 2π n=1 n∗ cn ρn n (ϑ)
1 b2
and
ρr
G 33 (r, ϕ; ρ, ψ) = H , ϕ − ψ + H , ϕ − ψ
c2 3 ρr
2
1 c r 1 b ρ
ln −H ,ϕ − ψ − H 2 ,ϕ − ψ
2 ρ ρ 3 c r
∞
1 (ρ2n −c2n )(3r 2n −b2n )
+ cos n (ϕ−ψ) ,
4π n=1 n∗ bn cn r n ρn
1 !
H (x, α) = ln 1 − 2x cos α + x 2
4π
and
3cn (b2n − 3c2n )
∗ = .
b2n
An important comment is appropriate as to the presented above elements of the
sought-after matrix of Green’s type G( p, ϕ; q, ψ). The comment reiterates actually
what we had already mentioned when the ODE matrices of Green’s type g0 ( p, q)
and gn ( p, q) were discussed. Namely, the peripheral (i = j) elements G i j (r, ϕ; ρ, ψ)
of G( p, ϕ; q, ψ) are defined in a single piece each. But as to the diagonal elements
G 11 (r, ϕ; ρ, ψ), G 22 (r, ϕ; ρ, ψ), and G 33 (r, ϕ; ρ, ψ), their forms presented above
are valid for ϑ ≤ τ and r ≤ ρ, respectively, while their expressions valid for ϑ ≥ τ
and r ≥ ρ can be obtained from those presented by interchanging ϑ with τ and r
with ρ, respectively.
The potential field presented in Fig. 4.7 was computed with the series compo-
nents in G i j (r, ϕ; ρ, ψ) truncated to just the fifth partial sums indicating their fast
convergence rate. This convincingly demonstrates effectiveness of our technique and
illustrates high computational capacity of the found above representations for the ele-
ments of the sought-after matrix of Green’s type G( p, ϕ; q, ψ) to the homogeneous
boundary value problem corresponding to (4.56)–(4.61). Parameters defining geo-
metrical and physical properties of fragments of the assembly are fixed as a = 1.0,
ϑ1 = 0.4π, c = 1.75, and λ1 = λ2 = λ3 , while the unit point source is released in
1 at (0.28π, 0.45π).
142 4 Compound Structures
The set of exercises that we offer below is designed to provide a further assistance
to the reader in the development of necessary skills required for constructing matrices
of Green’s type to boundary value problems that simulate potential fields induced in
thin-wall assemblies. Customarily, answers to the proposed exercises are available
in the appendix section of our volume.
Exercise 4.1
1 ∂ ∂u 1 (ϑ, ϕ) a 2 ∂ 2 u 1 (ϑ, ϕ)
D (ϑ) + = 0, in 1
D (ϑ) ∂ϑ ∂ϑ D 2 (ϑ) ∂ϕ2
1 ∂ ∂u 2 (θ, ϕ) b2 ∂ 2 u 2 (θ, ϕ)
D (θ) + = 0, in 2
D (θ) ∂θ ∂θ D 2 (θ) ∂ϕ2
∂u 1 (ϑ1 , ϕ) ∂u 2 (θ0 , ϕ)
u 1 (ϑ1 , ϕ) = u 2 (θ0 , ϕ) , =λ
∂ϑ ∂θ
4.5 Chapter Exercises 143
where
1 = {(ϑ, ϕ)| ϑ0 < ϑ < ϑ1 , 0 < ϕ < ϕ1 }
and
2 = {(θ, ϕ)| θ0 < θ < θ1 , 0 < ϕ < ϕ1 }
Exercise 4.2
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + = 0, in 1
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
1 ∂ ∂u 2 (r, ϕ) 1 ∂ 2 u 2 (r, ϕ)
r + 2 = 0, in 2
r ∂r ∂r r ∂ϕ2
∂u 1 (ϑ0 , ϕ)
= 0, u 2 (c, ϕ) = 0
∂ϑ
∂u 1 (ϑ, 0) ∂u 1 (ϑ, 2π)
u 1 (ϑ, 0) = u 1 (ϑ, 2π) , =
∂ϕ ∂ϕ
∂u 1 (ϑ1 , ϕ) ∂u 2 (b, ϕ)
u 1 (ϑ1 , ϕ) = u 2 (b, ϕ) , =λ
∂ϑ ∂r
where
1 = {(ϑ, ϕ)| ϑ0 < ϑ < ϑ1 , 0 ≤ ϕ < 2π}
2 = {b ≤r ≤ c, 0 ≤ ϕ < 2π}
a sin ϑ1 = b.
144 4 Compound Structures
Exercise 4.3
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + 2 2 = 0, in 1
a sin ϑ ∂ϑ
2 ∂ϑ a sin ϑ ∂ϕ2
1 ∂ ∂u 2 (θ, ϕ) b2 ∂ 2 u 2 (θ, ϕ)
D (θ) + = 0, in 2
D (θ) ∂θ ∂θ D 2 (θ) ∂ϕ2
lim u 1 (ϑ, ϕ) < ∞, u 2 (θ1 , ϕ) = 0
ϑ→0
∂u 1 (ϑ1 , ϕ) ∂u 2 (θ0 , ϕ)
u 1 (ϑ1 , ϕ) = u 2 (θ0 , ϕ) , =λ
∂ϑ ∂θ
where
1 = {(ϑ, ϕ)| 0 < ϑ < ϑ1 , 0 ≤ ϕ < 2π}
a sin ϑ1 = R + b sin θ0 .
Exercise 4.4
∂ 2 u 1 (z, ϕ) 1 ∂ 2 u 1 (z, ϕ)
+ = 0, in 1
∂z 2 a 2 ∂ϕ2
1 ∂ ∂u 2 (ϑ, ϕ) b2 ∂ 2 u 2 (ϑ, ϕ)
D (ϑ) + = 0, in 2
D (ϑ) ∂ϑ ∂ϑ D 2 (ϑ) ∂ϕ2
∂u 1 (l, ϕ) ∂u 1 (z, ϕ1 )
= 0, u 1 (z, 0) = 0, =0
∂z ∂ϕ
4.5 Chapter Exercises 145
∂u 2 (ϑ0 , ϕ) ∂u 2 (ϑ, ϕ1 )
= 0, u 2 (ϑ, 0) = 0, =0
∂ϑ ∂ϕ
∂u 1 (0, ϕ) ∂u 2 (ϑ1 , ϕ)
u 1 (0, ϕ) = u 2 (ϑ1 , ϕ) , =λ
∂z ∂θ
where
1 = {(z, ϕ)| 0 < z < l, 0 < ϕ < ϕ1 }
and
2 = {(ϑ, ϕ)| ϑ0 < ϑ < ϑ1 , 0 < ϕ < ϕ1 }
For methodological purposes, before exposing the reader to specifics of the com-
putational Green’s function approach [18, 23, 28, 30, 33, 43] to boundary value
problems, it is reasonable, in our opinion, to refer to the peculiarities of the classical
BIEM. Remind that this abbreviation was accepted in our volume for the bound-
ary integral equation method [12, 23, 36, 37, 41]. The reason for referring to this
well-known method is that our Green’s function approach represents its modified
version.
The BIEM represents one of the so-called meshless methods that reduce the
dimensionality of a targeted boundary value problem. A two-dimensional, for exam-
ple, problem reduces to a one-dimensional boundary integral equation. This brings
significantly down the required computer time. Providing arguments for this com-
ment, note that it is needless to advocate advantage of the dimension reduction in the
numerical analysis: just compare computational expenditures required for an ODE
boundary value problems, on one hand, and a two-dimensional PDE boundary value
problem, on the another hand.
The second critical argument, which stays behind the effectiveness of the BIEM,
flows out from the replacement of a differential operator in the original boundary
value problem with an integral operator in the problem to be actually numerically
tackled. It is commonly recognized that the numerical integration is a well-posed
procedure, in contrast to the numerical differentiation which is not [2] quite well-
posed computationally.
It is worth noting that, in addition to all the foregoing aspects of the BIEM that
contribute to its computational efficiency, the Green’s function version brings up
some extra features which notably enhance the overall computational potential of
the approach. We are going to focus on those enhancing features later in the current
section.
As to the terms regular and irregular which will be used in regard to a region’s
shape, the following comment is appropriate. We say that a region is regular in shape
if its boundary represents a set of segments of coordinate lines in a selected coordinate
system. If at least a part of the region’s boundary does not coincide with coordinate
lines, the shape will be referred to as irregular.
Let be a double-connected fragment of a surface of irregular configuration (see
Fig. 5.1). The outer boundary contour 0 of is supposed to be regular or is, in
other words, defined with coordinate lines of an accepted coordinate system on the
surface, whereas coordinate lines of that system do not define the inner contour
making the region’s shape irregular.
In , we consider the boundary value problem
u(P) = 0, P ∈ 0 (5.2)
5.1 Green’s Function Version of the BIEM 149
where is a second-order elliptic partial differential operator which makes the above
problem simulating a potential phenomenon in ; and h(P) is an integrable on
function. This specific problem statement is chosen on purpose. It is convenient, on
one hand, to highlight basics of our approach, and at the same time, it allows an easy
extension to other problem formulations that might emerge.
So, imagine now that the problem in (5.1)–(5.3) is treated by one of the standard
BIEM approaches, and its solution u(P) is respectively expressed in as the sum
of two line integrals:
u(P) = F(P, Q)μ0 (Q)d(Q) + F(P, Q)μ(Q)d(Q), P ∈ , (5.4)
0
where the kernel function F(P, Q) represents the fundamental solution to the gov-
erning equation in (5.1), and the weight functions μ0 (Q) and μ(Q) are integrable
on 0 and , respectively. Conventionally, the functions μ0 (Q) and μ(Q) will be
referred, in what follows, as the density functions, and they are yet to be determined.
Note that u(P) in (5.4) represents, due to F(P, Q), a solution to (5.1) with any
allowable density functions. So, (5.4) is an analytical form of the solution to the
problem in (5.1)–(5.3) defined up to μ0 (Q) and μ(Q). To determine the latter, we
take a limit of u(P) as the field point P approaches the outer boundary segment 0
yielding
F(P, Q)μ0 (Q)d0 (Q) + F(P, Q)μ(Q)d(Q) = 0, P ∈ 0 (5.5)
0
The relations in (5.5) and (5.6) represent a system of linear weakly singular integral
equations in the density functions μ0 (Q) and μ(Q).
The singularity of the above equations is associated with the fundamental solution
F(P, Q) and does not represent a significant issue for their numerical solution.
Another feature of these equations does represent, however, a problem: (5.5) and
(5.6) are integral equations of the first kind or, in other words, they are irregular. That
is why their numerical solution should be based on some regularizing algorithms.
At this stage in our presentation, we leave aside the regularity issue for the system
in (5.5) and (5.6), which we will, however, return to in our further discussion. Hence,
once approximate values of the density functions μ0 (Q) and μ(Q) are found, the
representation in (5.4) provides us with an approximate analytical solution to the
boundary value problem posed in (5.1)–(5.3).
In summary, two stages of a numerical procedure are in place to approximately
solve the problem setting in (5.1)–(5.3) within the scope of the discussed version of
the classical BIEM. First, a reliable numerical algorithm ought to be developed to
accurately solve the system of integral equations in (5.5) and (5.6). And second, an
efficient numerical integration is required to compute approximate values of u(P)
inside of .
Before turning to specifics of a procedure that we had developed for solution of
problems of the type in (5.1)–(5.3) by a Green’s function version of the BIEM, it
is with a great pleasure that the first author recollects a remarkable contribution of
professor Kupradze to the field. His guidance has been inspiring to our investigation
for decades. His priceless and wise recommendations had created a foundation for
our productive work in this challenging area of research.
While discussing advantages and shortcomings of different numerical methods,
traditionally used to solve boundary value problems for applied PDEs, the Master
had challenged us (a group of his pupils and followers) to think about possible incor-
poration of Green’s functions into the classical BIEM numerical schemes. Speaking
about this kind of perspectives, he had outlined its specific pros and cons focusing
on advantages that this incorporation could yield.
It had taken quite some time before we managed to translate into reality the
Master’s recommendations. But it was ultimately done, and our study in the field
resulted into numerous works, the first of which [30] was published in 1977 where
most of the Kupradze’s ideas had found a fertile soil. A reality significantly surpassed
most of our positive expectations. And to make a long story short, we are going to
proceed directly to our presentation of the Green’s function version of the BIEM.
Going to the point, we turn again to the setting of (5.1)–(5.3). And let G 0 (P, Q)
represent the Green’s function to the boundary value problem stated by (5.1) and (5.2)
in the simply connected region bounded with 0 (see Fig. 5.1). Due to its defining
properties, G 0 (P, Q), as a function of P, makes not only the governing equation
5.1 Green’s Function Version of the BIEM 151
in the density function μ(Q). This equation is singular because of its kernel function
G 0 (P, Q) which possesses the logarithmic singularity. But as well as in the case of
the equations in (5.5) and (5.6), the singularity is not the biggest issue for solution
of (5.8), whereas its irregularity is.
To avoid the foregoing hurdle, we had developed in [30] an alternative strategy for
solving problems of the type in (5.1)–(5.3). It implements another Kupradze’s rec-
ommendation proposed in [26], which provides in fact a regularizing effect. Namely,
instead of expressing the solution to (5.1)–(5.3) in the form of (5.7), we look for it
in a slightly different form as
u(P) = G 0 (P, Q)μ(Q)d (Q), P ∈ , (5.9)
where represents a contour that is outside of and closely located to the actual
contour of the aperture (see Fig. 5.2), while the density function μ(Q) is to be
found. will be referred to in our presentation as the fictitious contour. Satisfying
then the boundary condition in (5.3) by taking the field point P in (5.9) to , we have
h(P) = G 0 (P, Q)μ(Q)d (Q), P ∈ , (5.10)
contrast to (5.8), a regular functional equation, with its numerical solution promising
to be painless.
So, once the equation in (5.10) is solved, for a fixed form and location of the
fictitious contour , and approximate values of the density function μ(Q) are found,
one might consider the form in (5.9) as an analytical solution to the problem in
(5.1)–(5.3).
The authors of [26] suggested to call the just described regularizing approach as
the method of functional equations. It should be noted that this title is perfectly ade-
quate with the method’s nature. But more than twenty years after [26] was published,
a terminological lapse had happened. An absolutely pointless title-flip attempt was
undertaken in [7], where a different title, the fundamental solutions method, was
proposed for this very approach. And what strikes most is that the newly intro-
duced title is simply irrelevant to the method’s nature being neither informative nor
uniquely distinguishable. But it unfortunately happened misleading a notable part of
the computational mathematics community which accepted the title-flip. This had
subsequently created a very confusing if not distractive situation where two distinct
titles became associated with a single method.
This brief volume is not perhaps a suitable place to analyze the foregoing unfortu-
nate situation in more detail. However, for a reader who is not indifferent to justice or
is simply curious enough, we suggest to visit the introduction section of [36] where
at least some aspects of our viewpoint on the issue are clearly explained.
Moving further with our presentation, we turn back to the functional equation in
(5.10). Since its kernel function G 0 (P, Q) reveals no singularity, the only remain-
ing issue in the described procedure is the regularizing shape and location of the
fictitious contour . Our experience, gained through years of practical work in the
field, suggests that this issue can easily be resolved on a case-by-case basis. That is,
a brief numerical experiment should be conducted prior to the main computer work
for every class of particular problems.
5.1 Green’s Function Version of the BIEM 153
To give the reader a good sense of the computational power of the proposed
approach and to provide some reliable data that illustrate the accuracy level practically
attainable, we will formulate a sample problem of the type in (5.1)–(5.3). With a vast
number of computer-friendly representations of Green’s functions derived in this
volume, it appears doable to formulate such a sample problem whose exact solution
is available.
In doing so, let a double-connected region be the hemispherical surface
to the boundary value problem in (5.11) and (5.12) stated in the solid hemispherical
region . That is
F(ϑ, ϕ) = G(ϑ, ϕ; τ ∗ , ψ ∗ ).
The reader is referred to Chap. 2 where the Green’s function presented in (5.14)
was derived. Here (x) = tan(x/2), and the source is placed at a point (τ ∗ , ψ ∗ )
which belongs to but is out of . This clearly makes (τ ∗ , ψ ∗ ) belonging to the
simply connected region bounded with .
With all the above said, it becomes evident that the two-variable function
G(ϑ, ϕ; τ ∗ , ψ ∗ ) indeed represents the exact solution to the problem setting of (5.11)–
(5.13) in . Hence, this problem makes an excellent target for numerical experiments
whose purpose is to determine optimal values of all computational parameters in our
procedure.
To be specific, let the inner contour in the problem of (5.11)–(5.13) be a circle
of radius ρ = 0.3a centered at (π/4, π/2). And let also the source point in the right-
hand side of (5.13) be fixed at (π/4, π/2). This makes the profile G(ϑ, ϕ; π/4, π/2)
of the Green’s function in (5.14) the exact solution to the targeted sample problem.
Following the Green’s function-based strategy, developed for the problem in (5.1)–
(5.3), we use (5.14) as the resolving Green’s function, and express the solution to
(5.11)–(5.13) in the form
u(ϑ, ϕ) = G(ϑ, ϕ; τ , ψ)μ(τ , ψ)d (τ , ψ), (ϑ, ϕ) ∈ , (5.15)
where t is the angular coordinate in the local polar coordinate system whose origin
is the aperture’s center. The fictitious contour is parameterized in a similar way as
2π
G(t, ξ)μ(ξ) [τ (ξ)]2 + [ψ (ξ)]2 dξ = F(t), t ∈
0
where
G(t, ξ) = G(ϑ(t), ϕ(t); τ (ξ), ψ(ξ)),
m
A j G(t, ξ j )
μ(ξ j ) = F(t), t ∈ (5.17)
j =1
in the approximate values μ(ξ j ), j = 1, m of the density function μ(ξ) at the quadra-
ture nodes ξ j of the selected quadrature formula, for which A j are the quadrature
coefficients. So, the coefficients A j G(t, ξ j ) of (5.17) are functions of the parameter
t. Assigning now the parameter t the set of distinct values t1 , t2 , . . . , tm on , we
obtain the well-posed system
m
A j G(ti , ξ j )
μ(ξ j ) = F(ti ), i = 1, m
j=1
μ(ξ j ).
of linear algebraic equations in
Once approximate values μ(ξ j ) of the density function μ(ξ) are at hand, approx-
imate values of the solution u(ϑ, ϕ) to the problem of (5.11)–(5.13) in can be
computed as
m
u(ϑ, ϕ) ≈ A j G(ϑ, ϕ; τ (ξ j ), ψ(ξ j ))
μ(ξ j ), (ϑ, ϕ) ∈ . (5.18)
j =1
The data exposed in Tables 5.1 and 5.2 are presented to give the reader a good idea
about the accuracy level attainable within the scope of our procedure. A numerical
156 5 Irregular Configurations
experiment was conducted prior to that, revealing values of the regularizing para-
meters. It appears that the standard trapezoidal rule, with the number of quadrature
nodes m = 30 and the coefficient of proportionality k = 0.98, which determines the
radius of the fictitious contour , provides an optimal run of the procedure.
For methodological convenience, outcome data of the experience are put in two
separate tables that follow. Table 5.1 contains approximate values of u(ϑ, π/2) com-
puted at several points on the meridian ϕ = π/2 of passing through the aperture’s
center.
In Table 5.2, the reader finds approximate values of u(π/4, ϕ) computed at several
points on the parallel ϑ = π/4 of that also passes through the aperture’s center.
Based on a large number of relevant numerical experiments, we can comfortably
conclude that the data presented above provide the reader with a high degree of
confidence in the proposed approach. The upcoming sections deliver a number of
examples of a successful implementation of the just developed technique demon-
strating its broad range of applicability.
5.2 Regions of Irregular Shape 157
u (ϑ, 0) = 0, u ϑ, =0 (5.20)
2
u (ϑ, ϕ) = 0, on (5.21)
For the problem setting in (5.19)–(5.21), we are going to determine the profile
G (ϑ, ϕ; τ ∗ , ψ ∗ ) of its Green’s function, where (ψ ∗ , τ ∗ ) ∈ represents an arbitrarily
fixed in source point.
Let G 0 (ϑ, ϕ; τ , ψ) represent the Green’s function to the boundary value problem
in (5.19)–(5.20) set up in the simply connected region bounded with 0 . And we are
going, in what follows, to employ G 0 (ϑ, ϕ; τ , ψ) as a resolving Green’s function.
The sought-after profile G (ϑ, ϕ; τ ∗ , ψ ∗ ) of the Green’s function to the problem
setting in (5.19)–(5.21) is expressed as
G ϑ, ϕ; τ ∗ , ψ ∗ = G 0 ϑ, ϕ; τ ∗ , ψ ∗ + g ∗ (ϑ, ϕ) ,
where the additive term g ∗ (ϑ, ϕ) has to satisfy the following criteria: (i) be a solution
of the governing equation in (5.19); (ii) satisfy the boundary condition of (5.20); and
(iii) nullify the trace of the resolving Green’s function on . This implies that g ∗ (φ, θ)
has to be the solution of the boundary value problem
∂ ∂g ∗ (ϑ, ϕ) 1 ∂ 2 g ∗ (ϑ, ϕ)
sin ϑ + = 0, in
∂ϑ ∂ϑ sin ϑ ∂ϕ2
lim
g ∗ (ϑ, ϕ)
< ∞, lim
g ∗ (ϑ, ϕ)
< ∞,
ϑ→0 ϑ→π
158 5 Irregular Configurations
Fig. 5.4 The fields induced by multiple sources in the perforated and imperforate spherical sector
g ∗ (ϑ, 0) = 0, g ∗ ϑ, =0
2
g ∗ (ϑ, ϕ) = − G 0 ϑ, ϕ; τ ∗ , ψ ∗ , on .
It is evident that the above setting represents a problem of the type in (5.1)–(5.3)
treated in detail in the preceding section. That is why we skip all specifics of the
solution procedure, and just present some illustrative results convincingly showing
the effectiveness of our approach.
The potential field, depicted in the left fragment of Fig. 5.4, is induced in the
spherical sector of radius a perforated with a circular aperture of radius ρ = 0.25a
centered at (0.5π, 0.25π). The field is generated by unit sources released at three
distinct points (0.33π, 0.19π), (0.58π, 0.12π), and (0.75π, 0.28π). The field repre-
sents, by the way, a superposition of three precomputed profiles G (ϑ, ϕ; τ ∗ , ψ ∗ ) of
the Green’s function G (ϑ, ϕ; τ , ψ) for the problem in (5.19)–(5.21), with the source
points located as indicated above. To illustrate the effect that the aperture provides,
we showed in the right fragment of Fig. 5.4 the field that is generated by the same
set of unit sources acting in the non-perforated shell.
The attentive reader might notice a specific feature in all problem statements for
multiply connected regions of irregular configuration that we had dealt with thus far
in Sects. 5.1 and 5.2. That is, only homogeneous boundary conditions are imposed
on boundaries of the encountered regions. This feature does not, however, make a
limitation for our approach. A problem that follows is presented to illustrate the
point.
Let a multiply connected region on a toroidal surface of radii R and a represent
the quadrilateral
n
=− K i δ ϑ − τi∗ , ϕ − ψi∗ , in (5.22)
i=1
and
u (ϑ, ϕ) = Q j (ϑ, ϕ) , (ϑ, ϕ) ∈ j , j = 1, m, (5.25)
where the right-hand side of the governing differential equation is expressed in terms
of the generalized delta function, and each of the right-hand sides Q j (ϑ, ϕ) in the
boundary conditions of (5.25) represents a continuous on j function.
In compliance with our approach, the solution to the problem in (5.22)–(5.25) is
presented in the form
n
u (ϑ, ϕ) = K i G 0 ϑ, ϕ; τi∗ , ψi∗ + g ∗ (ϑ, ϕ) ,
i=1
where the two-variable function G 0 (ϑ, ϕ; τ ∗ , ψ ∗ ) stays for the profile of the Green’s
function G 0 (ϑ, ϕ; τ , ψ) to the boundary value problem (5.22)–(5.24) stated in the
simply connected region 0 , with its variables (τ , ψ), which represent geographical
coordinates of the source point, fixed at (τ ∗ , ψ ∗ ).
Note that the computer-friendly representation for the Green’s function
G 0 (ϑ, ϕ; τ , ψ) can be found in Chap. 3 and in the appendix.
As to the second additive component g ∗ (ϑ, ϕ) of u (ϑ, ϕ), it should possess the
following properties:
(i) be a solution of the homogeneous differential equation corresponding to (5.22)
everywhere in ;
(ii) satisfy the homogeneous boundary conditions of (5.23) and (5.24) imposed
on the outer part of the boundary of ;
(iii) compensate the trace of the first additive component of u(ϑ, ϕ) in the sense
that the boundary conditions of (5.25) are met.
160 5 Irregular Configurations
Given the above, we offer to express the additive term g ∗ (ϑ, ϕ) in the following
integral form:
m
g ∗ (ϑ, ϕ) = G 0 (ϑ, ϕ; τ , ψ) μi (τ , ψ) d
Si (τ , ψ) , (ϑ, ϕ) ∈ ,
i
i=1
n
− K i G 0 (ϑ, ϕ; τi∗ , ψi∗ ), (ϑ, ϕ) ∈ j ; j = 1, m. (5.26)
i =1
The above system is regular for any fixed form and location of each of the fictitious
contours i out of . The regularity is preconditioned by the structure of the integral
representation for g ∗ (ϑ, ϕ). The fact that the sets of observation and source points
never overlap in G 0 (ϑ, ϕ; τ , ψ) stays behind the regularity. This, in turn, ensures
the well-posedness of a system of linear algebraic equations in approximate values
of the density functions, which the system in (5.26) reduces to, once the integrals in
it are approximated with some finite sums by means of quadrature formulas.
Some numerical experiments are required prior to the main computer work with
the present algorithm. For every particular region configuration, it is necessary
to properly locate the fictitious contours i . Such experiments are not usually time
consuming and have to be conducted on a case-by-case basis.
Workability of the described algorithm is tested, in what follows, with an illus-
trative example. In Fig. 5.5, we depict the potential field induced by a unit point
source released at (−0.08π, 0.85π) in a thin toroidal shell whose defining radii are
a = 0.6 and R = 1.0. The shell is weakened with two circular apertures. The first
of them of radius ρ1 = 0.15 is centered at (0.1π, 0.2π), and its contour is kept at the
potential level Q 1 = 100, while the second aperture of radius ρ2 = 0.1 is centered
at (−0.05π, 0.5π), and its contour is kept at the potential level Q 2 = 1.0.
The just described technique can be extended to the case of assemblies of thin
shells. To be specific, consider a two-fragment composition of shells where one
of them is solid cylindrical, whose middle surface occupies the simply connected
region 2 , while another is spherical having a circular aperture, whose middle surface
5.2 Regions of Irregular Shape 161
occupies the double-connected region 1 (see Fig. 5.6). The assembly of 1 and 2
will be referred to as .
Potential field induced in the spherical shell is simulated in 1 by the equation
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + 2 2 = 0, in 1 , (5.27)
a sin ϑ ∂ϑ
2 ∂ϑ a sin ϑ ∂ϕ2
∂ 2 u 2 (z, ϕ) 1 ∂ 2 u 2 (z, ϕ)
+ 2 = 0, in 2 , (5.28)
∂z 2 a ∂ϕ2
∂u 1 (π/2, ϕ) ∂u 2 (0, ϕ)
u 1 (π/2, ϕ) = u 2 (0, ϕ), =λ (5.32)
∂ϑ ∂z
and
162 5 Irregular Configurations
Our goal is to develop an efficient algorithm for computing elements of the matrix
of Green’s type
G (ς, ϕ; ξ, ψ) = G i j (ς, ϕ; ξ, ψ) (5.34)
2×2
for the boundary value problem in (5.27)–(5.32), stated in the compound sim-
ply connected region whose fragments have no apertures, as a resolving matrix.
G (ς, ϕ; ξ, ψ) has already been obtained in this manual earlier (see Sect. 4.3).
For a fixed in location (ξ ∗ , ψ ∗ ) of the source point, we express the matrix of
Green’s type G (ς, ϕ; ξ ∗ , ψ ∗ ) in terms of the resolving matrix as
G ς, ϕ; ξ ∗ , ψ ∗ = G ς, ϕ; ξ ∗ , ψ ∗ + W∗ (ς, ϕ). (5.35)
where the fictitious contour represents a smooth closed line embraced with the
actual aperture contour .
From (5.36), it follows that the first component of the vector function V (ς, ϕ)
reads as
V1 (ϑ, ϕ) = G 11 (ϑ, ϕ; τ , ψ) M1 (τ , ψ) d (τ , ψ)
+ G 12 (ϑ, ϕ; ζ, ψ) M2 (ζ, ψ) d (ζ, ψ) , (ϑ, ϕ) in 1 .
Due to the way the vector functions V (ς, ϕ) and M (ξ, ψ) were introduced, the
above representation reduces to
∗
w11 (ϑ, ϕ) = G 11 (ϑ, ϕ; τ , ψ) μ∗1 (τ , ψ) d (τ , ψ) , (ϑ, ϕ) ∈ 1 . (5.37)
164 5 Irregular Configurations
= − G 11 ϑ, ϕ; τ ∗ , ψ ∗ , (ϑ, ϕ) ∈ , (5.38)
in the density function μ∗1 (τ , ψ). Due to the regular nature of the above equation, its
numerical solution is not expected to be problematic (for a fixed fictitious contour
). But finding optimal shape and location of is another issue representing a
regularizing stage of our algorithm. It has to be addressed on the case-by-case basis.
Our experience provides us with data ensuring a confidence in the efficiency of the
suggested approach.
Once an accurate approximation of the density function μ∗1 (τ , ψ) is found, the
∗
form in (5.37) gives us an explicit expression for the component w11 (ϑ, ϕ) of the
potential field generated in the double-connected region 1 by a unit source at a
∗
point (τ ∗ , ψ ∗ ) also located in 1 . To obtain the second fragment w21 (ϑ, ϕ) of the
potential field generated in the simply connected region 2 by the unit source located
at (τ ∗ , ψ ∗ ) ∈ 1 , we turn to the integral representation of (5.36), from which the
second component of the vector function V (ς, ϕ) appears in the form
V2 (z, ϕ) = G 21 (z, ϕ; τ , ψ) M1 (τ , ψ) d (τ , ψ)
+ G 22 (z, ϕ; ζ, ψ) M2 (ζ, ψ)d (ζ, ψ), (z, ϕ) ∈ 2 ,
spherical fragment is a = 1.0, length of the cylindrical fragment is l = 2.0, and the
aperture of radius ρ = 0.35 is centered at (0.25π, 0.2π). One point source is released
in the spherical fragment at (0.41π, 0.52π), while the second source is released in
the cylindrical fragment at (0.74, 0.25π).
The following comments are appropriate as to the computational aspects making
our procedure efficient:
(i) the regularizing effect is achieved by choosing the fictitious contour concen-
tric with circle of radius ρ = 0.98ρ of the latter;
(ii) the functional equation in (5.38) can accurately be solved by the quadrature
formula method (the regular trapezoid rule with 30 grid points);
(iii) the series-containing components of the matrix of Green’s type were truncated
to just the 5th term.
ation. In this regard, it is worth noting that the number of required iterations might
be rather massive, counting hundreds and even thousands. And this exact feature
stays behind the extreme cost of numerical algorithms used to solve most of inverse
problems.
From what was just outlined, it follows that the development of workable algo-
rithms for optimal shape design settings should focus on the minimization of compu-
tational expenditures for relevant direct problems. With this in mind, we will present,
in what follows, a convincing argumentation which supports the choice of our ver-
sion of the BIEM in the development of workable optimal shape design algorithms.
Remind that those Green’s functions, which are employed as kernels of correspond-
ing integral representations, are referred to, in the present study, as the resolving
Green’s functions, and their availability is an absolute prerequisite for the success of
our approach.
To provide the reader with necessary information regarding the efficiency of our
approach to the optimal shape design problems, we consider an illustrative exam-
ple. Let a thin shell whose middle surface occupies a double-connected region
representing the spherical belt
0 = ϑ0 ≤ ϑ ≤ ϑ1 , 0 ≤ ϕ < 2π
of radius a weakened with a circular aperture of radius ρ < a centered at (ϑc , 0).
The aperture’s contour will be denoted as .
In , we pose the boundary value problem
∂ ∂u (ϑ, ϕ) 1 ∂ 2 u (ϑ, ϕ)
sin ϑ + = 0, in (5.39)
∂ϑ ∂ϑ sin ϑ ∂ϕ2
∂u (ϑ0 , ϕ)
= 0, u (ϑ1 , ϕ) = 0, (5.40)
∂ϑ
∂u (ϑ, 0) ∂u (ϑ, 2π)
u (ϑ, 0) = u (ϑ, 2π) , = (5.41)
∂ϕ ∂ϕ
The technique proposed in Sect. 5.1 gives us a reliable background for obtaining a
highly accurate solution U (ϑ, ϕ) to the direct formulation of the problem in (5.39)–
(5.42) for any given set of its defining data. This includes values of the parameters
a, ϑ0 , ϑ1 , ρ, and ϑc that specify the shape of , as well as the function h(ϑ, ϕ).
We turn now to an inverse formulation of the above problem by assuming that a
part of its defining data is either missing or unavailable, and ought to be restored to
meet some constraints imposed on the sought-after solution. To be specific, let the
size and the shape of the spherical belt be fixed (implying the parameters a, ϑ0 , and
ϑ1 are given). Let also the function h (ϑ, ϕ) in (5.42) be fixed. With all the above,
we aim to determine the location and the size of the aperture (the values of ρ and ϑc
5.3 Optimal Shape Design Problems 167
∂U (ϑ, ϕ)
max U (ϑ, ϕ) ≤ A and max ≤B (5.43)
ϑ = ϑ0 ϑ=ϑ1 ∂ϑ
hold, where A and B are given constants, which will be referred to as the governing
parameters.
The above inverse formulation does not, of course, represent a well-posed prob-
lem. Indeed, it is quite possible that for a given set of its defining data, the problem
might either have no solution that complies with the constraints of (5.43) or allows
multiple solutions. To clearly underline the focus of our work, note that the existence
and uniqueness issues are left aside, and we are just going to concentrate on the
search of a single feasible solution (if any) of the inverse formulation, for a fixed set
of the defining data. Such a pragmatic approach to inverse problems is customarily
accepted in engineering practice.
Our strategy of handling the just described inverse problem is based on an iterative
solution of corresponding direct problems. This reduces the solution of the inverse
problem to the system of nonlinear equations
in ρ and ϑc , where
∂U (ϑ1 , ϕ)
f 1 (ρ, ϑc ) ≡ max U (ϑ0 , ϕ) and f 2 (ρ, ϑc ) ≡ max .
∂ϕ
An intricate feature of the system in (5.44) is that the functions f 1 (ρ, ϑc ) and
f 2 (ρ, ϑc ) depend on their arguments implicitly. Hence, direct application of standard
techniques becomes problematic unless a certain information is available concerning
properties of f 1 (ρ, ϑc ) and f 2 (ρ, ϑc ). In this regard, after studying the behavior of
the functions f 1 (ρ, ϑc ) and f 2 (ρ, ϑc ), we discovered their remarkable properties
that are very suggestive in the development of our strategy. That is, both of these
functions increase, if the variable ρ increases, whereas f 1 (ρ, ϑc ) decreases while
f 2 (ρ, ϑc ) increases, if the variable ϑc increases.
The indicated properties of the functions f 1 (ρ, ϑc ) and f 2 (ρ, ϑc ) allow us to
use the following instruments in the iterative procedure to achieve an appropriate
approximate solution of the system in (5.44):
(a) to increase both f 1 (ρ, ϑc ) and f 2 (ρ, ϑc ), we increment the value of ρ,
(b) to increase f 1 (ρ, ϑc ) and decrease f 2 (ρ, ϑc ) at the same time, we decrement
the value of ϑc ,
(c) to decrease f 1 (ρ, ϑc ) and increase f 2 (ρ, ϑc ) at the same time, the value of ϑc
should go up,
and
(d) if both f 1 (ρ, ϑc ) and f 2 (ρ, ϑc ) must decrease, then the value of ρ goes down.
168 5 Irregular Configurations
In Fig. 5.7, we depicted the recovered solution U (ϑ, ϕ) of the inverse problem
computed in the optimally shaped double-connected region . The parameters that
determine the shape of the spherical belt were chosen as a = 1.0, ϑ0 = 0.1π, and
ϑ1 = 0.5π, the right-hand side function in (5.42) was fixed as h(ϑ, ϕ) ≡ 1.0. The
governing parameters in the constraints of (5.43) were chosen as A = 0.85 and
B = 10.0.The initial values of the targeted parameters that specify the size and
location of the aperture are ρ = 0.05 and ϑc = 0.2π.
Note that the iterative process revealed a reasonably fast convergence. It took
just a couple of dozens of iterations to attain the accuracy level of order 10−4 for
the governing parameters A and B. Terminal values of the targeted parameters were
found as ρ ≈ 0.2011 and ϑc ≈ 1.1938.
For another optimal shape design problem, we turn to a composition of two thin-
shell fragments. Let middle surface of one of them occupy a double-connected region
1 on the spherical surface of radius a. The shape of
1 represents the quadrilateral
of radii b and R. Let also the condition R = a sin ϑ1 − b sin θ0 be met justifying the
contact of the fragments along the parallel ϑ1 = θ0 .
A direct boundary value problem of our interest is posed in the composed region
= 1 ∪ 2 as
5.3 Optimal Shape Design Problems 169
1 ∂ ∂u 1 (ϑ, ϕ) 1 ∂ 2 u 1 (ϑ, ϕ)
sin ϑ + 1
= 0, in (5.45)
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin ϑ
2 ∂ϕ2
1 ∂ ∂u 2 (θ, ϕ) b2 ∂ 2 u 2 (θ, ϕ)
D(θ) + 2 = 0, in 2 (5.46)
D(θ) ∂θ ∂θ D (θ) ∂ϕ2
∂u 1 (ϑ1 , ϕ) ∂u 2 (θ0 , ϕ)
u 1 (ϑ1 , ϕ) = u 2 (θ0 , ϕ) , =λ (5.49)
∂ϑ ∂θ
u 1 (ϑ, ϕ) = U (ϑ, ϕ) on , (5.50)
and
ϑ1 θ1
∂u 1 (ϑ, π/2) ∂u 2 (θ, π/2)
dϑ + dθ ≤ D,
ϑ0 ∂ϑ θ0 ∂ϕ
θ1 = 0.2π, and U (ϑ, ϕ) ≡ 1.0. This yields R ≈ 4.831. Values of the governing para-
meters in the optimizing criteria are chosen as A = 3.0, B = 0.1, C = 1.0, and
D = 2.0. The initial values of the sought-for parameters are set up as: ϑc = 0.26π,
ϕc = 0.1π, ρ = 0.35, and λ = 1.0.
It required forty-two iterations to attain the accuracy level of 10−4 for all the
governing parameters. The ultimate values of the targeted parameters are computed
as ϑc ≈ 1.1909, ϕc ≈ 0.9047, ρ ≈ 0.8787, and λ ≈ 22.018. The ultimate potential
field in the composed optimally shaped structure is shown in Fig. 5.8.
Our presentation on the optimal shape design analysis will be concluded with yet
another example. We consider a thin spherical shell of radius a composed of two
congruent fragments made of different materials, with λ representing their relative
coefficient of conductivity. Let the middle surface of the shell consist of two congruent
fragments
1 = {0 < ϑ < π/2; 0 ≤ ϕ < π/2}
and
2 = {π/2 < ϑ < π; 0 ≤ ϕ < π/2} .
at a minimum possible value. This inverse problem setting reflects the energy saving
strategy in keeping the potential on at zero level.
A solver for the described inverse problem was customarily developed on the
iterative basis. Within each single iteration, a corresponding direct problem is solved.
The required for that resolving matrix represents, in this case, the matrix of Green’s
type to the Dirichlet–Neumann boundary value problem set up in the compound
spherical sector. This matrix is available in the appendix. Upon running the solver
multiple times, we revealed (see Fig. 5.10) the relation between the parameters E
and λ, with the contour integral of (5.51) reaching a minimum value.
Appendix A
Catalogue of Green’s Functions
The reader, who is familiar with the content of this volume, can readily recognize
the following short-hand notations:
1 1
H (x, α) = ln ,
2π 1 − 2x cos α + x 2
1 1 − 2x cos α + x 2
H1(x, α, β) = ln ,
2π 1 − 2x cos β + x 2
1 (1 − 2x cos α + x 2 )(1 + 2x cos β + x 2 )
H2 (x, α, β) = ln ,
2π (1 + 2x cos α + x 2 )(1 − 2x cos β + x 2 )
and
x x π(ϕ+ψ) π(ϕ− ψ)
(x) = tanπ/ϕ1 , 0 (x) = tan , κ= , η=
2 2 ϕ1 ϕ1
which had earlier been introduced and then repeatedly used to make more compact
corresponding extensive developments. The above notations appeared also helpful
in making easily readable some of the heavy-loaded forms included in this appendix
as well.
Note that the list of Green’s functions and matrices of Green’s type, which could
potentially be obtained with the aid of our procedure for the considered in this
volume class of problems, is not limited to those shown below. If a required Green’s
function or matrix is not available in this book, the reader might try to construct it
by implementing correspondingly our recommendations.
Spherical Surface
Computer-friendly representations of Green’s functions G(ϑ, ϕ; τ , ψ) listed below
are constructed to a number of boundary value problems set up for the two-
dimensional partial differential equation
1 ∂ ∂u(ϑ, ϕ) 1 ∂ 2 u(ϑ, ϕ)
sin ϑ + =0
a 2 sin ϑ ∂ϑ ∂ϑ a 2 sin2 ϑ ∂ϕ2
1 0 (τ ) 0 (ϑ0 ) −1 0 (ϑ1 )
ln ln ln
2π 0 (ϑ1 ) 0 (ϑ) 0 (ϑ0 )
0 (ϑ) 0 (τ ) 20 (ϑ0 )
+H ,ϕ − ψ + H ,ϕ − ψ
0 (τ ) 0 (ϑ) 20 (ϑ1 )
20 (ϑ0 ) 0 (ϑ) 0 (τ )
−H ,ϕ − ψ − H ,ϕ − ψ
0 (ϑ) 0 (τ ) 20 (ϑ1 )
∞
1
+ Rn (ϑ, τ ) cos n (ϕ − ψ) ,
2π n=1
Appendix A: Catalogue of Green’s Functions 175
1 0 (ϑ0 ) 0 (ϑ) 0 (τ ) 20 (ϑ0 )
ln +H ,ϕ − ψ − H ,ϕ − ψ
2π 0 (ϑ) 0 (τ ) 0 (ϑ) 20 (ϑ1 )
20 (ϑ0 ) 0 (ϑ) 0 (τ )
−H ,ϕ − ψ + H , ϕ − ψ
0 (ϑ) 0 (τ ) 20 (ϑ1 )
∞
1
+ Rn (ϑ, τ ) cos n (ϕ − ψ) ,
2π n=1
+ Rn (ϑ, τ ) cos n (ϕ − ψ) ,
2π n=1
where
β
B=
sin ϑ1
β 0 (ϑ1 )
+ ln
sin ϑ1 0 (τ )
0 (ϑ) 0 (ϑ) 0 (τ )
+H ,ϕ − ψ − H ,ϕ − ψ
0 (τ ) 20 (ϑ1 )
∞
β
n0 (ϑ)n0 (τ )
+ cos n (ϕ−ψ)
2π n=1 (sin ϑ1 +nβ) 2n 0 (ϑ1 )
SC-2: u(ϑ1 , ϕ) = 0
0 (ϑ1 ) 0 (ϑ) 0 (ϑ) 0 (τ )
ln +H ,ϕ − ψ − H , ϕ − ψ
0 (τ ) 0 (τ ) 20 (ϑ1 )
Note that in the upcoming problem statements SQ-5 – SQ-8 and SQ-11,
the parameter k is expressed in terms of the series summation index n as
k = (2n − 1)/2.
SQ-5: u(ϑ0 , ϕ) = u(ϑ1 , ϕ) = 0, and u(ϑ, 0) = u ϕ (ϑ, ϕ1 ) = 0
⎛ ⎞
(ϑ) κ η (τ ) 2 (ϑ ) κ η
+ H2 ⎝ , , ⎠
0
H2 , ,
(τ ) 2 2 (ϑ) 2 (ϑ1 ) 2 2
⎛ ⎞
(ϑ) (τ ) κ η 2 (ϑ ) κ η
− H2 ⎝ , , ⎠
0
−H2 , ,
2 (ϑ1 ) 2 2 (ϑ) (τ ) 2 2
∞
1
⎛ ⎞
(ϑ) κ η (τ ) 2 (ϑ ) κ η
− H2 ⎝ , , ⎠
0
H2 , ,
(τ ) 2 2 (ϑ) 2 (ϑ1 ) 2 2
⎛ ⎞
(ϑ) (τ ) κ η 2 (ϑ ) κ η
+ H2 ⎝ , , ⎠
0
−H2 , ,
2 (ϑ1 ) 2 2 (ϑ) (τ ) 2 2
∞
1
⎛ ⎞
(ϑ) κ η (τ ) 2 (ϑ ) κ η
+ H2 ⎝ , , ⎠
0
H2 , ,
(τ ) 2 2 (ϑ) 2 (ϑ1 ) 2 2
⎛ ⎞
(ϑ) (τ ) κ η 2 (ϑ ) κ η
− H2 ⎝ , , ⎠
0
−H2 , ,
2 (ϑ1 ) 2 2 (ϑ) (τ ) 2 2
∞
B k (ϑ) 2k (τ ) − 2k (ϑ1 )
+ (cos kη − cos kκ)
ϕ + kπ B
n=1 1
k (τ ) 2k (ϑ1 )
∞
1
2πk B2k (ϑ1 ) + (ϕ1 + πk B) 2k (ϑ0 )
+ (cos kη − cos kκ)
2π n=1 kk (ϑ) k (τ ) 2k (ϑ1 )
182 Appendix A: Catalogue of Green’s Functions
2k
(τ ) − 2k (ϑ1 ) (ϕ1 − πk B) 2k (ϑ) − (ϕ1 + πk B) 2k (ϑ0 )
×
.
ϕ1 − (πk B)2 2k (ϑ1 ) − (ϕ1 + πk B)2 2k (ϑ0 )
Toroidal Surface
Readily computable representations of Green’s functions G(ϑ, ϕ; τ , ψ), listed in this
section, have been constructed to a number of boundary value problems set up for
the differential equation
1 ∂ ∂u(ϑ, ϕ) a 2 ∂ 2 u(ϑ, ϕ)
D(ϑ) + 2 =0
D(ϑ) ∂ϑ ∂ϑ D (ϑ) ∂ϕ2
2a a + R tan x/2
ω (x) = √ arctan √
R2 − a2 R2 − a2
1 q (τ , ϑ1 ) q (ϑ, ϑ0 )
2π q ϑ0, ϑ1
+H eq(ϑ,τ ) , ϕ − ψ + H eq(τ ,ϑ)+2q(ϑ0 ,ϑ1 ) , ϕ − ψ
−H eq(τ ,ϑ1 )+q(ϑ1 ,ϑ) , ϕ − ψ − H eq(ϑ0 ,τ )+q(ϑ0 ,ϑ) , ϕ − ψ
∞
1
sinh nq (ϑ, ϑ0 ) sinh nq (ϑ1 , τ )
+ cos n (ϕ − ψ)
π n=1 ne2nq(ϑ1 ,ϑ0 ) sinh nq (ϑ0 , ϑ1 )
1
q (ϑ, ϑ0 )
2π
+H eq(ϑ,τ ) , ϕ − ψ − H eq(τ ,ϑ)+2q(ϑ0 ,ϑ1 ) , ϕ − ψ
+H eq(τ ,ϑ1 )+q(ϑ1 ,ϑ) , ϕ − ψ − H eq(ϑ0 ,τ )+q(ϑ0 ,ϑ) , ϕ − ψ
∞
1
sinh nq (ϑ, ϑ0 ) cosh nq (ϑ1 , τ )
+ cos n (ϕ − ψ)
π n=1 ne2nq(ϑ1 ,ϑ0 ) cosh nq (ϑ0 , ϑ1 )
1
q (τ , ϑ1 )
2π
+H eq(ϑ,τ ) , ϕ − ψ − H eq(τ ,ϑ)+2q(ϑ0 ,ϑ1 ) , ϕ − ψ
−H eq(τ ,ϑ1 )+q(ϑ1 ,ϑ) , ϕ − ψ + H eq(ϑ0 ,τ )+q(ϑ0 ,ϑ) , ϕ − ψ
∞
1
cosh nq (ϑ, ϑ0 ) sinh nq (ϑ1 , τ )
+ cos n (ϕ − ψ) .
π n=1 ne2nq(ϑ1 ,ϑ0 ) cosh nq (ϑ0 , ϑ1 )
In the problems TQ-9 and TQ-10 that follow, the constant B is expressed in
terms of the parameters that define geometry of the region as B = βa/(R +
a sin ϑ0 ).
TQ-9: u (ϑ0 , ϕ) + βu ϑ (ϑ0 , ϕ) = u(ϑ1 , ϕ) = 0, and u(ϑ, 0) = u(ϑ, ϕ1 ) = 0
H1 e Pq(ϑ,τ ) , κ, η + H1 e P(q(τ ,ϑ)+2q(ϑ0 ,ϑ1 )) , κ, η
−H1 e P(q(τ ,ϑ1 )+q(ϑ1 ,ϑ)) , κ, η − H1 e P(q(ϑ0 ,τ )+q(ϑ0 ,ϑ)) , κ, η
∞
2B
+ eνq(ϑ,τ ) e2νq(τ ,ϑ1 ) − 1 sin νϕ sin νψ
n=1 1
ϕ + πn B
∞
1
eνq(ϑ,τ ) 2πn B + (ϕ1 + πn B) e2νq(ϑ0 ,ϑ1 )
+
sin νϕ sin νψ
π n=1 n ϕ1 −(πn B)2 +(ϕ1 +πn B)2 e2νq(ϑ0 ,ϑ1 )
× ϕ1 − πn B − (ϕ1 + πn B) e2νq(ϑ0 ,ϑ) 1 − e2νq(τ ,ϑ1 )
186 Appendix A: Catalogue of Green’s Functions
Joint Surfaces
The reader is exposed in this section to an extensive list of computer-friendly forms
for elements G i j (P, Q) of matrices of Green’s type G(P, Q) constructed to a number
of boundary value problems that simulate potential fields induced in compositions of
regions each of which belongs to a single surface of revolution. The interface lines in
the compositions represent parallels of the geographical coordinate systems, along
of which the ideal contact conditions are assumed.
One might get an impression that the presented below forms of G i j (P, Q) are
bulky and heavy-loaded. But this impression is not grounded, because it is hard to
deny the fact that all the elements G i j (P, Q) are written in closed analytical form,
expressed in terms of elementary functions and uniformly convergent series. And
this very feature is what makes the listed below representations easily accessible to
immediate computer implementations. In this regard, we suggest the reader to ignore
the inappropriate alertness.
1. Listed in this paragraph are elements of the 2-by-2 matrix of Green’s type obtained
for the problem, which was proposed as Exercise 4.1 in Chap. 4. A composition
of two toroidal quadrilateral-shaped regions
and
2 = {(θ, ϕ)| θ0 < θ < θ1 , 0 < ϕ < ϕ1 }
R1 + a sin ϑ1 = R2 + b sin θ0 .
Note that for the sake of compactness in all the elements of the matrix of Green’s
type presented in this paragraph, the following short-hand notations are accepted:
Appendix A: Catalogue of Green’s Functions 187
b
P = π/ϕ1 , ν = n P, = λ , q(x, y) = ω(x)−ω(y),
a
An = eνq(ϑ0 ,ϑ1 ) , and Bn = eνq(θ0 ,θ1 ) ,
(1+)An Bn + (1−)(A−1 −1
n Bn − An Bn )
Cn = .
(1+)2 A−2 −2 −2 −2
n Bn +(1− )(Bn − An )
2
G 22 (θ, ϕ; ς, ψ) = H1 e P(q(ς,θ)+2q(θ0 ,θ1 )+2q(ϑ0 ,ϑ1 )) , κ, η
+H1 e Pq(θ,ς) , κ, η − H1 e P(q(θ,θ1 )+q(ς,θ1 )) , κ, η
−H1 e P(q(θ0 ,θ)+q(θ0 ,ς)+2q(ϑ0 ,ϑ1 )) , κ, η
1 −
P(q(θ,ς)+2q(ϑ0 ,ϑ1 ))
+ H1 e , κ, η − H1 e P(q(θ0 ,θ)+q(θ0 ,ς)) , κ, η
1+
−H1 e P(q(θ,θ1 )+q(ς,θ1 )+2q(ϑ0 ,ϑ1 )) , κ, η + H1 e P(q(ς,θ)+2q(θ0 ,θ1 )) , κ, η
∞
2
Cn
+ sinh νq (θ1 , ς) sin νϕ sin νψ
π n=1 n
× eνq(ϑ1 ,ϑ0 ) (1 + ) eνq(θ,θ0 ) − (1 − ) eνq(θ0 ,θ)
+eνq(ϑ0 ,ϑ1 ) (1 − ) eνq(θ,θ0 ) + (1 + ) eνq(θ0 ,θ) .
2. Listed in this paragraph are elements of the 2-by-2 matrix of Green’s type obtained
for the problem, which has been proposed as Exercise 4.2 in Chap. 4. A compo-
sition of the spherical, of radius a, belt-shaped region
a sin ϑ1 = b.
1 c 1 0 (ϑ1 )
G 11 (ϑ, ϕ; τ , ψ) = ln + ln
2π b 2π 0 (τ )
2
0 (ϑ) b 0 (ϑ) 0 (τ )
+H ,ϕ − ψ − H , ϕ − ψ
0 (τ ) c2 20 (ϑ1 )
2
b 0 (τ ) 20 (ϑ0 ) 20 (ϑ0 )
+H 2 , ϕ−ψ − H , ϕ−ψ
c 0 (ϑ) 20 (ϑ1 ) 0 (ϑ) 0 (τ )
1− b2 20 (ϑ0 ) b2 0 (ϑ)
+ H 2 , ϕ−ψ − H 2 ,ϕ − ψ
1+ c 0 (ϑ) 0 (τ ) c 0 (τ )
0 (ϑ) 0 (τ ) 0 (τ ) 20 (ϑ0 )
+H , ϕ−ψ − H , ϕ−ψ
20 (ϑ1 ) 0 (ϑ) 20 (ϑ1 )
∞
1
Cn n0 (ϑ0 ) n0 (ϑ)
+ − n cos n (ϕ − ψ)
2π n=1 n n0 (ϑ) 0 (ϑ0 )
n
b n0 (τ ) n0 (ϑ1 )
× n (1 + ) n + (1 − ) n
c 0 (ϑ1 ) 0 (τ )
cn n0 (τ ) n0 (ϑ1 )
− n (1 − ) n + (1 + ) n ;
b 0 (ϑ1 ) 0 (τ )
1 c
G 12 (ϑ, ϕ; ρ, ψ) = ln
2π ρ
2 20 (ϑ0 ) b 0 (ϑ) b
+ H ,ϕ − ψ − H ,ϕ − ψ
1+ 0 (ϑ) 0 (ϑ1 ) ρ 0 (ϑ1 ) ρ
0 (ϑ0 )
2
ρb 0 (ϑ) ρb
+H ,ϕ − ψ − H ,ϕ − ψ
0 (ϑ) 0 (ϑ1 ) c2 0 (ϑ1 ) c2
∞ n
Cn n0 (ϑ0 ) n0 (ϑ) ρ cn
+ − − cos n (ϕ − ψ) ;
π n=1 n n0 (ϑ) n0 (ϑ0 ) cn ρn
1 c
G 21 (r, ϕ; τ , ψ) = ln
2π r
2 b0 (τ ) b20 (ϑ0 )
+ H ,ϕ − ψ − H ,ϕ − ψ
1+ r 0 (ϑ1 ) r 0 (τ ) 0 (ϑ1 )
r b20 (ϑ0 ) r b0 (τ )
+H 2 ,ϕ − ψ − H 2 ,ϕ − ψ
c 0 (τ ) 0 (ϑ1 ) c 0 (ϑ1 )
∞ n n
1 Cn 0 (ϑ0 ) 0 (τ )
n
r cn
+ − n − n cos n (ϕ − ψ) ;
π n=1 n n0 (τ ) 0 (ϑ0 ) cn r
190 Appendix A: Catalogue of Green’s Functions
1 c
G 22 (r, ϕ; ρ, ψ) = ln
2π ρ
2 2
r b 0 (ϑ0 )
+H ,ϕ − ψ − H , ϕ − ψ
ρ r ρ20 (ϑ1 )
rρ
ρb2 20 (ϑ0 )
−H 2 , ϕ − ψ + H ,ϕ − ψ
c r c2 20 (ϑ1 )
2 2
1− r 0 (ϑ0 ) b
+ H ,ϕ − ψ − H ,ϕ − ψ
1+ ρ20 (ϑ1 ) rρ
2
r ρ2 (ϑ0 ) ρb
−H 2 20 ,ϕ − ψ + H , ϕ − ψ
c 0 (ϑ1 ) r c2
∞
1
Cn ρn cn
+ − cos n (ϕ − ψ)
2π n=1 n cn ρn
n
b n0 (ϑ0 ) n0 (ϑ1 )
× n (1 + ) n + (1 − ) n
r 0 (ϑ1 ) 0 (ϑ0 )
rn n0 (ϑ0 ) n0 (ϑ1 )
− n (1 − ) n + (1 + ) n .
b 0 (ϑ1 ) 0 (ϑ0 )
In order to make the elements in the matrix of Green’s type, which are presented
in this paragraph, reasonably compact, we accepted the following short-hand
notations:
x sin ϑ1
0 (x) = tan , and = λ
2 b
while the coefficient Cn of the series components stays the same for all the four
elements of the matrix, and is expressed as
bn n0 (ϑ0 ) (1 + ) b2n 2n
0 (ϑ0 ) + (1 − ) An
Cn =
,
cn n0 (ϑ1 ) (1 + )2 Bn − 1 − 2 An
where
An = b2n 2n
0 (ϑ1 )−c 0 (ϑ0 )
2n 2n
and
Bn = c2n 2n
0 (ϑ1 )−b 0 (ϑ0 ) .
2n 2n
3. Elements of the 2-by-2 matrix of Green’s type are listed in this item, which are
obtained for the problem proposed as Exercise 4.3 in Chap. 4. A composition of
the spherical, of radius a, cap-shaped region
a sin ϑ1 = R + b sin θ0 .
1 1 0 (ϑ)
G 11 (ϑ, ϕ; τ , ψ) = q (θ0 , θ1 ) + ln
2π 2π 0 (ϑ1 )
0 (ϑ) 0 (ϑ) 0 (τ ) 2q(θ0 ,θ1 )
+H ,ϕ − ψ − H e , ϕ − ψ
0 (τ ) 20 (ϑ1 )
1− 0 (ϑ) 0 (τ ) 0 (ϑ) 2q(θ0 ,θ1 )
+ H , ϕ−ψ − H e , ϕ−ψ
1+ 20 (ϑ1 ) 0 (τ )
∞
1
Cn nq(θ0 ,θ1 ) (ϑ)
n
n (ϑ) n (τ )
+ e (1−) 0n + (1+) 0 2n 0
2π n=1 n 0 (τ ) 0 (ϑ1 )
(ϑ)
n
(ϑ) (τ )
n n
−enq(θ1 ,θ0 ) (1+) 0n +(1−) 0 2n 0 cos n (ϕ−ψ) ;
0 (τ ) 0 (ϑ1 )
1
G 12 (ϑ, ϕ; ς, ψ) = q (ς, θ1 )
2π
2 0 (ϑ) q(θ0 ,ς) 0 (ϑ) q(θ0 ,θ1 )+q(ς,θ1 )
+ H e , ϕ−ψ − H e , ϕ−ψ
1+ 0 (ϑ1 ) 0 (ϑ1 )
∞
Cn n0 (ϑ)
+ sinh nq (θ1 , ς) cos n (ϕ − ψ) ;
π n=1 n n0 (ϑ1 )
1
G 21 (θ, ϕ; τ , ψ) = q (θ, θ1 )
2π
2 0 (τ ) q(θ,θ1 )+q(θ0 ,θ1 ) 0 (ϑ) q(θ0 ,θ)
+ H e , ϕ−ψ − H e , ϕ−ψ
1+ 0 (ϑ1 ) 0 (ϑ1 )
∞
Cn 0 (τ )
n
+ sinh nq (θ1 , θ) cos n (ϕ − ψ) ;
π n=1 n n0 (ϑ1 )
192 Appendix A: Catalogue of Green’s Functions
1
G 22 (θ, ϕ; ς, ψ) = q (ς, θ1 )
q(θ,θ1 )+q(ς,θ1 ) 2π
+H e , ϕ − ψ − H eq(θ,ς) , ϕ − ψ
1 −
q(θ0 ,θ)+q(θ0 ,ς)
+ H e , ϕ − ψ − H eq(ς,θ)+2q(θ0 ,θ1 ) , ϕ − ψ
1+
∞
1
Cn
+ sinh nq (θ1 , ς) cos n (ϕ − ψ)
π n=1 n
× (1 − ) enq(θ0 ,θ) − (1 + ) enq(θ,θ0 ) .
enq(θ0 ,θ1 )
Cn = .
(1 + )2 e2nq(θ1 ,θ0 ) − 1 − 2
4. Elements of the 2-by-2 matrix of Green’s type, which are listed in this paragraph,
were supposed to be obtained in Exercise 4.4 of Chap. 4. A composition of the
cylindrical quadrilateral-shaped region
a = R + b sin ϑ0 .
Elements of the sought-after matrix of Green’s type for the foregoing boundary
value problem are written as follows:
Appendix A: Catalogue of Green’s Functions 193
P κ η P κ η
G 11 (z, ϕ; ζ, ψ) = H2 e 2 (z−ζ)/a , , + H2 e 2 (z+ζ−2l)/a , ,
P 2 2 2 2
((ζ−z−2l)/a+2q(ϑ ,ϑ )) κ η P
((−z−ζ)/a+2q(ϑ ,ϑ 1 ))
κ η
+H2 e 2 0 1
, , + H2 e 2 0
, ,
2 2 2 2
1 − P (−z−ζ)/a κ η P
((z+ζ−2l)/a+2q(ϑ ,ϑ )) κ η
+ H2 e 2 , , + H2 e 2 0 1
, ,
1+ 2 2 2 2
P κ η P κ η
((z−ζ)/a+2q(ϑ ,ϑ )) (ζ−z−2l)/a
+H2 e 2 0 1
, , + H2 e 2 , ,
2 2 2 2
∞
4
Cn ν(ζ −l)
+ cosh sin νϕ sin νψ
π n=1 2n − 1 a
× eνq(ϑ1 ,ϑ0 ) (1 + ) eνz/a + (1 − ) e−νz/a
+eνq(ϑ0 ,ϑ1 ) (1 − ) eνz/a + (1 + ) e−νz/a ;
2 P ((ζ−2l)/a+2ω(ϑ0 )−ω(ϑ)−ω(ϑ1 )) κ η
G 21 (ϑ, ϕ; ζ, ψ) = H2 e 2 , ,
1+ 2 2
P κ η
((ζ−2l)/a+ω(ϑ)−ω(ϑ )) P
(−ζ/a+ω(ϑ)−ω(ϑ )) κ η
+H2 e 2 1
, , + H2 e 2 1
, ,
P 2 2 2 2
(−ζ/a+2ω(ϑ0 )−ω(ϑ)−ω(ϑ1 )) κ η
+H2 e 2 , ,
2 2
∞
16 Cn ν(z −l)
+ cosh νq (τ , ϑ0 ) cosh sin νϕ sin νψ;
π n=1 2n − 1 a
P κ η
G 22 (ϑ, ϕ; τ , ψ) = H2 e 2 (q(τ ,ϑ1 )+q(ϑ,ϑ1 )−2l/a) , ,
P 2 2
(q(ϑ,τ )+2q(ϑ0 ,ϑ1 )−2l/a) κ η
+H2 e 2 , ,
P 2 2
κ η κ η
)
+ H2 e 2 (q(ϑ0 ,τ )+q(ϑ0 ,ϑ)) , ,
P
+H2 e 2 q(ϑ,τ
, ,
2 2 2 2
1− P (q(ϑ,τ )−2l/a) κ η P κ η
− H2 e 2 , , + H2 e 2 0 ,τ )+q(ϑ0 ,ϑ)−2l/a) , ,
(q(ϑ
1+ 2 2 2 2
194 Appendix A: Catalogue of Green’s Functions
P κ η P κ η
+H2 e 2 (q(τ ,ϑ1 )+q(ϑ,ϑ1 )) , , + H2 e 2 (q(ϑ,τ )+2q(ϑ0 ,ϑ1 )) , ,
2 2 2 2
∞
4 Cn
+ cosh νq (ϑ, ϑ0 ) sin νϕ sin νψ
π n=1 2n − 1
× eνl/a (1 + ) eνq(ϑ1 ,τ ) − (1 − ) eνq(τ ,ϑ1 )
−e−νl/a (1 − ) eνq(ϑ1 ,τ ) − (1 + ) eνq(τ ,ϑ1 ) .
For the sake of compactness, we have used in this paragraph the following short-
hand notations:
π (2n−1)
P= , ν= P, = λb, and q(x, y) = ω(x)−ω(y),
ϕ1 2
where the single-variable function ω can be found in Eq. (3.9) of Chap. 3. The
coefficient Cn of the series components stays the same for all the four elements
of the matrix, and is expressed as
(1+) eν(q(ϑ0 ,ϑ1 )−l/a) +(1−) eν(q(ϑ1 ,ϑ0 )−l/a) − eν(q(ϑ0 ,ϑ1 )+l/a)
Cn =
.
(1 + )2 e2ν(q(ϑ1 ,ϑ0 )+l/a) − 1 + 1 − 2 e2νq(ϑ1 ,ϑ0 ) − e2νl/a
5. In this paragraph, the reader is referred to one of the problems that we had encoun-
tered with earlier in Chap. 5. That is the boundary value problem stated in (5.45)–
(5.49) for a composition of two different quadrilateral-shaped regions. One of the
regions is spherical
1 = {(ϑ, ϕ)| ϑ0 ≤ ϑ ≤ ϑ1 , 0 ≤ ϕ ≤ ϕ1 }
2 = {(θ, ϕ)| θ0 ≤ θ ≤ θ1 , 0 ≤ ϕ ≤ ϕ1 }
is circular toroidal and belongs to the surface of radii b and R = a sin ϑ1−b sin θ0 .
Elements of the 2-by-2 matrix of Green’s type for the foregoing problem are
obtained as follows:
G 11 (ϑ, ϕ; τ , ψ) =
(ϑ) (τ ) 2 (ϑ0 ) Pq(ϑ0 ,ϑ1 )
H1 , κ, η + H1 e , κ, η
(τ ) (ϑ) 2 (ϑ1 )
2 (ϑ0 ) (ϑ) (τ ) Pq(ϑ0 ,ϑ1 )
−H1 , κ, η − H1 e , κ, η
(ϑ) (τ ) 2 (ϑ1 )
1− (ϑ) (τ ) 2 (ϑ0 ) Pq(ϑ0 ,ϑ1 )
+ H1 , κ, η + H1 e , κ, η
1+ 2 (ϑ1 ) (ϑ) (τ )
Appendix A: Catalogue of Green’s Functions 195
(ϑ) Pq(ϑ0 ,ϑ1 ) (τ ) 2 (ϑ0 )
−H1 e , κ, η − H1 , κ, η
(τ ) (ϑ) 2 (ϑ1 )
∞
1
Cn n (ϑ) n (ϑ0 )
+ − sin νϕ sin νψ
π n=1 n n (ϑ0 ) n (ϑ)
νq(θ1 ,θ0 ) n (ϑ1 ) n (τ )
× e (1 + ) n + (1 − ) n
(τ ) (ϑ1 )
νq(θ0 ,θ1 ) n
(ϑ 1 ) n
(τ )
−e (1 − ) n + (1 + ) n ;
(τ ) (ϑ1 )
G 12 (ϑ, ϕ; ς, ψ) =
2 (ϑ) Pq(θ0 ,ς) (ϑ) P(q(θ0 ,θ1 )+q(ς,θ1 ))
H1 e , κ, η − H1 e , κ, η
1+ (ϑ1 ) (ϑ1 )
(ϑ0 )
2
2 (ϑ0 )
+H1 e P(q(θ0 ,θ1 )+q(ς,θ1 )) , κ, η − H1 e Pq(θ0 ,ς) , κ, η
(τ ) (ϑ1 ) (τ ) (ϑ1 )
∞
4
Cn (ϑ) n
(ϑ0 )
n
+ − n sinh νq (θ1 , ς) sin νϕ sin νψ;
π n=1 n (ϑ0 )
n (ϑ)
G 21 (θ, ϕ; τ , ψ) =
2 (τ ) P(q(θ0 ,θ)+q(θ0 ,θ1 )) (ϑ) Pq(θ,θ1 )
H1 e , κ, η − H1 e , κ, η
1+ (ϑ1 ) (ϑ1 )
2 (ϑ0 ) 2 (ϑ0 )
+H1 e Pq(θ,θ1 ) , κ, η − H1 e P(q(θ0 ,θ1 )+q(θ,θ1 )) , κ, η
(τ ) (ϑ1 ) (τ ) (ϑ1 )
∞ n
4 Cn (ϑ0 ) (τ )
n
+ − n sinh νq (θ, θ0 ) sin νϕ sin νψ;
π n=1 n n (τ ) (ϑ0 )
G 22 (θ, ϕ; ς, ψ) =
2 (ϑ0 ) P(2q(θ0 ,θ1 )+q(ς,θ))
H1 e Pq(θ,ς) , κ, η + H1 e , κ, η
2 (ϑ1 )
2
P(q(θ,θ1 )+q(ς,θ1 )) (ϑ0 ) P(q(θ0 ,θ)+q(θ0 ,ς))
−H1 e , κ, η − H1 e , κ, η
2 (ϑ1 )
2
1− (ϑ0 ) Pq(θ,ς)
+ H1 e P(2q(θ0 ,θ1 )+q(ς,θ)) , κ, η + H1 e , κ, η
1+ 2 (ϑ1 )
2
(ϑ0 ) P(q(θ,θ1 )+q(ς,θ1 ))
−H1 e P(q(θ0 ,θ)+q(θ0 ,ς)) , κ, η − H1 e , κ, η
2 (ϑ1 )
∞
2
Cn
+ sinh νq (θ1 , ς) sin νϕ sin νψ
π n=1 n
196 Appendix A: Catalogue of Green’s Functions
n (ϑ1 ) n (ϑ0 )
× eνq(θ,θ0 ) (1 + ) n + (1 − ) n
(ϑ0 ) (ϑ1 )
(ϑ1 )
n
n (ϑ0 )
−eνq(θ0 ,θ) (1 − ) n + (1 + ) n .
(ϑ0 ) (ϑ1 )
π b
P= , ν = n P, = λ ,
ϕ1 a
and x
(x) = tan P , and q(x, y) = ω(x)−ω(y).
2
The single-variable function ω, in terms of which the function q(x, y) is defined,
was presented in Eq. (3.9) of Chap. 3. The coefficient Cn of the series components
stays the same for all the four elements of the matrix, and is expressed as
n (ϑ0 ) (1 − )K n (ϑ1 , ϑ0 , θ0 , θ1 ) + (1 + )2n (ϑ0 )eνq(θ0 ,θ1 )
Cn = n
,
(ϑ1 ) (1 + )2 n (ϑ1 , ϑ0 , θ1 , θ0 ) + (1 − 2 )n (ϑ0 , ϑ1 , θ1 , θ0 )
where
K n (u, v, x, y) = 2n (u)eνq(x,y) − 2n (v)eνq(y,x)
and
n (u, v, x, y) = 2n (u)e2νq(x,y) − 2n (v).
References
1. Abramovitz, M., & Stegun, I. (1972). Handbook of mathematical functions. Washington, D.C.:
National Bureau of Standards.
2. Arsenin, V. Ya. (1968). Basic equations and special functions of mathematical physics. London:
Iliffe Books Ltd.
3. Atkinson, K. (1989). An introduction to numerical analysis (2nd ed.). New York: Wiley.
4. Banerjee, P. K., & Butterfield, A. (1981). Boundary element method in engineering science.
London: McGraw-Hill.
5. Barton, G. (1989). Elements of Green’s functions and propagation. Oxford: Clarendon Press.
6. Berger, J. R. (1994). Boundary element analysis of anisotropic bimaterials with special Green’s
functions. Engineering Analysis Boundary Elements, 14(2), 123–131.
7. Bogomolny, A. (1985). Fundamental solutions method for elliptic boundary value problems.
SIAM Journal of Numerical Analysis, 22, 644–669.
8. Borodin, V. N., & Melnikov, Yu. A. (2013). Matrices of Green’s type for sets of Laplace
equations posed on joint surfaces of revolution weakened with apertures, In 13th International
Conference on Applied Mathematical Modeling and Computational Science, Waterloo, Canada.
9. Brebbia, C. A. (1978). The boundary element method for engineers. London-New York: Pentech
Press/Halstead Press.
10. Butkovsky, A. G. (1982). Green’s functions and transfer functions handbook. Translation from
Russian. New York: Halstead Press.
11. Chen-To, T. (1994). Dyadic Green’s functions in electromagnetic theory. New York: IEEE
Press.
12. Courant, R., & Hilbert, D. (1953). Methods of mathematical physics. New York: Interscience.
13. Cruse, T. A., Ewing, A. P., & Wikswo, L. (1999). Green’s function formulation of Laplace’s
equation for electromagnetic crack detection. Computational Mechanics, 23(5–6), 420–429.
14. Davydov, I. A., Melnikov, Yu A, & Nikulin, V. A. (1978). Green’s functions of the steady-state
heat conduction operator for some shells of revolution. Journal of Engineering Physics, 34,
723–728. (in Russian).
15. Dolgova, I. M., & Melnikov, Yu A. (1978). Construction of Green’s functions and matrices for
equations and systems of elliptic type. Translation from Russian PMM, 42, 740–746.
16. Duffy, D. (2001). Green’s functions with applications. Boca Raton: CRC Press.
17. Gavelya, S. P. (1965). Simply-supported and clamped shallow spherical shell with apertures.
Naukova Dumka, Kiev, 1, 93–98. (in Russian).
18. Gavelya, S. P. (1969). On one method of construction of Green’s matrices for joint shells,
Reports of the Ukrainian Academy of Science, Series A, (Vol. 12, pp. 1107–1111). (in Russian).
19. Gradstein, I. S., & Ryzhik, I. M. (1980). Tables of integrals. Series and Products. New York:
Academic Press.