Mathl. Comput. Modelling Vol. 25, No. 10, pp.

29-45, 1997
Pergamon Copyright@1997 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
0895-7177197 $17.00 + 0.00
PII: SO8957177(97)00072-l

Analytical Solution for Finite

Domain Laplacian Equation with
Coexistent Boundary Conditions
P. LI
Department of Exploration Geophysics, Curtin University of Technology
GPO Box U1987, Perth, WA 6845, Australia

F. STAGNITTI
School of Computing and Mathematics, Deakin University
P.O. Box 423, Warrnambool, Vie. 3280, Australia
N. F. UREN
Department of Exploration Geophysics, Curtin University of Technology
GPO Box U1987, Perth, WA 6845, Australia

(Received February 1997; accepted March 1997)

Abstract-This paper presents a new mathematical technique called the autocontrol image-
recharge method for deriving highly accurate, computationally efficient, analytical solutions to the
two-dimensional Finite Domain Laplacian Equation involving coexistent Neumann and Dirichlet
boundary conditions (e.g., with both transmission and reflection) and for very irregular boundary
shapes. This technique is an extension of recent work presented by Li et al. [l] and is based on devel-
oping an Expansion Coefficient Recurrence Relationship for two constant sequences for the coexistent
boundary conditions.

Keywords-Coexistent boundary condition, Finite domain Laplacian equation, Autocontrol

image-recharge method, Laplace equation’s basic solution.

1. INTRODUCTION

The application of high accuracy methods for the computational solution of Laplacian problems
is important in many fields of physics [2]. Consequently, many methods of solution have been
proposed. These methods range from the purely numerical, e.g., finite difference, finite element,
and boundary integral methods, through to analytical techniques such as conformal mapping
and series solutions. Numerical schemes for Laplacian problems often encounter the problems
of numerical dispersion and high computational effort. Although traditional analytical schemes
provide efficient means of producing computationally efficient and accurate solutions, they are
limited by the range and type of boundary conditions. Read and Volker [3,4] used eigenfunction
expansion methods to present a series solution. Their solution wss basically restricted to prob-
lems with very limited boundary shapes and boundary conditions. This paper presents a new
method of solution, which is analytical, computationally efficient, and can be applied to problems
involving coexistent boundary conditions and irregular boundary shapes.

The authors acknowledge the use of facilities at the Department of Exploration Geophysics at Curtin University,
and the assistance of B. Hartley in the preparation of the figures in this paper.

29
30 P. LI et al.

Li et al. [l] developed the image-recharge method for Laplacian flow problems. This method was
based on homogeneous boundary conditions, i.e., that the upper and lower boundary conditions
are either Neumann or Dirichlet. The method is very efficient with highly accurate results. This
paper extends the earlier work by Li et al. [l], to include mixed and coexistent boundary condition
problems. The new method is called the autocontrol image-recharge method.
Consider the most commonly encountered class of twodimensional steady-state Laplacian
problems as given in the following equation:
aa aa o
z+a,= 1 0)

where @ is the potential of water, heat, solute, etc. Equation (1) is subject to the boundary
condition
a(u) Em(u,f(4)
dn
+ b(u)@ (u, f(4) = c(u), (2)
where n is the unit normal and u is the free coordinate along on the boundary, a(~) is the reflection
control function, b(u) is the transmission control function, and c(u) is the boundary value function.
They are all piecewise continuous functions. The y = f(u) is a prescribed shape function on the
boundary Xl of the domain R. R is the region within of a simple closed curve 80. This function
y = f(u) must be continuous, unique, and its first derivative can be piecewise continuous. The
boundary shape can be quite irregular (see Figure 1). When a(u) = 0 and b(u) # 0, equation (2)
becomes a Dirichlet boundary problem and physically it means that the solution of this kind
of boundary value problem depends on the potential @ on the boundary Xl. When a(~) = 0,
b(u) # 0, and c(u) = 0, equation (2) becomes a total transmission boundary problem. On the
other hand, when b(u) = 0 and a(~) # 0, equation (2) becomes a Neumann boundary problem
and this means that the solution of this kind of boundary value problem depends on the normal
derivative of the potential @ on the boundary. When b(u) = 0, U(U) # 0, and C(U) = 0, the
boundary becomes a total reflector or barrier. If a(u) # 0 and b(u) # 0, equation (2) becomes
a coexistent boundary problem in which transmission and reflection occur. In this paper, the
autocontrol image-recharge method is developed to solve this more general kind of boundary
value problem.

Figure 1. Schematicof finite domain.

For convenience, we will use the “Neumann boundary”, and “Dirichlet boundary” to denote
that section of boundary where the Neumann boundary condition and Dirichlet boundary con-
dition are applied, respectively. Similarly, we use the “Neumann point” and “Dirichlet point” to
 Analytical Solution 31

denote those boundary points where the Neumann boundary condition and Dirichlet boundary
condition apply. Especially, when c(u) = 0, the corresponding boundaries and points are named
“solid Neumann boundary”, “solid Dirichlet boundary”, “solid Neumann point”, and “solid
Dirichlet point”. For conciseness in this paper, we will use following abbreviations. They are
FDLE (Finite Domain Laplacian Equation), AIRM (autocontrol image-recharge method), ECRR
(Expansion Coefficient Recurrence Relationship), UAIR (Upper Autocontrol Image Recharger),
LAIR (Lower Autocontrol Image Recharger), and FLBS (Finite Domain Laplace Equation’s Basic
Solution).

2. BASIC SOLUTION TO FDLE

Based on function sequence theory, any square integrable and piecewise continuous function
in L2 function space can be expanded by a linearly independent function sequence [5, pp. 285
3091. So, if we can discover a linear combination of suitable function sequences that satisfies
Laplace’s equation (1) and its boundary conditions (2), then the linear combination of function
sequences is also a solution to Laplace’s equation.
In fact, the image-recharge approach developed recently by Li et al. [l] provides a complete
solution to the problem of Laplace’s equation with arbitrary upper and lower boundary shapes,
subject to the homogeneous boundary conditions

a@
z= 0 7 (x = 0, s),
@(XYY)
= Y, (Yt = h(x)) 3 (4
aa
z =o, bb = fbb)) 9 (5)
where ft(x) and fb(X) are the upper and lower boundary shape functions, respectively. The
solution is
l&TX
Q(x,y) = e&expy cosy +eB, expy cos-,
s
(6)
n=O n=O
and the constant coefficients A,, and B,, can be determined by either of following recurrence
formulae [11:

ft(x) - 2 &,i-r&(x) = 2 An,&(x)

n=O n=O
i=l,2,3 ,..., m, (7)
2 &i&x) = 2 &&Xx),
n=O n=O
or
ft(x) - 2 Bn,i-l&(x) = E An,&(x)
n=O n=O
i= 1,2,3 ,..., m, (8)
2 An,i~i~(X) = g %,iq$(x),
n=O n=O
where

pk(2) = exp 7nnft(xl cos - -nnft (x) cos _n7rx

n7rx
s ’
d (4 = w s
s ’
(9)
ndb(X)
p:(x) = 5 [cos 7 + R(x) sin y] exp -, n=1,2,..., (10)

-kfb(X)
q:(x) = -s [cos 7 - R(x) sin y] exp n=1,2,..., (11)
s ’
TWX
&t(x) = exp + sin - 8 , n=1,2,..., (12)

-nnfb(X) sinE
qit(x) = - exp n= 1,2,.... (13)
9 s'
32 P. Lr et al.

The B,,,s are arbitrary initial constants. The first equations of recurrence relation (7) and (8)
are equivalent to the upper boundary condition (4), and the second equations of (7) and (8) are
equivalent to the lower boundary condition (5). When repetition number i = 1, the constants A,,1
can be obtained by use of the first equation in the recurrence relation (7) or (8). By using the
second equation in recurrence relation (7) or (8), the constants &,I can be obtained. Similarly,
when i = 2, by using the first equation of recurrence relation (7) or (8), A,.,,2 can be derived
from the equation found when i = 1. By using the second equation of recurrence (7) or (8),
B,,s can be derived in a similar way. Repeating this process until the convergence conditions,
lim,,, IA,,, - An,,,+11 + 0 and lim,,, IBn,m - &,,,,+I( -+ 0 are satisfied, results in the
required solution, i.e., in the limit the final values A,,, and Bn,,,, are the values of A, and B,
in equation (6).
Consider the more general case where the boundary conditions are coexistent as in equation (2).
We assume that, at the points A and B on the boundary Xl in Figure 1, the solid Neumann
boundary condition (3) is satisfied. Thus, in the Cartesian coordinate system the solution (6)
can be obtained using the classical method of separation of variables. This assumption suggests
that: if we can find two points A and B on the boundary Hl so that the straight line AB is
a suitable diameter for computational convenience, we can always utilise these points which are
assumed to satisfy solid Neumann boundary condition (3) to derive the FLBS (6) in the region R
regardless of the boundary condition type. In other words, for finite domain Laplacian problems
we can always derive the FLBS as in equation (6), since on any kind of boundary (Neumann,
Dirichlet, or both coexistent) we can always find points A and B such that the solid Neumann
boundary condition is assumed to be applied. We name equation (6) the finite domain Laplace
Equation’s Basic Solution (LEBS).

3. COMPOSITION OF THE FDLE’S BOUNDARY CONDITIONS

Analytical solutions baaed on homogeneous upper and lower boundary conditions which are
either solid Neumann or solid Dirichlet can easily be found as illustrated in Section 2. In Section 3,
we now consider coexistent boundary conditions where the Neumann and Dirichlet conditions
are satisfied on the upper and/or lower boundaries.
Note that finite two-dimensional closed region Cl is divided into two sections by the intersection
points (A and B) of a suitable diameter AB and its boundary curve dR. Since the region’s
orientation can be rotated into any position, for convenience, we define the shape of the upper
boundary yt = ft(z) and that of the lower boundary yb = f*(z) (see Figure 1). Note “t” means
top and “b” means bottom.
Along a Neumann boundary section we have,

(14)

where f’(x) is the gradient of the boundary shape function. Along a Dirichlet boundary the
f(z)) is known. So on the upper boundary y = ft(x), the boundary condition can
potential @(CC,
be expressed as

a”(x)

(a@(Gayft(x)) _ f;(z) awx,h(x))

8X >
+ qxp (x, ft(x)) = C”(X). (15)

We substitute ft(z) for y in equation (6) and then substitute the Q(x, ft(z)) into (15). The upper
boundary condition becomes

et(x) - 2 &P;(x) = 2 &J,-?;(x), (16)

n=O n=O
 Analytical Solution 33

where

PA(z) = [ye”(z) (cos y + $(z) sin y) + V(z) cos y] ap q, (17)

and

Q;(z) = [y u”(z) (- cos y + f;(z) sin y) + G(z) cos y] exp -nrt(z). (18)

In the same way, on the lower boundary y = fb(z), the boundary condition

ah(s) a@ (5, fbk:)) _ fbyzj a@ (5, fbk))

+ bb(X)@ (X, fb(X)) = cb(X), (1%
( SY 8X >

becomes
cb(z) - 2 &P,b(z) = 2 &Qf,(h (20)
n=O n=O

where

P,“(z) = [ye”(Z) (cos y + R(z) sin 5) + bb(z) cos y] exp *, (21)

and

Q:(Z) = [y oh(z) (- cos 7 + R(z) sin y) + bb(z) cos y] exp -n?rafb(2), (22)

where PA(x),p,$), Qi( z ), and Q:(z) are the function sequences with reflection control func-
tions a(z), transmission control functions b(z), and boundary value functions c(x). All of these
functions and sequences are continuous, square integrable, and uniquely valued. In any finite
region, since we can always find two points A and B, so that the connecting line AB is the
suitable diameter for computational convenience, we can use a coordinate rotation to make this
diameter line parallel to the z-axis. Therefore, the “3 in these equations is the length of the
suitable diameter of the region.

4. ECRR DEVELOPMENT FOR THE BOUNDARY CONDITIONS

The FLBS (6), satisfies Laplace’s equation exactly inside the boundaries, and the constant
coefficients A, and B,, are dependent on the boundary conditions with respect to x and y. For a
horizontal lower boundary and simple boundary conditions, the relationships between A, and B,
are independent of z and y, and can be deduced directly from the FLBS (6). If the boundaries
are arbitrarily shaped with simple boundary conditions, the relationships between A, and B, axe
dependent on z and Y, and a recurrence formula can be derived to describe the relationship [l].
When the boundary has a very irregular geometry with coexistent boundary conditions, obtaining
the relationships becomes much more difficult. The coefficients A, and B,, not only depend on
the arbitrary upper and lower boundary shapes but also depend on the coexistent boundary
conditions. We now develop another recurrence formula to reflect the interrelationships between
A, and B,, for an irregular boundary and coexistent boundary conditions.

4.1. Separation of Upper and Lower Bobndaries

In the previous work [l], the image-recharge method had been used to obtain a set of recurrence
formulae for the determination of the constants A, and B, for homogeneous upper and lower
boundary conditions. In this section, we present an AIRM to determine the constants A,, and B,,
for coexistent upper and lower boundary conditions.
 P. LI et al.

Figure 2. Downward image recharge for an infinite well.

In order to understand the operation of AIRM, consider the model expressed in Figure 2.
The purpose of the “Transfer” in Figure 2 is to provide information about at(z), b,(z), ct (CC),
d@(s, h(z))/& and WC, h(z)) on the boundary Xl to the “Recharger”. The purpose of the
“Recharger” is supply and update the image-recharge flow. The sole purpose of this image
recharger is to keep the boundary condition always satisfied. The image recharge flow can be wa
ter, heat, electrical current, etc. We name this physical model as a “autocontrol image-recharger”.
The function of this recharger is always to keep the boundary condition satisfied.
The FLBS of equation (1) is obtained as equation (6) using the solid Neumann points A and B.
If the value of arbitrary constants A, and B, can be determined using coexistent upper and lower
boundary conditions, the final solution can be determined.
The FLBS of (6) can be partitioned into two parts as follows:

where
$(z, y) = 2 A,exp y cos 7, (24)
n=O

and
@b(x)y) = 2 B, exp 7
n7rx
cos -. s (25)
n=O

Now, suppose that there is an infinite potential well with solid Neumann-type vertical bour&
aries, and a homogeneous and isotropic medium as shown in Figure 2. Above the curve AB
(Y = h(z)), th ere is an autocontrol downward image-recharge &(z). Its potential Qpt(z,y) is
governed by Laplace’s equation and the solution is expressed in (24) subject to the boundary
conditions

(26)
WG Y) = 0, (Y = --oo), (27)

and (15). The arbitrary constant coefficients A, will be determined using (15).
Further, suppose that there is another infinite potential well with solid Neumann vertical
boundaries, and a homogeneous and isotropic medium as shown in Figure 3. Under the curve
E (Y = fb(x)), an autocontrol upward image-recharge R,(x) exists. The potential @b(X) y) is
 Analytical Solution 35

00

R,(x)

a;
B

,x
Transfer

Figure 3. Upward image recharge for an infinite well.

also governed by Laplace’s equation and the solution is expressed in (25) subject to the following
boundary conditions:

a@b(x, 9)
ax = 0, (5 = 0, $1, (28)
@'a(? Y>= 0, (Y = cQ>, (29)
and (19). Similarly, the arbitrary constant coefficients B, can be determined using (19).

I
Image Recharger
?!I
Transfer

Figure 4. Downward and upward image recharge.

Superimposing Figures 2 and 3, results in Figure 4. The upper and lower boundaries, on
which the coexistent boundary conditions are applied, meet at the two points A and B which
are assumed to be solid Neumann points. Now Figure 4 shows the upward image-recharge effect
36 P. LI et al.

on a change in potential on the upper boundary and also the downward image-recharge effect
on a change in potential on the lower boundary. Based on the superposition theorem and using
this dynamic analysis method, the determination of the arbitrary constants A, and B, is now
achieved ss follows.

4.2. First Stage of the ECRR

In the previous section, we showed that the image-recharge &j(Z) from above and image-
recharge h(z) from below are interindependent and their potential governing equations are
Laplacian equations. The initial condition for the recurrence relations are

$(x, y) = e
n=O
A,,1 exp y cos y, (30)
12?rX
cos- (31)
s'

where the subscript 1 represents the first stage in the recurrence process. Now, the UAIR is
supplied with information about at(z), bt(x), and c”(z) in order to recharge the image flow. The
potential Qt on the boundary %% in Figure 4 is satisfied by the following equation:

a@t(xc,ft(x)>
at(x)

dn
+ bt(x)@t
(x,ft(x)) = C”(X).

Substituting (30) into (32) results in the equation

C”(X)= 2 -h,l~:(x),
n=O
(33)

where P,!,(x) is as in equation (17), and A,,1 is the first update in the recurrence for the con-
stant A, and can be calculated using the method that is described in [l, Appendix].
Because of the downward image-recharge &(z) potential $(z, y) from above, on the lower
boundary m (y = fb(z)) d oes not necessarily satisfy the lower boundary condition, the LAIR
uses the information on ub(x), bb(x), and cb(z) to update the upward image-recharge R,(X). The
total potential @ on the boundary %? is satisfied by

ah(x) 8% (2,fbb)> + 3% (x,h(x)) + bb(r) (@t (0, fb(x)) + @b (5, fb(x))) = c’(x). (34)
dn an >

In equations (30) and (31) substitute y = fb(X), and then substitute the expression for at (30)
and Qib (31) into (34) to get the following equation:

cb(x)- 2 An,&(x) = 2 ~n,&(x>, (35)

n=O n=O

where P,” and Qk are given by (21) and (22), respectively. Then the first stage recurrence
correction B,J can be calculated by (35).

4.3. The General ECRR Formula

In the region ACBDA of Figure 4, the total potential is given by substituting (30) and (31)
into (23). The constants A,,1 are determined with (30) using (33), and the constants Bn,i can be
evaluated by using (35). However, on the upper boundary, the total potential @(z, ft(z)) and its
normal derivative a@(~, ft(z))/a n, must be updated also. The upward and downward recharge
 Analytical Solution 37

potentials on the boundary 9 = ft(z) must satisfy the upper boundary condition. In other words,
the A,,1 have initially been determined for an infinitely deep well, and consequently, as a result
of the LAIR action the value must be modified. The potential on the upper boundary must now
satisfy the following equation:

42) - c %1&f&4 = c An,2%(4 (36)

n=O n=O

where PA(z) and Q;(z) are defined in (17) and (18). An,2 is the updated value of A,J.
The LAIR adjusts its recharge value so that the lower mixed boundary condition is always
satisfied. So substituting A,,,2 for A,,1 in (35) leads to a corrected set of values B,.,,2 for B,,J.
Likewise, the UAIR always adjusts its recharge so that the upper mixed boundary condition is
satisfied. So by replacing Bn,l with Bn,z in (36), a further set of corrected values An,3 for An,2
can be determined. Continuing in the same manner, the recurrence relationship of (35) and (36)
finally leads to two sets of constants A, and B,. For clarity, we rewrite the sequences as follows:

Q(Z) - 5 Bn,i-l&f(x) =2 4&W

n=O n=O
i = l,...,M, (37)
q,(z) - g Bn,iQf+) = 2 hJ’,bC+
n=O n=O

where M is the recurrence number which depends on the truncation number N, the boundary
shapes and the coexistent boundary conditions Pk, Qk, I’,“,and Q:(z) are given in (17), (18), (21),
and (22). The constants B,,o can be arbitrary real numbers. Note that when i = 1, equations
(30) and (31) are identical to (24) and (25), and the potential is given by equation (23). Also
when i = 1, the boundary conditions expressed in equation (37) are identical to equations (35)
and (36). Thus, according to the methods presented in [l, Appendix], the constants A,,1 can be
obtained by using the first equation in the recurrence relationship of (37) when i = 1. Similarly,
the constants B,J are obtained by applying the second equation of recurrence relation (37)
for i = 1. When i = 2, by using the first equation of recurrence relation (37), An,2 can be
derived, and by using the second equation of recurrence relation (37), Bn,2 can also be derived.
This process is repeated until the convergence conditions, lim,,, IA,,m - An,m_ll + 0 and
lim,+, ]Bn,m - Bn,,,+ I I + 0, are satisfied. In other words, the recurrence iteration ceases when

and
IBn,m - Bn,m-rl I c, (38)
where c is the arbitrary tolerance level. At this point, A,,, and B,,, are taken to be the values
A, and B,, in equation (6).
It is important to understand the relationship between the recurrence level, numerical trunccc
tion, and error propagation. The statement expressed by equation (38) is mathematically correct.
The recurrence level is dependent on the truncation number N and does not generate numerical
error in the solution process. The recurrence process is continued until the correct values for the
coefficients A, and B, are obtained. Now in practice, the conditions expressed in equation (38)
cannot be exactly matched and instead of the absolute error being exactly zero, we substitute
a desired tolerance level, e, say lo-” or similar. This tolerance level provides the convergence
criteria for the recurrence relations in the computation program. It is, however, important to
note that the recurrence level and convergence criteria play no real part in error propagation in
the solution. However, a numerical problem does exist in the fact that the analytical solution is
38 P. LI et al.

expressed as an infinite sequence, and in practice, this sequence must be truncated. Taking the
first N terms of this infinite series results in a truncation error of order N + 1. So in practice,
in order to find a sufficiently accurate solution for a Laplacian problem in a finite domain with
coexistent boundary conditions, a sufficiently large truncation number N must be selected from
experience. The truncation number depends on the complexity of the boundary shapes and the
coexistent boundary conditions. A large number of terms in the sequence may be required for
accurate solution of problems with domains having complex shapes. However, we will see in the
next section that even very complex domains, with sharp corners can be easily and quickly solved
by this method.
In summary, then the number of iterations must be large enough to satisfy condition (38)
adequately. The recurrence level will not really generate numerical errors unless it is prematurely
stopped before the convergence criteria are satisfied. Numerical errors are generated in truncating
the infinite series. The numerical accuracy is only dependent on the number of terms taken in
the series. The sequences for A, and B, are always convergent for properly defined boundary
conditions (i.e., the boundary conditions are uniquely defined on Xl and obey the usual physical
laws of conservation, etc.). Even in the case where the boundary conditions describe an extremely
complex domain, you are still able to generate two convergent sequences, even though the desired
accuracy in the solution may take a long time to be obtained.

5. REGION ORIENTATION AND

COORDINATE TRANSFORMATION
From the above sections, it is clear that choosing a suitable coordinate system is very important
in generating an accurate solution using this method. In order to use this method efficiently,
the following conditions should be observed. First, if there is no straight line solid Neumann
boundary then the biggest diameter must be parallel to the s-axis. Second, the upper and
the lower boundary functions must be uniquely-valued. Third, the solid Neumann boundary
condition must be applied at the intersection points A and B.
Sharp corners and complex domains will obviously be more difficult to handle than simple
domains. Now since the shape function and orientation of the boundary is to be determined
directly by a coordinate transformation, choosing a suitable coordinate transformation is very
important for deriving a “good” low-cost solution. Therefore, we propose the following procedure
for choosing a suitable coordinate system.
(a) When there are two parallel linear boundaries with solid Neumann boundary conditions
and the upper and lower boundary functions are unique, choose the coordinate system as
shown Figure 5.1. This situation frequently occurs in most porous media flow problems [l].
(b) When there is a linear boundary with a solid Neumann boundary condition as in Fig-
ure 5.2, wecJhoose t&coordinate system with the y-axis parallel to the line segment AC
if curves AB and CB are uniquely valued functions as shown Figure 5.2. Point B is
assumed to be a solid Neumann point.
(c) If the domain has no linear solid Neumann boundary, the biggest diameter should be
determined first and must be reoriented so that it is parallel to the z-axis of the Cartesian
coordinate system. However, in this case, if one of the upper or lower functions is not
uniquely defined in this coordinate system we must choose another coordinate system to
avoid the multiple value problem of the boundary function (Figure 5.3). For example,
in Figure 5.3, the coordinate system zoy has three values (ci,cs, and cs) for the lower
boundary the curve m at the point x = x0, but in the coordinate system x’o’y’ the
lower boundary function Am’ is uniquely defined.
Sometimes the multiple value problem of the boundary functions cannot be avoided, no matter
what coordinate system you choose as shown in Figure 5.4. The AIRM cannot handle these
situations.
 Analytical Solution 39

(AD and BD are solid Neumann Boundary)

/ *

Figure 5.1. Domain with parallel linear solid Neumann boundaries.

B
Figure 5.2. Domain with one linear solid Neumann boundary.

B
/

X
b

Figure 5.3. Domain with avoidable multiple value boundary.

 P. LI et al.

Figure 5.4. Domain with unavoidable multiple value boundary.

The points A and B are required to be solid Neumann points. However when choosing the
orientation of a coordinate system, the points A and B will probably be located at a Dirichlet
boundary. In this case, we still can assume that the points are solid Neumann points.

6. APPLICATION
In this section, we choose a porous-media flow problem to illustrate the solution to the finite
boundary Laplace equation. Three different hillslope regions were selected. The boundary shape
functions and conditions for these hillslope problems are shown in Tables 1 and 2.

Table 1. Boundary shape functions and conditions for Region 1.

Upper Boundary Function vt(x) = 0.25X + 20.0

0.4x
Bottom Boundary Function &(x) = (-X2 + 502 - 600)(“2) + 8
-0.42 + 20.0
Function case1 case2 Case 3
o’(x) 0.0 0.0 0.0

I 1.0 [O.O,20.01
1.0 [0.0,20.0]
cos (2 - 20.0)x
oh(x)

bt(x)
1.0

1.0
I 0.0
1.0
(20.0,30.0)
[30.0,50.0]

1.0
I ( 20.0
1.0
>

1.0
(20.0,30.0)

[30.0,50.0]

0.0 [O.O,20.01 ( 0.0 [0.0,20.0]

bb(x) 0.0 1.0 (20.0,30.0) { sin ( (x-2J?$o)n> (20.0,30.0)

0.0 [30.0,50.0]
I 0.0 [30.0,50.0]

ct (x) ft (2) f&l ft (xl

0.0 [O.O,
20.01
P(x) 0.0 0.0 1.0 (20.0,30.0)
0.0 130.0.50.01

In the first hillslope Region 1 shown in Figuresc-6.3, wzconsidezhree cases in which

the boundary conditions are different. The curves AB, BC, DE, and EF are solid Neumann
boundaries in each of Figures 6.1-6.3. The stream lines cannot pass though these boundaries.
But, upper boundary A^F is a Dirichlet boundary and the stream lines can pass though it. The
boundary c^o in Cases l-3 has different boundary conditions, so the stream lines are quite
 Analytical Solution 41

Table 2. Boundary shape functions and conditions for Regions 2 and 3.

f&l fbk)

0.4x
Region 2 0.25x + 20.0 - (-x2 + 50x - 600) (1’2) + 8
-0.4x + 20.0

I Region 3

Control FImction
0.12 + 20.0

Region 2
(I - 25.0)2
31.25
Region 3

at (2) 1.0 [O.O,15.01 1.0 [O.O,20.0)

0.0 (15.0,50.0] 0.0 (20.0,50.0]

ah(x) 1.0 1.0

bt (2) { 0.0 [O.O,15.01 { 0.0 [o.o, 20.01

1.0 (15.0,50.0] 1.0 (20.0,50.0]

bb(x) 0.0 0.0

ct (x) 0.0 [O.O,15.01 0.0 (0.0,20.0]

ft(z) (15.0,50.0] ft(x) (20.0,50.0]
cb(x) 0.0 0.0

Upper Function: y=O.25x+20.00

Lower Function:
y=o.4x
40 y=(-xh2+50.0x-600.O)A( l/2)+8.0
y=-0.4x+20.0
Stognotion Point (28.770, 27.192)

10 20 30
X-axis (m)
Figure 6.1. Steam line of Region 1, Class 1.

different for each case. In Case 1, Figure 6.1, E is a solid reflector and the stream lines cannot
pass though it. In Case 2, Figure 6.2, since c^o is total transmission boundary, the stream lines
can pass though it smoothly. However, in Case 3, Figure 6.3, c^o is a coexistent boundary and
42 P. LI et al.

Upper Function: y=O.25x+20.00

Lower Function:
y=o.4x
40 y=(-xA2+50.0x-6OO.Op( l/2)+8.0
=-0.4x+20.0
StagYnotion Point (23.460, 25.865)

10 20 30 40 50
X-axis (m)
Figure 6.2. Stream line of Region 1, Case 2.

the stream lines can partially pass thoughz. The value of the reflection control function o(z)
in Case 3 is decreased through the curve CD from point C to the point D. However, the value
of the transmission control function b(z) is increased along the same locus. This means that
in Figure 6.3, the ilow is more difficult through the boundary c^o closer to the point C on the
boundary c^o. It is found that when different boundary conditions are applied on the boundary
CD, the stagnation point is located at different point as shown in Figures 6.1-6.3.
In the second hillslope region shown in Figure, the boundaries m, m, and %% are
solid Neumann boundaries and the boundary GF is a solid Dirichlet boundary. On the upper
boundary A% mixed boundary conditions are applied. It is clearly shown that the stream lines
are reasonable in fitting the boundary shapes and conditions.
In the third hillslope region&own in Figure 8, the boundaries m and A^0 are solid Neu-
mann boundaries. Boundary DC is a solid Dirichlet boundary. Fluid cannot pass through the
boundary AD. Even at the sharp corners A and C, the values produced by the method in this
paper are still correct. The upper boundary is a mixed Neumann and Dirichlet boundary.
In both cases, in spite of the complex domain and boundary conditions, the AIRM algorithm
produced very smooth streamlines and a stable solution. The computational effort to produce
the results for these figures was minimal, taking just a few minutes using a Fortran program
running on a Spare Server 1OOOEunix computer system.

7. DISCUSSIONS AND CONCLUSIONS

An extension of the image-recharge method of [l], called the autocontrol image-recharge method
or AIRM has been initially detailed and illustrated. AIRM is capable of producing remarkably
 Analytical Solution 43

50

Upper Function: y=O.25x+20.00 ( 0.0,50.0)

Lower Function:
40 y=o.4x
y~l-d(~2x=52odod(-600.0)-( l/2)+8.0

Stoinotion Point’ (26.810, 26.702)

I

0
0 10 20 30 40 50
X-axis (m)
Figure 6.3. Stream line of Region 1, Case 3.

stable and computationally efficient solutions for two-dimensional homogeneous isotropic Lapla-
cian problems involving complex domains with irregular shapes and sharp corners, and with
coexistent Neumann and Dirichlet boundary conditions. The AIRM technique is based on an an-
alytical solution of a Laplacian equation with an infinite potential well and an inference technique
which results in a recursive correction procedure for the sequences of coefficients A,, and B,. The
recurrence iteration is carried out to a sufficient level such that acceptable accuracy in A, and B,
values is maintained within the given tolerance level. The accuracy, however, in the computation
procedure is determined by the level of truncation of the series expansion. Taking more terms in
the series, i.e., increasing the value of N, increases the accuracy of the solution, provided that
the convergence criteria are satisfied.
The boundary value problems solved in this paper are subject to the boundary condition (2)
which is different from the third type of boundary-value problem and the Robin [2] problems.
Though the boundary condition for the third-type of boundary-value problem is a coexistent
Dirichlet and Neumann boundary condition, it cannot reduce to the Neumann boundary condition
only. The Robin [2] boundary condition can reduce to the Dirichlet and Neumann boundary
conditions, but it is not a coexistent boundary condition. The boundary conditions introduced
in this paper are widely applicable to boundary value problems, since the equation (2) can be
reduced to Dirichlet, Neumann, the third type of boundary-value and Robin boundary conditions.
The utility of the AIRM for complex boundary value problems is dependent on the nature of
the problem and choice of a suitable coordinate transformation. A procedure for the flow domain
orientation has been described. Of course, the AIRM method, like any other method, cannot find
solutions to Laplacian problems with nonuniquely defined boundary functions, if the boundaries
44 P. LI et al.

Upper Function: y=O.25x+20.00 ( 0.0,50.0)

Lower Function:
40
y=(-xA2+50.0x-6OO.Op( l/2)+8.0
=-0.4x+20.0
Stoc$otion Point (30.670, 27.667)

0
0 10 20 30 40 50
X-axis (m)

Figure 7. Stream line of Region 2.

40 Upper Function y=O.lx+20.0

Lower Function y=(x-25.0?2/31.25
^c 30 Stognotion Point ( 33.766, 23.377)

.z
;:
I 20
>

X-oxis (m)
Figure 8. Stream line of Region 3.

are not transformed first. Some examples for suitable flow domain transformation have been
discussed.
In conclusion, we have presented a remarkably easy, computationally low cost, robust solution
procedure for twedimensional Laplacian problems involving complex domains with coexistent
boundaries. Actually it is an analytical solution (6) for the coexistent boundary condition defined
in this paper. Complete numerical solutions for the problems studied in this paper were generated
 Analytical Solution 45

in less then five minutes using a Fortran program on a Unix system running on a Spare Server
1OOOEcomputer. The simplicity of the method suggests the possible incorporation of the AIRM
algorithm into more complex mathematical models that are required in many branches of applied
physics and engineering.

REFERENCES
1. P. Li, F. Stagnitti and U. Das, An new analytical solution for Laplacian porou&media flow with arbitrary
boundary shapes and conditions, M&l. Comput. Modelling 24 (lo), 3-19, (1996).
2. M.-U. Tyn, Partial Differential Equations of Mathematical Physics, American Elsevier Science, (1973).
3. W.W. Read and R.E. Volker, Series solutions for steady seepage through hillsides with arbitrary flow bound-
aries, Water Resow. Res. 29 (8), 2871-2880, (1993).
4. W.W. Read, Series solutions for Laplace’s equation with nonhomogeneous mixed boundary conditions and
irregular boundaries, Mathl. Comput. Modelling 17 (12), 9-19, (1993).
5. B.Sz. Nagy, Intmduction to Real finctiona and Orthogonal Expansions, Oxford University Press, New York,
(1965).