Computers and Mathematics with Applications ( )

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

journal homepage: www.elsevier.com/locate/camwa

Haar wavelet collocation method for three-dimensional

elliptic partial differential equations
Imran Aziz a, , Siraj-ul-Islam b , Muhammad Asif a
a
Department of Mathematics, University of Peshawar, Pakistan
b
Department of Basic Sciences, University of Engineering and Technology Peshawar, Pakistan

article info abstract

Article history: A new collocation method based on Haar wavelet is presented for numerical solution of
Received 11 September 2016 three-dimensional elliptic partial differential equations with Dirichlet boundary condi-
Received in revised form 19 January 2017 tions. An important advantage of the method is that it can be applied to both linear as well
Accepted 26 February 2017
as nonlinear problems. The algorithm based on this new method is simple and can be easily
Available online xxxx
implemented in any programming language. Experimental rates of convergence of the pro-
posed method are calculated which are in agreement with theoretical results. The proposed
Keywords:
Haar wavelet
method is applied to several benchmark problems from the literature including linear and
Elliptic partial differential equations nonlinear elliptic problems as well as systems of elliptic partial differential equations. The
NavierStokes equations numerical experiments confirm the accuracy and diverse applicability of the method.
2017 Elsevier Ltd. All rights reserved.

1. Introduction

There are several applications of elliptic partial differential equations (EPDEs) in various fields of engineering and science.
Potential theory has played a paramount role in both analysis and computation for boundary value problems for elliptic
partial differential equations. Numerous applications can be found in fracture mechanics, fluid mechanics, elastodynamics,
electromagnetic and acoustics [1]. Many physical processes can be modeled using EPDEs, such as, the steady state thermal
conductance equations, the linear and non-linear convection equations that model the transport phenomena [24], the
Poisson equation that models the electric and gravitational potential [57]. A special case of fully nonlinear NavierStokes
equations, which describe the motion of fluid substances and represent the conservation of mass, momentum and energy,
are elliptic equations having variable coefficients associated with the second as well as first order derivative terms [812].
The numerical solution of EPDEs is an important problem in numerical analysis. A great deal of research work has been
published on the development of numerical methods for solution of such equations. Latest contribution in this stream
comprises meshless methods [1317], spline collocation methods [1820], finite-difference methods [2123], finite element
method [24], fast domain decomposition method [25], SincGalerkin method [26] and autocorrelation-based wavelets
methods [27].
Krishnaiah has worked on three-dimensional linear EPDEs and developed fourth order finite difference method without
mixed derivative terms [28]. Zhang [29], Gupta and Kouatchou [30], Iyengar and Manohar [31] proposed high order schemes
for 3D Poisson equation. High order compact finite difference schemes have been developed for steady state 3D convection
diffusion equations [3235]. Arabshahi and Dehgan developed a simple fourth order difference method for the nonlinear
second order 3D elliptic equation with constant coefficients [36]. Furthermore, in 1992, Jain presented fourth order compact

Corresponding author.
E-mail address: [email protected] (I. Aziz).

http://dx.doi.org/10.1016/j.camwa.2017.02.034
0898-1221/ 2017 Elsevier Ltd. All rights reserved.
2 I. Aziz et al. / Computers and Mathematics with Applications ( )

scheme using 19 grid points and a single computational cell for the solution of 3D nonlinear elliptic partial differential
equations with nonlinear first derivative terms [37]. In 1994, Jain and Mohanty presented a fourth order compact finite
difference discretization for solving the system of 3D nonlinear EPDEs with variable coefficients and mixed order derivative
terms [38].
Wavelets have numerous applications in approximation theory and have been extensively used in the context of numer-
ical approximation in the relevant literature during the last two decades [39]. A meager literature of some earlier work can
be found in [40]. Different types of wavelets have been used for the numerical solutions of integral equations, ordinary dif-
ferential equations, partial differential equations, fractional partial differential equations and numerical integration [4147].
Due to the simplicity, Haar wavelet has got the attention of many researchers. Applications of Haar wavelet for numerical
approximations can be found in [4853]. Functions in Haar wavelet family are made up of pairs of piecewise constant func-
tions and mathematically it is the simplest orthonormal wavelet family with a compact support. Because of simplicity Haar
wavelet method has turned out to be an effective tool for solving differential and integral equations. Lepik demonstrated
that Haar wavelet method is valuable tool also in the case of higher order differential and integral equations [5456].
The functions in the Haar wavelet family are piecewise constant functions and, therefore, their derivatives of any or-
der vanish. Due to this reason, when applying Haar wavelet to differential equations an integrated approach is used. This
approach was introduced first by Chen and Hsiao [48,49]. Later on, this approach was further developed by Lepik [54,57].
We extend this approach to numerical solution of three-dimensional elliptic PDEs. Our approach is new, more efficient and
different from all the previous works [39,57,58].
In this paper we will discuss numerical method based on the Haar wavelet to obtain the numerical solution of the
three-dimensional linear and nonlinear EPDEs, respectively, given as follows:

2u 2u 2u u u
1 (x, y, z ) + 2 (x , y, z ) + 3 ( x , y , z ) + 1 (x, y, z ) + 2 (x, y, z )
x 2 y 2 z 2 x y
u
+ 3 (x, y, z ) + (x, y, z )u(x, y, z ) = f (x, y, z ), (1)
z
and

2u 2u 2u u u u

, , , , , , u(x, y, z ), x, y, z = f (x, y, z ), (2)
x2 y2 z 2 x y z
where 1 , 2 , 3 , 1 , 2 , 3 , and f are functions of x, y and z or constants. The computational domain for the above
equations is [0, 1] [0, 1] [0, 1]. Moreover, the above equations are subject to the Dirichlet boundary conditions which
are given below.

u(0, y, z ) = g (0, y, z ), u(1, y, z ) = g (1, y, z ),

u(x, 0, z ) = g (x, 0, z ), u(x, 1, z ) = g (x, 1, z ), (3)
u(x, y, 0) = g (x, y, 0), u(x, y, 1) = g (x, y, 1).

In addition to the commonly used product of matrices, we will use two other types of matrix products as well. One is the
Kronecker tensor product and the other is Hadamard product.

Kronecker tensor product

Suppose that A is m n matrix and B is p q matrix, then the Kronecker tensor product of A and B denoted by A B is
the matrix of order mp nq and is defined as follows:

a11 B a12 B a1n B
a21 B a22 B a2n B
AB= .

an1 B an2 B ann B

Note that in MATLAB the Kronecker tensor product can be implemented easily and efficiently using the built-in function
kron().

Hadamard product

The Hadamard product of two vectors X and Y having same size is denoted by X Y and is defined below.

x1 y1
x2 y2
XY= ,

xn yn
 I. Aziz et al. / Computers and Mathematics with Applications ( ) 3

where
X = [x1 , x2 , . . . , xn ]T and Y = [y1 , y2 , . . . , yn ]T .
This paper is organized as follows. In Section 2, definition of Haar wavelet is given. In Section 3, formulation of the method
based on Haar wavelet is explained. In Section 4, performance of the method is demonstrated by applying it to some test
problems. Finally, some conclusions are drawn in Section 5.

2. Haar wavelet and their integrals

The Haar wavelet family for interval [0, 1) is defined as [45]:

1 for x [, ),

hi (x) = 1 for x [, ), i = 2, 3, . . . , (4)
0 elsewhere,
where
k k + 0.5 k+1
= , = , = .
m m m
In the above definition the integer m = 2j , j = 0, 1, . . . , shows the level of the wavelet and the integer k = 0, 1, . . . , m 1
is the translation parameter. The index i in Eq. (4) is computed utilizing the formula i = m + k + 1. In case of minimal values
m = 1, k = 0, we have i = 2. For i = 1, the function h1 (x) is the scaling function for the family of Haar wavelet which is
defined as
for x [0, 1),

1
h1 (x) =
0 elsewhere.

Any square integrable function f (x) defined on [0, 1) can be expressed as an infinite sum of functions in the Haar wavelet
family as follows:


f ( x) = ai hi (x).
i =1

For approximation purposes we consider a maximum value J of the integer j, level of the Haar wavelet in the above definition.
The integer J is then called maximum level of resolution. We additionally characterize the integer M = 2J . With these
notations any square integrable function f (x) defined on [0, 1) can be approximated as a finite sum as follows:
N

f ( x) = ai hi (x).
i=1

The following notations are introduced:

x
pi,1 (x) = hi (z )dz ,
0

and
x
pi,+1 (x) = pi, (z )dz , = 1, 2, . . .
0

These integrals can be calculated utilizing Eq. (4) and are given below.

for x [0, ),

0
1

n! (x ) for x [, ),

n


pi,n (x) = 1 (5)
(x )n 2(x )n for x [, ),



n!
1 (x )n 2(x )n + (x )n

for x [ , 1), n = 1, 2, . . .


n!
where i = 2, 3, . . . For i = 1, we have
xn
P1,n (x) = , n = 1, 2, 3, . . .
n!
Assume that N = 2M. For Haar wavelet approximations in case of three-dimensional functions, we will use the points:
(xi , yj , zk ), i, j, k = 1, 2, . . . , N , (6)
4 I. Aziz et al. / Computers and Mathematics with Applications ( )

as collocation points where

i 0.5
xi = yi = zi = , i = 1, 2, . . . , N . (7)
N
Additionally, we define the matrix H of values of Haar functions having order N N in the following manner.
h1 (x1 ) h2 ( x 1 ) hN (x1 )


h1 (x2 ) h2 ( x 2 ) hN (x2 )
H= .

h1 (xN ) h2 (xN ) hN ( x N )
In a similar way we define
p1,1 (x1 ) p2,1 (x1 ) pN ,1 (x1 )


p1,1 (x2 ) p2,1 (x2 ) pN ,1 (x2 )
P1 = ,

p1,1 (xN ) p2,1 (xN ) pN ,1 (xN )
and
p1,2 (x1 ) p2,2 (x1 ) pN ,2 (x1 )


p1,2 (x2 ) p2,2 (x2 ) pN ,2 (x2 )
P2 = .

p1,2 (xN ) p2,2 (xN ) pN ,2 (xN )

3. Numerical method

Let x = [x1 , x2 , . . . , xN ]T , y = [y1 , y2 , . . . , yN ]T , z = [z1 , z2 , . . . , zN ]T . Define the notations x, y, z as follows:

x = x 1N 1N ,
y = 1N y 1N ,
z = 1N 1N z,
where 1N denotes a column vector of size N having all entries equal to 1. Let us introduce the following notations:
2u 2u 2u
uxx = (x, y, z), uyy = (x, y, z), uzz = (x, y, z), (8)
x2 y2 z2
u u u
ux = (x, y, z), uy = (x, y, z), uz = (x, y, z),
x y z
and
u = u(x, y, z). (9)
Suppose that the second order partial derivative can be expressed by Haar wavelet series as under

2u N N N
= ai,j,k hi (x)hj (y)hk (z ). (10)
x2 i=1 j=1 k=1

Integrating and using the boundary conditions, we have

u N N N
= g (1, y, z ) g (0, y, z ) + ai,j,k (pi,1 (x) pi,2 (1))hj (y)hk (z ), (11)
x i=1 j=1 k=1

and
N
N
N
u(x, y, z ) = g (0, y, z ) + x(g (1, y, z ) g (0, y, z )) + ai,j,k (pi,2 (x) xpi,2 (1))hj (y)hk (z ). (12)
i=1 j=1 k=1

Substituting the collocation points as defined above we obtain

2u N N N
(x l , ym , zn ) = ai,j,k hi (xl )hj (ym )hk (zn ), l, m, n = 1, 2, . . . , N , (13)
x2 i=1 j=1 k=1

u
(xl , ym , zn ) = g (1, ym , zn ) g (0, ym , zn )
x
N N N
+ ai,j,k (pi,1 (x) pi,2 (1))hj (y)hk (z ), l, m, n = 1, 2, . . . , N , (14)
i=1 j=1 k=1
 I. Aziz et al. / Computers and Mathematics with Applications ( ) 5

and
u(xl , ym , zn ) = g (0, ym , zn ) + xl (g (1, ym , zn ) g (0, ym , zn ))
N
N
N
+ ai,j,k (pi,2 (xl ) xl pi,2 (1))hj (ym )hk (zn ), l, m, n = 1, 2, . . . , N . (15)
i=1 j=1 k=1

In matrix notations Eqs. (13)(15) can be expressed as

uxx = A2 a, (16)
ux = r11 + A1 a, (17)
u = r1 + Aa, (18)
where
A2 = H H H,
A1 = (P1 1N kT ) H H,
A = (P2 xkT ) H H,
r1 = g (0N 3 , y, z) + x (g (1N 3 , y, z) g (0N 3 , y, z)),
r11 = g (1N 3 , y, z) g (0N 3 , y, z),
a = [a1 , a2 , . . . , aN 3 ]T .
The notation 0N 3 is used to denote a column vector of size N 3 having all entries equal to zero and k is a column vector of
size N having following entries:

k = [p1,2 (1), p2,2 (1), . . . , pN ,2 (1)]T . (19)

Similarly, we obtain
uyy = B2 b, (20)
uy = r22 + B1 b, (21)
u = r2 + Bb, (22)
where
B2 = H H H,
B1 = H (P1 1N kT ) H,
B = H (P2 ykT ) H,
r2 = g (x, 0N 3 , z) + y (g (x, 1N 3 , z) g (x, 0N 3 , z)),
r22 = g (x, 1N 3 , z) g (x, 0N 3 , z),
b = [b1 , b2 , . . . , bN 3 ]T ,
and
uzz = C2 c, (23)
uz = r33 + C1 c, (24)
u = r3 + Cc, (25)
where
C2 = H H H,
C1 = H H (P1 1N kT ),
C = H H (P2 zkT ),
r3 = g (x, z, 0N 3 ) + z (g (x, y, 1N 3 ) g (x, y, 0N 3 )),
r33 = g (x, y, 1N 3 ) g (x, y, 0N 3 ),
c = [c1 , c2 , . . . , cN 3 ]T .
In order to express the Haar coefficients bi , i = 1, 2, . . . , N 3 in terms of Haar coefficients ai , i = 1, 2, . . . , N 3 , we compare
Eq. (18) with Eq. (22) and obtain
r1 + Aa = r2 + Bb. (26)
6 I. Aziz et al. / Computers and Mathematics with Applications ( )

Eq. (26) yields the following:

b = (B1 A)a + B1 (r1 r2 ). (27)

Using the above expression of b in terms of a, we can express the expressions of uyy and uy given in Eqs. (20)(21),
respectively, in terms of a and are given as follows:

uyy = B2 (B1 A)a + B1 (r1 r2 ), (28)

uy = r22 + B1 (B 1
A)a + B 1
(r1 r2 ). (29)
In a similar way, in order to express the Haar coefficients ci , i = 1, 2, . . . , N in terms of Haar coefficients ai , i =
3

1, 2, . . . , N 3 , Eq. (18) is compared with Eq. (25). Thus, we obtain the following relation.

c = (C1 A)a + C1 (r1 r3 ). (30)

Using the above relation, the expressions of uzz and uz given in Eq. (23) and Eq. (24), respectively, are expressed in terms of
a and are given as follows:

uzz = C2 (C1 A)a + C1 (r1 r3 ), (31)

uz = r33 + C1 (C 1
A)a + C 1
(r1 r3 ). (32)
Next we consider the EPDE given in Eq. (1). Substituting the Haar approximations, we obtain the following system in matrix
notations.
1 (x, y, z)(A2 a) + 2 (x, y, z)(B2 b) + 3 (x, y, z)(C2 c) + 1 (x, y, z)(r11 + A1 a)
+ 2 (x, y, z)(r22 + B1 b) + 3 (x, y, z)(r33 + C1 c) + (x, y, z)(r1 + Aa) = f (x, y, z). (33)
Replacing b and c by their expression given in Eqs. (27) and (30), respectively, we obtain the linear system
Ka = B, (34)
where
K = (1 (x, y, z) 1TN 3 ) A2 + (2 (x, y, z) 1TN 3 ) (B2 (B1 A))
+ (3 (x, y, z) 1TN 3 ) (C2 (C1 A)) + (1 (x, y, z) 1TN 3 ) A1
+ (2 (x, y, z) 1TN 3 ) (B1 (B1 A)) + (3 (x, y, z) 1TN 3 ) (C1 (C1 A)) + ( (x, y, z) 1TN 3 ) A, (35)
and
B = f (x, y, z) 2 (x, y, z) (B2 B1 (r1 r2 )) 3 (x, y, z) (C2 C1 (r1 r3 ))
1 (x, y, z) r11 2 (x, y, z) r22 2 (x, y, z) (B1 B1 (r1 r2 ))
3 (x, y, z) r33 3 (x, y, z) (C1 C1 (r1 r3 )) (x, y, z) r1 . (36)
From Eq. (34), we obtain

a = K1 B. (37)
Finally the numerical solution can be easily obtained using the vector a and Eq. (18). In a similar manner, for nonlinear EPDEs
given in Eq. (2), we substitute the Haar expressions for the unknown function u(x, y, z ) as well as its derivatives, and then
discretizing using collocation points we obtain the following nonlinear system in matrix form:

(A2 a, B2 b, C2 c, A1 a + r11 , B1 b + r22 , C1 c + r33 , Aa + r1 , x, y, z) = f (x, y, z). (38)

This nonlinear system may be solved by using Newtons method or Broydens method or any other iterative solver.

4. Numerical experiments

This section is devoted to numerical validation of the proposed theoretical work performed in the previous sections. The
test problems include linear and non-linear 3D elliptic partial differential equations. The experimental rate of convergence
Rc (T ) for some selected problems is also calculated which is defined as:
log[Ec (T /2)/Ec (T )]
Rc (N ) = ,
log(2)
where Ec (T ) denotes maximum absolute errors at the collocation points using T number of collocation points.
In all non-linear test problems, we have used Broydens method. The initial guess for Broydens method is taken to be
zero and the iterations are terminated when the convergent criterion 105 is satisfied.
 I. Aziz et al. / Computers and Mathematics with Applications ( ) 7

Table 1
Maximum absolute errors for Test Problem 1.
T Ec (T ) Rc (T )

222 8.9227 104

444 6.4125 104 0.4766
888 2.0157 104 1.6696
16 16 16 5.3267 105 1.9200
32 32 32 1.3501 105 1.9802

Test Problem 1. Consider the following 3D Poisson equation [58]:

2u 2u 2u

+ 2 + 2 = sin x sin y sin z , x, y, z ,
x y z
2 (39)
u(x, y, z ) = 0, x, y, z ,

where = [0, 1] [0, 1] [0, 1]. The exact solution is given by

1
u(x, y, z ) = sin x sin y sin z .
3 2

This a linear Poisson 3D equation. The proposed method is applied to this test problem and the numerical results are
shown in Table 1. The maximum absolute errors for this test problem are decreased to order of 105 for N = 32 32 32,
which is a reasonably good accuracy for practical purpose. More accurate solution can be obtained on a dense grid, however,
it will increase the computational cost as well. An important characteristic of the Haar wavelet collocation method is that
by increasing the number of collocation points, a more accurate solution can be obtained. The table also shows the rates
of convergence for different numbers of collocation points for this test problem. As the number of collocation points is
increased, the rates of convergence approach to 2, which can also be confirmed theoretically from the results proved by J.
Majak et al. [59,60].

Remarks

This problem has also been solved by Zhi Shi et al. [58]. They have assumed Haar approximation of the mixed derivative
given as follows:

6u 2M
= sin ( x ) aij hi (y)hj (z ) i, j = 1, 2, . . . , 2M .
x2 y2 z 2 i ,j = 1

This approximation can only be assumed in the case when the information that the mixed derivative given in the above
equation can be expressed in the form f (x)g (y, z ) is known. This means that their method can only be applicable to a very
small class of elliptic equations for which the above mentioned information is known. Moreover by imposing that condition
they have reduced a 3D problem to a 2D problem. Hence their work is not applicable to general form of 3D elliptic problems.
On the other hand our method is designed for general type of 3D elliptic equations.

Test Problem 2. Consider the following linear problem [61]:

2u 2y u
2
2z u
2
u u u

e2x
+ e + e + ex+y+z + + = f (x, y, z ), 0 x, y, z 1. (40)
x2 y2 z2 x y z
The exact solution is given by
u(x, y, z ) = cos x + cos y + cos z .
This is a another linear test problem with variable coefficients of the derivatives involved in the equation. We have computed
the maximum absolute errors for this test problem using different numbers of collocation points not only for the unknown
function u(x, y, z ) but also for the first and second order partial derivatives. This is another important characteristic of Haar
wavelet collocation method that after computing Haar coefficients one can easily compute the values of the derivatives at
any point. These numerical results are shown in Table 2. A small loss of accuracy is observed in the maximum absolute errors
for higher order derivatives.
In Table 3, approximate solutions, exact solutions and pointwise absolute errors at certain selected points are shown.

Test Problem 3. Consider the following linear problem [61]:

2u 1 2u 2u 1 u
+ 2 2 + 2 + = f (x, y, z ), 0 x, y, z 1. (41)
x 2 x y z x x
8 I. Aziz et al. / Computers and Mathematics with Applications ( )

Table 2
Maximum absolute errors for Test Problem 2.
T u(x, y, z )
Ec (T ) R c (T )

222 6.0041 104

444 1.0641 104 2.4963
888 3.3368 105 1.6731
16 16 16 8.5155 106 1.9703
ux (x, y, z ) uy (x, y, z ) u z (x , y , z )
Ec (T ) R c (T ) Ec (T ) Rc (T ) Ec (T ) Rc (T )

222 8.1836 103 8.1836 103 8.1836 103

444 2.2311 103 1.8749 2.2311 103 1.8749 2.2311 103 1.8749
888 6.0555 104 1.8814 6.0555 104 1.8814 6.0555 104 1.8814
16 16 16 1.5430 104 1.9725 1.5430 104 1.9725 1.5430 104 1.9725
uxx (x, y, z ) uyy (x, y, z ) uzz (x, y, z )
Ec (T ) R c (T ) Ec (T ) Rc (T ) Ec (T ) Rc (T )

222 1.4831 102 1.4831 102 1.4831 102

444 4.2498 103 1.8031 4.2498 103 1.8031 4.2498 103 1.8031
888 1.5735 103 1.4334 1.5735 103 1.4334 1.5735 103 1.4334
16 16 16 5.0434 104 1.6415 5.0434 104 1.6415 5.0434 104 1.6415

Table 3
Approximate and exact solutions for Test Problem 2.
(x, y, z ) Approximate solution Exact solution Absolute errors

(0, 0, 0) 3.000000000000000 3.000000000000000 0.0000000000

(0.1, 0.1, 0.1) 2.985008033543719 2.985012495834077 4.4622 106
(0.2, 0.2, 0.2) 2.940174508725575 2.940199733523725 2.5224 105
(0.3, 0.3, 0.3) 2.865973507539119 2.866009467376818 3.5959 105
(0.4, 0.4, 0.4) 2.763155701175666 2.763182982008655 2.7280 105
(0.5, 0.5, 0.5) 2.632875837728292 2.632891500168721 1.5662 105
(0.6, 0.6, 0.6) 2.476022958771090 2.476006844729035 1.6114 105
(0.7, 0.7, 0.7) 2.294553722381835 2.294526561853465 2.7160 105
(0.8, 0.8, 0.8) 2.090135734447257 2.090120128041496 1.5606 105
(0.9, 0.9, 0.9) 1.864824453236405 1.864829904811993 5.4515 106
(1, 1, 1) 1.620906917604419 1.620906917604419 0.0000000000

The exact solution is given by

u(x, y, z ) = x2 cos y cos z .

This is a singular linear problem. The numerical results in terms of maximum absolute errors are shown in Fig. 1. It is obvious
from this figure that the accuracy of the method is improved by increasing the number of collocation points.

Test Problem 4. Consider Convection diffusion equation [61]:

2u 2u 2u u u u
+ 2 + 2 + (x, y, z ) + (x, y, z ) + (x, y, z ) = f (x, y, z ), 0 x, y, z 1, (42)
x 2 y z x y z
where

(x, y, z ) = Re x(1 2y)(1 z ),

(x, y, z ) = Re y(1 2z )(1 x),
(x, y, z ) = Re z (1 2x)(1 y).
The exact solution is given by

u(x, y, z ) = sin x + sin y + sin z + sin 3 x + sin 3 y + sin 3 z .

This is a convection diffusion problem involving Reynolds number. The numerical results for different numbers of collocation
points using three different values of Reynolds number are shown in Table 4. Like every other numerical method, the
accuracy of the proposed method also deteriorates by increasing the Reynolds number. For Re = 10 and Re = 100, an
accuracy of order 103 is obtained for T = 32 32 32 number of collocation points, while it is decreased to order of 102
for Re = 1000 for the same number of collocation points.
 I. Aziz et al. / Computers and Mathematics with Applications ( ) 9

Fig. 1. Maximum absolute errors for Test Problem 3.

Table 4
Maximum absolute errors for Test Problem 4.
T Re = 10 Re = 100 Re = 1000
Ec (T ) R c (T ) Ec (T ) Rc (T ) Ec (T ) R c (T )

222
444 1.6800 101
888 1.1836 101 0.5053 1.7035 101
16 16 16 3.7039 102 1.6761 3.9912 102 2.0936 2.1565 101
32 32 32 9.7236 103 1.9295 9.8589 103 2.0173 2.9451 102 2.8723

Table 5
Maximum absolute errors for Test Problem 5.
T Ec (T ) No. of iterations Rc (T )

222 9.3314 10 4
5
444 9.2203 104 6 0.0173
888 2.5331 104 7 1.8639
16 16 16 6.4487 105 8 1.9738

Test Problem 5. Consider the non linear convection equation:

2u 2u 2u u u u

+ + = u + + + f (x, y, z ), 0 x, y, z 1. (43)
x2 y2 z2 x y z
The exact solution is
y z
u(x, y, z ) = ex sin sin .
2 2

This is a nonlinear test problem. Table 5 shows the maximum absolute errors, numbers of iterations taken by Broydens
method and rates of convergence for different numbers of collocation points. We observe that the proposed method have
equally good performance in case of non linear problems as well. The method is converging rapidly as the numbers of
iteration taken by the Broydens method are small.
In Table 6, approximate solutions, exact solutions and pointwise absolute errors at certain selected points are shown.

Test Problem 6. Consider the following non linear equation: [37]

2u 2u 2u u
2
+ 2 + 2 + = f (x, y, z ), 0 x, y, z 1, (44)
x2 y z x
where > 0 is a parameter. The exact solution is
u(x, y, z ) = ex + ey + ez .
10 I. Aziz et al. / Computers and Mathematics with Applications ( )

Table 6
Approximate and Exact solutions for Test Problem 5.
(x, y, z ) Approximate solution Exact solution Absolute errors

(0, 0, 0) 0.000000000000000 0.000000000000000 0.00000000000

(0.1, 0.1, 0.1) 0.028034665060509 0.027045457409953 9.8920 104
(0.2, 0.2, 0.2) 0.117981647072039 0.116633584916079 1.3480 103
(0.3, 0.3, 0.3) 0.275466851850289 0.278215853902863 2.7490 103
(0.4, 0.4, 0.4) 0.504117665461966 0.515412756720925 1.1295 102
(0.5, 0.5, 0.5) 0.844352171693876 0.824019248777620 2.0332 102
(0.6, 0.6, 0.6) 1.204287780188699 1.192592237740634 1.1695 102
(0.7, 0.7, 0.7) 1.601514234675673 1.598703425342831 2.8108 103
(0.8, 0.8, 0.8) 2.011124235911472 2.013020680659937 1.8964 103
(0.9, 0.9, 0.9) 2.398031342039269 2.399412338761300 1.3809 103
(1, 1, 1) 2.718281828459046 2.718281828459046 0.0000000000

Table 7
Maximum absolute errors for Test Problem 6.
T =5 = 10 = 15
Ec (T ) R c (T ) Ec (T ) Rc (T ) Ec (T ) Rc (T )

222 1.5069 103 3.0291 103 3.9780 103

444 2.5580 104 2.5599 8.5561 104 1.8238 1.3012 103 1.6122
888 5.8616 105 2.1241 2.0982 104 2.0278 3.3744 104 1.9471
16 16 16 1.4285 105 2.0367 5.2767 105 1.9914 8.4638 105 1.9952

Table 8
Maximum absolute errors for Test Problem 7.
T u(x, y, z ) v(x, y, z )
= 5, = 10
222 1.1402 103 8.2398 104
444 6.5799 104 3.6681 104
888 1.7957 104 9.9408 105
16 16 16 4.5196 105 2.4388 105
= = 100
222 4.7193 103 2.9920 103
444 2.2229 103 1.0347 103
888 6.7377 104 2.8239 104
16 16 16 1.3299 104 5.4251 105

This is another non linear test problem involving a parameter beta. The maximum absolute errors and rates of conver-
gence for different values of beta using different numbers of collocation points are reported in Table 7. The table shows very
good performance for the values, = 5, = 10 and = 15.

Test Problem 7. Consider the following non linear system of coupled PDEs [37]:

2u 2u 2u u v u v u v

2 2 + 3 2 + 4 2 = uv + + f (x, y, z ),
x y z x x y y z z
(45)
2v 2v 2v u v u v u v

5 2 + 3 2 + 4 2 = uv + + g (x, y, z ), 0 x, y, z 1
x y z x x y y z z

where , > 0, are parameters and the functions f (x, y, z ) and g (x, y, z ) are calculated such that the following are the
exact solutions of the above system.

u(x, y, z ) = ex cos y cos z ,

v(x, y, z ) = ex sin y sin z .

The numerical results for two different sets of parameters and are shown in Table 8. This table shows that the proposed
method can be applied to systems of elliptic equations as well with very good accuracy.
 I. Aziz et al. / Computers and Mathematics with Applications ( ) 11

Table 9
Maximum absolute errors for Test Problem 8.
T u(x, y, z ) v(x, y, z ) w(x, y, z )
Re = 1

222 2.8447 103 9.1306 103 2.8447 103

444 2.8317 103 6.0643 103 2.8317 103
888 1.4786 103 3.0308 103 1.4786 103
16 16 16 3.7790 104 7.8002 104 3.7787 104
Re = 10

222 2.4889 102 6.7929 102 24889 102

444 4.5338 103 1.1035 102 4.5337 103
888 1.5631 103 5.1186 103 1.5631 103
16 16 16 4.2250 104 1.3313 103 4.2251 104

Test Problem 8. Consider NavierStokes model equations in cartesian coordinates [61]:

2u 2u 2u u u u

1
+ + =u +v +w + f (x, y, z ),
Re x 2 y 2 z 2 x y z
v 2v 2v v v v
2
1
+ + =u +v +w + g (x, y, z ), (46)
Re x 2 y 2 z 2 x y z
w 2w 2w w w w
2
1
+ + =u +v +w + h(x, y, z ), 0 x, y, z 1,
Re x2 y2 z2 x y z
where Re > 0, is a constant called Reynolds number and the functions f (x, y, z ), g (x, y, z ) and h(x, y, z ) are calculated such
that the following are the exact solutions of the above system.

u(x, y, z ) = sin x cos y cos z ,

v(x, y, z ) = 2 cos x sin y cos z ,
w(x, y, z ) = cos x cos y sin z .
We have solved this problem for Re = 1 and Re = 10. The numerical results are shown in Table 9. A good performance of
the proposed method can be observed from this table.

5. Conclusion

In the present work, Haar wavelet collocation method is used to solve 3D elliptic linear and nonlinear partial differential
equations as well as systems involving such equations. The method is applied to several test problems including linear as
well as nonlinear problems. An important feature of the method is that the maximum absolute errors are decreased by
increasing the number of collocation points. The numerical rate of convergence is also calculated for different problems
which is approaching 2 in most of the cases, thus, confirming recently reported theoretical results [59,60].

References

[1] L. Ying, G. Biros, D. Zorin, A high-order 3D boundary integral equation solver for elliptic PDEs in smooth domains, J. Comput. Phys. 219 (2006) 247275.
[2] V. Arpaci, Convection Heat Transfer, Prentice Hall, New Jersey, 1984.
[3] P. Roache, Computational Fluid Dynamics, Hermosa Press, Albuquerque, New Mexico, 1972.
[4] T. Cebeci, Convection Heat Transfer, Horizons Publishing Inc., Springer, Long Beach, California, Heidelberg, 2002.
[5] J. Wang, W. Jhong, J. Zhang, A general meshsize fourth order compact difference discretization scheme for 3D Poisson equation, Appl. Math. Comput.
183 (2006) 804812.
[6] W. Spotz, G. Carey, A high-order compact formulation for the 3D Poisson equation, Numer. Methods Partial Differential Equations 183 (1996) 235243.
[7] T. Myint-U, L. Debnath, Linear Partial Differential Equations for Scientists and Engineers, fourth ed., Birkhauser, 2007.
[8] J. Anderson, Introduction to Computational Fluid Dynamics, McGraw Hill, New York, 1995.
[9] D. Anderson, J. Tannehill, R. Pletcher, Computational Fluid Mechanics and Heat Transfer, Hemisphere Publishing Co., 1984.
[10] T. Cebeci, J. Shao, F. Kafyeke, E. Laurendeace, Computational Fluid Dynamics for Engineers, Horizons Publishing Inc., Springer, Long Beach, California,
Heidelberg, 2005.
[11] H. Schliting, K. Gersten, Boundary Layer Theory, Springer-Verlag, 2001.
[12] J. Ferziger, M. Peric, Computational Methods for Fluid Dynamics, Springer-Verlag, Berlin, Heidelberg, 2002.
[13] G. Fasshauer, Meshfree approximation methods with Matlab, in: Interdisciplinary Mathematical Sciences, World Scientific Publishers, Singapore,
2007.
[14] H.-Y. Hu, Z.-C. Li, A.-D. Cheng, Radial basis collocation methods for elliptic boundary value problems, Comput. Math. Appl. 50 (2005) 289320.
[15] N. Duy, T. Cong, An integrated RBF technique based on Galerkin formulation for elliptic differential equations, Eng. Anal. Bound. Elem. 33 (2009)
191199.
[16] N. Duy, T. Cong, A Cartesian-grid discretisation scheme based on local integrated RBF for two-dimensional elliptic problems, CMES Comput. Model.
Eng. Sci. 51 (2009) 213238.
12 I. Aziz et al. / Computers and Mathematics with Applications ( )

[17] Q. Yang, J. Zheng, Y. Miao, Y. Sima, An improved hybrid boundary node method for solving steady fluid flow problems, Eng. Anal. Bound. Elem. 35
(2011) 1824.
[18] G. Fairweather, A. Karageorghis, J. Maack, Compact optimal quadratic spline collocation methods for the Helmholtz equation, J. Comput. Phys. 230
(2011) 28802895.
[19] C. Christara, Quadratic spline collocation methods for elliptic partial differential equations, BIT 34 (1994) 3361.
[20] A. Abushama, B. Bialecki, Modified nodal cubic spline collocation for Poisson equation, SIAM J. Numer. Anal. 46 (2008) 397418.
[21] S. Britt, S. Tsynkov, E. Turkel, A compact fourth order scheme for the Helmholtz equation in polar coordinates, J. Sci. Comput. 45 (2010) 2647.
[22] R. Boisvert, Families of high order accurate discretizations of some elliptic problems, SIAM J. Sci. Stat. Comput. 2 (1981) 268284.
[23] I. Singer, E. Turkel, Sixth order accurate finite difference schemes for the Helmholtz equation, J. Comput. Acoust. 14 (2006) 339351.
[24] P. Ciarlet, The Finite Element Method for Elliptic Problems, North Holland, 2002.
[25] B. Gustafsson, L. Hemmingsson-Franden, A fast domain decomposition high order Poisson solver, J. Sci. Comput. 14 (1999) 223243.
[26] N. Alonso, K. Bowers, An alternating direction SincGalerkin method for elliptic problems, J. Complexity 25 (2009) 237252.
[27] R. Gobbi, R. Spigler, Comparing Shannon to autocorrelation-based wavelets for solving singularly perturbed elliptic BV problems, BIT 51 (2011) 123.
[28] U. Krishnaiah, R. Manohar, J. Stephenson, Fourth order finite difference methods for three dimensional general linear elliptic problems with variable
coefficients, Numer. Methods Partial Differential Equations 3 (1987) 229240.
[29] J. Zhang, Fast and high accuracy multigrid solution of the 3D Poisson equation, Numer. Methods Partial Differential Equations 143 (1998) 449461.
[30] M. Gupta, J. Kouatchou, Symbolic derivation of finite difference approximations for the three dimensional Poisson equation, Numer. Methods Partial
Differential Equations 5 (1998) 593606.
[31] S. Iyengar, R. Manohar, High order difference methods for heat equation in polar cylindrical coordinates, J. Comput. Phys. 77 (1988) 425438.
[32] M. Gupta, J. Zhang, High accuracy multigrid solution of the 3D convectiondiffusion equation, Appl. Math. Comput. 113 (2000) 249274.
[33] J. Zhang, L. Ge, M. Gupta, Fourth order compact difference scheme for 3D convection diffusion equation with boundary layers on non uniform grids,
Neural Parallel Sci. Comp. 8 (2000) 373392.
[34] J. Zhang, An explicit fourth order compact finite difference scheme for three dimensional convection diffusion equation, Commun. Numer. Methods
Eng. 14 (1998) 209218.
[35] L. Ge, J. Zhang, Symbolic computation of high order compact difference schemes for three dimensional linear elliptic partial differential equations
with variable coefficients, J. Comput. Appl. Math. 143 (2002) 927.
[36] M. Dehgan, S. Arabshahi, A simple form for the fourth order difference method for 3D elliptic equations, Appl. Math. Comput. 184 (2007) 589598.
[37] M. Jain, R. Jain, R. Mohanty, Fourth order difference method for three dimensional elliptic equations with non-linear first derivative terms, Numer.
Methods Partial Differential Equations 8 (1992) 575591.
[38] R. Mohanty, M. Jain, The numerical solution of the system of 3D nonlinear elliptic equations with mixed derivatives and variable coefficients using
fourth order difference methods, Numer. Methods Partial Differential Equations 11 (1995) 187197.
[39] I. Aziz, Siraj-ul-Islam, B. Sarler, Wavelets collocation methods for the numerical solution of elliptic BV problems, Appl. Math. Model. 37 (2013) 676694.
[40] W. Dahmen, A. Kurdila, P. Oswald, Multiscale Wavelet Methods for Partial Differential Equations, Academic Press, 1997.
[41] K. Maleknejad, T. Lotfi, K. Mahdiani, Numerical solution of first kind fredholm integral equations with wavelets Galerkin method and wavelets
precondition, Appl. Math. Comput. 186 (2007) 794800.
[42] Siraj-ul-Islam, I. Aziz, F. Haq, A comparative study of numerical integration based on Haar wavelets and hybrid functions, Comput. Math. Appl. 59
(2010) 20262036.
[43] I. Aziz, Siraj-ul-Islam, W. Khan, Quadrature rules for numerical integration based on Haar wavelets and hybrid functions, Comput. Math. Appl. 61
(2011) 27702781.
[44] M. Dehghan, M. Lakestani, Numerical solution of nonlinear system of second order boundary value problems using cubic B spline scaling functions,
Int. J. Comput. Math. 85 (2008) 14551461.
[45] Siraj-ul-Islam, I. Aziz, B. Sarler, The numerical solution of second-order boundary-value problems by collocation method with the Haar wavelets,
Math. Comput. Modelling 50 (2010) 15771590.
[46] V. Comincioli, G. Naldi, T. Scapolla, A wavelet based method for numerical solution of nonlinear evolution equations, Appl. Numer. Math. 33 (2000)
291297.
[47] J. Wu, A wavelet operational method for solving fractional partial differential equations numerically, Appl. Math. Comput. 214 (2009) 3140.
[48] C. Chen, C. Hsiao, Haar wavelet method for solving lumped and distributed-parameter systems, IEE Proc., Control Theory Appl. 144 (1997) 8794.
[49] C. Hsiao, Haar wavelet approach to linear stiff systems, Math. Comput. Simulation 64 (2004) 561567.
[50] C. Hsiao, W. Wang, Haar wavelet approach to nonlinear stiff systems, Math. Comput. Simulation 57 (2001) 347353.
[51] U. Lepik, Haar wavelet method for nonlinear integro-differential equations, Appl. Math. Comput. 176 (2006) 324333.
[52] K. Maleknejad, F. Mirzaee, Using ratiionalized Haar wavelet for solving linear integral equations, Appl. Math. Comput. 160 (2005) 579587.
[53] E. Babolian, A. Shshsawaran, Numerical solution of nonlinear fredholm integral equations of the second kind using haar wavelets, J. Comput. Appl.
Math. 225 (2009) 8795.
[54] U. Lepik, Haar wavelet method for solving higher order differential equation, Int. J. Math. Comput. 8 (2008) 8494.
[55] U. Lepik, Solving integral and differential equations by the aid of non-uniform Haar wavelets, Appl. Math. Comput. 1 (2008) 326332.
[56] U. Lepik, Solving fractional integral equations by the Haar wavelet method, Appl. Math. Comput. 2 (2009) 468478.
[57] U. Lepik, Solving pdes with the aid of two-dimensional haar wavelets, Comput. Math. Appl. 61 (2011) 18731879.
[58] Z. Shi, Y. Cao, Q. Chen, Solving 2D and 3D Poisson equations and biharmonic equations by the Haar wavelet method, Appl. Math. Model. 36 (2012)
51435161.
[59] J. Majak, B. Shvartsman, M. Kirs, M. Pohlak, H. Herranen, Convergence theorem for the Haar wavelet based discretization method, Compos. Struct. 126
(2015) 227232.
[60] J. Majak, B. Shvartsman, K. Karjust, M. Mikola, A. Haavajoe, M. Pohlak, On the accuracy of the Haar wavelet discretization method, Composites B 80
(2015) 321327.
[61] R. Mohanty, N. Setia, A new high order compact off-step discretization for the system of 3D quasi-linear elliptic partial differential equations, Appl.
Math. Model. 37 (2013) 68706883.