Computers and Mathematics with Applications 61 (2011) 2770–2781

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

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

Quadrature rules for numerical integration based on Haar wavelets and

hybrid functions
Imran Aziz a,∗ , Siraj-ul-Islam b , Wajid Khan b
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: In this paper Haar wavelets and hybrid functions have been applied for numerical solution
Received 20 October 2010 of double and triple integrals with variable limits of integration. This approach is the
Received in revised form 9 March 2011 generalization and improvement of the methods (Siraj-ul-Islam et al. (2010) [9]) where
Accepted 9 March 2011
the numerical methods are only applicable to the integrals with constant limits. Apart from
generalization of the methods [9], the new approach has two major advantages over the
Keywords:
classical methods based on quadrature rule: (i) No need of finding optimum weights as
Haar wavelets
Hybrid functions
the wavelet and hybrid coefficients serve the purpose of optimal weights automatically
Quadrature rule (ii) Mesh points of the wavelets algorithm are used as nodal values instead of considering
Numerical method the n nodes as unknown roots of polynomial of degree n. The new methods are more
Double integrals efficient. The novel methods are compared with existing methods and applied to a number
Triple integrals of benchmark problems. Accuracy of the methods are measured in terms of absolute errors.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Numerical integration has several applications in science and engineering. A lot of work has been done in this area
in terms of the quadrature rule of numerical integration. The quadrature rule is based on polynomial interpolation.
Interpolating polynomials are used to find weights corresponding to nodes. Numerical quadrature bears some drawbacks,
these include: (i) The use of a large number of equally spaced nodes in the case of the Newton-Cotes quadrature rule
may cause erratic behavior with high degree polynomial interpolation (ii) The Gaussian quadrature rule is also based on
polynomial interpolation but the nodes as well as the weights are chosen to maximize the degree of accuracy of the resulting
rule. The Gaussian quadrature rule can be derived by the method of undetermined coefficients but the resulting equations
for the 2n unknown nodes and weights are nonlinear. This procedure is quite cumbersome for hand calculations and nodes
and weights are tabulated in advance before evaluating integrals numerically. A number of polynomials based quadrature
methods have been discussed in [1–8] and the references therein. In order to overcome some of the difficulties listed above,
we propose a new method based on Haar wavelets and hybrid functions to find numerical solutions of double and triple
integrals. This work should be considered as a logical continuation of our previous work of [9]. In the earlier paper [9],
Haar wavelets and hybrid functions are used to find numerical solution of definite integrals with constant limits and hence
the methods could be used only to a limited number of numerical integration problems. In the present paper we extend
the scope of applicability of the methods [9] to double and triple integrals with variable limits. In doing so we introduce a
new approach which not only widens the area of applicability of those methods but also reduces the computational time
and improves the accuracy of the algorithm. The approach used in [9] was to approximate the function using two and three-
dimensional Haar wavelets or two and three-dimensional hybrid functions in the case of double integrals and triple integrals

∗ Corresponding author. Tel.: +92 3328400520.

E-mail addresses: [email protected] (I. Aziz), [email protected] ( Siraj-ul-Islam), [email protected] (W. Khan).

0898-1221/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.camwa.2011.03.043
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2771

respectively. As we go to higher dimensions the number of coefficients increases exponentially and the computational cost
of the method increases considerably. To avoid the rising computational cost, the present approach is based on considering
one integral at a time and applying the Haar wavelet or hybrid function method for a single integral. After one integral has
been solved the same method is applied for the evaluation of other integrals repeatedly in a similar fashion.
Wavelets have been successfully used in the field of numerical approximations. Some of the wavelets applications
are related to finding numerical solutions of integral equations and numerical integration [10,9], ordinary differential
equations [11,12], partial differential equations [13] and fractional partial differential equations [14]. Various type
of wavelets have been used in such applications, for example, Daubechies [15], Battle-Lemarie [16], B-spline [11],
Chebyshev [17], Legendre [18,19] and Haar wavelets [20,21,12,9]. However because of their simplicity Haar wavelets have
received the attention of many researchers. Applications of Haar wavelets in the context of numerical approximations can
be found in the Refs. [21–23,20,24–29].
Hybrid functions have faster convergence than the Haar wavelets and can model discontinuities in a better manner than
Haar wavelets, [30]. Another useful property of hybrid functions is a special product matrix and a related coefficient matrix
with optimal order.The advantage of hybrid functions is that the order of block-pulse functions and Legendre polynomials
are adjustable to obtain highly accurate numerical solutions rather than the piecewise constant orthogonal function for the
solution of integral equations, [31]. Recently, hybrid functions have been successfully used for the numerical solution of
ordinary differential equations as well as integral equations, see [32,33,31,34,12,9].
Wavelets have also been applied for numerical integration [35] but their method is applicable to only those integrals
that have constant limits of integration. This paper proposes a new method based on the simple Haar wavelets and hybrid
functions. This approach has the following advantages:
(i) Provides accurate solution in comparison with the existing method.
(ii) Optimal weights are calculated using a built-in procedure in terms of wavelets or hybrid function coefficients. In the
new approach we do not need to consult a variety of tables for optimal weights.
(iii) No quadrature nodes are needed and the collocation points are used as nodal points.
(iv) The new method calculates the integrals explicitly and it does not need solving a nonlinear system resulting from the
unknown nodes and weights.
(vi) Simple and direct applicability with no need of other intermediate techniques is required.
The organization of this paper is as follows. In Section 2 the numerical integration using Haar wavelets for single, double
and triple integrals is described and in Section 3 hybrid functions are used for numerical integration of single, double and
triple integrals. Numerical results are reported in Section 4 and some conclusions are drawn in Section 5.

2. Numerical integration using Haar wavelets

In this and the subsequent section we use the same notations as used in [9]. Interpolation condition of the Haar function
and basic definitions of Haar scaling function and mother wavelets can be found in [9].

2.1. Numerical formula for single integrals using Haar wavelets

Numerical integration formulas using Haar wavelets were derived in [9] for single, double and triple integrals with
constant limits of integration. In this paper, we will apply Haar wavelets for integration of double and triple integrals with
variable limits of integration. The approach here would be different from that in [9]. In that case we used two- and three-
dimensional Haar wavelets functions, while in the present case we will use the same formula derived in [9] for numerical
integration of single integrals with constant limits repeatedly. It turns out that this new approach is computationally more
efficient. The formula derived for numerical integration for single integrals with constant limits in [9] is given by
b
( b − a) −
2M
( b − a) −
2M
(b − a)(i − 0.5)
∫  
f (x) dx ≈ f (xk ) = f a+ . (1)
a 2M i =1
2M i =1
2M

where M = 2J1 and J1 is the maximum level of resolution of Haar wavelets.

2.2. Numerical formula for double integrals with variable limits

Consider double integral with variable limits of the type

∫ b ∫ d(y)
F (x, y) dx dy. (2)
a c (y)

We apply formula (1) to the integral

∫ d(y)
F (x, y) dx, (3)
c (y)
2772 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781

while keeping y constant. Thus we obtain

d(y)
{d(y) − c (y)} −
2M
{d(y) − c (y)}(i − 0.5)
∫  
F (x, y) dx ≈ F c (y) + ,y . (4)
c (y) 2M i=1
2M

Assume that

{d(y) − c (y)} −
2M
{d(y) − c (y)}(i − 0.5)
 
G(y) = F c (y) + ,y , (5)
2M i=1
2M

and apply formula (1) once again, we get the following formula for numerical integration of double integrals with variable
limits
d(y)
b b
(b − a) −
2N
(b − a)(i − 0.5)
∫ ∫ ∫  
F (x, y) dx dy ≈ G(y)dy ≈ G a+ , (6)
a c (y) a 2N i=1
2N

where N = 2J2 and J2 is the maximum level of resolution of Haar wavelets. Note that we may take different levels of
resolution of Haar wavelets for different axis.

2.3. Numerical formula for triple integrals with variable limits

The formula for triple integrals with variable limits can be derived in a similar way and is given by
d(z ) f (y,z )
b
( b − a) −
2P
(b − a)(i − 0.5)
∫ ∫ ∫  
F (x, y, z ) dx dy dz ≈ H a+ , (7)
a c (z ) e(y,z ) 2P i =1
2P

where

{d(z ) − c (z )} −
2N
{d(z ) − c (z )}(i − 0.5)
 
H (z ) = G c (z ) + ,z , (8)
2N i=1
2N

and

{f (y, z ) − e(y, z )} −
2M
{f (y, z ) − e(y, z )}(i − 0.5)
 
G(y, z ) = F e(y, z ) + , y, z . (9)
2M i=1
2M

In the above formula P = 2J3 where J3 is the level of resolution of the Haar wavelets.

3. Numerical integration using hybrid function

Interpolation condition of the hybrid function and basic definitions of block-pulse function and Legendre polynomials
are given in [9].

3.1. Numerical formulas for single integrals using hybrid functions

The numerical integration formulas using hybrid functions for the integral
∫ 1
f (x)dx (10)
0

were derived for different values of m in [9]. We extend these formulas for integrals of the type
∫ b
f (x)dx (11)
a

where a, b ∈ R. Let us first consider the case where m = 1. For this value of m the formula is given by
∫ 1 n  
1− 2i − 1
f (x)dx ≈ f . (12)
0 n i=1 2n

Making the substitution x = a + (b − a)y in equation (11), we obtain

∫ b ∫ 1
f (x)dx = (b − a) f (a + (b − a)y) dy. (13)
a 0
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2773

Applying formula (12), the above integral becomes

b
( b − a) −
n
(b − a)(2i − 1)
∫  
f (x)dx ≈ f a+ . (14)
a n i =1
2n

In a similar way the formulas for m = 2, 3, 4, 5, 6 given in [9] can be modified for integrals with limits from a to b where
a, b ∈ R and are given below.
For m = 2,
b n
(b − a)(4i − 3) (b − a)(4i − 1)
∫ [    ]
b−a −
f (x) dx ≈ f a+ +f a+ . (15)
a 2n i=1
4n 4n

For m = 3,

b n
(b − a)(6i − 5) (b − a)(6i − 3)
∫    
b−a −
f (x) dx ≈ 3f a+ + 2f a+
a 8n i=1
6n 6n

(b − a)(6i − 1)

+ 3f a+ . (16)
6n

For m = 4,
b n
(b − a)(8i − 7) (b − a)(8i − 5)
∫ [    
b−a −
f (x) dx ≈ 13f a+ + 11f a+
a 48n i=1
8n 8n

(b − a)(8i − 3) (b − a)(8i − 1)
   ]
+ 11f a+ + 13f a+ . (17)
8n 8n
For m = 5,

b n
(b − a)(10i − 9) (b − a)(10i − 7)
∫    
b−a −
f (x) dx ≈ 275f a+ + 100f a+
a 1152n i=1 10n 10n
(b − a)(10i − 5) (b − a)(10i − 3)
   
+ 402f a+ + 100f a+
10n 8n

(b − a)(10i − 1)

+ 275f a+ . (18)
10n

For m = 6,

b n
(b − a)(12i − 11) (b − a)(12i − 9)
∫    
b−a −
f (x) dx ≈ 247f a+ + 139f a+
a 1280n i=1 12n 12n
(b − a)(12i − 7) (b − a)(12i − 5)
   
+ 254f a+ + 254f a+
12n 12n

(b − a)(12i − 3) (b − a)(12i − 1)
  
+ 139f a+ + 247f a+ . (19)
12n 12n

For m = 7,

b n
(b − a)(14i − 13) (b − a)(14i − 11)
∫    
b−a −
f (x) dx ≈ 24745f a+ + 56007f a+
a 138240n i=1 14n 14n
(b − a)(14i − 9) (b − a)(14i − 7)
   
+ 56007f a+ − 25028f a+
14n 14n
(b − a)(14i − 5) (b − a)(14i − 3)
   
+ 56007f a+ + 56007f a+
14n 14n

(b − a)(14i − 1)

+ 24745f a+ . (20)
14n
2774 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781

For m = 8,
b n [
(b − a)(16i − 15) (b − a)(16i − 13)
∫    
b−a −
f (x) dx ≈ 295627f a+ + 71329f a+
a 1935360n i=1 16n 16n
(b − a)(16i − 11)
(b − a)(16i − 9)
   
+ 471771f a+ + 128953f a +
16n 16n
(b − a)(16i − 7) (b − a)(16i − 5)
   
+ 128953f a + + 471771f a +
16n 16n
(b − a)(16i − 3) (b − a)(16i − 1)
   ]
+ 71329f a+ + 295627f a+ . (21)
16n 16n
For m = 9,
b n [
(b − a)(18i − 17)
∫  
b−a −
f (x) dx ≈ 832221f a+
a 5734400n i=1 18n
(b − a)(18i − 15)
(b − a)(18i − 13)
   
− 260808f a+ + 2903148f a +
18n 18n
(b − a)(18i − 11) (b − a)(18i − 9)
   
− 3227256f a + + 5239790f a +
18n 18n
(b − a)(18i − 7) (b − a)(18i − 5)
   
− 3227256f a + + 2903148f a +
18n 18n
(b − a)(18i − 3) (b − a)(18i − 1)
   ]
− 260808f a+ + 832221f a+ . (22)
18n 18n

3.2. Numerical formulas for double integrals

In [9] numerical integration formulas for double integrals with limits 0 and 1 were derived by approximating the
integrand F (x, y) with hybrid functions. Those formulas can only be applied when the area of integration is a rectangle.
Moreover, as we move from single integrals to higher integrals the computational cost of the approach used in [9] increases
considerably. We use an alternative procedure similar to the one used above for Haar wavelets. In this section hybrid
functions will be used to derive numerical integration formulas with variable limits, that is, where the area of integration
need not be a rectangle. The general form of such an integral would be
∫ b ∫ d(y)
F (x, y) dx dy. (23)
a c (y)

The formulas for numerical integration of single integrals with constant limits given in the last section will be applied twice
to this integral. First consider the integral
∫ d(y)
F (x, y) dx, (24)
c (y)

where y is treated as a constant. Applying formula (14) to the above integral, we have
d(y)
{d(y) − c (y)} −
n
{d(y) − c (y)}(2i − 1)
∫  
F (x, y) dx ≈ F c (y) + ,y . (25)
c (y) n i=1
2n

Therefore, integral equation (23) can be approximated as

d(y)
 
b b
{d(y) − c (y)} −
n
{d(y) − c (y)}(2i − 1)
∫ ∫ ∫ 
F (x, y) dx dy ≈ F c (y) + ,y dy. (26)
a c (y) a n i =1
2n

Now, let us assume that

{d(y) − c (y)} −
n
{d(y) − c (y)}(2i − 1)
 
G(y) = F c (y) + ,y . (27)
n i=1
2n
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2775

Applying formula (14) once again, now to the integral

∫ b
G(y)dy, (28)
a

we obtain
d(y)
b b n
(b − a)(2i − 1)
∫ ∫ ∫  
b−a −
F (x, y) dx dy ≈ G(y) dy ≈ G a+ . (29)
a c (y) a n i=1
2n

Hence the numerical approximation of the integral (23) using hybrid functions with m = 1 is given as

d(y)
b
( b − a) −
n
(b − a)(2i − 1)
∫ ∫  
F (x, y) dx dy ≈ G a+ , (30)
a c (y) n i =1
2n

where

{d(y) − c (y)} −
n
{d(y) − c (y)}(2i − 1)
 
G(y) = F c (y) + ,y . (31)
n i =1
2n

In a similar way we can derive formulas for other values of m. The formulas for m = 2, 3, . . . , 9 are given below.
For m = 2,

(d(y))
b
(b − a) −
n
(b − a)(4i − 3) (b − a)(4i − 1)
∫ ∫ [    ]
F (x, y) dx dy ≈ G a+ +G a+ , (32)
a c (y) 2n i=1
4n 4n

where

{d(y) − c (y)} −
n
{d(y) − c (y)}(4i − 3) {d(y) − c (y)}(4i − 1)
[    ]
G(y) = F c (y) + , y + F c (y) + ,y . (33)
2n i =1
4n 4n

For m = 3,
(d(y))
b
(b − a) −
n
(b − a)(6i − 5)
∫ ∫ [  
F (x, y) dx dy ≈ 3G a +
a c (y) 8n i=1
6n

(b − a)(6i − 3) (b − a)(6i − 1)
   ]
+ 2G a + + 3G a + , (34)
6n 6n

where

{d(y) − c (y)} −
n
{d(y) − c (y)}(6i − 5)
[  
G(y) = 3F c (y) + ,y
8n i=1
6n

{d(y) − c (y)}(6i − 3) {d(y) − c (y)}(6i − 1)

   ]
+ 2F c (y) + , y + 3F c (y) + ,y . (35)
6n 6n

For m = 4,
(d(y))
b
(b − a) −
n
(b − a)(8i − 7) (b − a)(8i − 5)
∫ ∫ [    
F (x, y) dx dy ≈ 13G a + + 11G a +
a c (y) 48n i=1
8n 8n

(b − a)(8i − 3) (b − a)(8i − 1)
   ]
+ 11G a + + 13G a + , (36)
8n 8n

where

{d(y) − c (y)} −
n
{d(y) − c (y)}(8i − 7) {d(y) − c (y)}(8i − 5)
[    
G(y) = 13F c (y) + , y + 11F c (y) + ,y
48n i=1
8n 8n

{d(y) − c (y)}(8i − 3) {d(y) − c (y)}(8i − 1)

   ]
+ 11F c (y) + , y + 13F c (y) + ,y . (37)
8n 8n
2776 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781

For m = 5,
(d(y))
b
(b − a) −
n
(b − a)(10i − 9)
∫ ∫ [  
F (x, y) dx dy ≈ 275G a +
a c (y) 1152n i=1 10n
(b − a)(10i − 7) (b − a)(10i − 5)
   
+ 100G a + + 402G a +
10n 10n
(b − a)(10i − 3) (b − a)(10i − 1)
   ]
+ 100G a + + 275G a + , (38)
10n 10n
where
{d(y) − c (y)} −
n
(d(y) − c (y))(10i − 9)
[  
G(y) = 275F c (y) + ,y
1152n i =1
10n
{d(y) − c (y)}(10i − 7) {d(y) − c (y)}(10i − 5)
   
+ 100F c (y) + , y + 402F c (y) + ,y
10n 8n
{d(y) − c (y)}(10i − 3) {d(y) − c (y)}(10i − 1)
   ]
+ 100F c (y) + , y + 275F c (y) + ,y . (39)
10n 10n
For m = 6,
(d(y))
b
(b − a) −
n
(b − a)(12i − 11)
∫ ∫ [  
F (x, y) dx dy ≈ 247G a +
a c (y) 1280n i=1 12n
(b − a)(12i − 9) (b − a)(12i − 7)
   
+ 139G a + + 254G a +
12n 12n
(b − a)(12i − 5) (b − a)(12i − 3)
   
+ 254G a + + 139G a +
12n 12n
(b − a)(12i − 1)
 ]
+ 247G a + . (40)
12n
where
{d(y) − c (y)} −
n
{d(y) − c (y)}(12i − 11)
[  
G(y) = 247F c (y) + ,y
1280n i =1
12n
{d(y) − c (y)}(12i − 9) {d(y) − c (y)}(12i − 7)
   
+ 139F c (y) + , y + 254F c (y) + ,y
12n 12n
{d(y) − c (y)}(12i − 5) {d(y) − c (y)}(12i − 3)
   
+ 254F c (y) + , y + 139F c (y) + ,y
12n 12n
{d(y) − c (y)}(12i − 1)
 ]
+ 247F c (y) + ,y . (41)
12n
For m = 7,
(d(y))
b
(b − a) −
n [
(b − a)(14i − 13)
∫ ∫  
F (x, y) dx dy ≈ 24745G a +
a c (y) 138240n i=1 14n
(b − a)(14i − 11) (b − a)(14i − 9)
   
+ 882G a + + 56007G a +
14n 14n
(b − a)(14i − 7) (b − a)(14i − 5)
   
− 25028G a + + 56007G a +
14n 14n
(b − a)(14i − 3) (b − a)(14n − 1)
   ]
+ 882G a + + 24745G a + (42)
14n 14n
where
{d(y) − c (y)} −
n
{d(y) − c (y)}(14i − 13)
[  
G(y) = 24745F c (y) + ,y
138240n i=1
14n
{d(y) − c (y)}(14i − 11) {d(y) − c (y)}(14i − 9)
   
+ 882F c (y) + , y + 56007F c (y) + ,y
14n 14n
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2777

{d(y) − c (y)}(14i − 7) {d(y) − c (y)}(14i − 5)

   
− 25028F c (y) + , y + 56007F c (y) + ,y
14n 14n
{d(y) − c (y)}(14i − 1) {d(y) − c (y)}(14i − 1)
   ]
+ 882F c (y) + , y + 24745F c (y) + ,y . (43)
14n 14n
For m = 8,
(d(y))
b n [
(b − a)(16i − 15)
∫ ∫  
b−a −
F (x, y) dx dy ≈ 295627G a +
a c (y) 1935360n i=1 16n

(b − a)(16i − 13) (b − a)(16i − 11)

   
+ 71329G a + + 471771G a +
16n 16n
(b − a)(16i − 9) (b − a)(16i − 7)
   
+ 128953G a + + 128953G a +
16n 16n
(b − a)(16i − 5) (b − a)(16i − 3)
   
+ 471771G a + + 71329G a +
16n 16n

(b − a)(16i − 1)
 ]
+ 295627G a + (44)
16n
where
{d(y) − c (y)} −
n
{d(y) − c (y)}(16i − 15)
[  
G(y) = 295627F c (y) + ,y
1935360n i=1
16n

{d(y) − c (y)}(16i − 13) {d(y) − c (y)}(16i − 11)

   
+ 71329F c (y) + , y + 471771F c (y) + ,y
16n 16n
{d(y) − c (y)}(16i − 9) {d(y) − c (y)}(16i − 7)
   
+ 128953F c (y) + , y + 128953F c (y) + ,y
16n 16n
{d(y) − c (y)}(16i − 5) {d(y) − c (y)}(16i − 3)
   
+ 471771F c (y) + , y + 71329F c (y) + ,y
16n 16n

{d(y) − c (y)}(16i − 1)
 ]
+ 295627F c (y) + ,y . (45)
16n
For m = 9,
(d(y))
b n [
(b − a)(18i − 17)
∫ ∫  
b−a −
F (x, y) dx dy ≈ 832221G a +
a c (y) 5734400n i=1 18n
(b − a)(18i − 15) (b − a)(18i − 13)
   
− 260808G a + + 2903148G a +
18n 18n
(b − a)(18i − 11) (b − a)(18i − 9)
   
− 3227256G a + + 5239790G a +
18n 18n
(b − a)(18i − 7) (b − a)(18i − 5)
   
− 3227256G a + + 2903148G a +
18n 18n
(b − a)(18i − 3) (b − a)(18i − 1)
   ]
− 260808G a + + 832221G a + (46)
18n 18n
where
{d(y) − c (y)} −
n
{d(y) − c (y)}(18i − 17)
[  
G(y) = 832221F c (y) + ,y
5734400n i=1
18n
{d(y) − c (y)}(18i − 15) {d(y) − c (y)}(18i − 13)
   
− 260808F c (y) + , y + 2903148F c (y) + ,y
18n 18n
{d(y) − c (y)}(18i − 11) {d(y) − c (y)}(18i − 9)
   
− 3227256F c (y) + , y + 5239790F c (y) + ,y
18n 18n
2778 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781

{d(y) − c (y)}(18i − 7) {d(y) − c (y)}(18i − 5)

   
− 3227256F c (y) + , y + 2903148F c (y) + ,y
18n 18n
{d(y) − c (y)}(18i − 3) {d(y) − c (y)}(18i − 1)
   ]
− 260808F c (y) + , y + 832221F c (y) + ,y . (47)
18n 18n

3.3. Numerical formula for triple integrals

The method used in the last section can be extended for numerical integration of triple or even integrals in higher
dimensions with variable limits of integration. We derive a formula for triple integrals using hybrid functions with m = 3.
Formulas for other values of m can be derived easily in a similar fashion.
Consider the triple integral
∫ b ∫ d(z ) ∫ f (y,z )
F (x, y, z ) dx dy dz . (48)
a c (z ) e(y,z )
Suppose m = 3. Applying formula (16) three times, we obtain
d(z ) f (y,z )
b
( b − a) −
n
(b − a)(6i − 5)
∫ ∫ ∫ [  
F (x, y, z ) dx dy dy dz ≈ 3H a+
a c (z ) e(y,z ) 8n i =1
6n

(b − a)(6i − 3) (b − a)(6i − 1)
   ]
+ 2H a + + 3H a + , (49)
6n 6n
where
{d(z ) − c (z )} −
n
{d(z ) − c (z )}(6i − 5)
[  
H (z ) = 3G c (z ) + ,z
8n i=1
6n
{d(z ) − c (z )}(6i − 3) {d(z ) − c (z )}(6i − 1)
   ]
+ 2G c (z ) + , z + 3G c (z ) + ,z , (50)
6n 6n

{f (y, z ) − e(y, z )} −
n
{f (y, z ) − e(y, z )}(6i − 5)
 
G(y, z ) = 3F e(y, z ) + , y, z
8n i=1
6n
{f (y, z ) − e(y, z )}(6i − 3)
 
+ 2F e(y, z ) + , y, z
6n

{f (y, z ) − e(y, z )}(6i − 1)

+ 3F e(y, z ) + , y, z . (51)
6n

4. Numerical examples

Example 1.
∫ 1 ∫ y
e|x+y−1| dx dy = −2 + e
0 0
This example has also been solved numerically in [8] by using symmetric Gauss Legendre quadrature formulas for composite
numerical integration over a triangular surfaces. As shown Table 2, the best results reported in [8] are accurate up to 6
decimal places for 10,000 nodal points by using a fifth-order Gauss Legendre polynomial. The new method presented in
the previous section achieves accuracy up to 7 decimal places corresponding to a third-order Legendre polynomial and 625
nodal points. The accuracy of the new method improves up to 10 decimal places for 10,000 nodal points by using an eighth-
order Legendre polynomial. Better accuracy is achieved with a considerably lower number of nodes in the case of hybrid
functions. These results are shown in Table 1. This clearly indicates the economic use of memory, efficiency and superiority
of the new method. Moreover, the method [8] can only be applied to integrals over triangular regions while our methods
can be applied to a variety of regions including triangular regions.

Example 2.
π sin y
π2
∫ ∫
4 1
dx dy = .
(1 − x2 )

0 0 32
Absolute errors are shown in Table 3. From Table 3 it is clear that the new method performs very well for this problem.
Like the previous case, performance of the hybrid function is much better than Haar wavelets due to their best interpolating
properties.
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2779

10
[9] 450
[9]
9 Present method
400 Present method
8
350
7

CPU time secs

CPU time secs
300
6
250
5
200
4
150
3
100
2
1 50

0 0
3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8
J n

Fig. 1. CPU time comparison of new and present methods using Haar wavelets (left) and hybrid functions (right).

Table 1
Absolute Errors of Example 1.
Haar Absolute errors Hybrid Absolute errors

M = 4, N = 5 1.6425E−03 m = 3, n = 25 9.0521E−07
M = 5, N = 10 4.9512E−04 m = 4, n = 50 1.4728E−08
M = 10, N = 15 1.9801E−04 m = 6, n = 100 1.3179E−09
M = 16, N = 16 1.1857E−04 m = 8, n = 100 2.9992E−10

Table 2
Selective absolute errors of Example 1 reported in [8].
Nodal points Order of Gauss Legendre polynomial Absolute errors

N = 180 × 180 2 7.995E−07

N = 100 × 100 3 2.322E−06
N = 100 × 100 4 6.284E−07
N = 100 × 100 5 2.845E−08

Example 3.
∫ ∫
1 2 √ √ √ 
(x + y)− 2 dx dy = 2 − 7 3 − 15 5 + 20 6 ≈ 3.549613026789713,
R 3
where R is a quadrilateral region connecting the points (−1, 2), (2, 1), (3, 3) and (1, 4). In our case, we divide the region R
into three subregions so that the given integral can be written as sum of three integrals as
1
∫ ∫ ∫ 2 ∫
2
(y+3)
− 12 1
(x + y) dx dy = (x + y)− 2 dx dy
R 1 (5−3y)
1
∫ 3 ∫
2
(y+3) ∫ 4 ∫ (9−2y)
− 21 1
+ (x + y) dx dy + (x + y)− 2 dx dy. (52)
2 (y−3) 3 y −3

This example has been solved numerically in [7] by using symmetric Gauss Legendre quadrature formulas for quadrilateral
regions only. Accuracy up to 15 decimal places is reported in [7] by using a Gauss Legendre polynomial of order 9
corresponding to 400 nodal points. Comparative performance of the present methods based on Haar wavelets, hybrid
functions and the method [7] are shown in Tables 4 and 5. It can be observed from these tables that the same accuracy is
attained in both cases. The added advantage of the new approach is that it requires considerably less manual labor involved
in the derivation of the new method in comparison with the existing methods [8,7]. Accuracy of the Haar wavelets based
algorithm is considerably lower than the hybrid functions in this case as well.

Example 4.
∫ 1 ∫ 1
1
 dxdy.
0 0 x2 + y2
The CPU time comparison of method [9] and present methods for different values of J in case of Haar wavelets and n in case
of hybrid functions are shown in Table 6. The numerical results show that the present methods are more efficient than the
methods in [9]. It is clear from the comparison that the present approach is more economical in terms of CPU time, especially
in the case of hybrid functions (the present approach is almost 10 times faster than the method in [9]). These results have
been obtained in Mathematica 7.
2780 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781

Table 3
Absolute Errors of Example 2.
Haar Absolute errors Hybrid Absolute errors

M = 3, N =3 1.7115E−04 m = 3, n = 10 2.5489E−08
M = 4, N =3 9.6650E−05 m = 5, n = 10 4.0726E−11
M = 5, N =4 6.2718E−05 m = 7, n = 10 1.0419E−13
M = 6, N =5 4.3844E−05 m = 9, n = 10 3.3307E−16

Table 4
Selective absolute errors of Example 3 reported in [7].
Nodal points Order of Gauss Legendre polynomial Absolute errors

N = 400 2 1.578E−07
N = 400 3 1.072E−10
N = 400 4 1.039E−13
N = 400 5 3.996E−15
N = 400 6 3.996E−15
N = 400 7 3.996E−15
N = 400 8 3.996E−15
N = 400 9 3.996E−15

Table 5
Absolute errors of Example 3.
Haar Absolute errors Hybrid Absolute errors

M =N =2 1.1497E−02 m = 3, n = 20 7.0503E−08
M =N =4 2.9874E−03 m = 4, n = 20 3.6491E−08
M =N =8 7.5501E−04 m = 5, n = 20 5.1215E−11
M =N = 16 1.8929E−04 m = 6, n = 20 3.0780E−11
M =N = 32 4.7355E−05 m = 7, n = 20 6.2172E−14
M =N = 64 1.1841E−05 m = 8, n = 20 3.8192E−14
M =N = 128 2.9604E−06 m = 9, n = 20 3.9968E−15

Table 6
CPU time comparison of methods in [9] and present methods for Example 4.
Haar Hybrid (m = 4)
J Method in [9] Present method n Method in [9] Present method

3 0.016 0.015 10 0.140 0.031

4 0.063 0.046 25 0.810 0.125
5 0.172 0.125 50 3.541 0.390
6 0.624 0.514 100 14.88 1.560
7 2.434 2.044 250 97.70 10.02
8 9.828 7.956 500 419.5 38.89

Table 7
Absolute errors of Example 5.
Haar Absolute errors Hybrid Absolute errors

M =N =P =8 3.5959E−03 m = 3, n = 20 2.4465E−07
M =N =P = 16 9.0291E−04 m = 4, n = 20 1.2654E−07
M =N =P = 32 2.2597E−04 m = 5, n = 20 4.2473E−11
M =N =P = 64 5.6508E−05 m = 6, n = 20 2.5509E−11

Example 5.
∫ π ∫ z∫ zy
 
1 x 1
sin dx dy dz = (4 + π 2 ).
0 0 0 y y 2

Absolute errors corresponding to three dimensional problem are shown in Table 7. Like all the previous cases, better
performance of new method for three dimensional integral (specially in the case of hybrid functions) is maintained
(see Fig. 1).
 I. Aziz et al. / Computers and Mathematics with Applications 61 (2011) 2770–2781 2781

5. Conclusion

In this paper a new method based on Haar wavelets and hybrid functions is applied for numerical integration of double
and triple integrals with variable limits. This method is the generalization and improvement of our earlier method [9]. This
method has several advantages over the conventional quadrature rule. The new method is tested on a variety of benchmark
problems. Comparison between the two methods shows that the hybrid functions method gives better results than the Haar
wavelets method. The hybrid functions method is also compared with symmetric Gauss Legendre quadrature [8,7] and it
has been shown that the present method is more accurate and is easy to implement. It can easily be observed that the new
method is explicit and easy to implement for a variety of problems arising in different applications.

Acknowledgement

We are thankful to the reviewer for the valuable suggestions which improved the first draft of the paper.

References

[1] R.L. Burden, J.D. Faires, Numerical analysis, PWS-KENT, USA, 1993.
[2] V. Rokhlin, End point corrected trapezoidal quadrature rules for singular functions, Comput. Math. Appl. 20 (1990) 51–62.
[3] J. Ma, V. Rokhlin, S. Vandzura, Generalised gaussian quadrature rules for asystems of artbitrary functions, SIAM J. Numer. Anal. 33 (1996) 971–996.
[4] H. Monien, Gaussian quadrature for sums: A rapidly convergent summation scheme, Math. Comput. 79 (2009) 857–869.
[5] B.K. Alpert, Hybrid gauss-trapezoidal quadrature rules, SIAM J. Sci. Comput. 20 (1999) 1551–1584.
[6] J. Place, J. Stach, Efficient numerical integration using gaussian quadrature, Simulation 73 (1999) 232–237.
[7] Md. Shafiqul Islam, M. Alamgir Hossain, Numerical integrations over an arbitrary quadrilateral region, Appl. Math. Comput. 210 (2009) 515–524.
[8] H.T. Rathod, K.V. Nagaraja, B. Venkatesudu, Symmetric gauss legendre quadrature formulas for composite numerical integration over a triangular
surface, Appl. Math. Comput. 188 (2007) 865–876.
[9] Siraj ul Islam, Imran Aziz, Fazal Haq, A comparative study of numerical integration based on haar wavelets and hybrid functions, Comput. Math. Appl.
59 (2010) 2026–2036.
[10] 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) 794–800.
[11] Mehdi Dehghan, Mehrdad Lakestani, Numerical solution of nonlinear system of second-order boundary value problems using cubic b-spline scaling
functions, Int. J. Comput. Math. 85 (2008) 1455–1461.
[12] Siraj-ul-Islam Imran Aziz, B. Šarler, The numerical solution of second-order boundary-value problems by collocation method with the haar wavelets,
Math. Comput. Modelling 50 (2010) 1577–1590.
[13] Valeriano Comincioli, Giovanni Naldi, Terenzio Scapolla, A wavelet-based method for numerical solution of nonlinear evolution equations, Appl.
Numer. Math. 33 (2000) 291–297.
[14] J.L. Wu, A wavelet operational method for solving fractional partial differential equations numerically, Appl. Math. Comput. 214 (2009) 31–40.
[15] Daubechies wavelet beam and plate finite elements, Lilliam alvarez diaz and maria t. martin and victoria vampa, Finite Elem. Anal. Des. 45 (2009),
200–209.
[16] X. Zhu, G. Lei, G. Pan, On application of fast and adaptive battle-lemarie wavelets to modelling of multiple lossy transmission lines, J. Comput. Phys.
132 (1997) 299–311.
[17] E. Babolian, F. Fattahzdeh, Numerical solution of differential equations by using chebyshev wavelet operational matrix of integration, Appl. Math.
Comput. 188 (2007) 417–426.
[18] E. Banifatemi, M. Razzaghi, S. Yousefi, Two-dimensional legendre wavelets method for the mixed volterra-fredholm integral equations, J. Vib. Control
13 (2007) 1667–1675.
[19] Xufeng Shang, Danfu Han, Numerical solution of fredholm integral equations of the first kind by using linear legendre multi-wavelets, Appl. Math.
Comput. 191 (2007) 440–444.
[20] Ü. Lepik, Numerical solution of evolution equations by the haar wavelet method, Appl. Math. Comput. 185 (2007) 695–704.
[21] C.F. Chen, C.H. Hsiao, Haar wavelet method for solving lumped and distributed-parameter systems, IEE Proc., Control Theory Appl. 144 (1997) 87–94.
[22] C.H. Hsiao, Haar wavelet approach to linear stiff systems, Math. Comput. Simulation 64 (2004) 561–567.
[23] Chun hui Hsiao, Wen-June Wang, Haar wavelet approach to nonlinear stiff systems, Math. Comput. Simulation 57 (2001) 347–353.
[24] Ü. Lepik, Haar wavelet method for nonlinear integro-differential equations, Appl. Math. Comput. 176 (2006) 324–333.
[25] Ü. Lepik, Numerical solution of differential equations using haar wavelets, Math. Comput. Simulation 168 (2005) 127–143.
[26] Ü. Lepik, E. Tamme, Solution of nonlinear fredholm integral equations via the haar wavelet method, Proc. Estonian Acad. Sci. Phys. Math. 56 (2007)
17–27.
[27] Ü. Lepik, E. Tamme, Application of the haar wavelets for solution of linear integral equations, Dynam. Sys. Appl., Antalaya, Proceedings (2004), 494–507.
[28] K. Maleknejad, F. Mirzaee, Using rationalized haar wavelet for solving linear integral equations, Appl. Math. Comput. 160 (2005) 579–587.
[29] E. Babolian, A. Shshsawaran, Numerical solution of nonlinear fredholm integral equations of the second kind using haar wavelets, J. Comput. Appl.
Math. 225 (2009) 87–95.
[30] Xing Tao, Wang, Numerical solution of optimal control for scaled systems by hybrid functions, Internat. J. Innovative Comput., Info. Cont. 4 (2008)
849–855.
[31] C.H. Hsiao, Hybrid function method for solving fredholm and volterra integral equations of the second kind, J. Comput. Appl. Math. 230 (2009) 59–68.
[32] H.R. Marzban, M. Razzaghi, Numerical solution of the controlled duffing oscillator by hybrid functions, Appl. Math. Comput. 140 (2003) 179–190.
[33] H.R. Marzban, H.R. Tabrizidooz, M. Razzaghi, Hybrid functions for nonlinear initial-value problems with applications to lane-emden type equations,
Phys. Lett. A 372 (2008) 5883–5886.
[34] H.R. Marzban, M. Razzaghi, Optimal control of linear delay systems via hybrid of block-pulse and legendre polynomials, J. Franklin Inst. 341 (2004)
279–293.
[35] H. Hashish, S.H. Behiry, N.A. Shamy, Numerical integration using wavelets, Appl. Math. Comput. 211 (2009) 480–487.