Malaysian Journal of Mathematical Sciences 16(1): 87–103 (2022)

Malaysian Journal of Mathematical Sciences

Journal homepage: https://einspem.upm.edu.my/journal

Modified HAM for solving linear system of Fredholm-Volterra Integral

Equations

Eshkuvatov, Z. K. ∗1,3 , Ismail, Sh. 2 , Mamatova, H. X. 1,3 , Viscarra, D. S. 4 , and Aloev, R. D. 3

1
Faculty of Ocean Engineering Technology and Informatics, Universiti Malaysia Terangganu, Malaysia
2
Faculty of Science and Technology, Universiti Sains Islam Malaysia, Malaysia
3
Faculty of Applied Mathematics and Intellectual Technologies, National University of Uzbekistan,
Uzbekistan
4
Departamento de Matemáticas, Universidad Simón I. Patiño, Bolivia

E-mail: [email protected]
∗
Corresponding author
Received: 8 March 2020
Accepted: 15 July 2021

Abstract

This paper considers systems of linear Fredholm-Volterra integral equations using a modified
homotopy analysis method (MHAM) and the Gauss-Legendre quadrature formula (GLQF) to
find approximate solutions. Standard homotopy analysis method (HAM), MHAM, and optimal
homotopy asymptotic method (OHAM) are compared for the same number of iterations. It is
noted from the chosen examples that MHAM with GLQF is comparable with standard HAM and
OHAM. In all cases, MHAM with GLQF approaches exact solutions, where residual rapidly
converges to zero when the number of iterations and quadrature nodes increases. The HAM
developed in this paper is better than the HAM developed by Shidfar & Molabahrami in “Solving
a system of integral equations by an analytic method”.

Keywords: homotopy analysis method, Gauss-Legendre quadrature formula, approximate so-

lutions, convergence.
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

1 Introduction

The homotopy analysis method (HAM) was first introduced by Liao [9] in his PhD thesis, where
he successfully applied it to nonlinear problems. A systematic and clear exposition on HAM is
given in Liao’s work [2]. After the publication of that work, a number of researchers have success-
fully applied this method to various nonlinear problems in science and engineering. For example,
see [10] discussed the non-linear problems, [11, 7] discussed fluid dynamics, [12] discussed the
non-linear problem and comparison between HPM and HAM, [1] elaborated on KdV equations,
[14] discussed non-linear differential problem, and [3] elaborated on engineering and biology.
The HAM contains an auxiliary parameter ~, which provides us with a simple way to adjust and
control the convergence region and the rate of convergence of the series solution. Moreover, em-
ploying the so-called ~-curve, it is easy to find the valid regions of ~ to gain a convergent series
solution. The main aim is to analyze the HAM from a mathematical point of view utilizing some
basic nonlinear and linear problems of the system of integral equations and hence show a pre-
sentable comparison by obtaining the exact solution of the homotopy equations. On the other
hand, employing the convergence-control parameter can always be avoided in the HAM. Thus,
HAM is an explicit semi-analytic solution of linear and nonlinear problems.

In numerical analysis, a quadrature rule approximates the definite integral of a function, usu-
ally stated as a weighted sum of function values at specific points within the domain of integration.
An n-point Gaussian quadrature rule, named after Carl Friedrich Gauss, is a quadrature rule con-
structed to give exact results for polynomials of degree 2n − 1 or less, employing a suitable choice
of nodes {xi } and weights {wi }, for i ∈ {1, 2, . . . , n}. Legendre polynomials Pn (x) are orthogonal
polynomials on the interval [−1, 1] with unit weight. The Gauss-Legendre quadrature formula
(GLQF) is a special case of Gaussian quadrature that allows us to approximate a function with
known asymptotic behavior at the edges of the integration interval. Applications of the GLQF can
be found in many references. For example, [4] derived error estimates for the Gauss Legendre and
the Gauss-Chebyshev quadrature formulas for analytic functions, [5] discussed Gauss-Legendre
quadrature formula (GLQF) extended to the kernel integrals of two variables and proved the
convergence of GLQF, while [8] discussed numerous methods on integration problems including
GLQF.

Consider a general system of linear Fredholm-Volterra integral equations (FVIEs) of the form
Z b Z x
Ū (x) = Ḡ(x) + λ1 K̄1 (x, t)Ū (t) dt + λ2 K̄2 (x, t)Ū (t) dt, (1)
a a

where
Ū (x) = (u1 (x), u2 (x), . . . , un (x)) ,
Ḡ(x) = (g1 (x), g2 (x), . . . , gn (x)) , (2)
K̄r (x, t) = (kr,ij (x, t)) , r ∈ {1, 2}, i, j ∈ {1, 2, . . . , n},

where x ∈ R denotes the independent variable; gi (x) are known analytic functions, and ui (x) are
unknown functions to be determined. For gi (x) = 0, (1) reduces to the homogeneous equation,
and the theory of eigenvalues and eigenfunctions can be applied.

Ghazanfari & Yari [6] considered (1) with λ2 = 0, and applied optimal homotopy asymp-
totic method (OHAM) to obtain an approximate solution. This work showed the effectiveness
of OHAM for (1) and provided easy tools to control the convergence region of approximating

88
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

solution series, wherever necessary. The results of OHAM were compared with the homotopy
perturbation method and the Taylor series expansion method. Shidfar & Molabahrami [13] pro-
posed HAM for solving systems of linear and nonlinear Volterra and Fredholm integral equations.
Using HAM, it is possible to find the exact solution or approximate solution for the problem. The
efficiency of the approach was illustrated by applying the procedure to several examples.

This paper is arranged in the following manner. In Section 2, we present the standard HAM
and modified HAM. Section 3 describes the implementation of standard HAM and modified HAM
for the problem (1)–(2). In Section 4, a GLQF is developed for kernel integration. Section 5 deals
with examples and shows a comparison of the proposed method with standard HAM and OHAM.
Finally, some conclusions and acknowledgements are given in Section 6.

2 Analysis of the HAM

To apply HAM to the i-th equation of (1), let us rewrite it as follows:

n
" #
X Z b Z x
ui (x) = gi (x) + λ1 k1,ij (x, t)uj (t) dt + λ2 k2,ij (x, t)uj (t) dt , i ∈ {1, 2, ..., n}. (3)
j=1 a a

The operator form of (3) is

Ni [ui (x)] = gi (x), i ∈ {1, 2, ..., n}, (4)

where
n 
X Z b Z x 
Ni [ui (x)] = ui (x) − λ1 k1,ij (x, t)uj (t) dt + λ2 k2,ij (x, t)uj (t) dt .
j=1 a a

In general, the operator Ni in (4) can be nonlinear or singular.

For the sake of clarity, we will first present a brief description of the standard HAM. From it,
we will derive the description of the algorithm of modified HAM (MHAM).

To generalize the traditional homotopy method, Liao [9] constructed the so-called zeroth-order
deformation equations,

(1 − q)L [φi (x, q) − ui,0 (x)] = q~ {Ni [φi (x, q)] − gi (x)} , (5)

where q ∈ [0, 1] is an embedding parameter, ~ is a nonzero auxiliary parameter, L is an auxil-

iary linear operator, ui,0 (x) are initial guesses of exact solutions ui (x), and φi (x, q) are unknown
functions.

Obviously, when q = 0 and q = 1, then

φi (x, 0) = ui,0 (x) and φi (x, 1) = ui (x)

89
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

hold, respectively. Thus, as q increases from 0 to 1, the function φi (x, q) varies from the initial
guess ui,0 (x) to the solution ui (x)—this is called deformation. Indeed, expanding φi (x, q) in Taylor
series with respect to q, one has
∞
X
φi (x, q) = ui,0 (x) + ui,k (x)q k , (6)
k=1

where
1 ∂ k [φi (x, q)] 

ui,k (x) = · . (7)
k! ∂q k 
q=0

As long as the auxiliary linear operator L, the initial guess ui,0 , the auxiliary parameter ~, and the
auxiliary functions are properly chosen, the series (6) converges at q = 1. Then, one has
∞
X
φi (x, 1) = ui (x) = ui,0 (x) + ui,k (x), (8)
k=1

which must be one of the solutions of the original system of nonlinear equations (4).

Based on (7), the governing equation can be deduced from the zeroth-order deformation equa-
tions (5). To do this, let us define the vectors

ūi,n = (ui,0 (x), ui,1 (x), . . . , ui,n (x)) .

Differentiating (5) m times with respect to the embedding parameter q, then setting q = 0, and
finally dividing by m!, we have the so-called m-th order deformation equation,

L [ui,m (x) − χm ui,m−1 (x)] = ~Ri,m (ūi,m−1 ) , (9)

where
∂ m−1 {Ni [φi (x, q)] − gi (x)} 

1
Ri,m [ūi,m−1 (x)] = · , (10)
(m − 1)! ∂q m−1 
q=0

and (
0 if m ≤ 1,
χm = (11)
1 if m > 1.

The modified form of the HAM can be obtained by the assumption that the function gi (x) can
be split into n terms, namely

gi (x) = si,0 (x) + si,1 (x) + · · · + si,n (x).

Expanding gi (x) = ϕi (x, q) into powers of the embedding parameter q, we have

n
X
ϕi (x, q) = si,0 (x)q 0 + si,1 (x)q 1 + · · · + si,n (x)q n = q j si,j (x). (12)
j=0

90
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

From (5) and (12), the new zeroth-order deformation equation has the form
(1 − q)L [φi (x, q) − ui,0 (x)] = q~ {Ni [φi (x, q)] − ϕi (x, q)} ,

and the m-th order deformation equation is

L [ui,m (x) − χm ui,m−1 (x)] = ~Ri,m (ūi,m−1 ) , (13)

where
∂ m−1 {Ni [φi (x, q)] − ϕi (x, q)} 

1
Ri,m [ūi,m−1 (x)] = · , (14)
(m − 1)! ∂q m−1 
q=0

and χm is defined by (11).

3 Implementation of Standard HAM and Modified HAM

For (3), let L = I and m = 1. Then (9) becomes

ui,1 (x) = ~Ri,1 (ūi,0 ) , (15)

where
Ri,1 (ūi,0 ) = {Ni [φi (x, q)] − gi (x)}q=0 .

 P∞ 
We can readily see that φi (x, q)|q=0 = ui,0 (x) + k=1 ui,k (x)q k q=0 = ui,0 (x), then
n
" Z #
X b Z x
Ri,1 (ūi,0 ) = ui,0 (x) − gi (x) − λ1 k1,ij (x, t)uj,0 (t) dt + λ2 k2,ij (x, t)uj,0 (t) dt . (16)
j=1 a a

Since ui,0 (x) = gi (x), from (15) and (16) it follows that
n
" Z #
X b Z x
ui,1 (x) = −~ λ1 K1,ij (x, t)gj (t) dt + λ2 K2,ij (x, t)gj (t) dt . (17)
j=1 a a

On the other hand, for m > 1, we have

ui,m (x) = ui,m−1 (x) + ~Ri,m (ūi,m−1 ) . (18)

Let m = 2, then, from (10) we obtain


∂ 
Ri,2 = [Ni [φi (x, q)] − gi (x)]
∂q q=0
( n
" Z
b
∂ X
= φi (x, q) − gi (x) − λ1 K1,ij (x, t)φj (x, q) dt (19)
∂q j=1 a
#)
Z x
+ λ2 K2,ij (x, t)φj (x, q) dt .
a
q=0

91
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103


d
It is not difficult to see that dq φi (x, q) = ui,1 (x). Therefore, from (18) and (19), it follows that

q=0

n 
#
X Z b Z x
ui,2 (x) = (1 + ~)ui,1 (x) − ~ λ1 K1,ij (x, t)uj,1 (t) dt + λ2 K2,ij (x, t)uj,1 (t) dt .
j=1 a a

By continuing this procedure, we obtain

n
"
X Z b
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ λ1 K1,ij (x, t)uj,m−1 (t) dt
j=1 a
# (20)
Z x
+ λ2 K2,ij (x, t)uj,m−1 (t) dt , m ≥ 2.
a

Eqs. (17) and (20) are the application of the standard HAM for problem (3).

Now, let us implement the MHAM for problem (3). To do this, the given function gi (x) is split
into n terms:
gi (x) = si,0 (x) + si,1 (x) + · · · + si,n (x),

and the function ϕi (x, q) is constructed in the form

n
X
ϕi (x, q) = si,0 (x)q 0 + si,1 (x)q 1 + · · · + si,n (x)q n = q j sij (x).
j=0

From (13) and (14), for m = 1, it follows that

ui,1 (x) = ~Ri,1 [ūi,0 (x)] = ~ {Ni [φi (x, q)] − ϕi (x, q)}q=0
( n
" Z
X b
= ~ φi (x, q) − ϕi (x, q) − λ1 k1,ij (x, t)φj (t, q) dt
j=1 a (21)
#)
Z x
+ λ2 k2,ij (x, t)φj (t, q) dt .
a
q=0

Since
∞
" #
X
k
φi (x, q)|q=0 = ui,0 (x) + ui,k (x)q = ui,0 (x) = gi (x),
k=1 q=0
 
n
X
ϕi (x, q)|q=0 = si,0 (x) + q j si,j (x) = si,0 (x),
j=1
q=0

upon substituting these into (21), we get

( n
" Z #)
X b Z x
ui,1 (x) = ~ gi (x) − si,0 (x) − λ1 k1,ij (x, t)gj (t) dt + λ2 k2,ij (x, t)gj (t) dt . (22)
j=1 a a

92
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Let m = 2. Then, from (13) and (14), and taking into account (11), we obtain

∂ 
ui,2 (x) = ui,1 (x) + ~Ri,2 [ūi,1 ] = ui,1 (x) + ~ [Ni [φi (x, q)] − ϕi (x, q)]
∂q q=0
" Z bX n
∂
= ui,1 (x) + ~ φi (x, q) − λ1 k1,ij (x, t)φj (t, q) dt
∂q a j=1
Z xX n
#

− λ2 k2,ij (x, t)φj (t, q) dt − ϕi (x, q)  .

a j=1

q=0

It is known that

∞
 " #
∂  ∂ X
k

φi (x, q) = ui,0 (x) + q ui,k (x)  = ui,1 (x),
 
∂q q=0 ∂q 
k=1 q=0
 
 n 
∂  ∂ X
j

ϕi (x, q)
 =  si,0 (x) + q si,j (x) 
 = si,1 (x).
∂q q=0 ∂q j=1 
q=0

Taking these into account, we have

n
" #
X Z b Z x
ui,2 (x) = (1 + ~) ui,1 (x) − ~si,1 (x) − ~ λ1 k1,ij (x, t)uj,1 (t) dt + λ2 k2,ij (x, t)uj,1 (t) dt .
j=1 a a

By continuing this procedure for any i ∈ {1, . . . , n} and any m ≥ 1, we get

( n
" #)
X Z b Z x
ui,1 (x) = ~ gi (x) − si,0 (x) − λ1 k1,ij (x, t)gj (t) dt + λ2 k2,ij (x, t)gj (t) dt
j=1 a a

n
"
X Z b
ui,m (x) = (1 + ~) ui,m−1 (x) − ~si,m−1 (x) − ~ λ1 k1,ij (x, t)uj,m−1 (t) dt
j=1 a
#
Z x
+ λ2 k2,ij (x, t)uj,m−1 (t) dt , 2 ≤ m ≤ n + 1, (23)
a
n
"
X Z b
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ λ1 k1,ij (x, t)uj,m−1 (t) dt
j=1 a
#
Z x
+ λ2 k2,ij (x, t)uj,m−1 (t) dt , m > n + 1.
a

It should be emphasized P∞that, for m ≥ 1, the families {ui,m (x)} are governed by equations (9)
and (13), and the series m=0 ui,m converge to the exact solutions with different rates based on
the choice of the initial guesses.

93
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

4 Gauss-Legendre Quadrature Formula

The Gauss-Legendre quadrature formula (GLQF) is a higher order quadrature formula given by
Kythe & Schäferkotter [8] in the form

Z 1 n+1
X
f (τ ) dτ = wi f (τi ) + Rn+1 (f ), (24)
−1 i=1

where
2
wi =  0 2 , (25)
(1 − τi2 ) Pn+1 (τi )

Pn+1
with i=1 wi = 2. It is exact for polynomials of degree 2n − 1 if weights wi are defined by (25),
and collocation points τi are chosen as the roots of Legendre polynomial Pn+1 (s), i.e.

Pn+1 (τi ) = 0, (26)

for i ∈ {1, 2, . . . , n + 1}.

The error term of the GLQF (24) is given by

4
22n+3 [(n + 1)!] (2n+2)
Rn+1 (f ) = 3f (ξ),
(2n + 3) [(2n + 2)!]

for −1 < ξ < 1.

The GLQF (24) can be applied to the kernel integral on an interval [a, b], as described by Eshku-
vatov et al. [5]. For Fredholm and Volterra integral we have, respectively,

b n+1
b−a X
Z
K1 (s, t)x(t) dt = W1,k (s)x(t1,k ) + Rn (x),
a 2
k=1
(27)
s n+1
s−a X
Z
K2 (s, t)x(t) dt = W2,k (s)x(t2,k ) + Rn (x),
a 2
k=1

where s ∈ [a, b], and

b−a b+a
W1,k (s) = k1 (s, t1,k )wk , t1,k = rk + ,
2 2 (28)
s−a s+a
W2,k (s) = k2 (s, t2,k )wk , t2,k = rk + ,
2 2

and rk are the roots of the Legendre polynomial defined by (26). Eqs. (27) and (28) are crucial in
computing the kernel integrals.

94
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Using GLQF (27) and (28), we can obtain the standard HAM for m ≥ 2 as follows
n
" n+1 n+1
#
X b−a X x−a X
ui,1 (x) = −~ λ1 W1,kij (x)g(t1,k ) + λ2 W2,kij (x)g(t2,k )
j=1
2 2
k=1 k=1
n
" n+1
X b−a X
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ λ1 W1,kij (x)ui,m−1 (t1,k ) (29)
j=1
2
k=1
n+1
#
x−a X
+ λ2 W2,kij (x)ui,m−1 (t2,k ) , m ≥ 2, i ∈ {1, . . . , n}.
2
k=1

On the other hand, for the MHAM, we have for i ∈ {1, . . . , n},
n
" n+1
X b−a X
ui,1 (x) = ~ [gi (x) − si,0 (x)] − ~ λ1 W1,kij (x)ui,0 (t1,k )
j=1
2
k=1
n+1
#
x−a X
+ λ2 W2,kij (x)ui,0 (t2,k ) ,
2
k=1
n
" n+1
X b−a X
ui,m (x) = (1 + ~)ui,m−1 (x) − ~si,m−1 (x) − ~ λ1 W1,kij (x)ui,m−1 (t1,k )
j=1
2
k=1
# (30)
n+1
x−a X
+ λ2 W2,kij (x)ui,m−1 (t2,k ) , 2 ≤ m ≤ n + 1,
2
k=1
n
" n+1
X b−a X
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ λ1 W1,kij (x)uj,m−1 (t1,k )
j=1
2
k=1
n+1
#
x−a X
+ λ2 W2,kij (x)uj,m−1 (t2,k ) , m > n + 1,
2
k=1

where
b−a b+a
W1,kij (x) = k1,ij (x, t1,k )wk , t1,k = rk + ,
2 2
x−a x+a
W2,k (x) = k2,ij (x, t2,k )wk , t2,k = rk + ,
2 2

and rk are the roots of Legendre polynomials defined by (26).

The MHAM with GLQF, as defined by (29) and (30), is very useful when the kernel integral
has no analytic solution, and it also is highly accurate.

95
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

5 Numerical Experiments

In this section, we present a comparison between the MHAM and the standard HAM, as well as
OHAM.

Example 1 (Ghazandi & Yari [6]): Let us consider the following system of Fredholm integral
equations:
 Z 1 Z 1
1 11 5 2 1 3 3
u1 (x) = 20 − 30 x + 3 x − 3 x + (x − t) u1 (t) dt + (x − t)2 u2 (t) dt,



0 0
Z 1 Z 1
 1 4 23 3 3 2 41 1 4
u2 (x) = − x + x + x − x − + (x − t) u1 (t) dt + (x − t)3 u2 (t) dt.


3 12 20 60 30 0 0

Comparing with (3) we have λ1 = 1, λ2 = 0, and the kernels and functions are
k1,11 (x, t) = (x − t)3 , k1,12 (x, t) = (x − t)2 ,
k1,21 (x, t) = (x − t)4 , k1,22 (x, t) = (x − t)3 ,
1 11 5 1
g1 (x) = − x + x2 − x3 ,
20 30 3 3
1 23 3 41 1
g2 (x) = − x4 + x3 + x2 − x − .
3 12 20 60 30

The exact solutions are (

u1 (x) = x2 ,
u2 (x) = −x + x2 + x3 .

As initial guess let us choose (

u1,0 (x) = g1 (x),
u2,0 (x) = g2 (x).

Then, using eq. (17) of the standard HAM, we obtain the first iteration as follows:
Z 1 Z 1 
u1,1 (x) = −~ (x − t)3 g1 (t) dt + (x − t)2 g2 (t) dt
0 0

61 3 161 2 11 89
= −~ x − x + x− ,
180 240 30 1800
Z 1 Z 1 
u2,1 (x) = −~ (x − t)4 g1 (t) dt + (x − t)3 g2 (t) dt
0 0

61 4 133 3 41 2 1601 209
= −~ x − x + x − x+ .
180 144 48 5040 6300

Using eq. (20) of the standard HAM, the second and third iterations can be computed as follows:
"Z
1
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ k1,i1 (x, t)u1,m−1 (t) dt
0
#
Z 1
+ k1,i2 (x, t)u2,m−1 (t) dt , i ∈ {1, 2}, m ≥ 2.
0

96
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Now, let us solve the same problem with the MHAM. To do this, let us rewrite g1 (x) and g2 (x)
as
   
5 2 11 1 3 1
g1 (x) = x − x + − x + = s1,0 (x) + s1,1 (x),
3 30 3 20
   
23 3 41 1 4 3 2 1
g2 (x) = x − x + − x + x − = s2,0 (x) + s2,1 (x).
12 60 3 20 30

The corresponding functions ϕi (x, q) can be written as

ϕ1 (x, q) = s1,0 (x) + s1,1 (x)q,

ϕ2 (x, q) = s2,0 (x) + s2,1 (x)q.

Since

φi (x, q)|q=0 = ui,0 (x) = gi (x),

ϕi (x, q)|q=0 = si,0 (x),

for i ∈ {1, 2}, then, for m = 1, eq. (22) of the MHAM gives us
 Z 1 Z 1 
3 2
u1,1 (x) = ~ g1 (x) − s1,0 (x) − (x − t) g1 (t) dt + (x − t) g2 (t) dt
0 0
 
121 3 161 2 11 179
= −~ x − x + x− ,
180 240 30 1800
 Z 1 Z 1 
u2,1 (x) = ~ g2 (x) − s2,0 (x) − (x − t)4 g1 (t) dt + (x − t)3 g2 (t) dt
0 0
 
121 4 133 3 169 2 1601 419
= −~ x − x + x − x+ .
180 144 240 5040 6300

Eq. (23) gives any iterations of the MHAM as follows:

"Z #
1 Z 1
3 2
ui,2 (x) = (1 + ~)ui,1 (x) − ~si,1 (x) − ~ (x − t) u1,1 (t) dt + (x − t) u2,1 (t) dt , i ∈ {1, 2},
0 0
"Z #
1 Z 1
3 2
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ (x − t) u1,m−1 (t) dt + (x − t) u2,m−1 (t) dt , i ∈ {1, 2}.
0 0

97
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Tables 1 and 2 summarize results of standard HAM and MHAM, as well as results from other
works.
Table 1: Numerical solutions of Example 1 for ~ = −1 and N = 3. [?] denotes data obtained from [6].

x ku1 − U1,3 k ku1 − U1,3 k ku1 − U1,3 k ku2 − U2,3 k ku2 − U2,3 k ku2 − U2,3 k
HAM MHAM OHAM [?] HAM MHAM OHAM [?]
0.0 1.98 × 10−6 3.10 × 10−4 2.24 × 10−7 1.32 × 10−6 2.09 × 10−4 1.54 × 10−7
−7 −4 −8 −7 −4
0.1 8.48 × 10 2.02 × 10 1.83 × 10 4.24 × 10 1.24 × 10 1.75 × 10−8
0.2 5.95 × 10−8 1.17 × 10−4 8.69 × 10−8 2.78 × 10−8 7.33 × 10−5 9.08 × 10−8
−7 −5 −8 −7 −5
0.3 4.22 × 10 5.28 × 10 3.32 × 10 1.77 × 10 4.66 × 10 8.76 × 10−8
−7 −6 −8 −7 −5
0.4 6.35 × 10 4.68 × 10 9.16 × 10 1.51 × 10 3.69 × 10 3.18 × 10−8
0.5 6.20 × 10−7 3.04 × 10−5 2.36 × 10−7 6.20 × 10−8 3.81 × 10−5 5.04 × 10−8
−7 −5 −7 −9 −5
0.6 4.17 × 10 5.57 × 10 3.49 × 10 6.61 × 10 4.56 × 10 1.30 × 10−7
0.7 6.45 × 10−8 7.48 × 10−5 3.78 × 10−7 6.50 × 10−8 5.60 × 10−5 1.77 × 10−7
−7 −5 −7 −7 −5
0.8 3.97 × 10 9.08 × 10 2.73 × 10 3.02 × 10 6.71 × 10 1.56 × 10−7
−7 −4 −8 −7 −5
0.9 9.28 × 10 1.07 × 10 1.80 × 10 7.65 × 10 7.84 × 10 3.33 × 10−8
1.0 1.49 × 10−6 1.28 × 10−4 5.46 × 10−7 1.49 × 10−6 9.04 × 10−5 2.31 × 10−7

Table 2: Numerical solutions of Example 1 for ~ = −1 and N = 10. [?] denotes data obtained from [13].

x ku1 − U1,10 k ku1 − U1,10 k ku1 − U1,10 k ku2 − U2,10 k ku2 − U2,10 k ku2 − U2,10 k
HAM MHAM HAM [?] HAM MHAM HAM [?]
0.0 1.71×10−17 2.86×10−15 − 3.02×10−18 1.06×10−14 −
−17 −15 −15
0.1 3.82×10 2.52×10 1.11×10 0.000 2.28×10−15 4.09×10−16
−17 −15 −15 −17 −15
0.2 5.55×10 7.34×10 1.28×10 2.78×10 1.92×10 2.43×10−16
0.3 4.16×10−17 1.12×10−14 3.94×10−15 0.000 2.83×10−15 4.86×10−16
−17 −14 −15 −17 −15
0.4 5.55×10 1.38×10 8.33×10 5.55×10 1.55×10 2.16×10−15
0.5 5.55×10−17 1.47×10−14 3.33×10−15 2.78×10−17 6.24×10−16 3.55×10−15
−14 −15 −17 −15
0.6 0.000 1.34×10 9.88×10 8.33×10 2.59×10 5.88×10−15
−15 −15 −17 −15
0.7 0.000 9.55×10 2.28×10 2.78×10 2.64×10 1.66×10−14
0.8 3.33×10−16 3.11×10−15 1.80×10−14 5.55×10−17 9.99×10−16 1.16×10−14
−16 −15 −16 −16 −15
0.9 2.22×10 6.88×10 2.22×10 1.11×10 9.88×10 3.46×10−14
1.0 2.22×10−16 2.05×10−14 1.38×10−14 0.000 2.61×10−14 1.86×10−14

5.1 Convergence Interval for Example 1

The convergence interval can be obtain by imposing the following conditions:


∂ ūiN (x, ~) 
= f (h), h ∈ [a, b], i ∈ {1, 2}. (31)
∂x 
x=0

In particular, for N = 5, (31) becomes


∂ ū15 (x, ~)  153188029 5 1047581 4 55217 3 3691 2 11 11
=− h − h − h − h − h− ,
∂x 
x=0 423360000 576000 15120 1008 6 30

∂ ū25 (x, ~)  16118857583 5 89753269 4 1925171 3 6415 2 1601 41
= h + h + h + h + h− .
∂x 
x=0 50803200000 56448000 604800 2016 1008 60

98
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

The graphs of these functions—called ~-curves—give the convergence region. Figure 1 shows the
~-curves on the interval ~ ∈ [−2.5, 0.5].

Figure 1: The ~-curves for Example 1 (HAM) for N = 5.

Example 2 (Shidfar & Molabahrami [13]): Consider the following system of linear Volterra in-
tegral equations of the second kind:
 Z x
3


 u1 (x) = −x − x + (u2 (t) + u3 (t)) dt,
0


 Z x
1 5 1 4 1 3


2
− − − − [(x − 1)u1 (t) + tu2 (t) − xu4 (t)] dt,

 u
 2
 (x) = x x x 3x 1 +
4 4 2 0
Z x (32)
1 31
u3 (x) = x6 − x3 + 2x2 + 3 + 2
  
 (x − t)u 1 (t) − 3t u 4 (t) dt,
2 6

0



 Z x

u4 (x) = x3 − 5 +

 (2x − 3t)u1 (t) dt.
0

The exact solutions of this system are

(
u1 (x) = x, u2 (x) = x2 − 1,
u3 (x) = 2x2 + 3, u4 (x) = x3 − 5.

Comparing (32) with (3), we have λ1 = 0, λ2 = 1, and

g1 (x) = −x − x3 ,
1 1 1
g2 (x) = x5 − x4 − x3 − 3x2 − 1,
4 4 2
1 6 31 3
g3 (x) = x − x + 2x2 + 3,
2 6
g4 (x) = x3 − 5,

with the kernels

K2,11 (x, t) = 0, K2,12 (x, t) = 1, K2,13 (x, t) = 1, K2,14 (x, t) = 0,

K2,21 (x, t) = x − 1, K2,22 (x, t) = t, K2,23 (x, t) = 0, K2,24 (x, t) = −x,
K2,31 (x, t) = x − t, K2,32 (x, t) = 0, K2,33 (x, t) = 0, K2,34 (x, t) = −3t2 ,
K2,41 (x, t) = 2x − 3t, K2,42 (x, t) = 0, K2,43 (x, t) = 0, K2,44 (x, t) = 0.

99
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Let us choose as initial guess

ui,0 (x) = gi (x), i ∈ {1, 2, 3, 4}.

Then, eqs. (17) and (20) of the standard HAM yields

4 Z
X 1
ui,1 (x) = −~ K2,ij (x, t)gj (t) dt,
j=1 0

and
Z x
ui,m (x) = (1 + ~)ui,m−1 − ~ K2,ij (x, t)uj,m−1 (t) dt,
0

for m ≥ 2 and i ∈ {1, 2, 3, 4}.

To apply eqs. (22) and (23) of the MHAM, let us decompose the functions gi (x) as follows:

g1 (x) = (−x) + (−x3 ) = s1,0 (x) + s1,1 (x),

 
2

1 5 1 4 1 3
g2 (x) = −3x − 1 + x − x − x = s2,0 (x) + s2,1 (x),
4 4 2
   
31 1 6
g3 (x) = − x3 + 2x2 + 3 + x = s3,0 (x) + s3,1 (x),
6 2
g4 (x) = (−5) + (x3 ) = s4,0 (x) + s4,1 (x).

Then, the corresponding functions ϕi (x, t) can be written as

ϕ1 (x, t) = s1,0 (x) + s1,1 (x)q, ϕ2 (x, t) = s2,0 (x) + s2,1 (x)q,
ϕ3 (x, t) = s3,0 (x) + s3,1 (x)q, ϕ4 (x, t) = s4,0 (x) + s4,1 (x)q.

It is easy to see that for i ∈ {1, 2, 3, 4}

φ1 (x, q)|q=0 = ui,0 (x) = gi (x),

ϕ1 (x, q)|q=0 = si,0 (x).

Then, from (23) we obtain

4 Z
!
X x
ui,1 (x) = ~ gi (x) − si,0 (x) − K2,ij (x, t)gj (t) dt ,
j=1 0

4 Z
X x
ui,2 (x) = (1 + ~)ui,1 (x) − ~si,1 (x) − ~ K2,ij (x, t)uj,1 (t) dt,
j=1 0

4 Z
X x
ui,m (x) = (1 + ~)ui,m−1 (x) − ~ K2,ij (x, t)uj,m−1 (t) dt, m ≥ 3.
j=1 0

100
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Tables 3 and 4 summarize results of standard HAM and MHAM, as well as results from other
works.

Table 3: Numerical solutions of u1,10 (x) and u2,10 (x) for Example 2 with ~ = −1 and N = 10. [?] denotes [13].

x ku1 − U1,10 k ku1 − U1,10 k ku1 − U1,10 k ku2 − U2,10 k ku2 − U2,10 k ku2 − U2,10 k
HAM MHAM HAM [?] HAM MHAM HAM [?]
0.0 0.000 0.000 − 0.000 0.000 0.000
0.1 5.55×10−17 2.78×10−17 2.14×10−15 1.11×10−16 1.11×10−16 3.33×10−15
0.2 3.89×10−16 4.16×10−16 2.56×10−14 5.55×10−16 3.33×10−16 2.75×10−14
−14 −14 −14 −14 −14
0.3 2.04×10 2.22×10 3.98×10 1.80×10 1.79×10 1.09×10−14
−13 −13 −13 −13 −13
0.4 2.58×10 2.92×10 3.57×10 2.54×10 2.52×10 2.38×10−13
0.5 1.96×10−12 2.18×10−12 2.45×10−12 1.32×10−12 1.29×10−12 9.09×10−13
−11 −11 −11 −12 −12
0.6 1.27×10 1.35×10 1.30×10 2.30×10 1.78×10 1.21×10−12
0.7 6.34×10−11 6.59×10−11 5.77×10−11 1.04×10−11 1.48×10−11 1.57×10−11
−10 −10 −10 −10 −10
0.8 1.96×10 2.25×10 1.48×10 1.22×10 1.44×10 1.38×10−10
−10 −10 −10 −10 −10
0.9 1.52×10 8.01×10 1.38×10 7.03×10 5.32×10 7.39×10−10
1.0 2.75 × 10−9 6.56 × 10−9 4.19 × 10−9 2.71 × 10−9 2.87 × 10−9 2.72 × 10−9

Table 4: Numerical solutions of u3,10 (x) and u4,10 (x) for Example 2 with ~ = −1 and N = 10. [?] denotes [13].

x ku3 − U3,10 k ku3 − U3,10 k ku3 − U3,10 k ku4 − U4,10 k ku4 − U4,10 k ku4 − U4,10 k
HAM MHAM HAM [?] HAM MHAM HAM [?]
0.0 0.000 0.000 − 0.000 0.000 −
0.1 4.44×10−16 4.44×10−16 4.44×10−16 8.88×10−16 0.000 0.000
−16 −15 −16
0.2 8.88×10 0.000 2.22×10 8.88×10 0.000 0.000
0.3 1.33×10−15 2.22×10−15 4.88×10−15 6.22×10−15 4.44×10−15 1.78×10−15
0.4 6.66×10−14 7.24×10−14 7.82×10−14 1.08×10−13 1.04×10−13 7.28×10−14
−13 −13 −12 −13 −13
0.5 8.87×10 9.93×10 1.02×10 8.90×10 8.40×10 6.38×10−13
−12 −12 −12 −12 −12
0.6 7.56×10 8.56×10 8.67×10 2.71×10 2.13×10 1.29×10−12
0.7 5.54×10−11 6.26×10−11 5.90×10−11 7.91×10−12 1.40×10−11 1.56×10−11
−10 −10 −10 −10 −10
0.8 3.63×10 4.06×10 3.62×10 1.45×10 2.09×10 1.81×10−10
0.9 1.92 × 10−9 1.94 × 10−9 1.83 × 10−9 1.15 × 10−9 1.92 × 10−9 1.28 × 10−9
−9 −9 −9 −9 −8
1.0 8.03 × 10 4.11 × 10 7.25 × 10 7.08 × 10 1.54 × 10 7.52 × 10−9

5.2 Convergence Interval for Example 2

∂
Figure 2 shows the four graphs of ∂x ui,10 (x, h)|x=0 , for i ∈ {1, 2, 3, 4}, in blue, red, green, yellow,
respectively. Note, graphs for red, green and yellow coincide with x-axis and cannot be clearly
observed. From this we can see that convergence interval of the ~-curve is [−8, 6].

101
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

Figure 2: The ~-curve for Example 2 (HAM) for N=10.

6 Conclusions

In this paper, we have developed standard and modified HAM for solving systems of linear Fredholm-
Volterra integral equations by combining Gauss-Legendre quadrature formulas. Numerical re-
sults showed that the developed method is comparable with OHAM in Ghazanfari & Yari [6] and
with HAM developed in Shidfar & Molabahrami [13]. Tables 1–4 revealed that absolute minimum
error is obtained in the most cases of modified HAM. The developed method can be used when
integration problems cannot be solved analytically.

Acknowledgement This work was supported by Universiti Sains Islam Malaysia (USIM) under
RMC Research Grant Scheme (FRGS, 2018), project code is USIM/ FRGS/ FST/ 055002/ 51118,
and under National University of Uzbekistan, project code: OT-F4-28.

Conflicts of Interest The authors declare no conflict of interest.

References

[1] S. Abbasbandy (2007). The application of homotopy analysis method to solve a generalized
Hirota-Satsuma coupled KdV equation. Physics Letters A, 361(6), 478–483.

[2] M. B. Adam & J. A. Tawn (2011). Modification of Pickands’ dependence function for ordered
bivariate extreme value distribution. Communications in Statistics - Theory and Methods, 40(9),
1687–1700.

[3] A. S. Bataineh, M. S. M. Noorani & I. Hashim (2009). Modified homotopy analysis method
for solving systems of second-order BVPs. Communications in Nonlinear Science and Numerical
Simulation, 14(2), 430–442.

[4] M. M. Chawla & M. K. Jain (1968). Error estimates for Gauss quadrature formulas for analytic
functions. Mathematics of Computation, 22(101), 82–90.

[5] Z. K. Eshkuvatov, H. H. Hameed, B. M. Taib & N. M. A. Nik Long (2019). General 2 × 2

system of nonlinear integral equations and its approximate solution. Journal of Computational
and Applied Mathematics, 361, 528–546.

102
Z. K. Eshkuvatov et al. Malaysian J. Math. Sci. 16(1): 87–103 (2022) 87 - 103

[6] B. Ghazanfari & N. Yari (2013). Optimal homotopy asymptotic method for solving system
of Fredholm integral equations. Communications in Numerical Analysis, 2013, Article ID: cna–
00186, 15 pages. https://doi.org/10.5899/2013/cna-00186.

[7] T. Hayat, Z. Abbas, M. Sajid & S. Asghar (2007). The influence of thermal radiation on MHD
flow of a second grade fluid. International Journal of Heat and Mass Transfer, 50(5-6), 931–941.
[8] P. K. Kythe & M. R. Schäferkotter (2004). Handbook of Computational Methods for Integration.
Chapman and Hall/CRC, New York.
[9] S. J. Liao (1992). Proposed Homotopy Analysis Techniques for the Solution of Nonlinear Problems.
PhD thesis, Shanghai Jiaotong University, Shanghai.
[10] S. J. Liao (1995). An approximate solution technique which does not depend upon small
parameters: A special example. International Journal of Non-Linear Mechanics, 30(3), 371–380.
[11] S. J. Liao (1997). A kind of approximate solution technique which does not depend upon
small parameters — II. an application in fluid mechanics. International Journal of Non-Linear
Mechanics, 32(5), 815–822.
[12] S. J. Liao (2005). Comparison between the homotopy analysis method and homotopy per-
turbation method. Applied Mathematics and Computation, 169(2), 1186–1194.
[13] A. Shidfar & A. Molabahrami (2011). Solving a system of integral equations by an analytic
method. Mathematical and Computer Modelling, 54(1-2), 828–835.
[14] M. Turkyilmazoglu (2011). Some issues on HPM and HAM methods: A convergence
scheme. Mathematical and Computer Modelling, 53(9-10), 1929–1936.

103