MATHEMATICAL COMMUNICATIONS 83

Math. Commun. 29(2024), 83–103

Sinc approximation for numerical solutions of

two-dimensional nonlinear Fredholm integral equations
Manochehr Kazemi∗
Department of Mathematics, Ashtian Branch, Islamic Azad University, Ashtian, Iran

Received September 1, 2021; accepted September 3, 2023

Abstract. A new efficient iterative method of successive approximations based on the

fixed point method and sinc quadrature is proposed for two-dimensional nonlinear Fred-
holm integral equations of the second kind (2DNFIEs). We have provided convergence
and error analysis of the suggested method. Besides, numerical stability of the method
concerning the choice of the first iteration is verified.

AMS subject classifications: 47H09, 47H10

Keywords: iterative approach, sinc quadrature, numerical stability, fixed point method

1. Introduction
In this research, a new iterative approach based on the fixed point method and sinc
quadrature is improved for the numerical solution of the (2DNFIEs):
Z d Z b
X(s, t) = f (s, t) + λ H(s, t, x, y)ψ(X(x, y))dxdy, (s, t) ∈ J, (1)
c a

where f and H are known functions on J = [a, b] × [c, d] and J × J, respectively,

X(s, t) is the unknown function to be determined, and λ is a constant. This type
of equations is applied in various fields of electromagnetism, plasma physics, tele-
graph equations, electrical engineering, queueing theory, economics, population, im-
age processing, reformulation of boundary value problems, and in many other areas
[13, 42, 31, 10, 30, 43]. Many different methods are usually used to solve Eq. (1) such
as the collocations and Galerkin methods [14, 9, 2], the Haar wavelet [17, 16], the
Bernoulli operational matrix method [6], triangular functions (TFs)[23, 12], Legen-
dre wavelets [18], the hybrid function method [15, 39], Bernoulli polynomials [5], the
reduced differential transform method [44], radial basis functions [32], the B-spline
collocation method [28], the hat basis function [33], block-pulse functions [38], ra-
tionalized Haar functions [3], the Bernstein polynomials [34], operational matrices
[29], and the degenerate kernel method [4].
In the approaches suggested above, the integral equation is converted into a
system of nonlinear algebraic equations which has to be accomplished with iterative
methods. It is cumbersome to solve these systems, or the solution may be unreliable.
∗ Corresponding author. Email address: univer ka@yahoo.com, m.kazemi@aiau.ac.ir (M. Kazemi)

https://www.mathos.unios.hr/mc
c 2024 School of Applied Mathematics and Informatics, University of Osijek
84 M. Kazemi

To reduce this problem, we attempt to provide a numerical method to approximate

a solution for Eq. (1) using the fixed point method and sinc quadrature. The sinc
method is a robust numerical tool for finding fast and accurate solutions in different
fields of problems [25, 40]. Numerical methods have been given for solving integral
equations including successive approximations methods based on quadrature rules
[7, 8, 20, 21, 22, 27]. The existence and uniqueness result quoted for Eq. (1) is
varied see [1, 11, 19, 26, 35]. The outline of the paper is as follows. In Section 2,
we assert basic theorems and properties of the sinc function, which are referred to
in the ensuing sections. In Section 3, the convergence of the method of successive
approximations is studied. Numerical stability of this method is proved in Section 4.
Numerical results are reported in Section 5 which confirm that the implementation
of the method is considerably fast and highly accurate.

2. Mathematical preliminaries

2.1. Modulus of continuity

Definition 1. Let g : J → R be a bounded function, the oscillation of g on J is the
quantity

ωJ (g, δ) = sup{|g(x, y) − g(x0 , y 0 )|; x, x0 ∈ [a, b];

p
y, y 0 ∈ [c, d]; ; (x − x0 )2 + (y − y 0 )2 ≤ δ}.


Furthermore, ωJ (g, δ) is called a uniform modulus of continuity of g if g ∈ C J .

Theorem 1. The following properties hold:

p
(i) |g(x, y) − g(x0 , y 0 )| ≤ ωJ (g, (x − x0 )2 + (y − y 0 )2 ) for all x, x0 ∈ [a, b] and
y, y 0 ∈ [c, d],

(ii) ωJ (g, δ) is a non-decreasing function in δ,

(iii) ωJ (g, 0) = 0,

(iv) ωJ (g, δ1 + δ2 ) ≤ ωJ (g, δ1 ) + ωJ (g, δ2 ) for any δ1 , δ2 ≥ 0,

(v) ωJ (g, nδ) ≤ nωJ (g, δ) for n ∈ N and any δ ≥ 0,

(vi) ωJ (g, αδ) ≤ (α + 1)ωJ (g, δ) for any δ, α ≥ 0,

(vii) ωJ (g, ·) is continuous at 0 iff g ∈ C(J),

(viii) If J ⊆ J 0 , then ωJ (g, δ) ≤ ωJ 0 (g, δ) for all δ ≥ 0.

Proof. See [24].

 Sinc approximation for solving two-dimensional integral equations 85

Theorem 2. Suppose that g : J → R is an integrable, bounded function. For each

a = x0 < x1 < ... < xn = b , c = y0 < y1 < ... < yn = d and each κi ∈ [xi−1 , xi ] ,
νj ∈ [yj−1 , yj ], we have
Z d Z b n X
X n
g(s, t)dsdt − (xi − xi−1 )(yj − yj−1 )g(κi , νj )
c a j=1 i=1
n X
X n q
≤ (xi − xi−1 )(yj − yj−1 )ω[xi−1 ,xi ]×[yj−1 ,yj ] (g, (xi − xi−1 )2 + (yj − yj−1 )2 ).
j=1 i=1

Proof. From Theorem 1, we have

Z dZ b n X
X n
g(s, t)dsdt − (xi − xi−1 )(yj − yj−1 )g(κi , νj )
c a j=1 i=1
n X
X n Z yj Z xi
≤ |g(s, t) − g(κi , νj )| dsdt
j=1 i=1 yj−1 xi−1

Xn X n q
≤ (xi − xi−1 )(yj − yj−1 )ω[xi−1 ,xi ]×[yj−1 ,yj ] (g, (xi − xi−1 )2 + (yj − yj−1 )2 ).
j=1 i=1

2.2. Sinc function

Definition 2 (see [37]). Let g be a function defined for all x on (∞, ∞) and h > 0.
The Whittaker cardinal is written as
∞
X
C(g, h)(x) = g(ih)S(i, h)(x),
−∞

whenever this series in (3) converges, and S(i, h)(x) are the sinc functions defined
by
sin[π( x−ih
h )]
S(i, h)(x) = x−ih
π( h )
Definition 3 (see [25]). Let h > 0. A(h) is the family of entire functions g such
that on the real line g ∈ L2 (R) and in the complex plane, g is of exponential type πh ,
i.e.,
π|z|
|g(z)| ≤ K exp( )
h
for ∃ K > 0.
Theorem 3 (see [25]). If g ∈ A(h), then for all z ∈ C
∞
X z − ih
g(z) = g(ih)sinc( ). (2)
i=−∞
h
86 M. Kazemi

Definition 4 (see [37]). A function f is said to decay double exponentially if ∃ α, K

so that

|f (t)| ≤ K exp(−αe|t| ), t ∈ R,

or a function g is said to decay double exponentially regarding the conformal map φ,

if ∃ α, K so that

|g(φ(t))φ0 (t)| ≤ K exp(−αe|t| ), t ∈ R, (3)

where φ is called the DE transformation. By [37, 36], we have the DE formula

for the definite integration of a function f :
Z b N
X −2πdN
f (x)dx = h f (ϕ(ih))ϕ0 (ih)) + O(e log(2πdN/α) ),
a i=−N

log( 2πdN )
with h = N
α
and
b−a π b+a
x = ϕ(t) = tanh( sinh(t)) + ,
2 2 2
b−a π/2 cosh(t)
ϕ0 (t) = , t∈R
2 cosh2 (π/2 sinh(t))

Corollary 1. If g ∈ A(h) and g ∈ L1 (R), then

X∞ Z ∞
h g(ih) = g(t)dt.
i=−∞ −∞

Proof. Replace z by t ∈ R in (2) and integrate the result over R (see [25]).
PN
Theorem 4. Let SN = h i=−N ϕ0 (ih). Then

lim SN = (b − a).
N →∞

Proof. By Corollary (1), we have

∞
X
lim SN = h ϕ0 (ih)
N →∞
i=−∞
Z ∞
0
= ϕ (t)dt
−∞
Z x
b−a π cosh(t)
= lim 2 dt
x→∞ −x 4 cosh (π/2 sinh(t))
b − a 4π sinh(π/2 sinh(x))
= lim
x→∞ 4 π cosh(π/2 sinh(x))
= b − a.
 Sinc approximation for solving two-dimensional integral equations 87

If g(x, y) decays double exponentially regarding

b−a π b+a
ϕ1 (s) = tanh( sinh(s)) +
2 2 2
and
d−c π d+c
ϕ2 (t) = tanh( sinh(t)) + ,
2 2 2
then the function g can be expanded as follows (see [37]):
∞
X ∞
X
g(ϕ1 (s), ϕ2 (t))ϕ0 (s)ϕ0 (t) = g(ϕ1 (ih1 ), ϕ2 (jh2 ))ϕ01 (ih1 )ϕ02 (jh2 )
i=−∞ j=−∞
t s
× sinc( − i)sinc( − j) + E(t, s, h1 , h2 ),
h1 h2

where h1 = log(2πdN1 N/α1 ) and h2 = log(2πdM2 N/α2 ) .

Integrating this expression regarding x and y, we have
Z b Z d Z ∞ Z ∞
g(x, y)dxdy = g(ϕ1 (s), ϕ2 (t))ϕ01 (s)ϕ02 (t)dsdt.
a c −∞ −∞

Further,
Z b Z d N
X M
X
g(x, y)dxdy =h1 h2 g(ϕ1 (ih1 ), ϕ2 (jh2 ))ϕ01 (ih1 )ϕ02 (jh2 )
a c i=−N j=−M
−2πd1 −2πd2
+ O(exp( )) + O(exp( )). (4)
h1 h2

3. Main results
3.1. The sequence of successive approximations
In the sequel, we shall prove the existence and uniqueness of the solution to Eq.
(1) by the Banach contraction principle. Consider Eq. (1) under the following
conditions:
a◦ . f ∈ C(J, R), ψ ∈ C(R, R) and H ∈ C(J × R, R),
b◦ . there exists ρ ≥ 0, such that |ψ(P1 ) − ψ(P2 )| ≤ ρ|P1 − P2 |, ∀P1 , P2 ∈ C(J).
c◦ . η = ρλNH (b − a)(d − c) < 1, where NH = max{ H(s, t, x, y) ; s, x ∈ [a, b], t, y ∈
[c, d], }, according to the continuity of H.
Now, we define the operator Γ : X → X by
Z d Z b
Γ(X)(s, t) := f (s, t) + λ H(s, t, x, y)ψ(X(x, y))dxdy, (s, t) ∈ J.
c a
88 M. Kazemi

Theorem 5. Under the above assumptions, Eq. (1) has a unique solution X ∗ ∈
C(J). Moreover, for any arbitrary choice of initial point X0 ∈ C(J), the Picard
sequence defined by

Xk = Γ(Xk−1 ), k≥1

with the initial value X0 := f0 converges with respect to the uniform norm to the
solution X(s, t) of the integral equation (1) and we have

ηk
X ∗ − Xk ku ≤ kX0 − Γ(X0 )ku , (5)
1−η
η
X ∗ − Xk ku ≤ kΓ(Xk−2 ) − Γ(Xk−1 )ku , (6)
1−η
hold for any k ∈ N. Furthermore, choosing X0 = f , (5) becomes

η k+1
X ∗ − Xk ku ≤ N0 , (7)
ρ(1 − η)
where
N0 = sup{|ψ(f (s, t))|; s ∈ [a, b], t ∈ [c, d]}.
Proof. For any X ∈ C(J) we have
 p 
ωJ (Γ(X), δ) ≤ sup f (s1 , t1 ) − f (s2 , t2 ) : (s2 − s1 )2 + (t2 − t1 )2 ≤ δ
(si ,ti )∈J, i=1,2
 Z dZ b
+ sup |λ H(s1 , t1 , x, y)ψ(X(x, y))dxdy
(si ,ti )∈J, i=1,2 c a
Z dZ b p 
−λ H(s2 , t2 , x, y)ψ(X(x, y))dxdy |: (s2 − s1 )2 + (t2 − t1 )2 ≤ δ
c a
Z dZ b
≤ ωJ (f, δ) + λ ωst H, δ)|ψ(X(x, y))|dxdy,
c a

where
 p 
ωst H, δ) = sup H(s1 , t1 , x, y) − H(s2 , t2 , x, y) ; (s2 − s1 )2 + (t2 − t1 )2 ≤ δ .
(xi ,yi )∈J,
i=1,2

Take the limit as δ → 0. Then by Theorem 1-vii, we have ωJ (Γ(X), δ) → 0, so Γ

maps C(J) onto itself.
Furthermore, for X, Y ∈ C(J), we have
Z d Z b
|Γ(X)(s, t) − Γ(Y )(s, t)| = λ K(s, t, x, y)[ψ(X(x, y)) − ψ(Y (x, y))]dxdy
c a
Z d Z b
≤λ |H(s, t, x, y)||ψ(X(x, y)) − ψ(Y (x, y))|dt
c a
≤ ρλNH (b − a)(d − c) k X − Y ku ,
 Sinc approximation for solving two-dimensional integral equations 89

for all (s, t) ∈ J. Thus,

k Γ(X) − Γ(Y ) ku ≤ η k X − Y ku .

By virtue of the Banach contraction principle and crucial condition c◦ , we infer that
Eq. (1) has a unique solution and the same Banach’s fixed point principle leads to
estimates (5) and (6).

Now, we introduce a numerical scheme to solve Eq. (1). Applying formula (4)
to Eq. (1), we have

X 0 (s, t) = r(s, t),

N
X M
X
X k (s, t) = r(s, t) + λh1 h2 H(s, t, ϕ1 (ih1 ), ϕ2 (jh2 ))ϕ01 (ih1 )ϕ02 (jh2 ) (8)
i=−N j=−M

× ψ(X̄k−1 (si , tj )).

To determine the unknown values X k (s, t), choosing the sinc points sp = ϕ1 (ph1 )
and tq = ϕ2 (qh2 ) as collocation points, we get

X 0 (sp , tq ) = r(sp , tq ),
N
X M
X
X k (sp , tq ) = r(sp , tq ) + λh1 h2 H(sp , tq , ϕ1 (ih1 ), ϕ2 (jh2 )) (9)
i=−N j=−M

× ψ(X k−1 (ϕ1 (ih1 ), ϕ2 (jh2 ))ϕ01 (ih1 )ϕ02 (jh2 )).

3.2. Error analysis

Theorem 6. Assume the conditions of Theorem 5 hold. Then the recurrence relation
(8) converges to the unique solution X ∗ of Eq. (1) and the following error estimate
holds:

η k+1 η θη 2
kX ∗ − X̄k k ≤ N0 + ωJ (f, δ) + ωst (H, δ)
ρ(1 − η) 1−η ρNH (1 − η)
ςη
+ ωxy H, δ),
ρNH (1 − η)

where 
 Nk = sup |ψ(X̄k (s, t))|,


 (s,t)∈J

 Γk = sup |ψ(Xk (s, t))|,


(s,t)∈J
(10)

 ς = max {Ni },


 i=0,k−1
 θ = max {Γi }.


i=0,k−2
90 M. Kazemi

Proof. By (8), ∀(s, t) ∈ J, we have

X 1 (s, t) − X 1 (s, t)
Z bZ d
≤λ H(s, t, x, y)ψ(X0 (x, y))dxdy
a c
N
X M
X
− λh1 h2 H(s, t, xi , yj )ϕ01 (ih1 )ϕ02 (jh2 )ψ(X̄0 (xi , yj ))
i=−N j=−M
Z b Z d
≤λ H(s, t, x, y)ψ(X0 (x, y))dxdy
a c
N
X M
X
− λh1 h2 H(s, t, x, y)ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))
i=−N j=−M
N
X M
X
+ λh1 h2 H(s, t, x, y)ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))
i=−N j=−M
N
X M
X
− λh1 h2 H(s, t, xi , yj )ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))
i=−N j=−M
N
X M
X
+ λh1 h2 H(s, t, xi , yj )ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))
i=−N j=−M
N
X M
X
− λh1 h2 H(s, t, xi , yj )ϕ01 (ih1 )ϕ02 (jh2 )ψ(X̄0 (xi , yj ))
i=−N j=−M
Z b Z d
≤λNH ψ(X0 (x, y))dxdy
a c
XN M
X
− λh1 h2 ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))|
i=−N j=−M
N
X M
X
+ λh1 h2 |H(s, t, x, y) − H(s, t, xi , yj )|ϕ01 (ih1 )ϕ02 (jh2 )ψ(X0 (xi , yj ))|
i=−N j=−M
N
X M
X
+ λh1 h2 |H(s, t, xi , yj )||ϕ01 (ih1 )ϕ02 (jh2 )|ψ(X0 (xi , yj ))−ψ(X̄0 (xi , yj ))|.
i=−N j=−M

So,
Z b Z d
X1 (s, t) − X 1 (s, t) ≤λNH | ψ(X0 (x, y))dxdy
a c
∞
X ∞
X
− (xi+1 − xi )(yi+1 − yi )ψ(X0 (xi , yj ))|
i=−∞ j=−∞
 Sinc approximation for solving two-dimensional integral equations 91

N
X M
X
+ N0 λωxy H, δ)h1 h2 ϕ01 (ih1 )ϕ02 (jh2 )
i=−N j=−M
N
X M
X
+ NH λh1 h2 ϕ01 (ih1 )ϕ02 (jh2 )|ψ(X0 (xi , yj )) − ψ(X̄0 (xi , yj ))|,
i=−N j=−M

where
 p 
ωxy H, δ) = sup H(s, t, x1 , y1 ) − H(s, t, x2 , y2 ) ; (x2 − x1 )2 + (y2 − y1 )2 ≤ δ .
(xi ,yi )∈J,
i=1,2

Based on Theorem 4, we have

∞
X ∞
X
ϕ01 (ih1 )ϕ02 (jh2 ) = (b − a)(d − c).
i=−∞ j=−∞

Furthermore, if we use the differential mean value theorem and Theorem 2, we have
si+1 − si = ϕ(κi )h1 ≤ b − a and ti+1 − ti = ϕ(νj )h2 ≤ d − c. So, we see that

X 1 (s, t) − X 1 (s, t)
∞
X ∞
X
≤λNH (xi+1 − xi )(yi+1 − yi )
i=−∞ j=−∞
p
× ω[xi ,xi+1 ]×[yi ,yi+1 ] (ψ(X0 ), ((xi+1 − xi )2 + (yi+1 − yi )2 )
+ N0 λ(b − a)(d − c)ωxy H, δ) + NH λ(b − a)(d − c)|ψ(f (xi , yj )) − ψ(f (xi , yj ))
r
(b − a)2 h1 + (d − c)2 h2
 
≤λNH (b − a)(d − c)ωJ ψ(f ),
16
+ N0 λ(b − a)(d − c)ωxy H, δ).

Now, for k = 2 and by condition (b◦ ), it follows that

r
(b − a)2 h1 + (d − c)2 h2
 
X2 (s, t) − X 2 (s, t) ≤λNH (b − a)(d − c)ωJ ψ(X1 ),
16
+ N1 λ(b − a)(d − c)ωxy H, δ)
+ ρNH λ(b − a)(d − c)kX1 (xi , yj ) − X̄1 (xi , yj )ku .

By induction for k ≥ 3, using condition (b◦ ), (6) and (9), we see that


XK (s, t) − X K (s, t) ≤λNH (b − a)(d − c)ωJ ψ(Xk−1 ), δ
+ Nk−1 λ(b − a)(d − c)ωxy H, δ)
+ ρNH λ(b − a)(d − c)kXK−1 − X̄K−1 k
η
η
≤ ωJ ψ(Xk−1 ), δ + Nk−1 ωxy H, δ) + ηkXK−1 − X̄K−1 ku ,
ρ ρNH
92 M. Kazemi
q
(b−a)2 h1 +(d−c)2 h2 b−a
√ d−c
√
where δ = 16 ≤ 4 h1 + 4 h2 . Taking the supremum for
(s, t) ∈ J, we have
η
η
kXk − X k ku ≤ ωJ ψ(Xk−1 ), δ + Nk−1 ωxy H, δ) + ηkXK−1 − X̄K−1 ku
ρ ρNH
η
η
Xk−1 − X k−1 u
≤ ωJ ψ(Xk−1 ), δ + Nk−2 ωxy H, δ) + ηkXK−2 − X̄K−2 ku
ρ ρNH
..
.
η
η
X2 − X 2 u
≤ ωJ ψ(X1 ), δ + N1 ωxy H, δ) + ηkX1 − X̄1 ku
ρ ρNH
η
η
X1 − X 1 u
≤ ωJ ψ(f ), δ + N0 ωxy H, δ) + ηkX0 − X̄0 ku ,
ρ ρNH
and therefore,
η
kXk − X k k ≤ ωJ (ψ(Xk−1 ), δ) + ηωJ (ψ(Xk−2 ), δ)
ρ

+ ... + η k−1 ωJ (ψ(f ), δ)

ωxy H, δ)η 2 k−1
+ Nk−1 + ηMk−2 + η Mk−3 + ... + η N0 . (11)
ρNH
p
Since for (s, t), (s0 , t0 ) ∈ J with (s − s0 )2 + (t − t0 )2 ≤ δ, we have

ψ(Xk (s, t)) − ψ(Xk (s0 , t0 )) ≤ρ Xk (s, t) − Xk (s0 , t0 )

Z dZ b
= f (s, t) + λ H(s, t, x, y)ψ(Xk−1 (x, y))dxdy
c a
Z d Z b
− f (s0 , t0 ) + λ H(s0 , t0 , x, y)ψ(Xk−1 (x, y))dxdy
c a
0 0
≤ρ f (s, t) − f (s , t )
Z dZ b
+ ρλ H(s, t, x, y)−H(s0 , t0 , x, y) ψ(Xk−1 (x, y)) dxdy
c a

≤ρ f (s, t) − f (s0 , t0 ) + ρλ(b − a)(d − c)ωst (H, δ)Γk−1 .

Consequently, we have
η
ωJ (ψ(Xm ), δ) ≤ ρωJ (f, δ) + ωst (H, δ)Γk−1 . (12)
NH
From (12) and (11), we get

kXk − X k ku ≤η 1 + η + η 2 + ... + η k−1 ωJ (f, δ)

 
η
+ ωst (H, δ) ηΓk−2 + η 2 Γk−3 + ... + B k Γ0
ρNH

ωxy H, δ)η
+ Nk−1 + ηMk−2 + η 2 Mk−3 + ... + η k−1 N0 .
ρNH
 Sinc approximation for solving two-dimensional integral equations 93

Since B < 1, by (10) we obtain

1 − ηk
 
kXk − X k ku ≤η ωJ (f, δ)
1−η
θη η(1 − η k−1 ) ςη (1 − η k )
+ ωst (H, δ) + ωxy H, δ).
ρNH 1−η ρNH 1 − η

Therefore
η
kXk − X k ku ≤ ωJ (f, δ)
(1 − η)
θη 2 ςη
+ ωst (H, δ) + ωxy H, δ). (13)
ρNH (1 − η) ρNH (1 − η)

Notice that η < 1. Thus from inequalities (7) and kX ∗ − X k ku ≤ kX ∗ − Xk ku +

kXk − X k ku , it is concluded that X k converges uniformly to X ∗ and the desired
error estimate is obtained.

4. Stability analysis
In order to study numerical stability of the computed values regarding small changes
in the initial iteration, we consider another initial iteration term Y0 (s, t) = g(s, t) ∈
C(J, R) such that ∃ε > 0, for which Y0 (s, t)−X0 (s, t) < ε, ∀t, s ∈ J. With Y0 (s, t) =
g(s, t), the new sequence {Yk (s, t)}∞
k=1 is:

Z d Z b
Yk (s, t) = g(s, t) + λ H(s, t, x, y)ψ(Yk−1 (x, y))dxdy, (s, t) ∈ J.
c a

Similarly to (9), we get

Ȳ0 (s, t) =g(s, t),

N
X M
X
Y k (s, t) =g(s, t) + λh1 h2 H(s, t, ϕ1 (ih1 ), ϕ2 (jh2 ))ϕ01 (ih1 )ϕ02 (jh2 )
i=−N j=−M

× ψ(Ȳk−1 (si , tj )).

Definition 5. The proposed numerical method is numerically stable regarding the

selection of the initial iteration if ∃ β1 , β2 > 0, which are independent of the stepsize
δ, and a continuous function η : (0, d] → [0, ∞) with limδ→0 η(δ) = 0 such that:

k X̄k − Ȳk k< β1 ε + β2 η(δ), k ∈ N ∪ {0},

p p
where d = b−a
4 log(2πd1 /α1 ) + d−c
4 log(2πd2 /α2 ).

Theorem 7. Under the assumptions of Theorem 6, scheme (9) is numerically stable

regarding the selection of the initial iteration.
94 M. Kazemi

Proof. In order to get numerical stability, we reintroduce the proof of Theorem 6

and we have  0
 Hk = sup |ψ(Ȳk (s, t))|,


 (s,t)∈J
 0
 Γk = sup |ψ(Yk (s, t))|,


(s,t)∈J

 ς 0 = max {Hi0 },

 i=0,k−1
0
max {Γ0i }.



 θ =
i=0,k−2

Similarly to (13), we get

η
kYk − Y k ku ≤ ωJ (g, δ)
(1 − η)
θ0 η2 ς 0η
+ ωst (H, δ) + ωxy H, δ).
ρNH (1 − η) ρNH (1 − η)

Using the triangle inequality, we obtain

kX k − Y k ku ≤kX k − Xk ku + kXk − Yk ku + kYk − Y k ku

η
≤kXk − Yk ku + ωJ (f, δ)
(1 − η)
θη 2 ςη
+ ωst (H, δ) + ωxy H, δ),
ρNH (1 − η) ρNH (1 − η)
η
+ ωJ (g, δ)
(1 − η)
θ0 η2 ς 0η
+ ωst (H, δ) + ωxy H, δ).
ρNH (1 − η) ρNH (1 − η)

Since
| X0 (s, t) − Y0 (s, t) |< ε, ∀(s, t) ∈ J,

we have
Z d Z b
X1 (s, t) − Y1 (s, t) ≤ X0 (s, t) + λ H(s, t, x, y)ψ(X0 (x, y))dxdy
c a
Z d Z b

− Y0 (s, t) − λ H(s, t, x, y)Ψ Y0 (x, y) dxdy
c a
Z d Z b
<ε + ρλNH | X0 (s, t) − Y0 (s, t) | dxdy
c a
<ε + ρλNH (b − a)(d − c)ε.

By induction, we get

1
kXk (s, t) − Yk (s, t)ku < ε,
1−η
 Sinc approximation for solving two-dimensional integral equations 95

for all (s, t) ∈ J and k ≥ 0. Then


1 η

kX k − Y k ku ≤ ε+ ωJ (f, δ) + ωJ (g, δ)
1−η 1−η
0
ς + ς0

(θ + θ )η
+ ωst (H, δ) + ωxy H, δ) ,
ρNH ρNH
where
1 η
β1 = , β2 = ,
1−η 1−η
and
(θ + θ0 )η ς + ς0
η(δ) = ωJ (f, δ) + ωJ (g, δ) + ωst (H, δ) + ωxy H, δ).
ρNH ρNH

Remark 1. Because η < 1, we see that

lim Xk − Y k u
= 0.
δ,ε→0

0
Remark 2 (Stopping criterion). For given ε > 0, the first k ∈ N is determined,
for which 0
| X̄k (sp , tq ) − X̄k−1 (sp , tq ) |< ε ,
and we end with this k by employing the approximations X̄k (s, t) of solution. Using
the triangle inequality, we see that

kX ∗ − X̄k ku ≤kX ∗ − Xk ku + kXk − X̄k ku

η η
≤ kXk − Xk−1 ku + ωJ (f, δ)
1−η 1−η
θη 2 ςη
+ ωst (H, δ) + ωxy H, δ),
ρNH (1 − η) ρNH (1 − η)
and

kXk − Xk−1 ku ≤kXk − X̄k ku + kX̄k − X̄k−1 ku + kX̄k−1 − Xk−1 ku

2η 2θη 2 2ςη
≤ ωJ (f, δ) + ωst (H, δ) + ωxy H, δ)
1−η ρNH (1 − η) ρNH (1 − η)
+ kX̄k − X̄k−1 ku .

Consequently,

η 2η 2
kX ∗ − X̄k ku ≤ kX̄k − X̄k−1 ku + ωJ (f, δ)
1−η (1 − η)2
2θη 3 2ςη 2
+ 2
ωst (H, δ) + ,
ρNH (1 − η) ρNH (1 − η)2
96 M. Kazemi

and therefore, in order to get | X ∗ (sp , tq ) − X̄k (sp , tq ) |< ε we require

2η 2 2θη 3 2ςη 2 3
2
ωJ (f, δ) + 2
ωst (H, δ) + = χ(δ) < ε, (14)
(1 − η) ρNH (1 − η) ρNH (1 − η)2 4
and
η ε
kX̄k − X̄k−1 ku < .
1−η 4
We can select the least N, M ∈ N, for which inequality (14) is kept. Lastly, we find
the lowest k ∈ N for which
ε 1−η 0
kX̄k − X̄k−1 k < . =ε.
4 η
0
With these, | X̄k (sp , tq ) − X̄k−1 (sp , tq ) |< ε leads to | X ∗ (sp , tq ) − X̄k (sp , tq ) |< ε,
and the desired accuracy ε is achieved.
Here, we propose an algorithm of numerical successive approximations to carry
out the proposed method.

The iterative algorithm

0
Input : a, b, c, d, λ, ε , N, M and the functions H, f .
Step 1: Put h1 = log(2πN N
)
, h2 = log(2πM
M
)
and X̄0 (s, t) := f0 (s, t).
Step 2: Compute X̄k (sp , tq ), p = −N, ..., N, q = −M, ..., M by (9).
Step 3: Compute T := max{| x̄k (sp , tq ) − x̄k−1 (sp , tq ) |},
p = −N, ..., N, q = −M, ..., M by (9).
0
Step 4: If T < ε , print k and print x̄k (sp , tq ),
p = −N, ..., N, q = −M, ..., M , STOP.
Otherwise, put k:= k + 1 and go to Step 2.

5. Numerical experiments
In the section, en (s, t) =| X ∗ (sp , tq ) − X̄k (sp , tq ) | is a pointwise error function in
(s, t) ∈ I and ken k∞ is a maximum absolute error, i.e., ken k∞ := max{| X ∗ (sp , tq ) −
X̄k (sp , tq ) | p = −N, ..., N, q = −M, ..., M }. Denote
 (i) 
ken k∞
ρn = log2 (i+1)
,
ken k∞
where X ∗ and X̄k are the exact and the approximate solution of Eq. (1), respectively.
(i)
In the above formula, ken k∞ denotes ken k∞ in the ith column of the tables.
We consider sinc grid points as:
S = {s−N , ..., s0 , ..., sN }
T = {t−M , ..., t0 , ..., tM }.
 Sinc approximation for solving two-dimensional integral equations 97

Let [a, b] × [c, d] = [0, 1] × [0, 1], λ = 1, and let ϕ1 = ϕ2 = ϕ be a conformal map
defined as:
1 π 1
s = ϕ(x) = tanh( sinh(x)) + , sp = ϕ(ph), p = −N, ..., N
2 2 2
1 π 1
t = ϕ(y) = tanh( sinh(y)) + , tq = ϕ(qh), q = −M, ..., M.
2 2 2
π π
Let us take α1 = α2 = 2 and d1 = d2 = 2, N = M for all examples. Then the step
size of sinc is given as:

log(2πN )
h1 = h2 = h = .
N
All computations were carried out with MAPLE 17 on a PC with 2.13 GHz frequency,
Intel Pentium processor with 2 GB RAM.

Example 1. Consider
1 1
stxy 3
Z Z
2 0.114
X(s, t) = s + t − + sin(X(x, y))dxdy,
(s + 1)(t2 + 1) 0 0 (s + 1)(t2 + 1)
(s, t) ∈ [0, 1] × [0, 1].

The exact solution is given by

X(s, t) = s2 + t.
0
Applying the proposed iterative method for N = M = 25, ε = 10−20 , we get
k = 11 iterations and in order to confirm the numerical stability, in the seventh
column, we include the differences between the effective computed values dp,q =
|X̄11 (sp , tq ) − Ȳ11 (sp , tq )|, p, q = −25, ..., 25, where the change in the first term of
the Picard sequence is 0.1 (f (s, t) := f (s, t) + 0.1). For N ∈ {5, 10, 15, 20, 25} and
0
ε = 10−15 , get k = 9. The results in Table 1 confirm the convergence of the algo-
rithm, that is, ken k∞ → 0 when h → 0.
98 M. Kazemi

(s, t) N=5 N=10 N=15 N=20 N=25 dp,q

(0.1,0.1) 1.668E-6 5.177E-8 1.328E-9 3.533E-11 3.944E-12 0.1000
(0.2,0.2) 3.059E-6 9.491E-8 2.434E-9 6.477E-11 7.230E-12 0.1000
(0.3,0.3) 4.235E-6 1.314E-7 3.370E-9 8.968E-11 1.001E-11 0.1001
(0.4,0.4) 5.244E-6 1.627E-7 4.173E-9 1.111E-10 1.239E-11 0.1001
(0.5,0.5) 6.118E-6 1.898E-7 4.868E-9 1.295E-10 1.446E-11 0.1001
(0.6,0.6) 6.882E-6 2.135E-7 5.477E-9 1.457E-10 1.627E-11 0.1001
(0.7,0.7) 7.557E-6 2.345E-7 6.014E-9 1.600E-10 1.786E-11 0.1011
(0.8,0.8) 8.157E-6 2.531E-7 6.491E-9 1.727E-10 1.928E-11 0.1011
(0.9,0.9) 8.693E-6 2.697E-7 6.918E-9 1.841E-10 2.055E-11 0.1021
ken k∞ 9.176E-6 2.847E-7 7.302E-9 1.943E-10 2.169E-11 -
ρn - 5.010 5.285 5.232 3.163 -
k 8 9 9 9 9 -
Time (s) 0.42 2.2 9.29 52.93 497.59 -
Table 1: Numerical results in Example 1.

Example 2. Consider [3]

Z 1 Z 1
X(s, t) = f (s, t) + H(s, t, x, y)ψ(X(x, y))dxdy, (s, t) ∈ [0, 1] × [0, 1],
0 0

where
1 s
f (s, t) = 2
− ,
(1 + s + t) 6(1 + t)
s
H(s, t, x, y) = (1 + y + x),
1+t
ψ(β) = β 2 ,

the exact solution is given by

1
X ∗ (s, t) = .
(1 + s + t)2

Tables 2-3 show numerical results for different numbers of meshes, and the proposed
method solutions are compared with results in [3]. Since they are from Table 3, the
errors of the proposed method are much smaller than the errors described in [3] at
the chosen points. Hence, our method has a moderate convergence rate.
 Sinc approximation for solving two-dimensional integral equations 99

(s, t) = ( 21l , 1
2l
) N=4 N=8 N=16 N=32 dp,q
1 2.319E-4 6.882E-6 2.135E-7 4.667E-10 0.1000
2 2.061E-4 6.117E-6 1.898E-7 4.149E-10 0.1000
3 1.767E-4 5.244E-6 1.627E-7 3.556E-10 0.1000
4 1.427E-4 4.235E-6 1.314E-7 2.872E-10 0.1001
5 1.031E-4 3.059E-6 9.491E-8 2.074E-10 0.1001
6 5.622E-5 1.668E-6 5.177E-8 1.131E-10 0.1011
ken k∞ 3.092E-4 9.176E-6 2.847E-7 6.223E-10 -
ρn - 5.07 5.01 8.83 -
k 7 7 7 7 -
Time (s) 0 0.37 2.03 24.59 -
Table 2: Numerical results in Example 2.

Method in [3] Present method

N =M ken k∞ ρn Iteration ken k∞ ρn Iteration
−2 −4
4 1.30 × 10 - 3 3.092 × 10 - 7
8 4.90 × 10−3 1.4 3 9.176 × 10−6 5.07 7
16 1.53 × 10−3 1.7 3 2.847 × 10−7 5.01 7
32 4.32 × 10−4 1.8 3 6.223 × 10−10 8.83 7
Table 3: Comparison of maximum absolute errors with the method in [3] for Exam-
ple 2.

Example 3. Consider [3, 16]

Z 1 Z 1
X(s, t) = f (s, t) + (s sin(y) + 1)(X(x, y))3 dxdy, (s, t) ∈ [0, 1] × [0, 1],
0 0

where
1 1
f (s, t) = s cos(t) + (cos4 (1) − 1) − sin(1)(cos2 (1) + 2),
20 12

the exact solution is given by

X ∗ (s, t) = s cos(t).

For N = 4, N = 8, N = 16, N = 32 and ε0 = 10−15 , the following results are

obtained (see tables 4-5 ). As we can see from numerical results, when the size of N
increases sinc method is much better than methods in [3] and [16].
100 M. Kazemi

(s, t) = ( 21l , 1
2l
) N=4 N=8 N=16 N=32 dp,q
1 1.217E-5 4.417E-7 1.289E-8 2.147E-11 0.1012
2 2.231E-5 8.098E-7 2.363E-8 3.937E-11 0.1011
3 3.090E-5 1.121E-6 3.272E-8 5.451E-11 0.1011
4 3.825E-5 1.388E-6 4.051E-8 6.749E-11 0.1001
5 4.463E-5 1.619E-6 4.726E-8 7.874E-11 0.1001
6 5.021E-5 1.822E-6 5.317E-8 8.859E-11 0.1000
ken k∞ 6.695E-5 2.429E-6 7.089E-8 1.181E-10 -
ρn - 5.07 5.01 9.22 -
k 6 6 6 7 -
Time (s) 0 0.37 2.03 24.59 -
Table 4: Numerical results in Example 3.

Method in [3] Method in [16]

N=M ken k∞ ρn Iteration ken k∞ ρn Iteration
4 5.20 × 10−2 - 4 4.3 × 10−2 - -
8 2.06 × 10−2 1.3 5 1.7 × 10−2 1.33 -
16 6.80 × 10−3 1.6 6 5.4 × 10−3 1.65 -
32 1.92 × 10−3 1.8 6 1.5 × 10−3 1.84 -

Present method
N=M ken k∞ ρn Iteration
4 6.695 × 10−5 - 6
8 2.429 × 10−6 4.79 6
16 7.089 × 10−8 5.10 6
32 1.181 × 10−10 9.22 7
Table 5: Comparison of maximum absolute errors in Example 3.

6. Conclusion

This research provided an iterative numerical method based on sinc quadrature for
solving nonlinear Eq. (1). This iterative method has two positive features. First,
we do not have to deal with any system of nonlinear equations. Second, it is quite
simple to apply and create an algorithm. The main results are Theorem 5, Theorem
6 and Theorem 7.

References
[1] M. Abdou, A. Badr, M. Soliman, On a method for solving a two-dimensional non-
linear integral equation of the second kind, J. Comput. Appl. Math. 253(2011), 3589–
3598.
[2] K. Atkinson, F. Potra, Projection and iterated projection methods for nonlinear
integral equations, SIAMJ. Numer. Anal. 24(1987), 1352–1373.
[3] E. Babolian, S. Bazm, P. Lima, Numerical solution of nonlinear two-dimensional
integral equations using rationalized Haar functions, Comm. Nonlinear. Sci. Numer.
Simulat. 16(2011), 1164–1175.
 Sinc approximation for solving two-dimensional integral equations 101

[4] M. Basseem, Degenerate Kernel Method for Three Dimension Nonlinear Integral
Equations of the Second Kind, Universal Journal of Integral Equations 3(2015), 61–66.
[5] S. Bazm, Bernoulli polynomials for the numerical solution of some classes of linear
and nonlinear integral equations, J. Comput. Appl. Math. 275(2015), 44–60.
[6] S. Bazm, A. Hosseini, Bernoulli operational matrix method for the numerical solution
of nonlinear two-dimensional Volterra-Fredholm integral equations of Hammerstein
type, Comput. Appl. Math. 39(2020), 49.
[7] A. M. Bica, The numerical method of successive interpolations for Fredholm func-
tional integral equations, Numer. Algorithms 58(2011), 351.
[8] A. M. Bica, M. Curila, S. Curila, About a numerical method of successive in-
terpolations for functional Hammerstein integral equations, J. Comput. Appl. Math.
236(2012), 2005–2024.
[9] H. Brunner, Collocation methods for Volterra integral and related functional differ-
ential equations, Cambridge University Press, Cambridge, 2004.
[10] W. C. Chew, M. S. Tong, B. Hu, Integral Equation Methods for Electromagnetic
and Elastic Waves, Morgan Claypool Pub., San Rafael, 2009.
[11] R. Ezzati, M. Kazemi, Existence of Solutions for Some Nonlinear Volterra Integral
Equations via Petryshyns Fixed Point Theorem, Int. J. Nonlinear Anal. Appl. 9(2018),
1–12.
[12] A. Fallahzadeh, Solution of two-dimensional Fredholm integral equation via RBF
triangular method, J. Interpolat. Approx. Sci. Comput. (2012), Art. ID 00002.
[13] R. Farengo, Y. Lee , P. Guzdar, Electromagnetic integral equation: application to
microtearing modes, Phys Fluids, 26(1983), 3515–3523.
[14] G. Han, R. Wang, The extrapolation method for two-dimensional Volterra integral
equations based on the asymptotic expansion of iterated Galerkin solutions, J. Integr.
Equ. Appl. 139(2001), 15–34.
[15] C. Hui Hsiao, Hybrid function method for solving Fredholm and Volterra integral
equations of the second kind, J. Comput. Appl. Math. 230(2009), 59–68.
[16] A. Imran, F. Khan, A new method based on Haar wavelet for the numerical solution
of two-dimensional nonlinear integral equations, J. Comput. Appl. Math. 272 (2014),
70–80.
[17] S. Inderdeep, S. Kumar, Haar wavelet method for some nonlinear Volterra integral
equations of the first kind, J. Comput. Appl. Math. 292 (2016), 541–552.
[18] A. Karimi, K. Maleknejad, R. Ezzati, Numerical solutions of system of two-
dimensional Volterra integral equations via Legendre wavelets and convergence, Appl.
Numer. Math. 156 (2020), 228–241.
[19] M. Kazemi, R. Ezzati, Existence of solution for some nonlinear two-dimensional
Volterra integral equations via measures of noncompactness, Appl. Math. Comput.
275(2016), 165–171.
[20] M. Kazemi, H. Mottaghi Golshan, R. Ezzati, M. Sadatrasoul, New approach
to solve two-dimensional Fredholm integral equations, J. Comput. Appl. Math.
354(2019), 66–79.
[21] M. Kazemi, V. Torkashvand, R. Ezzati, On a method based on Bernstein operators
for 2D nonlinear Fredholm-Hammerstein integral equations, U.P.B. Sci. Bull. Series A
83(2021), 178–198.
[22] M. M. Kazemi, Triangular functions for numerical solution of the nonlinear Volterra
integral equations, J. Appl. Math. Comput. (2021), 1–24.
[23] A. A. Khajehnasiri, Numerical solution of nonlinear 2D Volterra-Fredholm integro-
differential equations by two-dimensional triangular function, Int. J. Appl. Comput.
Math. 4(2016), 575–591.
102 M. Kazemi

[24] I. Kolodii, F. Khildebrand, Some properties of the modulus of continuity, Mathe-

matical notes of the Academy of Sciences of the USSR 9(1971), 285–288.
[25] J. Lund, K. Bowers, Sinc Methods for Quadrature and Differential Equations, SIAM,
Philadelphia, 1992.
[26] K. Maleknejad, R. Mollapourasl, K. Nouri, Study on existence of solutions for
some nonlinear functional-integral equations, Nonlinear Anal. Theory Methods Appl.
69(2008), 2582–2588.
[27] K. Maleknejad, P. Torabi, Application of fixed point method for solving nonlinear
Volterra-Hammerstein integral equation, U.P.B. Sci. Bull. Series A, 74(2012), 45–56.
[28] Z. Mahmoodi, J. Rashidinia, E. Babolian, B-Spline collocation method for lin-
ear and nonlinear Fredholm and Volterra integro-differential equations, Appl. Anal.
92(2013), 1787–1802.
[29] F. Mirzaee, S. Alipour, Solving two-dimensional non-linear quadratic integral equa-
tions of fractional order via operational matrix method, Multidiscipl. Model. Mater.
Struct. 15(2019), 1136–1151.
[30] F. Mirzaee, E. Hadadiyan, A computational method for nonlinear mixed Volterra-
Fredholm integral equations, Caspian J. Math. Sci. 24 (2014), 113–123.
[31] F. Mirzaee, E. Hadadiyan, S. Bimesl, Numerical solution for three-dimensional
nonlinear mixed Volterra-Fredholm integral equations via three-dimensional block-pulse
functions, Appl. Math. Comput. 237 (2014), 168–175.
[32] F. Mirzaee, N. Samadyar, Using radial basis functions to solve two dimensional
linear stochastic integral equations on non-rectangular domains, Eng. Anal. Bound.
Elem. 92(2018), 180–195.
[33] F. Mirzaee, N. Samadyar, Application of hat basis functions for solving two-
dimensional stochastic fractional integral equations, Comp. Appl. Math. 37(2018),
4899–4916.
[34] F. Mirzaee, N. Samadyar, Numerical solution based on twodimensional orthonormal
Bernstein polynomials for solving some classes of two-dimensional nonlinear integral
equations of fractional order, Appl. Math. Comput. J. 345(2019), 191–203.
[35] L. N. Mishra, H. M. Srivastava, M. Sen, Existence results for some nonlinear
functional-integral equations in Banach algebra with applications, Int. J. Anal. Appl.
11(2016), 1–10.
[36] M. Mori, M. Sugihara, The double-exponential transformation in numerical analy-
sis, J. Comput. Appl. Math. 127(2001), 287–296.
[37] M. Muhammad, A. Nurmuhammad, M. Mori, M. Sugihara, Numerical solution of
integral equations by means of the Sinc collocation method based on the double expo-
nential transformation, J. Comput. Appl. Math. 177(2005), 269–286.
[38] S. Najafalizadeh, R. Ezzati, Numerical methods for solving two-dimensional non-
linear integral equations of fractional order by using two-dimensional block pulse op-
erational matrix, Appl. Math. Comput. 280(2016), 46–56.
[39] M. A. Ramadan, M. R . Ali, An efficient hybrid method for solving Fredholm integral
equations using triangular functions, NTMSCI 5(2017), 213–224.
[40] F. Stenger, Numerical Methods Based on Sinc and Analytic Functions, Springer-
Verlag, New York, 1993.
[41] H. Takahasi, M. Mori, Double exponential formulas for numerical integration, Publ.
Res. Inst. Math. Sci. Kyoto University 9(1974), 721–741.
[42] Qi. Tang, D. Waxman, An integral equation describing an asexual population in a
changing environment, Nonlinear Anal. Theory Methods Appl. 53(2003), 683–699.
[43] K. F. Warnick, Numerical Analysis for Electromagnetic Integral Equations, Artech
House, London, 2008.
 Sinc approximation for solving two-dimensional integral equations 103

[44] A. Ziqan, S. Armiti, I. Suwan, Solving three-dimensional Volterra integral equation

by the reduced dierential transform method, Int. J. Appl. Math. Res. 5(2016), 103–106.