INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS

https://doi.org/10.1080/00207160.2023.2260007

RESEARCH ARTICLE

Numerical solution of nonlinear third-kind Volterra integral

equations using an iterative collocation method
Khedidja Kherchouchea , Azzeddine Bellour a and Pedro Limab
a Laboratoire de Mathématiques appliquées et didactique, Ecole Normale Superieure de Constantine, Constantine,
Algeria; b Department of Mathematics, Instituto Superior Técnico, University of Lisbon, Lisboa, Portugal

ABSTRACT ARTICLE HISTORY

In this paper, we discuss the application of an iterative collocation method Received 2 August 2022
based on the use of Lagrange polynomials for the numerical solution of a Revised 22 August 2023
class of nonlinear third-kind Volterra integral equations. The approximate Accepted 28 August 2023
solution is given by explicit formulas. The error analysis of the proposed KEYWORDS
numerical method is studied theoretically. Some numerical examples are Nonlinear third-kind Volterra
given to confirm our theoretical results. integral equation;
Collocation method; Iterative
method; Lagrange
polynomials; Convergence
analysis

MATHEMATICS SUBJECT
CLASSIFICATIONS
45E99; 45G05; 65R20

1. Introduction
The integral equations and their particular forms (integro-differential, delay integral, partial integro-
differential, . . . ) have become important in the mathematical modelling of many fields of sciences and
engineering (see, e.g. [2–5,8,10]).
Recently, there has been a growing interest in the numerical solution of integral equations (see,
e.g. [4,8,9,15–21,24]).
In this paper, we study a numerical method for nonlinear third-kind Volterra integral equations
of the form:
 t
β
t x(t) = f (t) + H(t, s, x(s)) ds, t ∈ I = [0, T], (1)
0

we assume that f (t) = tβ g(t)

and H(t, s, x(s)) = sμ−1 k(t, s, x(s)) where the functions f , k are suffi-
ciently smooth and β > 0, μ ≥ β + 1.
Equation (1) can be written as
 t  β−1
s 1 μ−β
x(t) = g(t) + s k(t, s, x(s)) ds, t ∈]0, T], (2)
0 t t

which is a particular case of the nonlinear of the so-called ‘cordial’ integral equations, which were
introduced by Vainikko in [23], for the core of the cordial equation is φ(s) = sβ−1 and the regular
kernel is sμ−β k(t, s, x).

CONTACT Azzeddine Bellour [email protected]

© 2023 Informa UK Limited, trading as Taylor & Francis Group

2 K. KHERCHOUCHE ET AL.

The existence and uniqueness result in Cm ([0, T]) for the nonlinear cordial equations was obtained
in [24], and the existence and uniqueness in Cm ([0, T]) of some classes of third-kind Volterra integral
equations was given in [1]. Nonlinear third-kind Volterra integral equations appear in some heat
conduction problems with mixed-type boundary conditions [2,3] and arise as mathematical models
in many models , see, for example [5,6,9].
An important contribution to the numerical solution of the nonlinear cordial integral equations
was given by Vainikko in [24] by using spline collocation–interpolation method, but the numerical
solution is given by an implicit formula.
Song et al. in [22] considered collocation methods in certain piecewise polynomial spaces for the
following nonlinear Volterra integral equations of the third kind
 t
t β x(t) = t β g(t) + k(t, s)G(x(s)) ds, t ∈ I = [0, T].
0

The numerical solutions of the linear third-kind Volterra integral equations have been studied in
many papers (see, e.g. [11–13,21–23,25]). A rigorous convergence analysis of the Chebyshev col-
location method was given in [11]. A spectral collocation method based on the generalized Jacobi
wavelets was given in [12]. An operational matrix is used to obtain for solving a class of third-kind
linear Volterra integral equations in [13]. Spectral collocation methods were established in [21]. The
spline collocation method was studied in [22].
The aim of this paper is to generalize the iterative collocation method in [7,14] to construct an
iterative collocation method in a suitable spline polynomials space. Under fairly simple conditions
on the data functions, we have obtained a high order of convergence; moreover, the approximate
solutions are given by explicit formulas.
This paper is organized as follows. In Section 2, we describe a numerical scheme for the solution
of Equation (1). In Section 3, we analyse the convergence and obtain error estimates. Numerical
examples that illustrate the performance of the method are presented in Section 4, and this paper
finishes with conclusion in Section 5.

2. Description of the collocation method

Let N be a uniform partition of the interval I = [0, T] defined by tn = nh, n = 0, . . . , N − 1, where
T
the stepsize is given by h = . Let the collocation parameters be 0 < c1 < · · · · · · < cm ≤ 1 and the
N
collocation points be tn,j = tn + cj h, j = 1, . . . , m, n = 0, . . . , N − 1. Define the subintervals σn =
[tn , tn+1 ], and σN−1 = [tN−1 , tN ].
Moreover, denote by πm the set of all real polynomials of degree not exceeding m.
We define the real polynomial spline space of degree m−1 as follows:

S(−1)
m−1 (I, N ) = {u : un = u|σn ∈ πm−1 , n = 0, .., N − 1}.

This is the space of piecewise polynomials of degree at most m−1. Its dimension is Nm. We consider
the space L∞ (I) with the norm

ϕ = inf {C ∈ R : |ϕ(t)| ≤ C fora.e. t ∈ I} < ∞.

It holds for any y ∈ Cm ([0, T]) that


m
ym (ζn (τ )) 
m
y(tn + τ h) = λj (τ )y(tn,j ) + n (τ ), n (τ ) = hm (τ − cj ), (3)
m!
j=1 j=1
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 3

m τ − cl
where τ ∈ [0, 1] and λj (τ ) = l=j are the Lagrange polynomials associated with the param-
c − cl
mj
eters cj , j = 1, . . . , m. Let m =  j=1 |λj | be the Lebesgue constants, such that
⎧ ⎫
m ⎨ m ⎬
 |λj |= max |λj (s)|, s ∈ [0, 1] .
⎩ ⎭
j=1 j=1

Inserting (3) for the function s −→ k(t, s, x(s))ds into (2), we obtain for each j = 1, . . . , m, , n =
0, . . . , N − 1
m  cj μ−β
 
h (n + τ )μ−1
x(tnj ) = g(tnj ) + k(tn + cj h, tn + cl h, x(tnl ))λl (τ ) dτ
0 (n + cj )β
l=1

 m  1
n−1  
hμ−β (i + τ )μ−1
+ k(tn + cj h, ti + cl h, x(til ))λl (τ ) dτ
0 (n + cj )β
i=0 l=1

+ o(hm ). (4)

Now, we approximate the solution of (1) by a certain function u ∈ S−1

m−1 (I, N ) such that


m
u(tn + τ h) = λl (τ )u(tn,l ), τ ∈ [0, 1]. (5)
l=1

Then the coefficients u(tn,j ) in the expansion (5) must satisfy the following system:
m 
 
cj hμ−β (n + τ )μ−1
u(tn,j ) = g(tnj ) + k(tn + cj h, tn + cl h, u(tnl ))λl (τ ) dτ
0 (n + cj )β
l=1

 m 
n−1  
1 hμ−β (i + τ )μ−1
+ k(tn + cj h, ti + cl h, u(til ))λl (τ ) dτ (6)
i=0 l=1 0 (n + cj )β

for j = 1, . . . , m, n = 0, . . . , N − 1. Since the above system is nonlinear, we will use an iterative collo-
cation sequence of splines uq ∈ S−1 m−1 (I, N ), q ∈ N, to approximate the exact solution of (1). These
splines are obtained by the formula


m
uq (tn + τ h) = λj (τ )uq (tn,j ), τ ∈ [0, 1] (7)
j=1

where the coefficients uq (tn,j ) are given by the following relation:

m 
 cj hμ−β (n + τ )μ−1
uq (tn,j ) = g(tnj ) + k(tn + cj h, tn + cl h, uq−1 (tnl ))λl (τ ) dτ
0 (n + cj )β
l=1

 m 
n−1  1 hμ−β (i + τ )μ−1
+ k(tn + cj h, ti + cl h, uq (til ))λl (τ ) dτ , (8)
0 (n + cj )β
i=0 l=1

where the initial values satisfy u0 (tn,j ) ∈ J (J is a bounded interval),

where the integrals on the right-hand side of (8) can be computed analytically using integration
by parts.
4 K. KHERCHOUCHE ET AL.

The above formula is explicit and the sequence of approximations uq is obtained without solving
any algebraic system.
In the next section, we will prove the convergence of the approximate solution uq to the exact
solution x of (1).

3. Convergence analysis
In this section, we assume that the function k satisfies the Lipschitz condition with respect to the third
variable: there exist L ≥ 0 such that

|k(t, s, y1 ) − k(t, s, y2 )| ≤ L|y1 − y2 |.

for all t, s ∈ I, where L is independent of t and s.

The following lemma will be used to prove the convergence of the current numerical method:

Lemma 3.1 ([4]): Let {kj }nj=0 be a given non-negative sequence and the sequence {εn } satisfies ε0 ≤ p0
and

n−1
εn ≤ p 0 + ki εi , n ≥ 1,
i=0

with p0 ≥ 0. Then εn can be bounded by

⎛ ⎞
n−1
εn ≤ p0 exp ⎝ kj ⎠ , n ≥ 1.
j=0

The following result gives the existence and the uniqueness of a solution for (1).

Lemma 3.2: Let g ∈ C([0, T]), k(t, s, u) ∈ C(T × R). We put k1 (t, s, u) = sμ−β k(t, s, u), so
k1 (t, s, u) ∈ C(T × R) where T = {(t, s) ∈ R2 : 0 ≤ t ≤ T, 0 ≤ s ≤ t}. Let
∂k1
∈ C(T × R).
∂u
Assume that equation
1
ξ = k1 (0, 0, ξ ) + g(0) (9)
β
has a unique solution ξ ∗ ∈ R, and that

a∗ (0, 0)
1 = , ∀λ : Re(λ) ≥ 0, (10)
λ+β

where

∂k1 
∗
a (0, 0) = (0, 0, u) .
∂u u=ξ ∗

Moreover, let k1 satisfy

k1 (t, s, u)| ≤ c0 + c1 |u|, (11)
with cβ1 < 1.
Then there is a unique solution x∗ ∈ C([0, T]) of (1), such that x∗ (0) = ξ ∗ .
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 5

Proof: The result follows from Theorems 7.1 and 7.5 of [24], taking into account that in our case
φ(x) = xβ−1 ,with β > 0, and therefore the linear integral operator Vφ (using the same notation as
in [24]) is defined by
 t μ−1
s
Vφ u(t) = u(s) ds,
0 tβ
hence the spectrum of this operator is
 
1
σ0 (Vφ ) = {0} ∪ : λ ∈ C, Re(λ) ≥ 0 ,
λ+β
which is used to obtain condition (10). 

Lemma 3.3: Let the conditions of Lemma 3.2 be satisfied and let x∗ be a solution of (1). Moreover, let
g ∈ Cm ([0, T]) and k ∈ Cm (T × R) for some natural m.
Then x∗ ∈ Cm ([0, T]).

Proof: The result follows from Theorem 8.1 of [24]. 

LT μ−β
Corollary 3.4: Under the condition β < 1, the following conditions of Lemma 3.2 are fulfilled:

(1) Equation (9) has a unique solution ξ ∗ ∈ R.

(2) Condition (10) is satisfied.
(3) Inequality (11) is satisfied, moreover cβ1 < 1.

Proof: (1) For all ξ ∈ R, we have k1 (0, 0, u) = 0, which implies that (9) has a unique solution ξ ∗ ∈ R
such that ξ ∗ = g(0)
(2) Since a∗ (0, 0) = 0 because ∂k
∂u (0, 0, u) = 0, ∀u ∈ R, then (10) is satisfied.
1

(3) We have,
μ−β
|k1 (t, s, u)| ≤ |k1 (t, s, u) − k1 (t, s, 0)| + |k1 (t, s, 0)| ≤ LT
   |u| + c0 ,
=c1

such that c0 = max{|k1 (t, s, 0)|, (t, s) ∈ I × I}.

c1
Hence the inequality (11) is satisfied, moreover β < 1.


Remark 3.1: Under the conditions of Lemmas 3.2, 3.3 and Corollary 3.4 the nonlinear weakly
singular third-kind Volterra integral Equation (1) has a unique solution in Cm ([0, T]).

The following lemma gives the existence and the uniqueness of the solution for the nonlinear
system (6).

Lemma 3.5: For sufficiently small h, the nonlinear system (6) has a unique solution u ∈ S−1
m−1 (I, N ).

Proof: We will use the induction combined with the Banach fixed point theorem.

(i) On the interval σ0 = [t0 , t1 ], the nonlinear system (6) becomes

m  cj μ−β 
h (τ )μ−1
u(t0,j ) = g(t0,j ) + k(t0 + cj h, t0 + cl h, u(t0l ))λl (τ ) dτ
0 (cj )β
l=1

We consider the operator  defined by

 : Rm −→ Rm
6 K. KHERCHOUCHE ET AL.

x = (x1 , . . . , xm ) −→ (x) = (1 (x), . . . , m (x)),

such that for j = 1, . . . , m, we have
m  
cj hμ−β (τ )μ−1
j (x) = g(t0,j ) + k(t0 + cj h, t0 + cl h, xl )λl (τ ) dτ .
0 (cj )β
l=1

Hence, for all x, y ∈ Rm , we have

hμ−β L m
(x) − (y)≤ x − y.
μ
μ−β
Since h μL m < 1 for sufficiently small h, then by Banach fixed point theorem, the nonlinear
system (6) has a unique solution u on σ0 .
(ii) Suppose that u exists and unique on the intervals σi , i = 0, . . . , n − 1 for n ≥ 1 and we show that
u exists and unique on the interval σn .
On the interval σn , the nonlinear system (6) becomes
m  cj μ−β
 
h (n + τ )μ−1
u(tn,j ) = G(tn,j ) + k(t n + cj h, t n + cl h, u(tnl ))λ l (τ ) dτ , (12)
0 (n + cj )β
l=1

where
 m 
n−1  
1 hμ−β (i + τ )μ−1
G(tn,j ) = g(tn,j ) + k(tn + cj h, ti + cl h, u(til ))λl (τ ) dτ .
i=0 l=1 0 (n + cj )β

We consider the operator  defined by

 : Rm −→ Rm
x = (x1 , . . . , xm ) −→ (x) = (1 (x), . . . , m (x)),
such that for j = 1, . . . , m, we have
m  cj μ−β 
h (n + τ )μ−1
j (x) = G(tn,j ) + k(tn + cj h, t n + cl h, x )λ
l l (τ ) dτ .
0 (n + cj )β
l=1

Hence, for all x, y ∈ Rm , we have

|j (x) − j (y)| ≤ cj hμ−β (n + cj )μ−β−1 L m x − y,
which implies that
(x) − (y)≤ hT μ−β−1 L m x − y.
Since hT μ−β−1 L m < 1 for sufficiently small h, then by Banach fixed point theorem, the
nonlinear system (12) has a unique solution u on σn . 

The following result gives the convergence of the approximate solution u to the exact solution x.

Theorem 3.6: Let f, k be m times continuously differentiable on their respective domains and assume
that
LT μ−β
< 1. (13)
β
Then,
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 7

(i) Equation (1) has a unique solution x ∈ Cm ([0, T]).

(ii) For sufficiently small h, the collocation solution u is well-defined, converges to the exact solution x
and the resulting error function ξ := x − u satisfies

ξ  ≤ Chm ,

where C is a finite constant independent of h.

Proof: (i) Since condition (13) is satisfied, existence and uniqueness of solution of (1) in Cm ([0, T])
is guaranteed according to Remark 3.1.
(ii) First of all, note that from Lemma 3.5 there exists a unique collocation solution u. We have,
from (4) and (6), for all n = 0, . . . , N − 1 and j = 1, . . . , m


m  
cj hμ−β (n + τ )μ−1
|ξnj | ≤ αhm + L |λ l (τ )| dτ |ξnl |
0 (n + cj )β
l=1


n−1 
m  
1 hμ−β (i + τ )μ−1
+ L |λl (τ )| dτ |ξil |,
0 (n + cj )β
i=0 l=1

where α is a positive number.

Since i + τ ≤ n + cj for all i = 0, . . . , n − 1, then


m  cj 
m μ−β μ−β−1
|ξnj | ≤ αh + Lh (n + cj ) |λl (τ )| dτ |ξnl |
l=1 0


n−1 
m  1 
+ Lhμ−β (i + cj )μ−β−1 |λl (τ )| dτ |ξil |
i=0 l=1 0


m  cj 
≤ αhm + LhT μ−β−1 |λl (τ )| dτ |ξnl |
l=1 0


n−1 
m  1 
μ−β−1
+ LhT |λl (τ )| dτ |ξil |. (14)
i=0 l=1 0

We consider the sequence ξn = max{|ξ(tn,l )|, l = 1, . . . ., m} for n = 0, . . . , N − 1.

Then, from (14), ξn satisfies for n = 0, . . . , N − 1,


n−1
ξn ≤ αhm + h LT μ−β−1 m ξn + h LT μ−β−1 m ξi .
     
a1 i=0 a1

1
Let h < a1 , Hence for all h ∈ (0, h], we have

α 
n−1
a1
m
ξn ≤ h +h ξi .
1 − ha 1 − ha
  1 i=0   1
a2 a3

Then, by Lemma 3.1, for all n = 0, . . . , N − 1

ξn ≤ a2 hm exp (Ta3 )
8 K. KHERCHOUCHE ET AL.

Moreover, by using (3) and (5), we obtain

x(m) 
ξ  ≤ m max(ξn , n = 0, . . . , N − 1) + hm
m!
m x(m) 
≤ m a2 h exp (Ta3 ) + hm .
m!
x(m) 
Thus, the proof is completed by taking C = m a2 exp(Ta3 ) + m! .


The following result gives the convergence of the iterative collocation solution uq to the exact
solution x.

Theorem 3.7: Consider the iterative collocation solution uq , q ≥ 1 defined by (7) and (8), then for any
initial condition u0 (tn,j ) ∈ J, the iterative collocation solution uq , q ≥ 1 converges to the exact solution
x. Moreover, the following error estimate holds:
ξ q  ≤ ahm + dbq hq ,
where the error ξ q (t) = uq (t) − x(t), provided that h is sufficiently small, where a, b, d are finite
constants independent of h.

Proof: We have, from (4) and (8), for all n = 0, . . . , N − 1 and j = 1, . . . , m


m  cj μ−β 
h (n + τ )μ−1
|ξ q (tn,j )| ≤ Chm + L |λ l (τ )| dτ |ξ q−1 (tnl )|
0 (n + cj )β
l=1


n−1 
m  
1 hμ−β (i + τ )μ−1
+ L |λl (τ )| dτ |ξ q (til )|,
0 (n + cj )β
i=0 l=1

where α is a positive number.

Since i + τ ≤ n + cj for all i = 0, . . . , n − 1, then

m  cj 
|ξ q (tn,j )| ≤ Chm + Lhμ−β (n + cj )μ−β−1 |λl (τ )| dτ |ξ q−1 (tnl )|
l=1 0


n−1 
m  1 
μ−β μ−β−1
+ Lh (i + cj ) |λl (τ )| dτ |ξ q (til )|
i=0 l=1 0


m  cj 
≤ Chm + LhT μ−β−1 |λl (τ )| dτ |ξ q−1 (tnl )|
l=1 0


n−1 
m  1 
μ−β−1
+ LhT |λl (τ )| dτ |ξ q (til )|.
i=0 l=1 0

q
Now, for each fixed q ≥ 1, we consider the sequence ξn = max{|ξ q (tn,l )|l = 1 · · · .m} for n =
0, . . . , N − 1, it follows that,

n−1
ξn ≤ Chm + h LT μ−β−1 m ξn h LT μ−β−1 m ξi .
q q−1 q
+ (15)
     
a1 i=0 a1
q
Next, we consider the sequence ξ q = max{ξn , n = 0, . . . ., N − 1}.
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 9

Then from (15), we have for n = 0, . . . , N − 1

q

n−1
q
ξn ≤ Chm + ha1 ξ q−1 + ha1 ξi .
i=0

Hence, by Lemma 3.1, for all n = 0, . . . , N − 1,

q
ξn ≤ C exp(Ta1 ) hm + h a1 exp(Ta1 ) ξ q−1 ,
     
a2 a3

which implies that, ξ q satisfies, for all q ≥ 1

ξ q ≤ a2 hm + ha3 ξ q−1
≤ a2 hm + a3 a2 hm+1 + h2 a23 ξ q−2
≤ a2 hm + a2 a3 hm+1 + a2 a23 hm+2 + h3 a33 ξ q−3 .

It is easy to show, by using the induction, that

q−1 m+q−1 q
ξ q ≤ a2 hm + a2 a3 hm+1 + a2 a23 hm+2 + · · · + a2 a3 h + hq a3 ξ 0
q−1 q−1 q
= a2 hm (1 + a3 h + a23 h2 + · · · + a3 h ) + hq a3 ξ 0
1 − (a3 h)q q
= a2 hm + hq a3 ξ 0 ,
1 − a3 h

1
Let h < a3 , Hence for all h ∈ (0, h], we have

1 q
ξ q ≤ a2 hm + hq a3 ξ 0 .
1−a h
  3 
a4

Since, u0 (tn,j ) ∈ J (bounded interval). Then the function u0 is bounded.

Hence, ξ 0 ≤ u0 − x ≤ u0  + x = a5 . It follows that
q
ξ q ≤ a4 hm + a5 a3 hq .

Moreover, by using (3) and (7), we obtain

q x(m) 
ξ  ≤ mξ + hm
m!
m q q x(m) 
≤ m a4 h + m a5 a3 h + hm .
m!

x(m) 
Thus the proof is completed by taking a = m a4 + m! , b = a3 , d = m a5 . 

Remark 3.2: From the error estimate in Theorem 3.7 it follows that for all q ≥ m the total error has
the order of O(hm ).
10 K. KHERCHOUCHE ET AL.

Table 1. Absolute errors with equidistant points for Example 4.1.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 2.22 × 10−4 9.37 × 10−5 2.77 × 10−5 1.17 × 10−5
0.1 8.88 × 10−4 9.37 × 10−5 1.94 × 10−4 1.17 × 10−5
0.2 1.55 × 10−3 9.37 × 10−5 3.61 × 10−4 1.17 × 10−5
0.3 2.22 × 10−3 9.37 × 10−5 5.28 × 10−4 1.17 × 10−5
0.4 2.89 × 10−3 9.37 × 10−5 6.95 × 10−4 1.17 × 10−5
0.5 3.56 × 10−3 9.36 × 10−5 8.62 × 10−4 1.17 × 10−5
0.6 4.22 × 10−3 9.33 × 10−5 1.02 × 10−3 1.16 × 10−5
0.7 4.88 × 10−3 9.29 × 10−5 1.19 × 10−3 1.16 × 10−5
0.8 5.52 × 10−3 9.21 × 10−5 1.35 × 10−3 1.16 × 10−5
0.9 6.12 × 10−3 9.12 × 10−5 1.51 × 10−3 1.15 × 10−5
1 6.38 × 10−3 9.60 × 10−5 1.61 × 10−3 1.18 × 10−5

Table 2. Absolute errors with Gaussian points for Example 4.1.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 1.66 × 10−4 4.99 × 10−5 2.08 × 10−5 6.24 × 10−6
0.1 6.66 × 10−4 5.00 × 10−5 1.45 × 10−4 6.24 × 10−6
0.2 1.16 × 10−3 5.00 × 10−5 2.70 × 10−4 6.25 × 10−6
0.3 1.66 × 10−3 5.00 × 10−5 3.95 × 10−4 6.24 × 10−6
0.4 2.16 × 10−3 5.00 × 10−5 5.21 × 10−4 6.25 × 10−6
0.5 2.66 × 10−3 4.99 × 10−5 6.46 × 10−4 6.24 × 10−6
0.6 3.16 × 10−3 4.97 × 10−5 7.70 × 10−4 6.23 × 10−6
0.7 3.66 × 10−3 4.94 × 10−5 8.95 × 10−4 6.22 × 10−6
0.8 4.14 × 10−3 4.88 × 10−5 1.01 × 10−3 6.18 × 10−6
0.9 4.62 × 10−3 4.80 × 10−5 1.14 × 10−3 6.13 × 10−6
1 4.88 × 10−3 5.21 × 10−5 1.23 × 10−3 6.38 × 10−6

4. Numerical examples
To illustrate the theoretical results obtained in the previous section, we present the following exam-
ples with T = 1. All the exact solutions x are already known. In each example, we calculate the error
between x and the iterative collocation solution uq . In all the examples, we choose N = 10, 20, m = 1,
2, 3, q = 5,u0 (tnj ) = 1, μ = 4 and β = 2, moreover, we calculate the experimental orders of conver-
gence (EOC) by using the maximum error eN  = max{|x(ti ) − uq (ti )|, i = 0, . . . , N} for N = 5, 10,
15, 20 and m = 1, 2, 3.
μ−β
Since the condition LT β < 1 is essential to guarantee the existence and uniqueness of the exact
solution, we checked that it is satisfied in all the numerical examples.
To check the influence of the choice of collocation points in the accuracy of the method, we have
j
tested two sets of collocation parameters : (a) equidistant points: cj = m+1 , j = 1, . . . , m; (b) Gaussian
√ √
points: for m = 2, c1 = 1
− 1
√ , c2 = 1
+ 1
√ ; for m = 3, c1 = 1
− √3 , c = 0, c3 = 1
+ √3 .
2 2 3 2 2 3 2 2 5 2 2 2 5

Example 4.1: Consider the following integral equation:

 t s3
x(t) = g(t) + k(t, s, x(s)) ds, t ∈ [0, 1].
0 t2

with k(t, s, x(s)) = (s+t) sin(x(s))

5 , g(t) is chosen such that the exact solution is x(t) = t 3 + 1. The abso-
lute errors for Example 4.1 are given in Table 1, for the case of equidistant points, and in Table 2, for
the case of Gaussian points. The experimental convergence orders for the two types of collocation
points are displayed in Table 3.
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 11

Table 3. Experimental orders of convergence (EOC) of Example 4.1.

N m=1 m=2 m=3
EOC with equidistant points
5
10 0.92 1.95 3.01
15 0.93 1.96 3.01
20 0.94 1.97 3.01

EOC with Gaussian points

5
10 0.92 1.95 3.04
15 0.93 1.96 3.03
20 0.94 1.96 3.03

Table 4. Absolute errors with equidistant points for Example 4.2.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 4.41 × 10−4 2.78 × 10−6 1.10 × 10−4 1.75 × 10−7
0.1 3.98 × 10−4 9.36 × 10−6 1.02 × 10−4 1.03 × 10−6
0.2 3.16 × 10−4 1.34 × 10−5 8.31 × 10−5 1.61 × 10−6
0.3 2.15 × 10−4 1.45 × 10−5 5.83 × 10−5 1.82 × 10−6
0.4 1.15 × 10−4 1.31 × 10−5 3.29 × 10−5 1.69 × 10−6
0.5 3.07 × 10−5 1.03 × 10−5 1.08 × 10−5 1.36 × 10−6
0.6 3.29 × 10−5 7.25 × 10−6 5.99 × 10−6 9.78 × 10−7
0.7 7.51 × 10−5 4.44 × 10−6 1.73 × 10−5 6.15 × 10−7
0.8 9.92 × 10−5 2.241 × 10−6 2.40 × 10−5 3.24 × 10−7
0.9 1.09 × 10−4 6.87 × 10−7 2.71 × 10−5 1.16 × 10−7
1 1.10 × 10−4 3.90 × 10−7 2.78 × 10−5 2.27 × 10−8

Table 5. Absolute errors with Gaussian points for Example 4.2.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 3.30 × 10−4 1.48 × 10−6 8.31 × 10−5 9.41 × 10−8
0.1 2.98 × 10−4 4.98 × 10−6 7.67 × 10−5 5.53 × 10−7
0.2 2.37 × 10−4 7.17 × 10−6 6.23 × 10−5 8.63 × 10−7
0.3 1.61 × 10−4 7.75 × 10−6 4.37 × 10−5 9.71 × 10−7
0.4 8.69 × 10−5 7.01 × 10−6 2.47 × 10−5 9.04 × 10−7
0.5 2.32 × 10−5 5.53 × 10−6 8.16 × 10−6 7.30 × 10−7
0.6 2.44 × 10−5 3.86 × 10−6 4.45 × 10−6 5.22 × 10−7
0.7 5.60 × 10−5 2.37 × 10−6 1.29 × 10−5 3.28 × 10−7
0.8 7.41 × 10−5 1.19 × 10−6 1.80 × 10−5 1.73 × 10−7
0.9 8.20 × 10−5 3.66 × 10−7 2.03 × 10−5 6.19 × 10−8
1 8.29 × 10−5 2.10 × 10−7 2.08 × 10−5 1.25 × 10−8

Example 4.2: Consider the following integral equation:

 t s3
x(t) = g(t) + k(t, s, x(s)) ds, t ∈ [0, 1],
0 t2

st 1
with k(t, s, x(s)) = 5(1+(x(s)) 2 ) , and g(t) is chosen so that the exact solution is x(t) = 5(t 2 +1) . The
absolute errors for Example 4.2 are given in Table 4, for the case of equidistant points, and in Table 5,
for the case of Gaussian points. The experimental convergence orders for the two types of collocation
points are displayed in Table 6.
12 K. KHERCHOUCHE ET AL.

Table 6. Experimental orders of convergence (EOC) of Example 4.2.

N m=1 m=2 m=3
EOC with equidistant points
5
10 0.97 1.96 2.95
15 0.98 1.97 2.98
20 0.98 1.97 2.97

EOC with Gaussian points

5
10 0.97 1.96 2.95
15 0.98 1.97 2.97
20 0.98 1.97 2.97

Table 7. Absolute errors with equidistant points for Example 4.3.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 2.21 × 10−3 6.98 × 10−6 5.55 × 10−4 4.46 × 10−7
0.1 2.10 × 10−3 2.41 × 10−5 5.33 × 10−4 2.65 × 10−6
0.2 1.88 × 10−3 3.70 × 10−5 4.83 × 10−4 4.38 × 10−6
0.3 1.59 × 10−3 4.40 × 10−5 4.12 × 10−4 5.39 × 10−6
0.4 1.27 × 10−3 4.53 × 10−5 3.33 × 10−4 5.69 × 10−6
0.5 9.64 × 10−4 4.25 × 10−5 2.53 × 10−4 5.41 × 10−6
0.6 6.78 × 10−4 3.71 × 10−5 1.80 × 10−4 4.78 × 10−6
0.7 4.35 × 10−4 3.07 × 10−5 1.18 × 10−4 3.99 × 10−6
0.8 2.37 × 10−4 2.44 × 10−5 6.67 × 10−5 3.19 × 10−6
0.9 8.30 × 10−5 1.87 × 10−5 2.64 × 10−5 2.46 × 10−6
1 3.99 × 10−5 1.74 × 10−5 4.95 × 10−6 2.06 × 10−5

Table 8. Absolute errors with Gaussian points for Example 4.3.

N = 10 N = 10 N = 20 N = 20
t m=2 m=3 m=2 m=3
0 1.65 × 10−3 3.72 × 10−6 4.16 × 10−4 2.31 × 10−7
0.1 1.57 × 10−3 1.28 × 10−5 3.99 × 10−4 1.41 × 10−6
0.2 1.41 × 10−3 1.97 × 10−5 3.62 × 10−4 2.33 × 10−6
0.3 1.19 × 10−3 2.34 × 10−5 3.09 × 10−4 2.87 × 10−6
0.4 9.59 × 10−4 2.41 × 10−5 2.50 × 10−4 3.03 × 10−6
0.5 7.23 × 10−4 2.26 × 10−5 1.90 × 10−4 2.88 × 10−6
0.6 5.09 × 10−4 1.98 × 10−5 1.35 × 10−4 2.55 × 10−6
0.7 3.26 × 10−4 1.64 × 10−5 8.85 × 10−5 2.13 × 10−6
0.8 1.77 × 10−4 1.30 × 10−5 4.99 × 10−5 1.70 × 10−6
0.9 6.18 × 10−5 1.00 × 10−5 1.96 × 10−5 1.31 × 10−6
1 2.94 × 10−5 9.30 × 10−6 3.56 × 10−6 1.10 × 10−6

Example 4.3: Consider the following integral equation:

 t s3
x(t) = g(t) + k(t, s, x(s)) ds, t ∈ [0, 1],
0 t2
(s+t) exp (x(s))
with k(t, s, x(s)) = 40(1+exp (x(s))) , and g(t) is chosen so that the exact solution is x(t) = ln (1 + t 2 ).
The absolute errors for Example 4.3 are given in Table 7, for the case of equidistant points, and in
Table 8, for the case of Gaussian points. The experimental convergence orders for the two types of
collocation points are displayed in Table 9.

In all the three examples, we observe that the experimental convergence orders correspond to
the theoretical ones, independently from the choice of the collocation parameters. Moreover, when
 INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS 13

Table 9. Experimental orders of convergence (EOC) of Example 4.3.

N m=1 m=2 m=3
EOC with equidistant points
5
10 0.99 1.98 2.97
15 0.99 1.98 2.98
20 0.99 1.98 2.98

EOC with Gaussian points

5
10 0.99 1.98 2.97
15 0.99 1.98 2.98
20 0.99 1.98 2.98

comparing the results obtained by the two sets of collocation points, it turns out that the absolute
errors are smaller in the case of the Gaussian points.

5. Conclusion
In this work, we have applied an iterative method for solving nonlinear third-kind Volterra integral
(−1)
equation (1) in the polynomial spline space Sm−1 (N ). The current method is easy to implement,
has high order of convergence and the coefficients of approximate solution are determined by using
explicit formula without solving any algebraic system. The numerical examples confirm that the
method is convergent, very effective and convenient.
Having tested two sets of collocation points (equidistant and Gaussian ones), we have concluded
from the numerical results that the use of Gaussian collocation points improves the accuracy but does
not change the convergence order of the method. Further research on this kind of problems will be
conducted by generalizing the work done to two-dimensional partial integro-differential equations
with weakly singular kernels.

Acknowledgments
The third author (P. Lima) acknowledges financial support from FCT, through projects UIDB/04621/2020,
UIDP/04621/2020.

Disclosure statement
No potential conflict of interest was reported by the author(s).

ORCID
Azzeddine Bellour http://orcid.org/0000-0002-3644-0804

References
[1] S.S. Allaei, Z.W. Yang, and H. Brunner, Existence, uniqueness and regularity of solutions to a class of third-kind
Volterra integral equations, J. Integral Equ. Appl. 27 (2015), pp. 325–342.
[2] M.A. Bartoshevich, On a heat conduction problem, Inz̆.-Fiz. Z̆. 2 (1975), pp. 340–346. (in Russian).
[3] M.A. Bartoshevich, Expansion in one orthogonal system of Watson operators for solving heat conduction problems,
Inz̆.-Fiz. Z̆. 3 (1975), pp. 516–522. (in Russian).
[4] H. Brunner, Collocation Methods for Volterra Integral and Related Functional Differential Equations, Cambridge
University Press, Cambridge, 2004.
[5] H. Brunner, Volterra Integral Equations: An Introduction to Theory and Applications, Cambridge University Press,
Cambridge, 2017.
[6] P. Grandits, A regularity theorem for a Volterra equation of the third kind, J. Integral Equ. Appl. 4 (2008), pp. 1–21.
[7] K. Kherchouche, A. Bellour, and P. Lima, Iterative collocation method for solving a class of nonlinear weakly singular
Volterra integral equations, Dolomites Res. Notes Approx 14 (2021), pp. 33–41.
14 K. KHERCHOUCHE ET AL.

[8] H. Laib, A. Bellour, and A. Boulmerka, Taylor collocation method for a system of nonlinear Volterra delay integro-
differential equations with application to COVID-19 epidemic, Int. J. Comput. 99 (2022), pp. 852–876.
[9] H. Laib, A. Boulmerka, A. Bellour, and F. Birem, Numerical solution of two-dimensional linear and nonlinear
Volterra integral equations using Taylor collocation method, J. Comput. Appl. Math. 417 (2023), pp. 1–21.
[10] M.J. Lighthill, Contributions to the theory of heat transfer through a laminar boundary layer, Proc. R. Soc. Lond.
Ser. A. 202 (1950), pp. 359–377.
[11] X. Ma and C. Huang, Recovery of high order accuracy in spectral collocation method for linear Volterra integral
equations of the third-kind with non-smooth solutions, J. Comput. Appl. Math. 392 (2021), pp. 113458.
[12] S. Nemati, P. Lima, and D. Torres, Numerical solution of a class of third-kind Volterra integral equations using Jacobi
wavelets, Numer. Algorithms 86 (2021), pp. 675–691.
[13] S. Nemati and P. Lima, Numerical solution of a third-kind Volterra integral equation using an operational matrix
technique, in 2018, European Control Conference (ECC), Limassol. (2018), pp. 3215–3220.
[14] K. Rouibah, A. Bellour, P. Lima, and E. Rawashdeh, Iterative continuous collocation method for solving nonlinear
volterra integral equations, Kragujev. J. Math. 46 (2022), pp. 635–648.
[15] Y. Rostami, A new wavelet method for solving a class of nonlinear partial integro-differential equations with weakly
singular kernels, Math. Sci. 16 (2022), pp. 225–235.
[16] Y. Rostami, An effective computational approach based on hermite wavelet Galerkin for solving parabolic Volterra
partial integro differential equations and its convergence analysis, Math. Model. Anal. 28 (2023), pp. 163–179.
[17] Y. Rostami, Two approximated techniques for solving of system of two-dimensional partial integral differential
equations with weakly singular kernels, Comput. Appl. Math. 40 (2021), pp. 1–31.
[18] Y. Rostami, Operational matrix of two dimensional Chebyshev wavelets and its applications in solving nonlinear
partial integro-differential equations, Eng. Comput. 38 (2020), pp. 745–761.
[19] Y. Rostami and K. Maleknejad, Comparison of two hybrid functions for numerical solution of nonlinear mixed
partial integro-differential equations, Iran. J. Sci. Technol. Trans. Sci. 46 (2022), pp. 645–658.
[20] Y. Rostami and K. Maleknejad, The solution of the nonlinear mixed partial integro-differential equation via two-
dimensional hybrid functions, Mediterr. J. Math. 19 (2022), pp. 1–20.
[21] H. Song, Y. Xiao, and M. Chen, Collocation methods for third-kind Volterra integral equations with proportional
delays, Appl. Math. Comput. 388 (2020), pp. 1–11.
[22] H. Song, Z. Yang, and H. Brunner, Analysis of collocation methods for nonlinear Volterra integral equations of the
third kind, Calcolo 56 (2019), pp. 1–29.
[23] G. Vainikko, Cordial Volterra integral equations 1, Numer. Funct. Anal. Optim. 30 (2009), pp. 1145–1172.
[24] G. Vainikko, Spline collocation-interpolation method for linear and nonlinear cordial Volterra integral equations,
Numer. Funct. Anal. Optim. 32 (2011), pp. 83–109.
[25] Z. Wang, M. Zhou, and Y. Guo, An hp-version Jacobi spectral collocation method for the third-kind VIEs, J. Sci.
Comput. 87 (2021), pp. 1–15.