1.
Introduction
There are several problems in nonlinear analysis that can be modeled by fixed point equations involving certain nonlinear operators. However, there are many fixed point equations that cannot be solved analytically, for instance cosx=x. To overcome this problem, iterative processes provide useful tools to approximate the fixed point of nonlinear operators. For instance, the equilibrium problem, the variational inequality problem, the saddle point problem, the problem of finding the roots of polynomials, signal processing, image restoration, tomography and intensity-modulated radiation therapy are some well-known problems whose solutions are approximated with some suitable iterative processes. For more details, we refer to [9,10,11,12,24,31,37,38].
The convergence of the iterative process, known as the Picard iterative process, is used to prove the well-known Banach contraction principle [5]. This principle solves a fixed point problem for contraction mappings defined on a complete metric space, and it has become a useful tool for proving the existence and approximation of solutions to nonlinear functional equations such as differential equations, partial differential equations and integral equations. To approximate the solution of linear and nonlinear equations, as well as inclusion, several authors have used various iterative approaches. On the one hand, a study of nonexpansive mappings is crucial due to:
(1) Existence of fixed points of such mappings rely on the geometric properties of the underlying space instead of compactness properties.
(2) These mappings are an important generalization of contraction mappings.
(3) This class of mappings is used as the transition operators for certain initial value problems of differential inclusions involving accretive or dissipative operators.
(4) Different problems appearing in many real life situations involve nonexpansive mappings (see Bruck[7]).
There are several iterative algorithms of practical interest which are generated using nonexpansive mappings. For instance, the successive projection approach is utilized to address intersection problems arising in tomography and signal processing. The proximity operator of a convex function is a nonexpansive mapping that acts as projection in the context of optimization. This is employed in image processing problems like total variation denoising. Nonexpansive mappings are used to model the flow of traffic and congestion dynamics. In signal processing, nonexpansive mappings are used for signal recovery and reconstruction. This is why the study of nonexpansive mappings has attracted the attention of several mathematicians. Notice that the Banach contraction principle states that the sequence generated by Picard iterative scheme converges to a unique fixed point in the case of contraction mappings. The Picard iteration, on the other hand, may not converge to a fixed point of nonexpansive mapping. For instance, if we take the mapping Tp=1−p on [0,1], then the Picard iteration does not converge to its fixed point (which is 12), for any choice of p∈[0,1] other than 12. Motivated by this fact, several authors proposed and implemented various iterative schemes to approximate fixed points of nonexpansive mappings.
Mann [25] proposed an iterative scheme to approximate the fixed point for nonexpansive mappings. The proposed scheme is given as follows: Let p1 be an initial guess, then,
where {ηn} is an appropriate sequence in (0,1) and Z+ represents the set of positive integers.
Later, Ishikawa [16] introduced an iterative scheme to estimate the fixed point of pseudo-contractive mapping. The sequence {pn} proposed by Ishikawa iterative scheme is given as: If p1 is an initial guess, then,
where {ηn} and {ρn} are appropriate sequences in (0,1). Similarly, Noor [27], Agarwal et al.[4], Abbas and Nazir [2], Thakur et al.[35], Ullah and Arshad [36] proposed different iterative schemes for approximating the solution of nonlinear problems involving operators satisfying certain contraction conditions. Let p1 be an initial guess then the following schemes are given as Table 1:
Where {ηn}, {ρn} and {σn} are appropriate sequences of parameters in (0,1).
Later, Abbas et al.[1] proposed an iterative scheme called the AA-iterative scheme that converges faster than the iterative schemes mentioned above. The sequence is defined as follows: For an initial guess p1,
where {ηn},{ρn} and {σn} are appropriate sequences in (0,1).
Most of the iterative processes for a certain class of mappings are primarily defined on Banach spaces along with some appropriate geometric structure, most frequently on uniformly convex spaces. Studying iterative processes on modular spaces is a recent trend and is attracting the attention of several researchers now. This is because modular spaces such as Orlicz spaces or Lebesgue spaces constitute a suitable framework to solve nonlinear problems arising in different branches of mathematics. Kassab and ţurcanu [18] used the Thakur et al. iterative scheme in the structure of modular spaces. Their mapping in a modular context satisfy the condition (E) in the modular version given by García-Falset et al.[14] (see also, Khan [19] and references mentioned therein). Moreover, several authors have developed and studied different iterative schemes to solve fixed point problems and nonlinear equations (for further details, we refer to [17,23,33]). Furthermore, modular spaces provide an appropriate framework to model certain nonlinear problems and hence are being considered by many authors, for instance, see [13,22,28] and references mentioned therein.
The study of generalized αm-nonexpansive mappings has become a very active area of research these days and several interesting results have been obtained in this direction (for example, [15,32]). Motivated by the work in [18,19], we obtained the approximation results using the AA -iterative scheme (1.3) for the mapping required to satisfy a modular counterpart of the generalized αm-nonexpansive mappings in [30]. Our results extend, unify and generalize the corresponding results that exist in literature.
This paper is structured as follows: Section 1 contains some introductory materials needed in the sequel. In Section 2, we reviewed the definitions regarding modular spaces and their basic properties. Section 3 deals with the mappings satisfying the modular version of generalized αm -nonexpansive mappings with an example of one of such mappings. In Section 4, we discuss the convergence of the iterative scheme defined in (1.3). The fifth section focuses on the investigation on stability and data dependence. Finally, numerical examples are presented to support the results proved herein.
2.
Preliminaries
To make the section self-contained, some basic concepts of modular spaces are presented. Most of these materials are taken from [3,20,21,26].
Definition 2.1. [20] Let V be a vector space over R. A mapping m: V→[0,+∞] is called modular if it satisfies the following: For any p,q∈V,
(1) m(p)=0 if and only if p=0,
(2) m(αp)=m(p) for |α|=1,
(3) m(αp+(1−α)q)≤m(p)+m(q), where α∈[0,1].
If for any α∈[0,1] and p,q∈V, the condition (3) is replaced with the following condition:
then m is called a convex modular.
Throughout this paper we shall presume that m is a convex modular.
Example 2.2. Let V=R and m: R→[0,+∞] be defined by m(p)=p2. Note that all the conditions of Definition 1 are satisfied. Also, m is even and convex and hence is a convex modular. Clearly, m does not satisfy the triangular inequality. Indeed, if we take p=12 and q=2, then
Definition 2.3. [26] Let m be a convex modular defined on V. The set
is called a modular space with the norm ‖.‖m defined as follows:
Definition 2.4. [20] Assume that V is a vector space and m a modular function on V. Then:
(1) A sequence {pn}⊂Vm is said to be m-convergent to p∈Vm if limn→+∞m(pn−p)=0.
(2) A sequence {pn}⊂Vm is said to be m -Cauchy if limn,r→+∞m(pn−pr)=0.
(3) We say Vm is m-complete if any m-Cauchy sequence in Vm is m-convergent.
(4) A set C⊂Vm is said to be m-closed if any sequence {pn}⊂C which is m-convergent to some point p implies that p∈C.
(5) A set C⊂Vm is said to be m-bounded if the m-diameter of C is finite.
(6) A set K⊂Vm is said to be m-compact if any sequence {pn}⊂K has a subsequence which is m -convergent to some p∈K.
(7) m satisfies the Fatou property if for any p,q,qn∈Vm,
whenever {qn} m- converges to q.
Note that the Fatou property is crucial to understand the geometric characteristics of the modular in the framework of normed spaces and modular spaces.
Definition 2.5. [20] The uniformly convex type properties of m are given as:
(a) For ϵ>0,r>0, define
If D1(r,ϵ)≠∅, let
If D1(r,ϵ)=∅, then take ρ1(r,ϵ)=1. We say that m fulfills the condition (UUC1) if for every s,ϵ>0, there exists δ1(s,ϵ)>0, depending on s and ϵ such that
(b) For ϵ>0,r>0, define
If D2(r,ϵ)≠∅, let
If D2(r,ϵ)=∅, then take ρ2(r,ϵ)=1. We say that m satisfies the condition (UUC2) if for every s,ϵ>0, there exists δ2(s,ϵ)>0, depending on s and ϵ such that
Lemma 2.6. [21] Assume that m is a convex modular which satisfies the condition (UUC1) and let {αn}∈(0,1) be a sequence bounded away from 0 and 1. Suppose there exists r>0 such that
where {pn} and {qn} are sequences in Vm. Then, limn→+∞m(pn−qn)=0.
Definition 2.7. [18] Let {pn} be a sequence in a modular space Vm. Suppose C⊂Vm is nonempty. The function ϕ: C→[0,+∞] is defined by
is known as the m-type function related to the sequence {pn}.
A sequence {xn} in C is said to be a minimizing sequence of ϕ if limn→+∞ϕ(xn)=infx∈Cϕ(x).
Example 2.8. Consider the set of real numbers which is a modular vector space with convex modular m(p)=p2. Take C=Q⊂R and the sequence {pn}={1√n},n≥1.
The m-type function in this case is
which is clearly unbounded. The minimizing sequence {xn} of ϕ is given by xn=1n,n≥1.
Lemma 2.9. [3] Suppose that Vm is m -complete and m satisfies the Fatou property. Let C be a nonempty convex and m-closed subset of Vm and ϕ: C→[0,+∞] the m-type function related to the sequence {pn} in Vm.
Assume that ϕ0=infp∈Cϕ(p)<+∞.
(a) If m satisfies the condition (UUC1), then all minimizing sequences of ϕ are m -convergent to the same point.
(b) If m satisfies the condition (UUC2) and {xn} is a minimizing sequence of ϕ, then the sequence {xn2} m-converges to a point which is independent of {xn}.
Definition 2.10. [18] Suppose that Vm is a modular space. We say that the modular m fulfills the Δ2 condition if there exists a constant J≥0 such that m(2p)≤Jm(p) for any p∈Vm. We denote the smallest such constant J by π2.
Note that the modular defined in Example 2.1 satisfies the Δ2 condition with J=4.
Lemma 2.11. [6] Let {un} and {tn} be sequences of positive real numbers that fulfill the following inequality:
where υn∈(0,1) for all n∈Z+ with +∞∑n=0υn=+∞. If limn→+∞tnυn=0, then limn→+∞un=0.
Lemma 2.12. [34] Let {un} and {tn} be sequences of nonnegative real numbers such that there exists n0 so that for n≥n0, it satisfies the following inequality:
where υn∈(0,1) for all n∈Z+ with +∞∑n=0υn=+∞. Then,
3.
Generalized αm-nonexpansive mappings
Pant and Shukla [30] introduced the class of generalized αm-nonexpansive mappings in the context of Banach spaces. Here, we adopt the definition from [30] in the framework of modular spaces.
Definition 3.1. Suppose that C is a nonempty subset of the modular space Vm. A mapping T: C→Vm is called the generalized αm-nonexpansive mappings if there exists α∈(0,1) such that for all p,q∈C,
implies that
Example 3.2. The modular m established in Example 2.2 presents R with the modular space. Take the subset of R that is C=[0,+∞). Define a mapping T: C→C as follows
Then T is a generalized αm-nonexpansive mapping. Indeed, take α=13.
Case (Ⅰ) Let p>2 and q>2. Then, we have
Case (Ⅱ) Let p>2 and q∈[0,2]. Then, we have
Case (Ⅲ) Let p∈[0,2] and q∈[0,2]. Then, we have
So, T is a generalized αm-nonexpansive mapping.
Throughout this paper we denote the set of fixed points of T by F(T).
4.
Convergence analysis
Theorem 4.1. Let C be a nonempty m-closed convex subset of a modular space Vm and T: C→C be a generalized αm-nonexpansive mapping with F(T)≠∅. Choose any p1∈C and any p∗∈F(T). If {pn} is the sequence defined by the AA-iterative scheme (1.3), and m(pj−p∗)<+∞ for some j≥1, then limn→+∞m(pn−p∗) exists for all p∗∈F(T).
Proof. Let p∗∈F(T). Since T is a generalized αm-nonexpansive mapping it satisfies the condition of m-nonexpansive given in (Definition 4.1 in [3]). Using the iterative scheme (1.3) and the convexity of m, we have
Recall that T is a generalized αm-nonexpansive mapping with T(p∗)=p∗, hence, we have
Using (4.2) in (4.1), we obtain
If
then,
Now, since T(p∗)=p∗,
Also,
and again using T(p∗)=p∗,
Putting (4.7) and (4.3) in (4.6), we get
Using (4.8) and (4.5), we have
By (4.9) and (4.4) we get
Now, taking
so, T(un)=qn. Then,
Also,
and
Using (4.13) and (4.7) in (4.12), we obtain
Putting (4.14) in (4.11), we get
Now,
and
Putting (4.17) in (4.16), we have
From (4.18) and (4.15), we obtain
This shows that {m(pn−p∗)} is decreasing and bounded below, hence, limn→+∞m(pn−p∗) exists. □
Lemma 4.2. Let C be a nonempty subset of a modular space Vm and T: C→C be a generalized αm-nonexpansive mapping. Assume that there exists a bounded sequence {pn}⊂C such that
and let ϕ be the m-type function defined by the sequence {pn}. Then T leaves the minimizing sequence invariant, i.e., if {xn} is a minimizing sequence for ϕ, then, so is {Txn}.
Proof. Assume that {pn}⊂C is such that
For any p∈C, we have
which implies that
Now, assume that {xn} is a minimizing sequence. Using (4.21), we get
This implies that
So, {T(xn)} is a minimizing sequence for ϕ. □
Lemma 4.3. Assume that C is a nonempty m-closed and convex subset of a m-complete modular space Vm and m is (UUC1) which fulfill the Δ2 condition and the Fatou property. Assume that the m-type function ϕ: C→[0,+∞] is defined by a sequence {pn} in Vm and that
Let {xn} and {yn} be two minimizing sequences for ϕ. Then,
(i) each convex combination of {xn} and {yn} is a minimizing sequence for ϕ,
(ii) limn→+∞m(xn−yn)=0.
Proof. (i) Let zn=λxn+(1−λ)yn,λ∈(0,1), n≥1. Then for any p∈C, we get
which gives
That is,
Taking limit and using the fact that {xn} and {yn} are minimizing sequences, we obtain
as required.
(ii) Note that for λ=12, zn=12xn+12yn, n≥1, we have
using (i), {zn} is a minimizing sequence and by using Lemma 2.9, each minimizing sequence m-converge to the same point, say r. Thus,
We get limn→+∞m(zn−yn)=0. Now, using Δ2 condition we have
By taking limit as n→+∞ we get the required result.
□
Theorem 4.4. Let C be a nonempty m-closed, m-bounded and convex subset of a m-complete modular space Vm. Suppose that m satisfies the condition (UUC1), the Δ2 condition and the Fatou property. Let T: C→C be a generalized αm -nonexpansive mapping and {pn} be a sequence given in the AA -iterative scheme (1.3). Then F(T)≠∅ if and only if
and {pn} is bounded.
Proof. Suppose F(T)≠∅ and p∗∈F(T). By the above Theorem 4.1, limn→+∞m(pn−p∗) exists and {pn} is bounded. Put
From (4.3), (4.10) and (4.15), we have
It follows from (4.2) that
Thus,
By taking lim inf as n→+∞, we get
From (4.26) and (4.29), we have
Now, from (4.28), we obtain that
which on taking lim inf as n→+∞ gives that
By (4.25) and (4.32), we get
From (4.31), we have
On taking lim inf as n→+∞, we obtain that
Thus, from (4.3) and (4.34), we get
Also,
Hence,
From (4.23), (4.27), (4.35) and Lemma 2.6, we get
Conversely, assume that {pn} is a bounded sequence and
Let ϕ: C→[0,+∞] be the m-type function generated by {pn} and suppose that {xn} is a minimizing sequence for ϕ converging to a point r∈C. Using Lemmas 4.2 and 4.3, we get
On the other hand using (4.20) as n→+∞, we obtain
This gives that {pn} converges to T(r). As we know, limit is always unique, we get T(r)=r. □
5.
Stability and data dependence
Ostrowski [29] established the concept of stability for a fixed point iterative technique. The analogue of the Ostrowski definition in the framework of modular spaces is given as follows:
Definition 5.1. Assume that C is a nonempty subset of a modular space Vm and {ϱn} is an approximate sequence of {pn} in C. Then, the iterative process pn+1=ℏ(T,pn) for a function ℏ, converging to a fixed point p∗ of T: C→C is called T -stable or stable w.r.t. T if the condition that
is equivalent to the condition that {ϱn} is m-convergent to p∗. Here {ℓn} is defined as
Theorem 5.2. Assume that C is a nonempty m-closed and convex subset of a m-complete modular space Vm and T: C→C is a m-contraction mapping with contraction constant c. Then, the iterative scheme defined in (1.3) is T-stable.
Proof. Let {ϱn} be an approximate sequence of {pn} in C. The sequence defined by the iteration (1.3) is:
and
We show that limn→+∞ℓn=0 if and only if
If limn→+∞ℓn=0, then, using the convexity modular m, Δ2 condition, it follows from (1.3) that
Also,
Let
then,
Since, limn→+∞ℓn=0, we get
Hence, by Lemma 2.11, we have
On the other hand, if
then, we have
This implies that limn→+∞ℓn=0. Hence, the iterative scheme (1.3) is T-stable. □
Definition 5.3. Assume that T,ˆT: C→C are two mappings. Then ˆT is known as an approximate mapping of T, if there exists ϵ>0 such that, for all p∈C, we have
Theorem 5.4. Assume that ˆT is an approximate operator of a m-contraction T with maximum acceptable error ϵ. Let {pn} be an iterative sequence generated by (1.3) and define an iterative scheme ˆpn as follows:
with real sequences {ηn}, {ρn} and {σn} in (0,1) satisfying ηnρnσn≥12 for all n∈N. If T(p∗)=p∗ and ˆT(^p∗)=^p∗, such that
then,
Proof. By convexity and the Δ2 property
Now, let
So,
Using similar arguments as in (5.4) we get
Using (5.4) and (5.6) in (5.5) we get
Now, suppose that
then,
Following arguments similar to those given above, we get
Therefore (5.7) becomes
Now, using similar arguments as in (5.4), we have
Taking
and
in Lemma 2.12, we get
Also,
Taking limn→+∞ and using the inequality (5.11) we obtain
□
6.
Numerical experiments
In this section, we present the numerical experiment for the applicability of our results. We used the MATLAB version R2018a for all of the numerical calculations. We compare the iterative scheme (1.3) with the existing methods for generalized αm- nonexpansive mapping and used the Example 3.2 to implement our results. Moreover, we take different initial guesses and parameters for comparison.
Example 6.1. Let T be the generalized αm-nonexpansive mapping defined in Example 3.2. We now discuss a numerical experiment to substantiate the convergence of iteration (1.3). Take an initial value p1=150 and take different sequences of parameters, i.e., those we used for Figure 1,
for Figure 2,
for Figure 3,
and for Figure 4,
Note that by the graphs, the convergence of iterative schemes (1.3) converges faster than the other schemes for generalized αm-nonexpansive mapping for all choices of the used parameters. On the other hand, the other iterative schemes change their convergence behaviors by changing the parameters. Moreover, we have considered different initial guesses and observed that the whole space becomes the basin of attraction for the scheme considered herein. Thus, the iterative scheme we used is superior in this framework of study.
Example 6.2. Define T: R2→R2 defined as follows:
Clearly, the mapping T is a generalized αm-nonexpansive mapping with α=14 and m=‖.‖2. We illustrate the convergence behavior of the different iterative schemes along with the iteration (1.3) and take different initial values as shown in Figures 5–8 for the convergence and comparison.
7.
Conclusions
Our aim of this work is to study the AA-iterative scheme proposed by Abbas et al. [1] to approximate the fixed points of generalized αm-nonexpansive mappings in the structure of modular spaces. We proposed adequate requirements regarding the convergence of the iterative scheme to approximate the solution of a fixed point equation involving αm-nonexpansive mappings in the framework of uniformly convex type modular spaces. Numerical examples are given and demonstrate that the AA-iterative method converges faster than certain known schemes for generalized αm-nonexpansive mappings in the context of modular spaces. For future direction, one may apply the AA-iterative algorithm in image processing problems involving αm-nonexpansive mappings such as image denoising and reconstruction. One may also enhance the convergence speed and preserve the image features as compared to existing methods.
Use of AI tools declaration
The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.
Acknowledgments
The third and fourth authors gratefully acknowledge technical and financial support from the Agency for Research and Innovation, Ministry of Education and University of Hafr Al Batin, Saudi Arabia. Moreover, all authors are thankful to the reviewers for their time spent to give critical and useful comments which helped in improving the initial version of the paper.
This research was funded by the University of Hafr Al Batin, Institutional Financial Program under project number IFP-A-2022-2-1-09.
Conflicts of interest
The authors declare that they have no conflicts of interest.