Posts by Stefan Soltuz

Abstract

Authors

Stefan M. Soltuz
Tiberiu Popoviciu Institute of Numerical Analysis , Romanian Academy

Keywords

Krasnoselskij iteration; Mann iteration; Ishikawa iteration; quasi-contractive operators

Paper coordinates

Ş.M. Şoltuz, The equivalence between Krasnoselskij, Mann, Ishikawa, Noor and multistep iterations, Math. Commun. 12 (2007): 1, 53-61.

PDF

About this paper

Journal

Mathematical Communications

Publisher Name

Sveučilište Josipa Jurja Strossmayera u Osijeku Fakultet primijenjene matematike i informatike
Trg Ljudevita Gaja 6, Osijek

Print ISSN

1331-0623

Online ISSN

1848-8013

google scholar link

??

Paper (preprint) in HTML form

The equivalence between Krasnoselskij, Mann, Ishikawa, Noor and multistep iterations

Ştefan M. Şoltuz
Abstract

We prove that Krasnoselskij, Mann, Ishikawa, Noor and multistep iterations are equivalent when applied to quasi-contractive operators.

Key words: Krasnoselskij iteration, Mann iteration, Ishikawa iteration, quasi-contractive operators

AMS subject classifications: 47H10
Received February 21, 2007
Accepted March 12, 2007

1. Introduction

Let XX be a real Banach space, DD a nonempty, convex subset of XX, and TT a selfmap of DD, let x0=u0∈Dx_{0}=u_{0}\in D. The Mann iteration, (see [5]), is defined by

un+1=(1−αn)​un+αn​T​un,u_{n+1}=\left(1-\alpha_{n}\right)u_{n}+\alpha_{n}Tu_{n}, (1)

where {αn}⊂(0,1)\left\{\alpha_{n}\right\}\subset(0,1). The Krasnoselskij iteration, (see [4]), is defined by

xn+1=(1−λ)​xn+λ​T​xn,x_{n+1}=(1-\lambda)x_{n}+\lambda Tx_{n}, (2)

where λ∈(0,1)\lambda\in(0,1).
Definition 1. [7] The operator T:X→XT:X\rightarrow X satisfies condition ZZ (or is a quasicontraction) if and only if there exist real numbers a,b,ca,b,c satisfying 0<a<1,0<b,c<1/20<a<1,0<b,c<1/2 such that for each pair x,yx,y in XX, at least one condition is true

  • •

    (z1)​‖T​x−T​y‖≤a​‖x−y‖\left(z_{1}\right)\|Tx-Ty\|\leq a\|x-y\|,

  • •

    (z2)​‖T​x−T​y‖≤b​(‖x−T​x‖+‖y−T​y‖)\left(z_{2}\right)\|Tx-Ty\|\leq b(\|x-Tx\|+\|y-Ty\|),

  • •

    (z3)​‖T​x−T​y‖≤c​(‖x−T​y‖+‖y−T​x‖)\left(z_{3}\right)\|Tx-Ty\|\leq c(\|x-Ty\|+\|y-Tx\|).

It is known, see Rhoades [8], that (z1),(z2)\left(z_{1}\right),\left(z_{2}\right) and (z3)\left(z_{3}\right) are independent conditions. Note that a map satisfying condition ZZ is independent, see Rhoades [7], of the class of strongly pseudocontractive maps.

00footnotetext: *Institute of Numerical Analysis "T. Popoviciu", P.O. Box 68-1, 400110 Cluj-Napoca, Romania, e-mail: smsoltuz@gmail.com

In [ 9,?9,? ] the following conjecture was given: "if the Mann iteration converges, then so does the Ishikawa iteration". In a series of papers [9], [10], [11], [12], [13], Professor B. E. Rhoades and the author, we have given a positive answer to this Conjecture, showing the equivalence between Mann and Ishikawa iterations for strongly and uniformly pseudocontractive maps.

In [2], the following open question was given: "are Krasnoselskij iteration and Mann iteration equivalent (in the sense of [9]) for enough large classes of mappings?" We shall give a positive answer to this question: if Krasnoselskij iteration converges, then Mann (and the corresponding Ishikawa iteration) also converges and conversely, dealing with maps satisfying condition ZZ. Note that Professor B. E. Rhoades and the author have already given a positive answer in [15] for the class of pseudocontractive maps.

Lemma 1 [[18]]. Let {an}\left\{a_{n}\right\} be a nonnegative sequence which satisfies the following inequality

an+1≤(1−λn)​an+σn,a_{n+1}\leq\left(1-\lambda_{n}\right)a_{n}+\sigma_{n}, (3)

where λn∈(0,1),∀n≥n0,∑n=1∞λn=∞\lambda_{n}\in(0,1),\forall n\geq n_{0},\sum_{n=1}^{\infty}\lambda_{n}=\infty, and σn=o​(λn)\sigma_{n}=o\left(\lambda_{n}\right). Then limn→∞an=0\lim_{n\rightarrow\infty}a_{n}=0.

2. Main results

Let F​(T)F(T) denote the fixed point set with respect to DD for the map TT. Suppose that x∗∈F​(T)x^{*}\in F(T).

Theorem 1. Let XX be a normed space, DD a nonempty, convex, closed subset of XX and T:D→DT:D\rightarrow D an operator satisfying condition ZZ. If u0=x0∈Du_{0}=x_{0}\in D, then the following are true: if the Mann iteration (1) converges to x∗x^{*}, then the Krasnoselskij iteration (2) converges to x∗x^{*}. Conversely, if the Krasnoselskij iteration (2) converges to x∗x^{*}, then the Mann iteration (1) converges to x∗x^{*}, provided that αn≥A>0,∀n∈ℕ\alpha_{n}\geq A>0,\forall n\in\mathbb{N}.

Proof. Consider x,y∈Dx,y\in D. Since TT satisfies condition ZZ, at least one of the conditions from (z1),(z2)\left(z_{1}\right),\left(z_{2}\right) and (z3)\left(z_{3}\right) is satisfied. If (z2)\left(z_{2}\right) holds, then

‖T​x−T​y‖\displaystyle\|Tx-Ty\| ≤b​(‖x−T​x‖+‖y−T​y‖)\displaystyle\leq b(\|x-Tx\|+\|y-Ty\|)
≤b​(‖x−T​x‖+(‖y−x‖+‖x−T​x‖+‖T​x−T​y‖)),\displaystyle\leq b(\|x-Tx\|+(\|y-x\|+\|x-Tx\|+\|Tx-Ty\|)),

thus

(1−b)​‖T​x−T​y‖≤b​‖x−y‖+2​b​‖x−T​x‖.(1-b)\|Tx-Ty\|\leq b\|x-y\|+2b\|x-Tx\|.

From 0≤b<10\leq b<1 one obtains,

‖T​x−T​y‖≤b1−b​‖x−y‖+2​b1−b​‖x−T​x‖.\|Tx-Ty\|\leq\frac{b}{1-b}\|x-y\|+\frac{2b}{1-b}\|x-Tx\|.

If ( z3z_{3} ) holds, then one gets,

‖T​x−T​y‖\displaystyle\|Tx-Ty\| ≤c​(‖x−T​y‖+‖y−T​x‖)\displaystyle\leq c(\|x-Ty\|+\|y-Tx\|)
≤c​(‖x−T​x‖+‖T​x−T​y‖+‖x−y‖+‖x−T​x‖),\displaystyle\leq c(\|x-Tx\|+\|Tx-Ty\|+\|x-y\|+\|x-Tx\|),

hence,

(1−c)​‖T​x−T​y‖\displaystyle(1-c)\|Tx-Ty\| ≤c​‖x−y‖+2​c​‖x−T​x‖​ i.e.\displaystyle\leq c\|x-y\|+2c\|x-Tx\|\text{ i.e. }
‖T​x−T​y‖\displaystyle\|Tx-Ty\| ≤c1−c​‖x−y‖+2​c1−c​‖x−T​x‖\displaystyle\leq\frac{c}{1-c}\|x-y\|+\frac{2c}{1-c}\|x-Tx\|

Denote

δ:=max⁡{a,b1−b,c1−c}\delta:=\max\left\{a,\frac{b}{1-b},\frac{c}{1-c}\right\}

to obtain

0≤δ<10\leq\delta<1

Finally, we get

‖T​x−T​y‖≤δ​‖x−y‖+2​δ​‖x−T​x‖,∀x,y∈D\|Tx-Ty\|\leq\delta\|x-y\|+2\delta\|x-Tx\|,\forall x,y\in D (4)

Formula (4) was obtained as in [1].
We will prove the implication (i)⇒(i​i)(i)\Rightarrow(ii). Use (1) (2) and (4) with

x:=un\displaystyle x=u_{n}
y:=yn\displaystyle y=y_{n}

to obtain

‖un+1−xn+1‖=‖xn+1−un+1‖\displaystyle\left\|u_{n+1}-x_{n+1}\right\|=\left\|x_{n+1}-u_{n+1}\right\|
=‖xn−un−λ​xn+λ​un−λ​un+αn​un+λ​T​xn−λ​T​un+λ​T​un−αn​T​un‖\displaystyle=\left\|x_{n}-u_{n}-\lambda x_{n}+\lambda u_{n}-\lambda u_{n}+\alpha_{n}u_{n}+\lambda Tx_{n}-\lambda Tu_{n}+\lambda Tu_{n}-\alpha_{n}Tu_{n}\right\|
≤(1−λ)​‖un−xn‖+|αn−λ|​‖un−T​un‖+λ​‖T​un−T​xn‖\displaystyle\leq(1-\lambda)\left\|u_{n}-x_{n}\right\|+\left|\alpha_{n}-\lambda\right|\left\|u_{n}-Tu_{n}\right\|+\lambda\left\|Tu_{n}-Tx_{n}\right\|
≤(1−λ)​‖un−xn‖+|αn−λ|​‖un−T​un‖+λ​δ​‖un−xn‖+2​λ​δ​‖un−T​un‖\displaystyle\leq(1-\lambda)\left\|u_{n}-x_{n}\right\|+\left|\alpha_{n}-\lambda\right|\left\|u_{n}-Tu_{n}\right\|+\lambda\delta\left\|u_{n}-x_{n}\right\|+2\lambda\delta\left\|u_{n}-Tu_{n}\right\|
=(1−λ​(1−δ))​‖un−xn‖+(|αn−λ|+2​λ​δ)​‖un−T​un‖.\displaystyle=(1-\lambda(1-\delta))\left\|u_{n}-x_{n}\right\|+\left(\left|\alpha_{n}-\lambda\right|+2\lambda\delta\right)\left\|u_{n}-Tu_{n}\right\|.

Denote

an\displaystyle a_{n} :=‖un−xn‖\displaystyle=\left\|u_{n}-x_{n}\right\|
λn\displaystyle\lambda_{n} :=λ​(1−δ)⊂(0,1)\displaystyle=\lambda(1-\delta)\subset(0,1)
σn\displaystyle\sigma_{n} :=(|αn−λ|+2​λ​δ)​‖un−T​un‖\displaystyle=\left(\left|\alpha_{n}-\lambda\right|+2\lambda\delta\right)\left\|u_{n}-Tu_{n}\right\|

Since limn→∞‖un−x∗‖=0,T\lim_{n\rightarrow\infty}\left\|u_{n}-x^{*}\right\|=0,T satisfies condition ZZ, and x∗∈F​(T)x^{*}\in F(T), from (4) one has

0\displaystyle 0 ≤‖un−T​un‖\displaystyle\leq\left\|u_{n}-Tu_{n}\right\|
≤‖un−x∗‖+‖x∗−T​un‖\displaystyle\leq\left\|u_{n}-x^{*}\right\|+\left\|x^{*}-Tu_{n}\right\|
≤(δ+1)​‖un−x∗‖→0​ as ​n→∞\displaystyle\leq(\delta+1)\left\|u_{n}-x^{*}\right\|\rightarrow 0\text{ as }n\rightarrow\infty

Hence limn→∞‖un−T​un‖=0\lim_{n\rightarrow\infty}\left\|u_{n}-Tu_{n}\right\|=0; that is σn=o​(λn)\sigma_{n}=o\left(\lambda_{n}\right). Lemma 1 leads to limn→∞‖un−xn‖=\lim_{n\rightarrow\infty}\left\|u_{n}-x_{n}\right\|= 0 . Use

0≤‖x∗−xn‖≤‖un−x∗‖+‖xn−un‖0\leq\left\|x^{*}-x_{n}\right\|\leq\left\|u_{n}-x^{*}\right\|+\left\|x_{n}-u_{n}\right\|

to deduce

limn→∞xn=x∗.\lim_{n\rightarrow\infty}x_{n}=x^{*}.

We will prove (i​i)⇒(i)(ii)\Rightarrow(i). That is, if Krasnoselskij iteration converges, then Mann iteration does converge. Use (4) with

x:=xn,\displaystyle x=x_{n},
y:=un,\displaystyle y=u_{n},

to obtain

‖xn+1−un+1‖\displaystyle\left\|x_{n+1}-u_{n+1}\right\|
=‖xn−un−αn​xn+αn​un+αn​xn−λ​xn+λ​T​xn−αn​T​xn+αn​T​xn−αn​T​un‖\displaystyle=\left\|x_{n}-u_{n}-\alpha_{n}x_{n}+\alpha_{n}u_{n}+\alpha_{n}x_{n}-\lambda x_{n}+\lambda Tx_{n}-\alpha_{n}Tx_{n}+\alpha_{n}Tx_{n}-\alpha_{n}Tu_{n}\right\|
=‖(1−αn)​(xn−un)+(αn−λ)​xn−(αn−λ)​xn​T​xn+αn​(T​xn−T​un)‖\displaystyle=\left\|\left(1-\alpha_{n}\right)\left(x_{n}-u_{n}\right)+\left(\alpha_{n}-\lambda\right)x_{n}-\left(\alpha_{n}-\lambda\right)x_{n}Tx_{n}+\alpha_{n}\left(Tx_{n}-Tu_{n}\right)\right\|
≤(1−αn)​‖xn−un‖+|αn−λ|​‖xn−T​xn‖+αn​‖T​xn−T​un‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\left|\alpha_{n}-\lambda\right|\left\|x_{n}-Tx_{n}\right\|+\alpha_{n}\left\|Tx_{n}-Tu_{n}\right\|
≤(1−αn)​‖xn−un‖+|αn−λ|​‖xn−T​xn‖+αn​δ​‖xn−un‖+2​αn​δ​‖xn−T​xn‖\displaystyle\leq\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\left|\alpha_{n}-\lambda\right|\left\|x_{n}-Tx_{n}\right\|+\alpha_{n}\delta\left\|x_{n}-u_{n}\right\|+2\alpha_{n}\delta\left\|x_{n}-Tx_{n}\right\|
=(1−αn​(1−δ))​‖xn−un‖+(|αn−λ|+2​αn​δ)​‖xn−T​xn‖.\displaystyle=\left(1-\alpha_{n}(1-\delta)\right)\left\|x_{n}-u_{n}\right\|+\left(\left|\alpha_{n}-\lambda\right|+2\alpha_{n}\delta\right)\left\|x_{n}-Tx_{n}\right\|.

Denote

an\displaystyle a_{n} :=‖xn−un‖\displaystyle=\left\|x_{n}-u_{n}\right\|
λn\displaystyle\lambda_{n} :=αn​(1−δ)⊂(0,1)\displaystyle=\alpha_{n}(1-\delta)\subset(0,1)
σn\displaystyle\sigma_{n} :=(|αn−λ|+2​αn​δ)​‖xn−T​xn‖\displaystyle=\left(\left|\alpha_{n}-\lambda\right|+2\alpha_{n}\delta\right)\left\|x_{n}-Tx_{n}\right\|

Since limn→∞‖xn−x∗‖=0,T\lim_{n\rightarrow\infty}\left\|x_{n}-x^{*}\right\|=0,T satisfies condition ZZ, and x∗∈F​(T)x^{*}\in F(T), from (4) one has,

0\displaystyle 0 ≤‖xn−T​xn‖\displaystyle\leq\left\|x_{n}-Tx_{n}\right\|
≤‖xn−x∗‖+‖x∗−T​xn‖\displaystyle\leq\left\|x_{n}-x^{*}\right\|+\left\|x^{*}-Tx_{n}\right\|
≤(δ+1)​‖xn−x∗‖→0​ as ​n→∞,\displaystyle\leq(\delta+1)\left\|x_{n}-x^{*}\right\|\rightarrow 0\text{ as }n\rightarrow\infty,

Hence limn→∞‖xn−T​xn‖=0\lim_{n\rightarrow\infty}\left\|x_{n}-Tx_{n}\right\|=0, that is σn=o​(λn)\sigma_{n}=o\left(\lambda_{n}\right). Lemma 1 leads to limn→∞‖xn−un‖=\lim_{n\rightarrow\infty}\left\|x_{n}-u_{n}\right\|= 0 . Thus,

‖x∗−un‖≤‖xn−un‖+‖xn−x∗‖→0​ as ​n→∞.\left\|x^{*}-u_{n}\right\|\leq\left\|x_{n}-u_{n}\right\|+\left\|x_{n}-x^{*}\right\|\rightarrow 0\text{ as }n\rightarrow\infty.

The Ishikawa iteration is defined (see [3]) by

xn+1\displaystyle x_{n+1} =(1−αn)​xn+αn​T​yn,\displaystyle=\left(1-\alpha_{n}\right)x_{n}+\alpha_{n}Ty_{n}, (5)
yn\displaystyle y_{n} =(1−βn)​xn+βn​T​xn,\displaystyle=\left(1-\beta_{n}\right)x_{n}+\beta_{n}Tx_{n},

where {αn}⊂(0,1),{βn}⊂[0,1)\left\{\alpha_{n}\right\}\subset(0,1),\left\{\beta_{n}\right\}\subset[0,1).
The following result is from [17].
Theorem 2 [[17]]. Let XX be a normed space, DD a nonempty, convex, closed subset of XX and T:D→DT:D\rightarrow D an operator satisfying condition ZZ. If u0=x0∈Du_{0}=x_{0}\in D, then the following are equivalent:
(i) the Mann iteration (1) converges to x∗x^{*},
(ii) the Ishikawa iteration (5) converges to x∗x^{*}.

Theorems 1 and 2 lead to the following corollary.
Corollary 1. Let XX be a normed space, DD a nonempty, convex, closed subset of XX and T:D→DT:D\rightarrow D an operator satisfying condition ZZ. If u0=x0∈D,αn≥A>0,∀n∈ℕu_{0}=x_{0}\in D,\alpha_{n}\geq A>0,\forall n\in\mathbb{N}, then the following are equivalent:
(i) the Mann iteration (1) converges to x∗x^{*},
(ii) the Ishikawa iteration (5) converges to x∗x^{*}.
(iii) the Krasnoselskij iteration (2) converges to x∗x^{*}.

3. Further results

For v1∈Dv_{1}\in D, Noor introduced in [6] the following three-step procedure,

tn\displaystyle t_{n} =(1−γn)​vn+γn​T​vn,\displaystyle=\left(1-\gamma_{n}\right)v_{n}+\gamma_{n}Tv_{n}, (6)
wn\displaystyle w_{n} =(1−βn)​vn,+βn​T​tn,\displaystyle=\left(1-\beta_{n}\right)v_{n},+\beta_{n}Tt_{n},
vn+1\displaystyle v_{n+1} =(1−αn)​vn+αn​T​wn.\displaystyle=\left(1-\alpha_{n}\right)v_{n}+\alpha_{n}Tw_{n}.

The multi-step procedure of arbitrary fixed order p≥2p\geq 2, see [14], is defined by

ynp−1\displaystyle y_{n}^{p-1} =(1−βnp−1)​xn+βnp−1​T​xn,\displaystyle=\left(1-\beta_{n}^{p-1}\right)x_{n}+\beta_{n}^{p-1}Tx_{n}, (7)
yni\displaystyle y_{n}^{i} =(1−βni)​xn+βni​T​yni+1,i=1,…,p−2;\displaystyle=\left(1-\beta_{n}^{i}\right)x_{n}+\beta_{n}^{i}Ty_{n}^{i+1},i=1,\ldots,p-2;
xn+1\displaystyle x_{n+1} =(1−αn)​xn+αn​T​yn1,\displaystyle=\left(1-\alpha_{n}\right)x_{n}+\alpha_{n}Ty_{n}^{1},

where {αn}⊂(0,1),{βni}⊂[0,1),1≤i≤p−1\left\{\alpha_{n}\right\}\subset(0,1),\left\{\beta_{n}^{i}\right\}\subset[0,1),1\leq i\leq p-1.
We shall generalize the above Theorem 2, see also [17], by proving that (7) and (1) are equivalent.

Theorem 3. Let XX be a normed space, DD a nonempty, convex, closed subset of XX and T:D→DT:D\rightarrow D an operator satisfying condition ZZ. If u0=x0∈Du_{0}=x_{0}\in D, then the following are equivalent:
(i) the Mann iteration (1) converges to x∗x^{*},
(ii) the iteration (7) converges to x∗x^{*}.

Proof. We shall use (4) :

‖T​x−T​y‖≤δ​‖x−y‖+2​δ​‖x−T​x‖,∀x,y∈D.\|Tx-Ty\|\leq\delta\|x-y\|+2\delta\|x-Tx\|,\forall x,y\in D.

We will prove the implication (i)⇒(i​i)(i)\Rightarrow(ii). Suppose that limn→∞un=x∗\lim_{n\rightarrow\infty}u_{n}=x^{*}. Using limn→∞‖xn−un‖=0\lim_{n\rightarrow\infty}\left\|x_{n}-u_{n}\right\|=0, and 0≤‖x∗−xn‖≤‖un−x∗‖+‖xn−un‖0\leq\left\|x^{*}-x_{n}\right\|\leq\left\|u_{n}-x^{*}\right\|+\left\|x_{n}-u_{n}\right\| we get

limn→∞xn=x∗\lim_{n\rightarrow\infty}x_{n}=x^{*}

Using now (1) (7) and (4) with

x:=un,\displaystyle x=u_{n},
y:=yn1,\displaystyle y=y_{n}^{1},

we have

‖un+1−xn+1‖≤\displaystyle\left\|u_{n+1}-x_{n+1}\right\|\leq ‖(1−αn)​(un−xn)+αn​(T​un−T​yn1)‖\displaystyle\left\|\left(1-\alpha_{n}\right)\left(u_{n}-x_{n}\right)+\alpha_{n}\left(Tu_{n}-Ty_{n}^{1}\right)\right\| (8)
≤\displaystyle\leq (1−αn)​‖un−xn‖+αn​‖T​un−T​yn1‖\displaystyle\left(1-\alpha_{n}\right)\left\|u_{n}-x_{n}\right\|+\alpha_{n}\left\|Tu_{n}-Ty_{n}^{1}\right\|
≤\displaystyle\leq (1−αn)​‖un−xn‖+αn​δ​‖un−yn1‖+\displaystyle\left(1-\alpha_{n}\right)\left\|u_{n}-x_{n}\right\|+\alpha_{n}\delta\left\|u_{n}-y_{n}^{1}\right\|+
+2​αn​δ​‖un−T​un‖.\displaystyle+2\alpha_{n}\delta\left\|u_{n}-Tu_{n}\right\|.

Using (4) with x:=un,y:=yn1x:=u_{n},y:=y_{n}^{1}, we have

‖un−yn1‖≤\displaystyle\left\|u_{n}-y_{n}^{1}\right\|\leq ‖(1−βn1)​(un−xn)+βn1​(un−T​xn)‖\displaystyle\left\|\left(1-\beta_{n}^{1}\right)\left(u_{n}-x_{n}\right)+\beta_{n}^{1}\left(u_{n}-Tx_{n}\right)\right\| (9)
≤\displaystyle\leq (1−βn1)​‖un−xn‖+βn1​‖un−T​xn‖\displaystyle\left(1-\beta_{n}^{1}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}^{1}\left\|u_{n}-Tx_{n}\right\|
≤\displaystyle\leq (1−βn1)​‖un−xn‖+βn1​‖un−T​un‖+\displaystyle\left(1-\beta_{n}^{1}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}^{1}\left\|u_{n}-Tu_{n}\right\|+
+βn1​‖T​un−T​xn‖\displaystyle+\beta_{n}^{1}\left\|Tu_{n}-Tx_{n}\right\|
≤\displaystyle\leq (1−βn1)​‖un−xn‖+βn1​‖un−T​un‖+\displaystyle\left(1-\beta_{n}^{1}\right)\left\|u_{n}-x_{n}\right\|+\beta_{n}^{1}\left\|u_{n}-Tu_{n}\right\|+
+βn1​δ​‖un−xn‖+2​δ​βn1​‖un−T​un‖\displaystyle+\beta_{n}^{1}\delta\left\|u_{n}-x_{n}\right\|+2\delta\beta_{n}^{1}\left\|u_{n}-Tu_{n}\right\|
=\displaystyle= (1−βn1​(1−δ))​‖un−xn‖+\displaystyle\left(1-\beta_{n}^{1}(1-\delta)\right)\left\|u_{n}-x_{n}\right\|+
+βn1​‖un−T​un‖​(1+2​δ).\displaystyle+\beta_{n}^{1}\left\|u_{n}-Tu_{n}\right\|(1+2\delta).

Relations (8) and (9) lead to

‖un+1−xn+1‖≤\displaystyle\left\|u_{n+1}-x_{n+1}\right\|\leq (1−αn)​‖un−xn‖+\displaystyle\left(1-\alpha_{n}\right)\left\|u_{n}-x_{n}\right\|+ (10)
+αn​δ​(1−βn1​(1−δ))​‖un−xn‖+\displaystyle+\alpha_{n}\delta\left(1-\beta_{n}^{1}(1-\delta)\right)\left\|u_{n}-x_{n}\right\|+
+αn​βn1​δ​‖un−T​un‖​(1+2​δ)+\displaystyle+\alpha_{n}\beta_{n}^{1}\delta\left\|u_{n}-Tu_{n}\right\|(1+2\delta)+
+αn​δ​‖un−yn‖\displaystyle+\alpha_{n}\delta\left\|u_{n}-y_{n}\right\|
=\displaystyle= (1−αn​(1−δ​(1−βn1​(1−δ))))​‖un−xn‖+\displaystyle\left(1-\alpha_{n}\left(1-\delta\left(1-\beta_{n}^{1}(1-\delta)\right)\right)\right)\left\|u_{n}-x_{n}\right\|+
+αn​δ​‖un−T​un‖​(βn1​(1+2​δ)+2​δ).\displaystyle+\alpha_{n}\delta\left\|u_{n}-Tu_{n}\right\|\left(\beta_{n}^{1}(1+2\delta)+2\delta\right).

Denote by

an\displaystyle a_{n} :=‖un−xn‖\displaystyle=\left\|u_{n}-x_{n}\right\|
λn\displaystyle\lambda_{n} :=αn​(1−δ​(1−βn1​(1−δ)))⊂(0,1),\displaystyle=\alpha_{n}\left(1-\delta\left(1-\beta_{n}^{1}(1-\delta)\right)\right)\subset(0,1),
σn\displaystyle\sigma_{n} :=αn​δ​‖un−T​un‖​(βn1​(1+2​δ)+2​δ).\displaystyle=\alpha_{n}\delta\left\|u_{n}-Tu_{n}\right\|\left(\beta_{n}^{1}(1+2\delta)+2\delta\right).

Since limn→∞‖un−x∗‖=0,T\lim_{n\rightarrow\infty}\left\|u_{n}-x^{*}\right\|=0,T satisfies condition ZZ, and x∗∈F​(T)x^{*}\in F(T), from (4) we obtain

0\displaystyle 0 ≤‖un−T​un‖\displaystyle\leq\left\|u_{n}-Tu_{n}\right\|
≤‖un−x∗‖+‖x∗−T​un‖\displaystyle\leq\left\|u_{n}-x^{*}\right\|+\left\|x^{*}-Tu_{n}\right\|
≤(δ+1)​‖un−x∗‖→0​ as ​n→∞.\displaystyle\leq(\delta+1)\left\|u_{n}-x^{*}\right\|\rightarrow 0\text{ as }n\rightarrow\infty.

Hence limn→∞‖un−T​un‖=0\lim_{n\rightarrow\infty}\left\|u_{n}-Tu_{n}\right\|=0; that is σn=o​(λn)\sigma_{n}=o\left(\lambda_{n}\right). Lemma 1 leads to limn→∞‖un−xn‖=\lim_{n\rightarrow\infty}\left\|u_{n}-x_{n}\right\|= 0 .

We will prove now that if multistep iteration converges then Mann iteration does. Using (4) with

x:=yn1,\displaystyle x=y_{n}^{1},
y:=un,\displaystyle y=u_{n},

we obtain

‖xn+1−un+1‖≤\displaystyle\left\|x_{n+1}-u_{n+1}\right\|\leq ‖(1−αn)​(xn−un)+αn​(T​yn1−T​un)‖\displaystyle\left\|\left(1-\alpha_{n}\right)\left(x_{n}-u_{n}\right)+\alpha_{n}\left(Ty_{n}^{1}-Tu_{n}\right)\right\| (11)
≤\displaystyle\leq (1−αn)​‖xn−un‖+αn​‖T​yn1−T​un‖\displaystyle\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}\left\|Ty_{n}^{1}-Tu_{n}\right\|
≤\displaystyle\leq (1−αn)​‖xn−un‖+αn​δ​‖yn1−un‖+\displaystyle\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}\delta\left\|y_{n}^{1}-u_{n}\right\|+
+2​αn​δ​‖yn1−T​yn1‖.\displaystyle+2\alpha_{n}\delta\left\|y_{n}^{1}-Ty_{n}^{1}\right\|.

The following relation holds

‖yn1−un‖≤\displaystyle\left\|y_{n}^{1}-u_{n}\right\|\leq ‖(1−βn1)​(xn−un)+βn1​(T​xn−un)‖\displaystyle\left\|\left(1-\beta_{n}^{1}\right)\left(x_{n}-u_{n}\right)+\beta_{n}^{1}\left(Tx_{n}-u_{n}\right)\right\| (12)
≤\displaystyle\leq (1−βn1)​‖xn−un‖+βn1​‖T​xn−un‖\displaystyle\left(1-\beta_{n}^{1}\right)\left\|x_{n}-u_{n}\right\|+\beta_{n}^{1}\left\|Tx_{n}-u_{n}\right\|
≤\displaystyle\leq (1−βn1)​‖xn−un‖+βn1​‖T​xn−xn‖+\displaystyle\left(1-\beta_{n}^{1}\right)\left\|x_{n}-u_{n}\right\|+\beta_{n}^{1}\left\|Tx_{n}-x_{n}\right\|+
+βn1​‖xn−un‖\displaystyle+\beta_{n}^{1}\left\|x_{n}-u_{n}\right\|
≤\displaystyle\leq ‖xn−un‖+βn1​‖T​xn−xn‖.\displaystyle\left\|x_{n}-u_{n}\right\|+\beta_{n}^{1}\left\|Tx_{n}-x_{n}\right\|.

Substituting (12) in (11), we obtain

‖xn+1−un+1‖≤\displaystyle\left\|x_{n+1}-u_{n+1}\right\|\leq (1−αn)​‖xn−un‖+\displaystyle\left(1-\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+ (13)
+αn​δ​(‖xn−un‖+βn1​‖T​xn−xn‖)+\displaystyle+\alpha_{n}\delta\left(\left\|x_{n}-u_{n}\right\|+\beta_{n}^{1}\left\|Tx_{n}-x_{n}\right\|\right)+
+2​αn​δ​‖yn1−T​yn1‖\displaystyle+2\alpha_{n}\delta\left\|y_{n}^{1}-Ty_{n}^{1}\right\|
≤\displaystyle\leq (1−(1−δ)​αn)​‖xn−un‖+αn​βn1​δ​‖T​xn−xn‖+\displaystyle\left(1-(1-\delta)\alpha_{n}\right)\left\|x_{n}-u_{n}\right\|+\alpha_{n}\beta_{n}^{1}\delta\left\|Tx_{n}-x_{n}\right\|+
+2​αn​δ​‖yn1−T​yn1‖.\displaystyle+2\alpha_{n}\delta\left\|y_{n}^{1}-Ty_{n}^{1}\right\|.

Denote by

an\displaystyle a_{n} :=‖xn−un‖\displaystyle=\left\|x_{n}-u_{n}\right\|
λn\displaystyle\lambda_{n} :=αn​(1−δ)⊂(0,1)\displaystyle=\alpha_{n}(1-\delta)\subset(0,1)
σn\displaystyle\sigma_{n} :=αn​βn1​δ​‖T​xn−xn‖+2​αn​δ​‖yn1−T​yn1‖\displaystyle=\alpha_{n}\beta_{n}^{1}\delta\left\|Tx_{n}-x_{n}\right\|+2\alpha_{n}\delta\left\|y_{n}^{1}-Ty_{n}^{1}\right\|

Since limn→∞‖xn−x∗‖=0,T\lim_{n\rightarrow\infty}\left\|x_{n}-x^{*}\right\|=0,T satisfies condition ZZ, and x∗∈F​(T)x^{*}\in F(T), from (4) we obtain

0\displaystyle 0 ≤‖xn−T​xn‖\displaystyle\leq\left\|x_{n}-Tx_{n}\right\|
≤‖xn−x∗‖+‖x∗−T​xn‖\displaystyle\leq\left\|x_{n}-x^{*}\right\|+\left\|x^{*}-Tx_{n}\right\|
≤(δ+1)​‖xn−x∗‖→0​ as ​n→∞\displaystyle\leq(\delta+1)\left\|x_{n}-x^{*}\right\|\rightarrow 0\text{ as }n\rightarrow\infty

Note that βni∈[0,1),∀n≥1,1≤i≤p−1\beta_{n}^{i}\in[0,1),\forall n\geq 1,1\leq i\leq p-1, and use (4) to obtain

0\displaystyle 0 ≤‖yn1−T​yn1‖\displaystyle\leq\left\|y_{n}^{1}-Ty_{n}^{1}\right\|
≤‖yn1−x∗‖+‖x∗−T​yn1‖\displaystyle\leq\left\|y_{n}^{1}-x^{*}\right\|+\left\|x^{*}-Ty_{n}^{1}\right\|
≤(δ+1)​‖yn1−x∗‖≤(δ+1)​[(1−βn1)​‖xn−x∗‖+βn1​‖T​yn2−x∗‖]\displaystyle\leq(\delta+1)\left\|y_{n}^{1}-x^{*}\right\|\leq(\delta+1)\left[\left(1-\beta_{n}^{1}\right)\left\|x_{n}-x^{*}\right\|+\beta_{n}^{1}\left\|Ty_{n}^{2}-x^{*}\right\|\right]
≤(δ+1)​[‖xn−x∗‖+δ​‖yn2−x∗‖]\displaystyle\leq(\delta+1)\left[\left\|x_{n}-x^{*}\right\|+\delta\left\|y_{n}^{2}-x^{*}\right\|\right]
≤(δ+1)​[‖xn−x∗‖+‖yn2−x∗‖]\displaystyle\leq(\delta+1)\left[\left\|x_{n}-x^{*}\right\|+\left\|y_{n}^{2}-x^{*}\right\|\right]
≤(δ+1)​[‖xn−x∗‖+(1−βn2)​‖xn−x∗‖+βn2​‖T​yn3−x∗‖]\displaystyle\leq(\delta+1)\left[\left\|x_{n}-x^{*}\right\|+\left(1-\beta_{n}^{2}\right)\left\|x_{n}-x^{*}\right\|+\beta_{n}^{2}\left\|Ty_{n}^{3}-x^{*}\right\|\right]
≤(δ+1)​[‖xn−x∗‖+‖xn−x∗‖+‖T​yn3−x∗‖]\displaystyle\leq(\delta+1)\left[\left\|x_{n}-x^{*}\right\|+\left\|x_{n}-x^{*}\right\|+\left\|Ty_{n}^{3}-x^{*}\right\|\right]
≤(δ+1)​[2​‖xn−x∗‖+δ​‖yn3−x∗‖]\displaystyle\leq(\delta+1)\left[2\left\|x_{n}-x^{*}\right\|+\delta\left\|y_{n}^{3}-x^{*}\right\|\right]
≤(δ+1)​[2​‖xn−x∗‖+‖yn3−x∗‖]​…\displaystyle\leq(\delta+1)\left[2\left\|x_{n}-x^{*}\right\|+\left\|y_{n}^{3}-x^{*}\right\|\right]\ldots
≤(δ+1)​[(p−2)​‖xn−x∗‖+‖ynp−1−x∗‖]\displaystyle\leq(\delta+1)\left[(p-2)\left\|x_{n}-x^{*}\right\|+\left\|y_{n}^{p-1}-x^{*}\right\|\right]
≤(δ+1)​[(p−2)​‖xn−x∗‖+(1−βnp−1)​‖xn−x∗‖+βnp−1​‖T​xn−x∗‖]\displaystyle\leq(\delta+1)\left[(p-2)\left\|x_{n}-x^{*}\right\|+\left(1-\beta_{n}^{p-1}\right)\left\|x_{n}-x^{*}\right\|+\beta_{n}^{p-1}\left\|Tx_{n}-x^{*}\right\|\right]
≤(δ+1)​[(p−1)​‖xn−x∗‖+‖T​xn−x∗‖]\displaystyle\leq(\delta+1)\left[(p-1)\left\|x_{n}-x^{*}\right\|+\left\|Tx_{n}-x^{*}\right\|\right]
≤(δ+1)​[(p−1)​‖xn−x∗‖+δ​‖xn−x∗‖]\displaystyle\leq(\delta+1)\left[(p-1)\left\|x_{n}-x^{*}\right\|+\delta\left\|x_{n}-x^{*}\right\|\right]
=(δ+1)​‖xn−x∗‖​[(p−1)+δ]→0​ as ​n→∞,\displaystyle=(\delta+1)\left\|x_{n}-x^{*}\right\|[(p-1)+\delta]\rightarrow 0\text{ as }n\rightarrow\infty,

Hence limn→∞‖xn−T​xn‖=0\lim_{n\rightarrow\infty}\left\|x_{n}-Tx_{n}\right\|=0 and limn→∞‖yn1−T​yn1‖=0\lim_{n\rightarrow\infty}\left\|y_{n}^{1}-Ty_{n}^{1}\right\|=0 that is σn=o​(λn)\sigma_{n}=o\left(\lambda_{n}\right). Lemma 1 and (13) lead to limn→∞‖xn−un‖=0\lim_{n\rightarrow\infty}\left\|x_{n}-u_{n}\right\|=0. Thus, we get ‖x∗−un‖≤‖xn−un‖+‖xn−x∗‖→0\left\|x^{*}-u_{n}\right\|\leq\left\|x_{n}-u_{n}\right\|+\left\|x_{n}-x^{*}\right\|\rightarrow 0.

Theorem 3 and Corollary 1 lead to the following result.
Corollary 2. Let XX be a normed space, DD a nonempty, convex, closed subset of XX and T:D→DT:D\rightarrow D an operator satisfying condition ZZ. If the initial point is the same for all iterations, αn≥A>0,∀n∈ℕ\alpha_{n}\geq A>0,\forall n\in\mathbb{N}, then the following are equivalent:
(i) the Mann iteration (1) converges to x∗x^{*};
(ii) the Ishikawa iteration (5) converges to x∗x^{*};
(iii) the iteration (7) converges to x∗x^{*}.
(iii) the Noor iteration (6) converges to x∗x^{*},
(iv) the Krasnoselskij iteration (2) converges to x∗x^{*}.

Acknowledgment. The author is indebted to referee for carefully reading the paper and for making useful suggestions.

References

[1] V. Berinde, On the convergence of the Ishikawa iteration in the class of quasi contractive operators, Acta Math. Univ. Comenianae LXXIII(2004), 119-126.
[2] V. Berinde, M. Berinde, The fastest Krasnoselskij iteration for approximating fixed points of strictly pseudo-contractive mappings, Carpatian J. Math. 21(2005), 13-20.
[3] S. Ishikawa, Fixed points by a new iteration method, Proc. Amer. Math. Soc. 44(1974), 147-150.
[4] M. A. KrasnoselskiJ, Two remarks on the method of succesive approximations, Uspehi Mat. Nauk. 10(1955), 123-127.
[5] W. R. Mann, Mean value methods in iteration, Proc. Amer. Math. Soc. 4(1953), 506-510.
[6] M. A. Noor, New approximation schemes for general variational inequalities, J. Math. Anal. Appl. 251(2000), 217-229.
[7] B. E. Rhoades, Fixed point iterations using infinite matrices, Trans. Amer. Math. Soc. 196(1974), 161-176.
[8] B. E. Rhoades, A comparison of various definitions of contractive mappings, Trans. Amer. Math. Soc. 226(1977), 257-290.
[9] B. E. Rhoades, Ş. M. Şoltuz, On the equivalence of Mann and Ishikawa iteration methods, Int. J. Math. Math. Sci. 2003(2003), 451-459.
[10] B. E. Rhoades, Ş. M. Şoltuz, The equivalence of the Mann and Ishikawa iteration for non-Lipschitzian operators, Int. J. Math. Math. Sci. 42(2003), 2645-2652.
[11] B. E. Rhoades, Ş. M. Şoltuz, The equivalence between the convergences of Ishikawa and Mann iterations for asymptotically pseudocontractive map, J. Math. Anal. Appl. 283(2003), 681-688.
[12] B. E. Rhoades, Ş. M. Şoltuz, The equivalence of Mann and Ishikawa iteration for a Lipschitzian psi-uniformly pseudocontractive and psi-uniformly accretive maps, Tamkang J. Math., 35(2004), 235-245.
[13] B. E. Rhoades, Ş. M. Şoltuz, The equivalence between the convergences of Ishikawa and Mann iterations for asymptotically nonexpansive in the intermediate sense and strongly successively pseudocontractive maps, J. Math. Anal. Appl. 289(2004), 266-278.
[14] B. E. Rhoades, Ş. M. Şoltuz, The equivalence between Mann-Ishikawa iterations and multistep iteration, Nonlinear Analysis 58(2004), 219-228.
[15] B. E. Rhoades, Ş. M. Şoltuz, The equivalence between the Krasnoselskij, Mann and Ishikawa iterations, Revue d’analyse numerique et de theorie de l’approximation 𝟑𝟔​(2006)\mathbf{36}(2006), to appear.
[16] B. E. Rhoades, Ş. M. Şoltuz, The equivalence between the TT-stabilities of Mann and Ishikawa iterations J. Math. Anal. Appl. 318(2006), 472-475.
[17] Ş. M. Şoltuz, The equivalence of Picard, Mann and Ishikawa iterations dealing with quasi-contractive operators, Math. Comm. 10(2005), 81-89.
[18] X. Weng, Fixed point iteration for local strictly pseudocontractive mapping, Proc. Amer. Math. Soc. 113(1991), 727-731.

Related Posts

Upon the Pseudodecreasing Sequences

Abstract AuthorsStefan M. Soltuz Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy Keywords? Paper coordinatesM. Soltuz, Upon the Pseudodecreasing Sequences,…

Upon the convergence of subconvex sequences

Abstract AuthorsStefan M. Soltuz Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy Keywords? Paper coordinatesŞ.M. Şoltuz, Upon the convergence of subconvex…

A Stability For Multivalued Optimization,

Abstract AuthorsStefan M. Soltuz Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy Keywords? Paper coordinatesS. M. Soltuz, A Stability For Multivalued…

The Integral of a Multivalued Mapping

Abstract AuthorsStefan M. Soltuz Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy Keywords? Paper coordinatesS. M. Soltuz, The Integral of a…

The multivalued form of a classic result

Abstract AuthorsStefan M. Soltuz Tiberiu Popoviciu Institute of Numerical Analysis Romanian Academy Keywords? Paper coordinatesŞ.M. Şoltuz, The multivalued form of a…