Fixed Point Theory and Applications
Volume 2010 (2010), Article ID 734181, 21 pages
doi:10.1155/2010/734181
Research Article

Robustness of Mann Type Algorithm with Perturbed Mapping for Nonexpansive Mappings in Banach Spaces

1Department of Mathematics, Shanghai Normal University, Shanghai 200234, China
2Scientific Computing Key Laboratory, Shanghai Universities, Shanghai, China
3Department of Information Management, Cheng Shiu University, no.840, Chengcing Road, Niaosong Township, Kaohsiung County 833, Taiwan
4Department of Applied Mathematics, National Sun Yat-Sen University, Kaohsiung 804, Taiwan

Received 30 October 2009; Accepted 10 January 2010

Academic Editor: Simeon Reich

Copyright © 2010 L. C. Ceng et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

The purpose of this paper is to study the robustness of Mann type algorithm in the sense that approximately perturbed mapping does not alter the convergence of Mann type algorithm. It is proven that Mann type algorithm with perturbed mapping 𝑥 𝑛 + 1 = 𝜆 𝑛 𝑥 𝑛 + ( 1 𝜆 𝑛 ) ( 𝑇 𝑥 𝑛 + 𝑒 𝑛 ) 𝜆 𝑛 𝜇 𝑛 𝐹 ( 𝑥 𝑛 ) remains convergent in a Banach space setting where 𝜆 𝑛 , 𝜇 𝑛 [ 0 , 1 ] , 𝑇 a nonexpansive mapping, 𝑒 𝑛 , 𝑛 = 0 , 1 , , errors and 𝐹 a strongly accretive and strictly pseudocontractive mapping.

1. Introduction

Let 𝐶 be a nonempty closed convex subset of a real Banach space 𝑋 , and 𝑇 𝐶 𝐶 a nonexpansive mapping (i.e., 𝑇 𝑥 𝑇 𝑦 𝑥 𝑦 for all 𝑥 , 𝑦 𝐶 ). We use F i x ( 𝑇 ) to denote the set of fixed points of 𝑇 ; that is, F i x ( 𝑇 ) = { 𝑥 𝐶 𝑇 𝑥 = 𝑥 } . Throughout this paper it is assumed that F i x ( 𝑇 ) . Construction of fixed points of nonlinear mappings is an important and active research area. In particular, iterative methods for finding fixed points of nonexpansive mappings have received vast investigation since these methods find applications in a variety of applied areas of variational inequality problems, equilibrium problems, inverse problems, partial differential equations, image recovery, and signal processing (see, e.g., [117]).

In 1953, Mann [18] introduced an iterative algorithm which is now referred to as Mann's algorithm. Most of the literature deals with the special case of the general Mann's algorithm; that is, for an arbitrary initial guess 𝑥 0 𝐶 , the sequence { 𝑥 𝑛 } is generated by the recursive manner

𝑥 𝑛 + 1 = 𝜆 𝑛 𝑥 𝑛 + 1 𝜆 𝑛 𝑇 𝑥 𝑛 , 𝑛 0 , ( 1 . 1 ) where 𝐶 is a convex subset of a Banach space 𝑋 , 𝑇 𝐶 𝐶 is a mapping and { 𝜆 𝑛 } is a sequence in the interval [ 0 , 1 ] . It is well known that Mann's algorithm can be employed to approximate fixed points of nonexpansive mappings and zeros of (strongly) accretive mappings in Hilbert spaces and Banach spaces. Many convergence theorems have been announced and published by a large number of authors. A typical convergence result in connection with the Mann's algorithm is the following theorem proved by Ishikawa [19].

Theorem IS (see [19])
Let 𝐶 be a nonempty subset of a Banach space 𝑋 and let 𝑇 𝐶 𝑋 be a nonexpansive mapping. Let { 𝜆 𝑛 } be a real sequence satisfying the following control conditions:(a) 𝑛 = 0 𝜆 𝑛 = ;(b) 0 𝜆 𝑛 𝜆 < 1 . Let { 𝑥 𝑛 } be defined by (1.1) such that 𝑥 𝑛 𝐶 for all 𝑛 0 . If { 𝑥 𝑛 } is bounded then 𝑥 𝑛 𝑇 𝑥 𝑛 0 as 𝑛 .

The interest and importance of Theorem IS lie in the fact that strong or weak convergence of the sequence { 𝑥 𝑛 } can be achieved under certain appropriate assumptions imposed on the mapping 𝑇 , the domain 𝐷 ( 𝑇 ) or the space 𝑋 . In an infinite-dimensional space 𝑋 , Mann's algorithm has only weak convergence, in general. In fact, it is known that if the sequence { 𝜆 𝑛 } is such that 𝑛 = 0 𝜆 𝑛 ( 1 𝜆 𝑛 ) = , then Mann's algorithm converges weakly to a fixed point of 𝑇 provided that the underlying space 𝑋 is a Hilbert space or more general, a uniformly convex Banach space which has a Fréchet differentiable norm or satisfies Opial's property. See, for example, [20, 21].

The study of the robustness of Mann's algorithm is initiated by Combettes [22] where he considered a parallel projection method algorithm in signal synthesis (design and recovery) problems in a real Hilbert space 𝐻 as follows:

𝑥 𝑛 + 1 = 𝑥 𝑛 + 𝜆 𝑛 𝑚 𝑖 = 1 𝑤 𝑖 𝑃 𝑖 𝑥 𝑛 + 𝑐 𝑖 , 𝑛 𝑥 𝑛 , ( 1 . 2 ) where for each 𝑖 , 𝑃 𝑖 ( 𝑥 ) is the (nearest point) projection of a signal 𝑥 𝐻 onto a closed convex subset 𝑆 𝑖 of 𝐻 [23] ( 𝑆 𝑖 is interpreted as the 𝑖 th constraint set of the signals), { 𝜆 𝑛 } 𝑛 0 is a sequence of relaxation parameters in ( 0 , 2 ) , { 𝑤 𝑖 } 𝑚 𝑖 = 1 are strictly positive weights such that 𝑚 𝑖 = 1 𝑤 𝑖 = 1 , and 𝑐 𝑖 , 𝑛 stands for the error made in computing the projection onto 𝑆 𝑖 at iteration 𝑛 . Then he proved the following robustness result of algorithm (1.2).

Theorem 1.1 (see [22]). Assume 𝐺 = 𝑚 𝑖 = 1 𝑆 𝑖 . Assume also (i) 𝑛 = 0 𝜆 𝑛 ( 2 𝜆 𝑛 ) = , (ii) 𝑛 = 0 𝜆 𝑛 𝑚 𝑖 = 1 𝑤 𝑖 𝑐 𝑖 , 𝑛 < . Then the sequence { 𝑥 𝑛 } generated by (1.2) converges weakly to a point in 𝐺 .
Define a mapping 𝑇 𝐻 𝐻 by
𝑇 𝑥 = 2 𝑚 𝑖 = 1 𝑤 𝑖 𝑃 𝑖 ( 𝑥 ) 𝑥 , 𝑥 𝐻 , ( 1 . 3 ) and put 𝑒 𝑛 = 2 𝑚 𝑖 = 1 𝑤 𝑖 𝑐 𝑖 , 𝑛 , 𝛼 𝑛 𝜆 = 𝑛 2 ( 0 , 1 ) , 𝑛 0 . ( 1 . 4 ) Since 𝑃 𝑖 is a projection, the mapping 𝑉 𝑖 = 2 𝑃 𝑖 𝐼 is nonexpansive. Thus 𝑃 𝑖 = ( 𝐼 + 𝑉 𝑖 ) / 2 and algorithm (1.2) can be rewritten as 𝑥 𝑛 + 1 = 1 𝛼 𝑛 𝑥 𝑛 + 𝛼 𝑛 𝑇 𝑥 𝑛 + 𝑒 𝑛 , ( 1 . 5 ) where 𝑇 is given by (1.3). Note that 𝑇 can be written as 𝑇 = 𝑚 𝑖 = 1 𝑤 𝑖 𝑉 𝑖 and thus 𝑇 is nonexpansive. Note also that F i x ( 𝑇 ) = 𝑚 𝑖 = 1 F i x ( 𝑉 𝑖 ) = 𝐺 . Furthermore, conditions (i) and (ii) in Theorem 1.1 can be stated as ( i ) 𝑛 = 0 𝛼 𝑛 ( 1 𝛼 𝑛 ) = ( i i ) 𝑛 = 0 𝛼 𝑛 𝑒 𝑛 < .

Very early, some authors had considered Mann iterations in the setting of uniformly convex Banach spaces and established strong and weak convergence results for Mann iterations; see, e.g., [24, 25]. Recently, Kim and Xu [26] studied the robustness of Mann's algorithm for nonexpansive mappings in Banach spaces and extended Combettes' robustness result (Theorem 1.1 above) for projections from Hilbert spaces to the setting of uniformly convex Banach spaces.

Theorem 1.2 (see [26, Theorem 3.3]). Assume that 𝑋 is a uniformly convex Banach space. Assume, in addition, that either 𝑋 has the KK- property or 𝑋 satisfies Opial's property. Let 𝑇 𝑋 𝑋 be a nonexpansive mapping such that F i x ( 𝑇 ) . Given an initial guess 𝑥 0 𝑋 . Let { 𝑥 𝑛 } be generated by the following perturbed Mann's algorithm: 𝑥 𝑛 + 1 = 1 𝛼 𝑛 𝑥 𝑛 + 𝛼 𝑛 𝑇 𝑥 𝑛 + 𝑒 𝑛 , 𝑛 0 , ( 1 . 6 ) where { 𝛼 𝑛 } ( 0 , 1 ) and { 𝑒 𝑛 } 𝑋 satisfy the following properties: (i) 𝑛 = 0 𝛼 𝑛 ( 1 𝛼 𝑛 ) = ,(ii) 𝑛 = 0 𝛼 𝑛 𝑒 𝑛 < . Then the sequence { 𝑥 𝑛 } converges weakly to a fixed point of 𝑇 .

Further, Kim and Xu [26] also extended the robustness to nonexpansive mappings which are defined on subsets of a Hilbert space and to accretive operators.

Theorem 1.3 (see [26, Theorem 4.1]). Let 𝐶 be a nonempty closed convex subset of a Hilbert space 𝐻 and 𝑇 𝐶 𝐶 a nonexpansive mapping with F i x ( 𝑇 ) . Let { 𝑥 𝑛 } be generated from an arbitrary 𝑥 0 𝐶 via one of the following algorithms (1.7) and (1.7): 𝑥 𝑛 + 1 = 1 𝛼 𝑛 𝑥 𝑛 + 𝛼 𝑛 𝑃 𝐶 𝑇 𝑥 𝑛 + 𝑒 𝑛 𝑥 , 𝑛 0 , 𝑛 + 1 = 𝑃 𝐶 1 𝛼 𝑛 𝑥 𝑛 + 𝛼 𝑛 T 𝑥 𝑛 + 𝑒 𝑛 , 𝑛 0 , ( 1 . 7 ) where the sequences { 𝛼 𝑛 } ( 0 , 1 ) and { 𝑒 𝑛 } 𝑋 are such that (i) 𝑛 = 0 𝛼 𝑛 ( 1 𝛼 𝑛 ) = ,(ii) 𝑛 = 0 𝛼 𝑛 𝑒 𝑛 < . Then { 𝑥 𝑛 } converges weakly to a fixed point of 𝑇 .

Theorem 1.4 (see [26, Theorem 5.1]). Let 𝑋 be a uniformly convex Banach space. Assume in addition that either 𝑋 has the KK- property or 𝑋 satisfies Opial's property. Let 𝐴 be an 𝑚 -accretive operator in 𝑋 such that 𝐴 1 ( 0 ) . Moreover, assume that { 𝛼 𝑛 } ( 0 , 1 ) , { 𝑐 𝑛 } ( 0 , ) , and { 𝑒 𝑛 } 𝑋 satisfy the following properties: (i) 𝑛 = 0 𝛼 𝑛 ( 1 𝛼 𝑛 ) = ;(ii) 𝑛 = 0 𝛼 𝑛 𝑒 𝑛 < ;(iii) 0 < 𝑐 < 𝑐 𝑛 < 𝑐 < , where 𝑐 and 𝑐 are two constants;(iv) 𝑛 = 0 | 𝑐 𝑛 + 1 𝑐 𝑛 | < . Then the sequence { 𝑥 𝑛 } generated from an arbitrary 𝑥 0 𝑋 by 𝑥 𝑛 + 1 = 1 𝛼 𝑛 𝑥 𝑛 + 𝛼 𝑛 𝐽 𝑐 𝑛 𝑥 𝑛 + 𝑒 𝑛 , 𝑛 0 , ( 1 . 8 ) converges weakly to a point of 𝐴 1 ( 0 ) .

Let 𝑋 be a real reflexive Banach space. Let 𝑇 𝑋 𝑋 be a nonexpansive mapping with F i x ( 𝑇 ) . Assume that 𝐹 𝑋 𝑋 is 𝛿 -strongly accretive and 𝜆 -strictly pseudocontractive with 𝛿 + 𝜆 1 where 𝛿 , 𝜆 ( 0 , 1 ) . In this paper, inspired by Combettes' robustness result (Theorem 1.1 above) and Kim and Xu's robustness result (Theorem 1.2 above) we will consider the robustness of Mann type algorithm with perturbed mapping, which generates, from an arbitrary initial guess 𝑥 0 𝑋 , a sequence { 𝑥 𝑛 } by the recursive manner

𝑦 𝑛 = 𝜆 𝑛 𝑥 𝑛 + 1 𝜆 𝑛 𝑇 𝑥 𝑛 + 𝑒 𝑛 , 𝑥 𝑛 + 1 = 𝑦 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 𝑥 𝑛 , 𝑛 0 , ( 1 . 9 ) where { 𝜆 𝑛 } , { 𝜇 𝑛 } , and { 𝑒 𝑛 } are sequences in [ 0 , 1 ] and in 𝑋 , respectively, such that

(i) 𝑛 = 0 𝜆 𝑛 ( 1 𝜆 𝑛 ) = ;(ii) 𝑛 = 0 ( 1 𝜆 𝑛 ) 𝑒 𝑛 < ;(iii) 𝑛 = 0 𝜆 𝑛 𝜇 𝑛 < .

More precisely, we will prove under conditions (i)–(iii) the weak convergence of the algorithm (1.9) in a uniformly convex Banach space 𝑋 which either has the KK-property or satisfies Opial's property. This theorem extends Kim and Xu's robustness result (Theorem 1.2 above) from Mann's algorithm (1.6) with errors to Mann type algorithm (1.9) with perturbed mapping 𝐹 . On the other hand, we also extend Kim and Xu's robustness results (Theorems 1.3 and 1.4 above) for nonexpansive mappings which are defined on subsets of a Hilbert space and accretive operators in a uniformly convex Banach space from Mann's algorithm with errors to Mann type algorithm with perturbed mapping.

Throughout this paper, we use the following notations:

(i) stands for weak convergence and for strong convergence,(ii) 𝜔 𝑤 ( { 𝑥 𝑛 } ) = { 𝑥 𝑥 𝑛 𝑘 𝑥 } denotes the weak 𝜔 -limit set of { 𝑥 𝑛 } .

2. Preliminaries

Let 𝑋 be a real Banach space. Recall that the norm of 𝑋 is said to be Fréchet differentiable if, for each 𝑥 𝑆 ( 𝑋 ) , the unit sphere of 𝑋 , the limit

l i m 𝑡 0 𝑥 + 𝑡 𝑦 𝑥 𝑡 ( 2 . 1 ) exists and is attained uniformly in 𝑦 𝑆 ( 𝑋 ) . In this case, we have

1 2 𝑥 2 1 + , 𝐽 ( 𝑥 ) 2 𝑥 + 2 1 2 𝑥 2 ) + , 𝐽 ( 𝑥 ) + 𝑏 ( ( 2 . 2 ) for all 𝑥 , 𝑋 , where 𝐽 is the normalized duality map from 𝑋 to 𝑋 defined by

𝑥 𝑗 ( 𝑥 ) = 𝑋 𝑥 , 𝑥 = 𝑥 2 = 𝑥 2 , ( 2 . 3 ) , is the duality pairing between 𝑋 and X , and 𝑏 is a function defined on [ 0 , ) such that l i m 𝑡 0 𝑏 ( 𝑡 ) / 𝑡 = 0 . Examples of Banach spaces which have a Fréchet differentiable norm include 𝑙 𝑝 and 𝐿 𝑝 for 1 < 𝑝 < (these spaces are actually uniformly smooth).

It is known that a Banach space 𝑋 is Fréchet differentiable if and only if the duality map 𝐽 is single-valued and norm-to-norm continuous.

We need the concept of the KK-property. A Banach space 𝑋 is said to have the KK-property (the Kadec-Klee property) if, for any sequence { 𝑧 𝑛 } in 𝑋 , the conditions 𝑧 𝑛 𝑧 and 𝑧 𝑛 𝑧 imply that 𝑧 𝑛 𝑧 . It is known [27, Remark 3.2] that the dual space of a reflexive Banach space with a Fréchet differentiable norm has the KK-property.

Recall now that 𝑋 satisfies Opial's property [28] provided that, for each sequence { 𝑥 𝑛 } in 𝑋 , the condition 𝑥 𝑛 𝑥 implies

l i m s u p 𝑛 𝑥 𝑛 𝑥 < l i m s u p 𝑛 𝑥 𝑛 𝑦 , 𝑦 𝑋 , 𝑦 𝑥 . ( 2 . 4 ) It is known [28] that each 𝑙 𝑝 ( 1 𝑝 < ) enjoys this property, while 𝐿 𝑝 does not unless 𝑝 = 2 . It is known [29] that any separable Banach space can be equivalently renormed so that it satisfies Opial's property.

Recall that a Banach space 𝑋 is said to be uniformly convex if, for each 0 < 𝜀 2 , the modulus of convexity 𝛿 𝑋 ( 𝜀 ) of 𝑋 defined by

𝛿 𝑋 ( 𝜀 ) = i n f 1 𝑥 + 𝑦 2 𝑥 , 𝑦 𝑋 , 𝑥 1 , 𝑦 1 , a n d 𝑥 𝑦 𝜀 ( 2 . 5 ) is positive.

We need an inequality characterization of uniform convexity.

Lemma 2.1 (see [30]). Given a number 𝑟 > 0 . A real Banach space 𝑋 is uniformly convex if and only if there exists a continuous strictly increasing function 𝜑 [ 0 , ) [ 0 , ) , 𝜑 ( 0 ) = 0 , such that 𝜆 𝑥 + ( 1 𝜆 ) 𝑦 2 𝜆 𝑥 2 + ( 1 𝜆 ) 𝑦 2 ) 𝜆 ( 1 𝜆 ) 𝜑 ( 𝑥 𝑦 ( 2 . 6 ) for all 𝜆 [ 0 , 1 ] and 𝑥 , 𝑦 𝑋 such that 𝑥 𝑟 and 𝑦 𝑟 .

A mapping 𝐹 with domain 𝐷 ( 𝐹 ) and range 𝑅 ( 𝐹 ) in 𝑋 is called 𝛿 -strongly accretive if for each 𝑥 , 𝑦 𝐷 ( 𝐹 ) ,

𝐹 𝑥 𝐹 𝑦 , 𝑗 ( 𝑥 𝑦 ) 𝛿 𝑥 𝑦 2 , 𝑗 ( 𝑥 𝑦 ) 𝐽 ( 𝑥 𝑦 ) ( 2 . 7 )

for some 𝛿 ( 0 , 1 ) . 𝐹 is called 𝜆 -strictly pseudocontractive if for each 𝑥 , 𝑦 𝐷 ( 𝐹 ) ,

𝐹 𝑥 𝐹 𝑦 , 𝑗 ( 𝑥 𝑦 ) 𝑥 𝑦 2 𝜆 𝑥 𝑦 ( 𝐹 𝑥 𝐹 𝑦 ) 2 , 𝑗 ( 𝑥 𝑦 ) 𝐽 ( 𝑥 𝑦 ) ( 2 . 8 ) for some 𝜆 ( 0 , 1 ) . It is easy to see that (2.8) can be rewritten as

( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 , 𝑗 ( 𝑥 𝑦 ) 𝜆 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 2 . ( 2 . 9 )

The following proposition will be used frequently throughout this paper. For the sake of completeness, we include its proof.

Proposition 2.2. Let 𝑋 be a real Banach space and 𝐹 𝐷 ( 𝐹 ) 𝑋 a mapping. (i)If 𝐹 is a 𝜆 -strictly pseudocontractive, then 𝐹 is Lipschitz continuous with constant ( 1 + 1 / 𝜆 ) . (ii)If 𝐹 is 𝛿 -strongly accretive and 𝜆 -strictly pseudocontractive with 𝛿 + 𝜆 1 , then for each fixed 𝜇 [ 0 , 1 ] , the mapping 𝐼 𝜇 𝐹 has the following property: ( 𝐼 𝜇 𝐹 ) 𝑥 ( 𝐼 𝜇 𝐹 ) 𝑦 1 𝜇 1 1 𝛿 𝜆 𝑥 𝑦 , 𝑥 , 𝑦 𝐷 ( 𝐹 ) . ( 2 . 1 0 )

Proof. (i) From (2.9), we derive 𝜆 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 2 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 , 𝑗 ( 𝑥 𝑦 ) ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 𝑥 𝑦 , 𝑗 ( 𝑥 𝑦 ) 𝐽 ( 𝑥 𝑦 ) ( 2 . 1 1 ) which implies that 1 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 𝜆 𝑥 𝑦 . ( 2 . 1 2 ) Thus 1 𝐹 𝑥 𝐹 𝑦 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 + 𝑥 𝑦 1 + 𝜆 𝑥 𝑦 , ( 2 . 1 3 ) and so 𝐹 is Lipschitz continuous with constant ( 1 + 1 / 𝜆 ) .
(ii) From (2.8) and (2.9), we obtain
𝜆 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 2 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 , 𝑗 ( 𝑥 𝑦 ) = 𝑥 𝑦 2 𝐹 𝑥 𝐹 𝑦 , 𝑗 ( 𝑥 𝑦 ) ( 1 𝛿 ) 𝑥 𝑦 2 . ( 2 . 1 4 ) Since 𝛿 + 𝜆 1 ( 1 𝛿 ) / 𝜆 ( 0 , 1 ] , we have ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 1 𝛿 𝜆 𝑥 𝑦 , 𝑥 , 𝑦 𝐷 ( 𝐹 ) . ( 2 . 1 5 ) Consequently, for each fixed 𝜇 [ 0 , 1 ] , we have [ ] 𝑥 𝑦 𝜇 ( 𝐹 ( 𝑥 ) 𝐹 ( 𝑦 ) ) = ( 1 𝜇 ) ( 𝑥 𝑦 ) + 𝜇 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 ( 1 𝜇 ) 𝑥 𝑦 + 𝜇 ( 𝐼 𝐹 ) 𝑥 ( 𝐼 𝐹 ) 𝑦 ( 1 𝜇 ) 𝑥 𝑦 + 𝜇 1 𝛿 𝜆 = 𝑥 𝑦 1 𝜇 1 1 𝛿 𝜆 𝑥 𝑦 , 𝑥 , 𝑦 𝐷 ( 𝐹 ) . ( 2 . 1 6 ) This shows that inequality (2.10) holds.

Proposition 2.3. Let 𝑋 be a uniformly convex Banach space and 𝐶 a nonempty closed convex subset of 𝑋 . (i)Reference [31] (demiclosedness principle). If 𝑇 𝐶 𝐶 is a nonexpansive mapping and if { 𝑥 𝑛 } is a sequence in 𝐶 such that 𝑥 𝑛 𝑥 and ( 𝐼 𝑇 ) 𝑥 𝑛 𝑦 , then ( 𝐼 𝑇 ) 𝑥 = 𝑦 .(ii)Reference [32]. If 𝐶 is also bounded, then there exists a continuous, strictly increasing, and convex function 𝛾 [ 0 , ) [ 0 , ) (depending only on the diameter of 𝐶 ) with 𝛾 ( 0 ) = 0 and such that 𝛾 ( 𝑇 ( 𝜆 𝑥 + ( 1 𝜆 ) 𝑦 ) ( 𝜆 𝑇 𝑥 + ( 1 𝜆 ) 𝑇 𝑦 ) ) 𝑥 𝑦 𝑇 𝑥 𝑇 𝑦 ( 2 . 1 7 ) for all 𝑥 , 𝑦 𝐶 , 𝜆 [ 0 , 1 ] , and nonexpansive mappings 𝑇 𝐶 𝑋 .

We also use the following elementary lemma.

Lemma 2.4 (see [33]). Let { 𝑎 𝑛 } and { 𝑏 𝑛 } be sequences of nonnegative real numbers such that 𝑛 = 0 𝑏 𝑛 < and 𝑎 𝑛 + 1 𝑎 𝑛 + 𝑏 𝑛 for all 𝑛 1 . Then l i m 𝑛 𝑎 𝑛 exists.

3. Robustness of Mann Type Algorithm with Perturbed Mapping

Let 𝑋 be a real reflexive Banach space. Let 𝑇 𝑋 𝑋 be a nonexpansive mapping with F i x ( 𝑇 ) . Assume that 𝐹 𝑋 𝑋 is 𝛿 -strongly accretive and 𝜆 -strictly pseudocontractive with 𝛿 + 𝜆 1 . We now discuss the robustness of Mann type algorithm with perturbed mapping, which generates, from an initial guess 𝑥 0 𝑋 , a sequence { 𝑥 𝑛 } as follows:

𝑦 𝑛 = 𝜆 𝑛 𝑥 𝑛 + 1 𝜆 𝑛 𝑇 𝑥 𝑛 + 𝑒 𝑛 , 𝑥 𝑛 + 1 = 𝑦 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 𝑥 𝑛 , 𝑛 0 , ( 3 . 1 ) where { 𝜆 𝑛 } , { 𝜇 𝑛 } , and { 𝑒 𝑛 } are sequences in [ 0 , 1 ] and in 𝑋 , respectively, such that

(i) 𝑛 = 0 𝜆 𝑛 ( 1 𝜆 𝑛 ) = ;(ii) n = 0 ( 1 𝜆 𝑛 ) 𝑒 𝑛 < ;(iii) 𝑛 = 0 𝜆 𝑛 𝜇 𝑛 < .

We remark that Mann type algorithm with perturbed mapping is based on Mann iteration method and steepest-descent method. Indeed, in algorithm (3.1), one iteration step “ y n = 𝜆 n x n + ( 1 𝜆 n ) ( T x n + e n ) ” is taken from Mann iteration method, and another iteration step “ x n + 1 = y n 𝜆 n 𝜇 n F ( x n ) ” is taken from steepest-descent method.

We first discuss some properties of algorithm (3.1).

Lemma 3.1. Let { 𝑥 𝑛 } be generated by algorithm (3.1) and let 𝑝 F i x ( 𝑇 ) . Then l i m 𝑛 𝑥 𝑛 𝑝 exists.

Proof. We have 𝑥 𝑛 + 1 = 𝑦 𝑝 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 𝑥 𝑛 = 𝜆 𝑝 𝑛 𝑥 𝑛 + 1 𝜆 𝑛 𝑇 𝑥 𝑛 + 𝑒 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 𝑥 𝑛 𝜆 𝑝 𝑛 𝐼 𝜇 𝑛 𝐹 𝑥 𝑛 + 𝑝 1 𝜆 𝑛 𝑇 𝑥 𝑛 + 𝑝 1 𝜆 𝑛 𝑒 𝑛 1 𝜆 𝑛 𝑇 𝑥 𝑛 𝑝 + 𝜆 𝑛 𝐼 𝜇 𝑛 𝐹 𝑥 𝑛 + 𝑝 1 𝜆 𝑛 𝑒 𝑛 1 𝜆 𝑛 𝑥 𝑛 𝑝 + 𝜆 𝑛 𝐼 𝜇 𝑛 𝐹 𝑥 𝑛 𝐼 𝜇 𝑛 𝐹 𝑝 + 𝐼 𝜇 𝑛 𝐹 + 𝑝 𝑝 1 𝜆 𝑛 𝑒 𝑛 1 𝜆 𝑛 𝑥 𝑛 𝑝 + 𝜆 𝑛 1 𝜇 𝑛 1 1 𝛿 𝜆 𝑥 𝑛 𝑝 + 𝜆 𝑛 𝜇 𝑛 𝐹 ( 𝑝 ) + 1 𝜆 𝑛 𝑒 𝑛 = 1 𝜆 𝑛 𝜇 𝑛 1 1 𝛿 𝜆 𝑥 𝑛 𝑝 + 𝜆 𝑛 𝜇 𝑛 ( 𝐹 𝑝 ) + 1 𝜆 𝑛 𝑒 𝑛 𝑥 𝑛 𝑝 + 𝜆 𝑛 𝜇 𝑛 ( 𝐹 𝑝 ) + 1 𝜆 𝑛 𝑒 𝑛 . ( 3 . 2 ) The conclusion of the lemma is a consequence of Lemma 2.4.

Proposition 3.2. Let 𝑋 be a uniformly convex Banach space. (i)For all 𝑝 , 𝑞 F i x ( 𝑇 ) and 0 𝑡 1 , l i m 𝑛 𝑡 𝑥 𝑛 + ( 1 𝑡 ) 𝑝 𝑞 exists.(ii)If, in addition, the dual space 𝑋 of 𝑋 has the 𝐾 𝐾 -property, then the weak 𝜔 -limit set of { 𝑥 𝑛 } , 𝜔 𝑤 ( 𝑥 𝑛 ) , is a singleton.

Proof. (i) For integers 𝑛 , 𝑚 1 , define the mappings 𝑇 𝑛 and 𝑆 𝑛 , 𝑚 as follows:
𝑇 𝑛 𝑥 = 𝜆 𝑛 𝑥 + 1 𝜆 𝑛 𝑇 𝑥 + 1 𝜆 𝑛 𝑒 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 ( 𝑥 ) , 𝑥 𝑋 , ( 3 . 3 ) and 𝑆 𝑛 , 𝑚 = 𝑇 𝑛 + 𝑚 1 𝑇 𝑛 + 𝑚 2 𝑇 𝑛 . It is easy to see that 𝑥 𝑛 + 𝑚 = 𝑆 𝑛 , 𝑚 𝑥 𝑛 . First, let us show that 𝑇 𝑛 and 𝑆 𝑛 , 𝑚 are nonexpansive. Indeed, for all 𝑥 , 𝑦 𝑋 , using Proposition 2.2 no. (ii) we have 𝑇 𝑛 𝑥 𝑇 𝑛 𝑦 = 𝜆 𝑛 𝑥 + 1 𝜆 𝑛 𝑇 𝑥 + 1 𝜆 𝑛 𝑒 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 𝜆 ( 𝑥 ) 𝑛 𝑦 + 1 𝜆 𝑛 𝑇 𝑦 + 1 𝜆 𝑛 𝑒 𝑛 𝜆 𝑛 𝜇 𝑛 = 𝜆 𝐹 ( 𝑦 ) 𝑛 𝐼 𝜇 𝑛 𝐹 𝑥 + 1 𝜆 𝑛 𝜆 𝑇 𝑥 𝑛 𝐼 𝜇 𝑛 𝐹 𝑦 + 1 𝜆 𝑛 𝑇 𝑦 𝜆 𝑛 𝐼 𝜇 𝑛 𝐹 𝑥 𝐼 𝜇 𝑛 𝐹 𝑦 + 1 𝜆 𝑛 𝑇 𝑥 𝑇 𝑦 𝜆 𝑛 1 𝜇 𝑛 1 1 𝛿 𝜆 𝑥 𝑦 + 1 𝜆 𝑛 = 𝑥 𝑦 1 𝜆 𝑛 𝜇 𝑛 1 1 𝛿 𝜆 𝑥 𝑦 . ( 3 . 4 ) Thus 𝑇 𝑛 𝑋 𝑋 is nonexpansive (due to 𝜆 𝑛 𝜇 𝑛 [ 0 , 1 ] ) and so is 𝑆 𝑛 , 𝑚 .
Second, let us show that for each 𝑣 F i x ( 𝑇 ) ,
𝑆 𝑛 , 𝑚 𝑣 𝑣 𝑛 + 𝑚 1 𝑗 = 𝑛 1 𝜆 𝑗 𝑒 𝑗 + 𝜆 𝑗 𝜇 𝑗 . 𝐹 ( 𝑣 ) ( 3 . 5 ) Indeed, whenever 𝑚 = 1 , we have 𝑆 𝑛 , 1 = 𝑇 𝑣 𝑣 𝑛 = 𝜆 𝑣 𝑣 𝑛 𝑣 + 1 𝜆 𝑛 𝑇 𝑣 + 1 𝜆 𝑛 𝑒 𝑛 𝜆 𝑛 𝜇 𝑛 𝐹 ( 𝑣 ) 𝑣 1 𝜆 𝑛 𝑒 𝑛 + 𝜆 𝑛 𝜇 𝑛 ( = 𝐹 𝑣 ) 𝑛 + 1 1 𝑗 = 𝑛 1 𝜆 𝑗 𝑒 𝑗 + 𝜆 𝑗 𝜇 𝑗 . 𝐹 ( 𝑣 ) ( 3 . 6 ) This implies that inequality (3.5) holds for 𝑚 = 1 . Assume that inequality (3.5) holds for some 𝑚 1 . Consider the case of 𝑚 + 1 . Observe that 𝑆 𝑛 , 𝑚 + 1 = 𝑇 𝑣 𝑣 𝑛 + 𝑚 𝑆 𝑛 , 𝑚 = 𝜆 𝑣 𝑣 𝑛 + 𝑚 𝑆 𝑛 , 𝑚 𝑣 + 1 𝜆 𝑛 + 𝑚 𝑇 𝑆 𝑛 , 𝑚 𝑣 + 1 𝜆 𝑛 + 𝑚 𝑒 𝑛 + 𝑚 𝜆 𝑛 + 𝑚 𝜇 𝑛 + 𝑚 𝐹 𝑆 𝑛 , 𝑚 𝑣 = 𝜆 𝑣 𝑛 + 𝑚 𝐼 𝜇 𝑛 + 𝑚 𝐹 𝑆 𝑛 , 𝑚 + 𝑣 𝑣 1 𝜆 𝑛 + 𝑚 𝑇 𝑆 𝑛 , 𝑚 + 𝑣 𝑣 1 𝜆 𝑛 + 𝑚 𝑒 𝑛 + 𝑚 1 𝜆 𝑛 + 𝑚 𝑇 𝑆 𝑛 , 𝑚 𝑣 𝑣 + 𝜆 𝑛 + 𝑚 𝐼 𝜇 𝑛 + 𝑚 𝐹 𝑆 𝑛 , 𝑚 + 𝑣 𝑣 1 𝜆 𝑛 + 𝑚 𝑒 𝑛 + 𝑚 1 𝜆 𝑛 + 𝑚 𝑆 𝑛 , 𝑚 𝑣 𝑣 + 𝜆 𝑛 + 𝑚 𝐼 𝜇 𝑛 + 𝑚 𝐹 𝑆 𝑛 , 𝑚 𝑣 𝐼 𝜇 𝑛 + 𝑚 𝐹 𝑣 + 𝐼 𝜇 𝑛 + 𝑚 𝐹 + 𝑣 𝑣 1 𝜆 𝑛 + 𝑚 𝑒 𝑛 + 𝑚 1 𝜆 𝑛 + 𝑚 𝑆 𝑛 , 𝑚 𝑣 𝑣 + 𝜆 𝑛 + 𝑚 1 𝜇 𝑛 + 𝑚 1 1 𝛿 𝜆 𝑆 𝑛 , 𝑚 𝑣 𝑣 + 𝜆 𝑛 + 𝑚 𝜇 𝑛 + 𝑚 𝐹 ( 𝑣 ) + 1 𝜆 𝑛 + 𝑚 𝑒 𝑛 + 𝑚 = 1 𝜆 𝑛 + 𝑚 𝜇 𝑛 + 𝑚 1 1 𝛿 𝜆 𝑆 𝑛 , 𝑚 𝑣 𝑣 + 𝜆 𝑛 + 𝑚 𝜇 𝑛 + 𝑚 𝐹 ( 𝑣 ) + 1 𝜆 𝑛 + 𝑚