Weak Convergence of Ishikawa Iterates for
Nonexpansive Maps
Abdul Rahim Khan
1, Hafiz Fukhar-ud-din
2 ∗Abstract—We establish weak convergence of the
Ishikawa iterates of nonexpansive maps under a vari-ety of new control conditions and without employing any of the properties: (i) Opial’s property (ii) Fr´echet differentiable norm (iii) Kadec-Klee property.
Keywords: uniformly convex Banach space, nonexpan-sive map, weak convergence, Ishikawa iterates, Kadec-Klee property
1
Introduction
LetE be a real Banach space and let C be a nonempty closed convex subset of E. A map T : C → C is non-expansive if kT x−T yk ≤ kx−yk for all x, y ∈ C. We denote byF(T) the set of fixed points ofT.
Numerous problems in mathematics and physical sciences can be formulated in a fixed point problem for noexpan-sive maps.In view of practical importance of these prob-lems, methods of finding fixed points of nonexpansive maps continue to be a flourishing topic in fixed point the-ory. Iterative construction of fixed points of these maps is a fascinating field of research (see, [1, 4, 7, 9, 10]). In 1967, Browder [1] studied the iterative construction of fixed points of nonexpansive maps on closed and convex subsets of a Hilbert space (see also [3]).
For a map T of C into itself, we consider the Ishikawa iteration scheme: x1∈C,and
(
xn+1=αnT yn+ (1−αn)xn,
yn =βnT xn+ (1−βn)xn, n≥1.
(1.1)
where{αn}and{βn} are sequences in [0,1].
Set
δ(r) = inf
½
1−1
2kx+yk:kxk ≤1,kyk ≤1,kx−yk ≥r
¾
.
A Banach space E is uniformly convex if for each r ∈
(0,2],the numberδ(r)>0.
∗The author A. R. Khan is grateful to King Fahd
Univer-sity of Petroleum & Minerals and SABIC for supporting RE-SEARCH PROJECT SB10012. Address1: Department of
Mathe-matics and Statistics, King Fahd University of Petroleum and Min-erals, Dahran, 31261, Saudia Arabia Email: arahim@kfupm.edu.sa. Address2: Department of Mathematics, The Islamia
Univer-sity of Bahawalpur, Bahawalpur, 63100, Pakistan. Email: hfdin@yahoo.com
For a sequence, the symbol → (resp.⇀) denotes norm (resp. weak) convergence. The space E is said to sat-isfy : (i) Opial’s property [8] if for any sequence {xn}
in E, xn ⇀ x implies that lim supn→∞kxn−xk < lim supn→∞kxn−yk for all y ∈ E with y 6= x; (ii) Kadec-Klee property [6] if for every sequence{xn} inE,
xn ⇀ x and kxnk → kxk together imply xn → x as
n→ ∞.
LetS ={x∈E :kxk= 1}and letE∗
be the dual ofE, that is, the space of all continuous linear functionalsf on E.The norm ofE is : (iii)Gˆateaux differentiable [10] if
lim
t→0
kx+tyk − kxk
t
exists for each x and y in S and (iv) Fr´echet differen-tiable [10] if for each xin S, the above limit is attained uniformly fory∈S.
A mapping T : C → E is demiclosed at y ∈ E if for each sequence {xn} in C and each x∈E, xn ⇀ x and
T xn→y imply thatx∈C andT x=y.
One of the fundamental and celebrated results in the theory of nonexpansive maps is Browder’s demiclosed principle[1] which states that if C is a nonempty closed convex subset of a uniformly convex Banach space E, then for every nonexpansive map T : C → E, I−T is demiclosed at zero, i.e., for any {xn} ⊂ C, xn ⇀ x and
(I−T)xn→0 imply thatT x=x.
The above stated demiclosed principle has played an im-portant role in the study of approximation of fixed points of nonexpansive maps through weak(strong) convergence of certain iterates.
A suitable varient of Lemma 3.1 due to G´ornicki [5] for nonexpansive maps in uniformly convex Banach space is as follows:
Lemma 1.1. Let Cbe a nonempty bounded closed convex
subset of a uniformly convex Banach space Eand let T be a nonexpansive map of C into itself. If xn⇀ x({xn} ⊂
C, x∈C),then there exists strictly increasing convex map g: [0,∞)→[0,∞)with g(0) = 0 such that
g(kx−T xk)≤lim inf
n→∞ kxn−T xnk.
Note that Browder’s demiclosed principle is a simple con-sequence of Lemma 1.1.
Proceedings of the World Congress on Engineering and Computer Science 2010 Vol II WCECS 2010, October 20-22, 2010, San Francisco, USA
ISBN: 978-988-18210-0-3
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Tan and Xu[10] and Takahashi and Tamura[9], respec-tively, proved the following interesting results.
Theorem A. Let C be a nonempty bounded closed
con-vex subset of a uniformly concon-vex Banach space E which satisfies Opial’s condition or whose norm is Fr´echet dif-ferentiable and let T be a nonexpansive map of C into itself. Then the sequence {xn} given by (1.1) converges
weakly to a fixed point of T provided the following con-dition is satisfied:
(C1) P∞
n=1αn(1−αn) =∞,P
∞
n=1βn(1−αn)<∞ and
lim supn→∞βn<1.
Theorem B. Let C be a nonempty closed convex
sub-set of a uniformly convex Banach space E which satisfies Opial’s condition or whose norm is Fr´echet differentiable and let T be a nonexpansive map of C into itself. Sup-pose that {xn} in (1.1) satisfies the condition:
(C2) αn ∈ [a,1] and βn ∈ [a, b] or αn ∈ [a, b] and
βn∈[0, b]for some a, b∈[0,1].
Then{xn} converges weakly to a fixed point ofT.
Note that Tan and Xu’s result is applicable to the case: αn= 1−1/nandβn= 1/nfor alln≥1,while Takahashi
and Tamura’s result is applicable to the case: αn=βn =
1/2 for all n ≥ 1. Moreover, in both the results, the demiclosed principle based on strong convegence of the approximate sequence{xn−T xn}to 0 has been utilized.
Using Lemma 1.1, we establish weak convergence of the Ishikawa iterates of nonexpansive maps under a variety of new parametric control conditions and without using any of the properties: (i) Opial’s property (ii) Fr´echet differentiable norm (iii) Kadec-Klee property.
In the sequel, we need the following lemmas.
Lemma 1.2 [11, Theorem 2]. Let r > 0 be a fixed real
number. Then a Banach space E is uniformly convex if and only if there is a continuous strictly increasing convex map g : [0,∞)→[0,∞)with g(0) = 0 such that for all x, y∈Br[0] ={x∈E:kxk ≤r},
kλx+ (1−λ)yk2≤λkxk2+(1−λ)kyk2−λ(1−λ)g(kx−yk)
for all λ∈[0,1].
Lemma 1.3 [12, Lemma 2.2]. Let g : [0,∞) → [0,∞)
with g(0) = 0 be a strictly increasing map. If a se-quence {xn} in [0,∞) satisfies limn→∞g(xn) = 0, then limn→∞xn= 0.
2
Weak Convergence
We establish a pair of lemmas for the development of our convergence result.
Lemma 2.1. LetCbe a nonempty closed convex subset
of a uniformly convex Banach spaceE.LetT :C→Cbe nonexpansive map with at least one fixed point.Suppose
{xn} is given by (1.1). Then limn→∞kxn−pkexists for eachp∈F(T).
Proof. For anyp∈F(T),utilizing (1.1), we have
kxn+1−pk ≤ kαn(T yn−p) + (1−αn)(xn−p)k
≤ αnkT yn−pk+ (1−αn)kxn−pk
≤ αnkβn(T xn−p) + (1−βn)(xn−p)k
+(1−αn)kxn−pk
≤ (αnβn+αn(1−βn) + 1−αn)kxn−pk
= kxn−pk.
This proves that {kxn−pk} is a non-increasing and
bounded sequence and hence limn→∞kxn−pkexists.
Lemma 2.2. LetCbe a nonempty closed convex subset
of a uniformly convex Banach space E and let T be a nonexpansive map ofCinto itself with at least one fixed point. Let{αn} and{βn}be sequences in [0,1] and
sat-isfy one of the following three sets of conditions: (C3) :P∞
n=1αn(1−αn) =∞, lim supn→∞βn <1; (C4) :P∞
n=1βn(1−βn) =∞, lim infn→∞αn>0;
(C5) : 0≤αn ≤b <1,P
∞
n=1αn=∞, βn→0 asn→ ∞.
Then lim infn→∞kxn−T xnk= 0.
Proof. Letp∈F(T). With the help of Lemma 1.2 and
the scheme (1.1), we have:
kxn+1−pk 2
≤ kαn(T yn−p) + (1−αn)(xn−p)k
2
≤ αnkT yn−pk
2
+ (1−αn)kxn−pk
2
−αn(1−αn)g(kxn−T ynk)
≤ αnkyn−pk
2
+ (1−αn)kxn−pk
2
−αn(1−αn)g(kxn−T ynk)
≤ αn[βnkxn−pk
2
+ (1−βn)kxn−pk
2
−βn(1−βn)g(kxn−T xnk)]
+(1−αn)kxn−pk
2
−αn(1−αn)g(kxn−T ynk)
≤ kxn−pk
2
−αn(1−αn)g(kxn−T ynk)
−αnβn(1−βn)g(kxn−T xnk)
From the above estimate, we have the following two im-portant inequalities:
αn(1−αn)g(kxn−T ynk)≤ kxn−pk
2
− kxn+1−pk 2
(2.1) and
αnβn(1−βn)g(kxn−T xnk)≤ kxn−pk
2
− kxn+1−pk 2
. (2.2) Case I:αn andβn satisfy (C3).
Letm≥1. Then from the inequality (2.1), we have
Xm
n=1αn(1−αn)g(kxn−T ynk)≤
kx1−pk 2
− kxm+1−pk 2
<∞ Proceedings of the World Congress on Engineering and Computer Science 2010 Vol II
WCECS 2010, October 20-22, 2010, San Francisco, USA
ISBN: 978-988-18210-0-3
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
.
Whenm→ ∞in the above inequality, we have
P∞
n=1αn(1−αn)g(kxn−T ynk)<∞.Since
P∞
n=1αn(1−
αn) =∞,therefore we have lim infn→∞g(kxn−T ynk) =
0.
From Lemma 1.3, we get that lim infn→∞kxn−T ynk=
0.
Since
kxn−T xnk ≤ kxn−T ynk+kT xn−T ynk
≤ kxn−T ynk+kxn−ynk
= kxn−T ynk+kxn−ynk
= kxn−T ynk+βnkxn−T xnk,
so we have
(1−βn)kxn−T xnk ≤ kxn−T ynk.
Therefore, from lim infn→∞kxn−T ynk = 0 and lim supn→∞βn<1, we deduce that
lim inf
n→∞
kxn−T xnk= 0.
Case II:αn andβn satisfy (C4).
From the inequality (2.2), we have
Xm
n=1αnβn(1−βn)g(kxn−T xnk) ≤ kx1−pk
2
− kxm+1−pk 2
<∞.
Letting m → ∞, we get that P∞
n=1αnβn(1 −
βn)g(kxn−T xnk)<∞.
Since P∞
n=1βn(1 − βn) = ∞, therefore
lim infn→∞αng(kxn−T xnk) = 0.
That is, (lim infn→∞αn) (lim infn→∞g(kxn−T xnk)) = 0.
As lim infn→∞αn > 0, therefore
lim infn→∞g(kxn−T xnk) = 0.
By Lemma 1.3, we get that
lim inf
n→∞ kxn−T xnk= 0. Case III:αn andβn satisfy (C5).
Using the condition ”0 ≤ αn ≤ b < 1” in the
inequal-ity(2.1), we have
αn(1−b)g(kxn−T ynk)≤ kxn−pk
2
− kxn+1−pk 2
.
Summing the first m terms of the above inequality, we have that
(1−b)Xm
n=1αng(kxn−T ynk) ≤ kx1−pk
2
− kxm+1−pk 2
<∞.
Whenm→ ∞,we get (1−b)P∞
n=1αng(kxn−T ynk)< ∞. Since P∞
n=1αn = ∞, therefore
lim infn→∞g(kxn−T ynk) = 0. Again from Lemma
1.3, we get that lim infn→∞kxn−T ynk= 0.
Since kxn−T xnk ≤ kxn−T ynk +βnkxn−T xnk ≤
kxn−T ynk+βnM for someM >0 andβn→0,therefore
we get
lim inf
n→∞
kxn−T xnk= 0. (2.3)
Now we prove our convergence result.
Theorem 2.3: LetCbe a nonempty closed convex
sub-set of a uniformly convex Banach space E and let T be a nonexpansive map of C into itself with at least one fixed point. Let {αn} and {βn} be sequences in [0,1]
and satisfy one of the three sets of conditions of Lemma 2.2. Then the sequence {xn} defined by (1.1), converges
weakly to a fixed point ofT.
Proof. Let ωw(xn), the weak ω-limit set of {xn}, be
given by:
ωw(xn) ={y∈E:xnk⇀ y for{xnk} ⊆ {xn}}.
Since limn→∞kxn−pk exists for each p∈F(T),
there-fore the sequence {xn} is bounded.Without any loss of
generality, we can suppose that C is bounded. This gives that there exists a subsequence {xni} of {xn}
such that xni ⇀ p ∈ ωw(xn) as i → ∞ and vice
versa. This shows that ωw(xn) 6= φ and so by
Lemma 1.1, g(kp−T pk) ≤ lim infk→∞kxnk−T xnkk.
But lim infk→∞kxnk−T xnkk = 0 by Lemma 2.2. That
is, g(kp−T pk) = 0. By the properties ofg,we get that
kp−T pk = 0. That is p ∈ F(T) and hence ωw(xn) ⊂
F(T). Next, we follow Chang et. al[2]to prove the weak convergence of the sequence. For any p∈ωw(xn), there
exists a subsequence{xnj} of{xn}such that
xnj ⇀ pasj→ ∞. (2.4)
Hence from (2.4) and continuity ofT, it follows that
T xni ⇀ p. (2.5)
Now from (1.1), (2.4) and (2.5), we get that
yni = (1−βni)xni+βniT xni⇀ p. (2.6)
From (2.4), (2.6) and the continuity ofT, we have that
T yni = (T yni−xni) +xni ⇀ p. (2.7)
Again from (1.1) and (2.7), we conclude that
xni+1= (1−αni)xni+αniT yni⇀ p
Continuing in this way, by induction, we can prove that, for anym>0,
xni+m⇀ p. Proceedings of the World Congress on Engineering and Computer Science 2010 Vol II
WCECS 2010, October 20-22, 2010, San Francisco, USA
ISBN: 978-988-18210-0-3
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
By induction, we get that S∞
m=0{xnj+m}
con-verges weakly to p as j → ∞; in fact {xn}∞n=n1 =
S∞
m=0{xnj+m}
∞
j=1 gives thatxn⇀ pasn→ ∞.
Remark 2.4. A comparison of Theorem 2.3 with
Theo-rem A reveals that the assumptionP∞
n=1βn(1−αn)<∞
in Theorem A is superflous. Also, it is obvious that the assumption (C2) in Theorem B implies (C3) −(C4). Also Theorem A and Theorem B are established under the Opial’s property or Fr´echet differentiable norm. Moreover, Kadec-Klee property is required in Theorem 4.1[4] to establish the weak convergence of the Ishikawa iterates. The weak convergence theorem presented in this paper is valid in any uniformly convex Banach space.
References
[1] F. E. Browder,Convergence of approximates to fixed points of nonexpansive nonlinear mappings in Ba-nach spaces, Arch. Rational Mech. Anal.24 (1967), 82-90.
[2] S.S. Chang, Y.J. Cho, H. Zhou, Demi-closed prin-cipal and weak convergence problems for asymptoti-cally nonexpansive mappings,J. Korean Math. Soc. 38 (2001), No. 6, 1245–1260.
[3] J. B. Diaz, F. T. Metcalf,On the structure of the set of subsequential limit points of successive approxima-tions, Bull. Amer. Math. Soc.73(1967), 516-519.
[4] H. Fukhar-ud-din, A. R. Khan,Approximating com-mon fixed points of asymptotically nonexpansive maps in uniformly convex Banach spaces, Comput. Math. Appl. 53 (2007), 1349-1360.
[5] J. Gornicki,Nonlinear ergodic theorems for asymp-totically non-expansive mappings in Banach spaces satisfying opial’s condition, J. Math. Anal. Appl. 161(1991), 440–446.
[6] W. Kaczor,Weak convergence of almost orbits of as-ymptotically nonexpansive commutative semigroups, J. Math. Anal. Appl. 272(2002), 565-574.
[7] A. R. Khan, N. Hussain,Iterative approximation of fixed points of nonexpansive maps, Sci. Math. Jpn. 54(2001), 503-511.
[8] Z. Opial, Weak convergence of the sequence of suc-cessive approximations for nonexpansive mappings, Bull. Amer. Math. Soc. 73 (1967), 591–597.
[9] W. Takahashi, T. Tamura,Convergence theorems for a pair of nonexpansive mappings, J. Convex Anal. 5 (1995), 45–58.
[10] K.K.Tan, H.K. Xu, Approximatig fixed points of nonexpansive mappings by the Ishikawa iteration process, J. Math. Anal. Appl. 178 (1993), 301–308.
[11] H. K. Xu, Inequalities in Banach spaces with appli-cations, Nonlinear Anal. TMA 16(12) (1991), 1127– 1138.
[12] H.Y. Zhou, G. T. Guo, H. J. Hwang, Y. J. Cho, On the iterative methods for nonlinear operator equations in Banach spaces, PanAmer Math. J., 14(4)(2004), 61-68.
Proceedings of the World Congress on Engineering and Computer Science 2010 Vol II WCECS 2010, October 20-22, 2010, San Francisco, USA
ISBN: 978-988-18210-0-3
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)