Common Fixed Points of a New Three-Step Iteration with Errors of ...
Common Fixed Points of a New Three-Step Iteration with Errors of ...
Common Fixed Points of a New Three-Step Iteration with Errors of ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Journal <strong>of</strong> Nonlinear Analysis and Optimization<br />
Vol. 1 No. 1 (2010), 169-182<br />
ISSN : 1906-9605<br />
http://www.sci.nu.ac.th/jnao<br />
<strong>Common</strong> <strong>Fixed</strong> <strong>Points</strong> <strong>of</strong> a <strong>New</strong> <strong>Three</strong>-<strong>Step</strong> <strong>Iteration</strong> <strong>with</strong> <strong>Errors</strong> <strong>of</strong><br />
Asymptotically Quasi-Nonexpansive Nonself-Mappings in Banach<br />
spaces<br />
Utith Inprasit 1,2 and Hathaikarn Wattanataweekul 1,2<br />
ABSTRACT: In this paper, we introduce an iterative method for finding a common element <strong>of</strong><br />
the set <strong>of</strong> solutions <strong>of</strong> a generalized mixed equilibrium problem and the set <strong>of</strong> common fixed<br />
points <strong>of</strong> a finite family <strong>of</strong> nonexpansive mappings in a real Hilbert space. Then, we prove that<br />
the sequence converges strongly to a common element <strong>of</strong> the above two sets. Furthermore,<br />
we apply our result to prove three new strong convergence theorems in fixed point problems,<br />
mixed equilibrium problems, generalized equilibrium problems and equilibrium problems.<br />
1. Introduction<br />
We assume that X is a normed space and C is a nonempty subset <strong>of</strong> X. A mapping T : C →<br />
C is said to be asymptotically nonexpansive [3] if there exists a sequence {k n } <strong>of</strong> real numbers<br />
<strong>with</strong> k n ≥ 1 and lim k n = 1 such that ‖T n x − T n y‖ ≤ k n ‖x − y‖ for all x, y ∈ C and each<br />
n→∞<br />
n ≥ 1. The class <strong>of</strong> asymptotically nonexpansive mappings is a natural generalization <strong>of</strong> the<br />
important class <strong>of</strong> nonexpansive mappings. Goebel and Kirk [3] proved that if C is a nonempty<br />
closed and bounded subset <strong>of</strong> a uniformly convex Banach space, then every asymptotically<br />
nonexpansive self-mapping has a fixed point. A mapping T : C → C is called asymptotically<br />
quasi-nonexpansive if F(T) ̸= ∅ and there exists a sequence {k n } <strong>of</strong> real numbers <strong>with</strong> k n ≥ 1<br />
and lim k n = 1 such that ‖T n x − q‖ ≤ k n ‖x − q‖ for all x ∈ C, q ∈ F(T), n ≥ 1, where F(T)<br />
n→∞<br />
is the set <strong>of</strong> fixed points <strong>of</strong> T. The mapping T is called uniformly L-Lipschitzian if there exists a<br />
positive constant L such that ‖T n x − T n y‖ ≤ L‖x − y‖ for all x, y ∈ C and each n ≥ 1. It is easy<br />
to see that an asymptotically nonexpansive mapping must be uniformly L-Lipschitzian as well<br />
as asymptotically quasi-nonexpansive but the converse does not hold.<br />
In 2000, Noor [9] introduced a three-step iterative sequence and studied the approximate<br />
solutions <strong>of</strong> variational inclusions in Hilbert spaces. Glowinski and Le Tallec [4] applied<br />
three-step iterative sequences for finding the approximate solutions <strong>of</strong> the elastoviscoplasticity<br />
problem, eigenvalue problems and in the liquid crystal theory. It has been shown in [1],<br />
that three-step method performs better than two-step and one-step methods for solving variational<br />
inequalities. The three-step schemes are natural generalization <strong>of</strong> the splitting methods<br />
to solve partial differential equations; see, Noor [9, 10, 11]. This signifies that Noor three-step<br />
Corresponding author: utith@sci.ubu.ac.th(U. Inprasit)<br />
Manuscript received March 16, 2010. Accepted July 31, 2010.
170 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
methods are robust and more efficient than the Mann (one-step) and Ishikawa (two-step) type<br />
iterative methods to solve problems <strong>of</strong> pure and applied sciences.<br />
In 2001, Khan and Takahashi [5] have approximated common fixed points <strong>of</strong> two asymptotically<br />
nonexpansive mappings by the modified Ishikawa iteration. Recently Shahzad and<br />
Udomene [15] established convergence theorems for the modified Ishikawa iteration process<br />
<strong>of</strong> two asymptotically quasi-non expansive mappings to a common fixed point <strong>of</strong> the mappings.<br />
For related results <strong>with</strong> error terms, we refer to [2, 6, 13] and [15].<br />
The purpose <strong>of</strong> this paper is to establish strong and weak convergence theorems <strong>of</strong> a<br />
new three-step iteration for three asymptotically quasi-nonexpan sive non-self mappings in a<br />
uniformly convex Banach space. This scheme can be viewed as an extension <strong>of</strong> Xu and Noor<br />
[18], Suantai [16] and Nilsrakoo and Saejung [8].<br />
Let X be a normed space. A subset C <strong>of</strong> X is said to be a retract <strong>of</strong> X if there exists a continuous<br />
map P : X → C such that Px = x for all x ∈ C. Every closed convex set <strong>of</strong> a uniformly<br />
convex Banach space is a retract. A map P : X → C is said to be a retraction if P 2 = P. It follows<br />
that if a map P is a retraction, then Py = y for all y in the range <strong>of</strong> P. A mapping T : C → X is<br />
said to be asymptotically quasi-nonexpansive if F(T) ̸= ∅ and there exists a sequence {k n } <strong>of</strong> real<br />
numbers <strong>with</strong> k n ≥ 1 and lim<br />
n→∞<br />
k n = 1 such that<br />
‖T(PT) n−1 x − q‖ ≤ k n ‖x − q‖<br />
for all x ∈ C, q ∈ F(T), n ≥ 1, where F(T) is the set <strong>of</strong> fixed points <strong>of</strong> T and (PT) 0 = I, the<br />
identity operator on C.<br />
The mapping T : C → X is called uniformly L-Lipschitzian if there exists a positive constant<br />
L such that<br />
‖T(PT) n−1 x − T(PT) n−1 y‖ ≤ L‖x − y‖<br />
for all x, y ∈ C and n ∈ N.<br />
Let C be a nonempty closed convex subset <strong>of</strong> X and P : X → C a nonexpansive retraction<br />
<strong>of</strong> X onto C, and let T 1 , T 2 , T 3 : C → X be asymptotically quasi-nonexpansive mappings and F<br />
⋂<br />
is the set <strong>of</strong> all common fixed points <strong>of</strong> T i i.e., F = 3 F(T i ), where F(T i ) = {x ∈ C : T i x = x}<br />
for all i = 1, 2, 3. Then, for arbitrary x 1 ∈ C, compute the sequences {x n }, {y n }, {z n } by the<br />
iterative scheme<br />
i=1<br />
z n = P[a n T 1 (PT 1 ) n−1 x n + (1 − a n − δ n )x n + δ n u n ],<br />
y n = P[b n T 2 (PT 2 ) n−1 z n + c n T 1 (PT 1 ) n−1 x n + (1 − b n − c n − σ n )x n + σ n v n ],<br />
(1)<br />
x n+1 = P[α n T 3 (PT 3 ) n−1 y n + β n T 2 (PT 2 ) n−1 z n + γ n T 1 (PT 1 ) n−1 x n<br />
+ (1 − α n − β n − γ n − ρ n )x n + ρ n w n ]<br />
for all n ≥ 1, where {a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } are appropriate sequences<br />
in [0, 1] and {u n }, {v n }, {w n } are bounded sequences in C.<br />
Without errors (δ n = σ n = ρ n ≡ 0), and T 1 , T 2 , T 3 are self-maps <strong>of</strong> C, the iterative scheme<br />
(1) reduces to the following iterative scheme:<br />
(2)<br />
z n = a n T n 1 x n + (1 − a n )x n ,<br />
y n = b n T n 2 z n + c n T n 1 x n + (1 − b n − c n )x n ,<br />
x n+1 = α n T n 3 y n + β n T n 2 z n + γ n T n 1 x n + (1 − α n − β n − γ n )x n , n ≥ 1,<br />
where {a n }, {b n }, {c n }, {α n }, {β n }, {γ n } are appropriate sequences in [0, 1].<br />
If T := T 1 = T 2 = T 3 , then (2) reduces to the iterative scheme defined by Nilsrakoo and<br />
Saejung [8].<br />
If γ n ≡ 0 and T := T 1 = T 2 = T 3 , then (2) reduces to the iterative scheme defined by<br />
Suantai [16].<br />
If c n = β n = γ n ≡ 0 and T := T 1 = T 2 = T 3 , then (2) reduces to the iterative scheme<br />
defined by Xu and Noor [18].
171 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
(3)<br />
If a n = b n = c n ≡ 0, then (2) reduces to the following iterative scheme:<br />
x n+1 = α n T n 3 x n + β n T n 2 x n + γ n T n 1 x n + (1 − α n − β n − γ n )x n<br />
for all n ≥ 1, where {α n }, {β n }, {γ n } are appropriate sequences in [0, 1].<br />
To study strong and weak convergence theorems <strong>of</strong> the iterative scheme 1, we recall<br />
some useful well-known concepts and results.<br />
Recall that a Banach space X is said to satisfy Opial’s condition [12] if for each sequence<br />
{x n } and x, y ∈ X <strong>with</strong> x n → x weakly as n → ∞ and x ̸= y imply that<br />
lim sup ‖x n − x‖ < lim sup ‖x n − y‖.<br />
n→∞ n→∞<br />
In what follows, we shall make use <strong>of</strong> the following lemmas.<br />
Lemma 1.1. [17, Lemma 1]. Let {a n }, {b n }, {δ n } be sequences <strong>of</strong> nonnegative real numbers satisfying<br />
the inequality<br />
a n+1 ≤ (1 + δ n )a n + b n for all n = 1, 2, . . . .<br />
If<br />
∞<br />
∑ δ n < ∞ and ∑<br />
∞ b n < ∞, then<br />
n=1<br />
n=1<br />
(i) lim a n exists, and<br />
n→∞<br />
(ii) lim a n = 0 whenever lim inf a n = 0.<br />
n→∞ n→∞<br />
Lemma 1.2. [7, Lemma 1.4]. Let X be a uniformly convex Banach space and let B r = {x ∈ X :<br />
‖x‖ ≤ r}, r > 0 be a closed ball <strong>of</strong> X. Then there exists a continuous, strictly increasing convex<br />
function g : [0, ∞) → [0, ∞), g(0) = 0 such that<br />
‖λx + µy + ξz + ϑw‖ 2 ≤ λ‖x‖ 2 + µ‖y‖ 2 + ξ‖z‖ 2 + ϑ‖w‖ 2 − λµg ( ‖x − y‖ )<br />
for all x, y, z, w ∈ B r and all λ, µ, ξ, ϑ ∈ [0, 1] <strong>with</strong> λ + µ + ξ + ϑ = 1.<br />
Similar to Lemma 1.2, we can prove the next lemma.<br />
Lemma 1.3. Let X be a uniformly convex Banach space and let B r be a closed ball <strong>of</strong> X. Then there<br />
exists a continuous, strictly increasing convex function g : [0, ∞) → [0, ∞), g(0) = 0 such that<br />
‖λx + µy + ξz + ϑw + ζs‖ 2 ≤ λ‖x‖ 2 + µ‖y‖ 2 + ξ‖z‖ 2 + ϑ‖w‖ 2 + ζ‖s‖ 2 − λµg ( ‖x − y‖ )<br />
for all x, y, z, w, s ∈ B r and all λ, µ, ξ, ϑ, ζ ∈ [0, 1] <strong>with</strong> λ + µ + ξ + ϑ + ζ = 1.<br />
Lemma 1.4. [16, Lemma 2.7]. Let X be a Banach space which satisfies Opial’s condition and let {x n }<br />
be a sequence in X. Let u, v ∈ X be so that lim ‖x n − u‖ and lim ‖x n − v‖ exist. If {x nk } and {x mk }<br />
n→∞ n→∞<br />
are subsequences <strong>of</strong> {x n } which converge weakly to u and v, respectively, then u = v.<br />
2. Main Results<br />
In this section, we prove strong and weak convergence theorems for the iterative scheme<br />
(1) for asymptotically quasi-nonexpansive nonself-mappings in a Banach space. In order to<br />
prove our main results, the following lemmas are needed.<br />
Lemma 2.1. Let X be a uniformly convex Banach space and C a nonempty closed convex nonexpansive<br />
retract <strong>of</strong> X <strong>with</strong> P as a nonexpansive retraction. Let T 1 , T 2 , T 3 : C → X be asymptotically<br />
quasi-nonexpansive mappings <strong>with</strong> respect to sequences {k n }, {l n }, {m n }, respectively, such that<br />
F ̸= ∅, k n ≥ 1, l n ≥ 1, m n ≥ 1,<br />
∞<br />
∑<br />
n=1<br />
(k n − 1) < ∞,<br />
∞<br />
∑<br />
n=1<br />
(l n − 1) < ∞ ∑<br />
∞ (m n − 1) < ∞. Let<br />
{a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } be real sequences in [0, 1] such that a n + δ n ,<br />
b n + c n + σ n and α n + β n + γ n + ρ n are in [0, 1] for all n ≥ 1,<br />
∞<br />
∑<br />
n=1<br />
δ n < ∞,<br />
n=1<br />
∞<br />
∑<br />
n=1<br />
σ n < ∞,<br />
∞<br />
∑<br />
n=1<br />
ρ n < ∞<br />
and let {u n }, {v n }, {w n } be bounded sequences in C. For a given x 1 ∈ C, let {x n }, {y n }, {z n } be the
172 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
sequences defined as in (1). Then<br />
(i) lim ‖x n − q‖ exists for all q ∈ F.<br />
n→∞<br />
(ii) If one <strong>of</strong> the following conditions (a), (b), (c) and (d) holds, then<br />
lim ‖T 1(PT 1 ) n−1 x n − x n ‖ = 0.<br />
n→∞<br />
(a) lim inf β n > 0 and 0 < lim inf a n ≤ lim sup(a n + δ n ) < 1.<br />
n→∞ n→∞ n→∞<br />
(b) lim inf α n , lim inf b n > 0 and 0 < lim inf a n ≤ lim sup(a n + δ n ) < 1.<br />
n→∞ n→∞ n→∞ n→∞<br />
(c) 0 < lim inf γ n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞<br />
(d) 0 < lim inf α n and 0 < lim inf c n ≤ lim sup(b n + c n + σ n ) < 1.<br />
n→∞ n→∞ n→∞<br />
(iii) If either (a) 0 < lim inf β n ≤ lim sup(α n + β n + γ n + ρ n ) < 1<br />
n→∞ n→∞<br />
or (b) lim inf α n > 0 and 0 < lim inf b n ≤ lim sup(b n + c n + σ n ) < 1,<br />
n→∞ n→∞ n→∞<br />
then lim ‖T 2 (PT 2 ) n−1 z n − x n ‖ = 0.<br />
n→∞<br />
(iv) If 0 < lim inf α n ≤ lim sup(α n + β n + γ n + ρ n ) < 1,<br />
n→∞ n→∞<br />
then lim ‖T 3 (PT 3 ) n−1 y n − x n ‖ = 0.<br />
n→∞<br />
Pro<strong>of</strong>. (i) Let q ∈ F. By (1), we obtain<br />
(4)<br />
‖z n − q‖ = ‖P[a n T 1 (PT 1 ) n−1 x n + (1 − a n − δ n )x n + δ n u n ] − P(q)‖<br />
≤ a n ‖T 1 (PT 1 ) n−1 x n − q‖ + (1 − a n − δ n )‖x n − q‖ + δ n ‖u n − q‖<br />
≤ (1 + a n (k n − 1) − δ n )‖x n − q‖ + δ n ‖u n − q‖<br />
and<br />
(5)<br />
‖y n − q‖ = ‖P[b n T 2 (PT 2 ) n−1 z n + c n T 1 (PT 1 ) n−1 x n + (1 − b n − c n − σ n )x n<br />
+ σ n v n ] − P(q)‖<br />
≤<br />
≤<br />
b n ‖T 2 (PT 2 ) n−1 z n − q‖ + c n ‖T 1 (PT 1 ) n−1 x n − q‖<br />
+ (1 − b n − c n − σ n )‖x n − q‖ + σ n ‖v n − q‖<br />
b n l n ‖z n − q‖ + c n k n ‖x n − q‖ + (1 − b n − c n − σ n )‖x n − q‖<br />
+ σ n ‖v n − q‖.<br />
By (4) and (5), we obtain<br />
‖x n+1 − q‖ = ‖P[α n T 3 (PT 3 ) n−1 y n + β n T 2 (PT 2 ) n−1 z n + γ n T 1 (PT 1 ) n−1 x n<br />
+ (1 − α n − β n − γ n − ρ n )x n + ρ n w n ] − P(q)‖<br />
≤ α n ‖T 3 (PT 3 ) n−1 y n − q‖ + β n ‖T 2 (PT 2 ) n−1 z n − q‖<br />
+ γ n ‖T 1 (PT 1 ) n−1 x n − q‖ + (1 − α n − β n − γ n − ρ n )‖x n − q‖<br />
+ ρ n ‖w n − q‖<br />
≤ α n m n ‖y n − q‖ + β n l n ‖z n − q‖ + γ n k n ‖x n − q‖<br />
+ (1 − α n − β n − γ n − ρ n )‖x n − q‖ + ρ n ‖w n − q‖<br />
≤ (α n m n b n l n + β n l n )‖z n − q‖ + α n m n c n k n ‖x n − q‖<br />
+ (α n m n − α n m n b n − α n m n c n − α n m n σ n )‖x n − q‖<br />
+ α n m n σ n ‖v n − q‖ + γ n k n ‖x n − q‖<br />
+ (1 − α n − β n − γ n − ρ n )‖x n − q‖ + ρ n ‖w n − q‖<br />
≤ ‖x n − q‖ + ((l n − 1)(α n m n b n + β n ) + (k n − 1)(γ n + α n m n c n<br />
+ (α n m n b n l n + β n l n )a n ) + α n (m n − 1))‖x n − q‖<br />
+ (m n l n + l n )δ n ‖u n − q‖ + m n σ n ‖v n − q‖ + ρ n ‖w n − q‖.
173 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
Since {l n }, {m n }, {u n }, {v n }, {w n } are bounded, there exists a constant K > 0 such that α n m n b n +<br />
β n ≤ K, γ n + α n m n c n + (α n m n b n l n + β n l n )a n ≤ K, (m n l n + l n )‖u n − q‖ ≤ K, m n ‖v n − q‖ ≤<br />
K, ‖w n − q‖ ≤ K and α n ≤ K for all n ≥ 1. Then<br />
(6)<br />
‖x n+1 − q‖ ≤<br />
(<br />
1 + K ( (k n − 1) + (l n − 1) + (m n − 1) )) ‖x n − q‖<br />
+ K(δ n + σ n + ρ n )<br />
By Lemma 1.1, we obtain lim<br />
n→∞<br />
‖x n − q‖ exists.<br />
Next, we want to prove (ii), (iii) and (iv). It follows from (i) that {x n − q}, {T 1 (PT 1 ) n−1 x n −<br />
q}, {y n − q}, {T 3 (PT 3 ) n−1 y n − q}, {z n − q} and {T 2 (PT 2 ) n−1 z n − q} are all bounded. Let<br />
{<br />
M = max<br />
sup<br />
n≥1<br />
sup<br />
n≥1<br />
sup<br />
n≥1<br />
‖x n − q‖, sup<br />
n≥1<br />
‖T 3 (PT 3 ) n−1 y n − q‖, sup<br />
n≥1<br />
‖T 2 (PT 2 ) n−1 z n − q‖, sup<br />
n≥1<br />
‖T 1 (PT 1 ) n−1 x n − q‖, sup ‖y n − q‖,<br />
n≥1<br />
‖z n − q‖, sup ‖u n − q‖,<br />
n≥1<br />
}<br />
‖v n − q‖, sup ‖w n − q‖<br />
n≥1<br />
By Lemma 1.3, there exists a continuous, strictly increasing convex function g : [0, ∞) → [0, ∞)<br />
<strong>with</strong> g(0) = 0 such that<br />
.<br />
(7)<br />
‖λx + µy + ξz + ϑw + ζs‖ 2 ≤ λ‖x‖ 2 + µ‖y‖ 2 + ξ‖z‖ 2 + ϑ‖w‖ 2 + ζ‖s‖ 2<br />
− λµg ( ‖x − y‖ )<br />
for all x, y, z, w, s ∈ B r and all λ, µ, ξ, ϑ, ζ ∈ [0, 1] <strong>with</strong> λ + µ + ξ + ϑ + ζ = 1. By (7), we have<br />
(8)<br />
‖z n − q‖ 2 = ‖P[a n T 1 (PT 1 ) n−1 x n + (1 − a n − δ n )x n + δ n u n ] − P(q)‖ 2<br />
≤ ‖a n (T 1 (PT 1 ) n−1 x n − q) + (1 − a n − δ n )(x n − q) + δ n (u n − q)‖ 2<br />
≤ a n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + (1 − a n − δ n )‖x n − q‖ 2<br />
+ δ n ‖u n − q‖ 2 − a n (1 − a n − δ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ a n k 2 n‖x n − q‖ 2 + (1 − a n − δ n )‖x n − q‖ 2 + δ n ‖u n − q‖ 2<br />
− a n (1 − a n − δ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ (1 + a n (k 2 n − 1) − δ n )‖x n − q‖ 2 + δ n ‖u n − q‖ 2<br />
and<br />
(9)<br />
‖y n − q‖ 2 = ‖P[b n T 2 (PT 2 ) n−1 z n + c n T 1 (PT 1 ) n−1 x n + (1 − b n − c n − σ n )<br />
x n + σ n v n ] − P(q)‖ 2<br />
≤ ‖b n (T 2 (PT 2 ) n−1 z n − q) + c n (T 1 (PT 1 ) n−1 x n − q)<br />
+ (1 − b n − c n − σ n )(x n − q) + σ n (x n − q)‖ 2<br />
≤ b n ‖T 2 (PT 2 ) n−1 z n − q‖ 2 + (1 − b n − c n − σ n )‖x n − q‖ 2<br />
+ c n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + σ n ‖v n − q‖ 2<br />
− b n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
≤ b n l 2 n‖z n − q‖ 2 + (1 − b n − c n − σ n )‖x n − q‖ 2 + c n k 2 n‖x n − q‖ 2<br />
+ σ n ‖v n − q‖ 2 − b n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )
174 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
By (7) , (8) and (9), we obtain<br />
‖x n+1 − q‖ 2 = ‖P[α n T 3 (PT 3 ) n−1 y n + β n T 2 (PT 2 ) n−1 z n + γ n T 1 (PT 1 ) n−1 x n<br />
+ (1 − α n − β n − γ n − ρ n )x n + ρ n w n ] − P(q)‖ 2<br />
≤ α n ‖T 3 (PT 3 ) n−1 y n − q‖ 2 + β n ‖T 2 (PT 2 ) n−1 z n − q‖ 2<br />
+ γ n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
− α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ )<br />
+ ρ n ‖w n − q‖ 2<br />
≤ α n m 2 n‖y n − q‖ 2 + β n l 2 n‖z n − q‖ 2 + γ n k 2 n‖x n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
+ (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
− α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ )<br />
≤ α n m 2 nc n k 2 n‖x n − q‖ 2 + α n m 2 n(1 − b n − c n − σ n )‖x n − q‖ 2<br />
+ γ n k 2 n‖x n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
+ (α n m 2 nb n l 2 n + β n l 2 n)‖z n − q‖ 2 + α n m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nb n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
− α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ )<br />
≤ α n m 2 nc n k 2 n‖x n − q‖ 2 + α n m 2 n(1 − b n − c n )‖x n − q‖ 2<br />
+ γ n k 2 n‖x n − q‖ 2 + (1 − α n − β n − γ n )‖x n − q‖ 2 + (α n m 2 nb n l 2 n<br />
+ β n l 2 n)‖x n − q‖ 2 + (α n m 2 nb n l 2 n + β n l 2 n)(a n (k 2 n − 1))‖x n − q‖ 2<br />
− (α n m 2 nb n l 2 n + β n l 2 n)a n (1 − a n − δ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
+ (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nb n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
− α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ )<br />
= ‖x n − q‖ 2 + ( (k 2 n − 1)(α n m 2 nc n + γ n + (α n m 2 nb n l 2 n + β n l 2 n)a n )<br />
+ (l 2 n − 1)(α n m 2 nb n + β n ) + α n (m 2 n − 1) ) ‖x n − q‖ 2<br />
+ (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nb n l 2 na n (1 − a n − δ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
− β n l 2 na n (1 − a n − δ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
− α n m 2 nb n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
− α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ )<br />
Since {k n }, {l n }, {m n }, {u n }, {v n }, {w n } are bounded and {x n } is bounded, there exist constants<br />
K 0 > 0 such that<br />
(α n m 2 nc n + γ n + (α n m 2 nb n l 2 n + β n l 2 n)a n )‖x n − q‖ 2 ≤ K 0 ,<br />
(α n m 2 nb n + β n )‖x n − q‖ 2 ≤ K 0 , α n ‖x n − q‖ 2 ≤ K 0 , (m 2 nl 2 n + l 2 n)‖u n − q‖ 2 ≤ K 0 ,<br />
m 2 n‖v n − q‖ 2 ≤ K 0 and ‖w n − q‖ 2 ≤ K 0 for all n ≥ 1. Thus<br />
α n m 2 nb n l 2 na n (1 − a n − δ n )g ( ||T 1 (PT 1 ) n−1 x n − x n || ) ≤ ||x n − q|| 2 − ||x n+1 − q|| 2<br />
(10)<br />
(11)<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).<br />
β n l 2 na n (1 − a n − δ n )g ( ||T 1 (PT 1 ) n−1 x n − x n || ) ≤ ||x n − q|| 2 − ||x n+1 − q|| 2<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).
175 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
α n m 2 nb n (1 − b n − c n − σ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ ) ≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
(12)<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).<br />
α n (1 − α n − β n − γ n − ρ n )g ( ‖T 3 (PT 3 ) n−1 y n − x n ‖ ) ≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
(13)<br />
Again (7), we obtain<br />
(14)<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).<br />
‖y n − q‖ 2 = ‖P[b n T 2 (PT 2 ) n−1 z n + c n T 1 (PT 1 ) n−1 x n + (1 − b n − c n − σ n )x n<br />
+ σ n v n ] − P(q)‖ 2<br />
By (7), (8) and (14), we have<br />
≤ c n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + (1 − b n − c n − σ n )‖x n − q‖ 2<br />
+ b n ‖T 2 (PT 2 ) n−1 z n − q‖ 2 + σ n ‖v n − q‖ 2<br />
− c n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ c n k 2 n‖x n − q‖ 2 + (1 − b n − c n − σ n )‖x n − q‖ 2 + b n l 2 n‖z n − q‖ 2<br />
+ σ n ‖v n − q‖ 2 − c n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
‖x n+1 − q‖ 2 = ‖P[α n T 3 (PT 3 ) n−1 y n + β n T 2 (PT 2 ) n−1 z n + γ n T 1 (PT 1 ) n−1 x n<br />
+ (1 − α n − β n − γ n − ρ n )x n + ρ n w n ] − P(q)‖ 2<br />
≤ α n ‖T 3 (PT 3 ) n−1 y n − q‖ 2 + β n ‖T 2 (PT 2 ) n−1 z n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
+ γ n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
− β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
≤ α n m 2 n‖y n − q‖ 2 + β n l 2 n‖z n − q‖ 2 + γ n k 2 n‖x n − q‖ 2<br />
+ ρ n ‖w n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
+ β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
≤ α n m 2 nc n k 2 n‖x n − q‖ 2 + α n m 2 n(1 − b n − c n − α n )‖x n − q‖ 2<br />
+ γ n k 2 n‖x n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
+ (α n m 2 nb n l 2 n + β n l 2 n)‖z n − q‖ 2 + α n m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nc n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
− β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
≤ α n m 2 nc n k 2 n‖x n − q‖ 2 + α n m 2 n(1 − b n − c n )‖x n − q‖ 2<br />
+ γ n k 2 n‖x n − q‖ 2 + (1 − α n − β n − γ n )‖x n − q‖ 2 + (α n m 2 nb n l 2 n<br />
+ β n l 2 n)‖x n − q‖ 2 + (α n m 2 nb n l 2 n + β n l 2 n)(a n (k 2 n − 1))‖x n − q‖ 2<br />
+ (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nc n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
− β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
= ‖x n − q‖ 2 + ( (k 2 n − 1)(α n m 2 nc n + γ n + (α n m 2 nb n l 2 n + β n l 2 n)a n )<br />
+ (l 2 n − 1)(α n m 2 nb n + β n ) + α n (m 2 n − 1) ) ‖x n − q‖ 2<br />
+ (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− α n m 2 nc n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
− β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )
176 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
Thus<br />
α n m 2 nc n (1 − b n − c n − σ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) ≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
(<br />
+ K 0 (k<br />
2<br />
n − 1) + (ln 2 − 1) + (m 2 n − 1) )<br />
(15)<br />
+ K 0 (δ n + σ n + ρ n ).<br />
β n (1 − α n − β n − γ n − ρ n )g ( ‖T 2 (PT 2 ) n−1 z n − x n ‖ )<br />
(16)<br />
By (7) , (8) and (9), we obtain<br />
≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
(<br />
+ K 0 (k<br />
2<br />
n − 1) + (ln 2 − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).<br />
Thus<br />
‖x n+1 − q‖ 2 = ‖P[α n T 3 (PT 3 ) n−1 y n + β n T 2 (PT 2 ) n−1 z n + γ n T 1 (PT 1 ) n−1 x n<br />
+ (1 − α n − β n − γ n − ρ n )x n + ρ n w n ] − P(q)‖ 2<br />
≤ γ n ‖T 1 (PT 1 ) n−1 x n − q‖ 2 + α n ‖T 3 (PT 3 ) n−1 y n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
+ (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2 + β n ‖T 2 (PT 2 ) n−1 z n − q‖ 2<br />
− γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ γ n k 2 n‖x n − q‖ 2 + α n m 2 n‖y n − q‖ 2 + β n l 2 n‖z n − q‖ 2<br />
+ ρ n ‖w n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
− γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ γ n k 2 n‖x n − q‖ 2 + β n l 2 n‖z n − q‖ 2 + α n m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
+ α n m 2 nb n l 2 n‖z n − q‖ 2 + α n m 2 n(1 − b n − c n − σ n )‖x n − q‖ 2<br />
+ α n m 2 nc n k 2 n‖x n − q‖ 2 + (1 − α n − β n − γ n − ρ n )‖x n − q‖ 2<br />
− γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
≤ γ n k 2 n‖x n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
+ (α n m 2 nb n l 2 n + β n l 2 n)‖x n − q‖ 2 + (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2<br />
+ ( (α n m 2 nb n l 2 n + β n l 2 n)a n (k 2 n − 1) ) ‖x n − q‖ 2 + α n m 2 nc n k 2 n‖x n − q‖ 2<br />
+ (1 − α n − β n − γ n )‖x n − q‖ 2 + α n m 2 n(1 − b n − c n )‖x n − q‖ 2<br />
− γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ )<br />
= ‖x n − q‖ 2 + ( (k 2 n − 1)(α n m 2 nc n + γ n + (α n m 2 nb n l 2 n + β n l 2 n)a n )<br />
+ (l 2 n − 1)(α n m 2 nb n + β n ) + α n (m 2 n − 1) ) ‖x n − q‖ 2<br />
+ (m 2 nl 2 n + l 2 n)δ n ‖u n − q‖ 2 + m 2 nσ n ‖v n − q‖ 2 + ρ n ‖w n − q‖ 2<br />
− γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) .<br />
γ n (1 − α n − β n − γ n − ρ n )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) ≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
(17)<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n ).<br />
(ii) (a) Let lim inf β n > 0 and 0 < lim inf a n ≤ lim sup(a n + δ n ) < 1. Then there exists<br />
n→∞ n→∞ n→∞<br />
a positive integer n 0 and η, η ′ ∈ (0, 1) such that 0 < η < β n , 0 < η < a n and a n + δ n < η ′ < 1
177 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
for all n ≥ n 0 . This implies by (11) that<br />
(18)<br />
η 2 (1 − η ′ )g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) ≤ ‖x n − q‖ 2 − ‖x n+1 − q‖ 2<br />
+ K 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1) )<br />
+ K 0 (δ n + σ n + ρ n )<br />
for all n ≥ n 0 . It follows from (18) that for r ≥ n 0 ,<br />
(19)<br />
r (<br />
∑ g ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) ≤<br />
n=n 0<br />
≤<br />
1 ( r<br />
η 2 (1 − η ′ )<br />
∑ (‖x n − q‖ 2 − ‖x n+1 − q‖ 2 )<br />
n=n 0<br />
+ K 0<br />
r<br />
∑<br />
n=n 0<br />
(<br />
(k<br />
2<br />
n − 1) + (l 2 n − 1) + (m 2 n − 1)<br />
+ δ n + σ n + ρ n<br />
) )<br />
1<br />
(‖x<br />
η 2 (1 − η ′ n0 − q‖ 2 r (<br />
+ K 0<br />
)<br />
∑ (k<br />
2<br />
n − 1)<br />
n=n 0<br />
+ (l 2 n − 1) + (m 2 n − 1) )) .<br />
Since 0 ≤ t 2 − 1 ≤ 2t(t − 1) for all t ≥ 1 and ∑<br />
∞ (k n − 1) < ∞,<br />
∞, we get<br />
∞<br />
∑<br />
n=1<br />
∞<br />
∑<br />
(k 2 n − 1) < ∞,<br />
∞<br />
∑<br />
n=1<br />
(l 2 n − 1) < ∞,<br />
n=1<br />
∞<br />
∑<br />
n=1<br />
(l n − 1) < ∞,<br />
∞<br />
∑ (m n − 1) <<br />
n=1<br />
∞<br />
∑ (m 2 n − 1) < ∞. By inequality (19), let r → ∞.<br />
n=1<br />
We get g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) < ∞. Thus lim g ( ‖T 1 (PT 1 ) n−1 x n − x n ‖ ) = 0. Since g is<br />
n=n 0 n→∞<br />
strictly increasing and continuous at 0 <strong>with</strong> g(0) = 0, it follows that lim ‖T 1 (PT 1 ) n−1 x n −<br />
n→∞<br />
x n ‖ = 0.<br />
By using a similar method as in (ii) part (a) together <strong>with</strong> (10), (17), (15), (16), (12) and<br />
(13), the results in (ii) (b,c,d), (iii) (a,b) and (iv), respectively, can be proved. □<br />
Lemma 2.2. Let X be a uniformly convex Banach space and C a nonempty closed convex nonexpansive<br />
retract <strong>of</strong> X <strong>with</strong> P as a nonexpansive retraction. Let T 1 , T 2 , T 3 : C → X be asymptotically<br />
quasi-nonexpansive mappings <strong>with</strong> respect to sequences {k n }, {l n }, {m n }, respectively, such that<br />
∞<br />
∞<br />
∞<br />
F ̸= ∅, k n ≥ 1, l n ≥ 1, m n ≥ 1, ∑ (k n − 1) < ∞, ∑ (l n − 1) < ∞, ∑ (m n − 1) < ∞. Let<br />
n=1<br />
n=1<br />
n=1<br />
{a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } be real sequences in [0, 1] such that a n + δ n ,<br />
b n + c n + σ n , α n + β n + γ n + ρ n are in [0, 1] for all n ≥ 1,<br />
∞<br />
∑ δ n < ∞,<br />
n=1<br />
∞<br />
∑<br />
n=1<br />
σ n < ∞,<br />
∞<br />
∑<br />
n=1<br />
ρ n < ∞<br />
and let {u n }, {v n }, {w n } be bounded sequences in C. For a given x 1 ∈ C, let {x n }, {y n }, {z n } be the<br />
sequences defined as in (1). Suppose T 1 , T 2 , T 3 are uniformly L-Lipschitzian. If lim ‖T 1 (PT 1 ) n−1 x n −<br />
n→∞<br />
x n ‖ = 0, lim ‖T 2 (PT 2 ) n−1 z n − x n ‖ = 0, lim ‖T 3 (PT 3 ) n−1 y n − x n ‖ = 0, then<br />
n→∞ n→∞<br />
(i)<br />
(ii)<br />
(iii)<br />
lim ‖T 1 x n − x n ‖ = 0,<br />
n→∞<br />
lim ‖T 2 x n − x n ‖ = 0, and<br />
n→∞<br />
lim ‖T 3 x n − x n ‖ = 0.<br />
n→∞<br />
Pro<strong>of</strong>. Since<br />
‖x n+1 − x n ‖ ≤ α n ‖T 3 (PT 3 ) n−1 y n − x n ‖ + β n ‖T 2 (PT 2 ) n−1 z n − x n ‖<br />
+ γ n ‖T 1 (PT 1 ) n−1 x n − x n ‖ + ρ n ‖w n − x n ‖<br />
−→ 0 as n → ∞,
179 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
It follows that,<br />
Hence (iii) is satisfied.<br />
‖T 3 x n − x n ‖ ≤ ‖T 3 (PT 3 ) n−1 x n − x n ‖ + ‖T 3 (PT 3 ) n−1 x n − T 3 x n ‖<br />
≤ ‖T 3 (PT 3 ) n−1 x n − x n ‖ + L‖T 3 (PT 3 ) n−2 x n − x n ‖<br />
−→ 0 as n → ∞.<br />
Theorem 2.3. Let X be a uniformly convex Banach space and C a nonempty closed convex nonexpansive<br />
retract <strong>of</strong> X <strong>with</strong> P as a nonexpansive retraction. Let T 1 , T 2 , T 3 : C → X be asymptotically<br />
quasi-nonexpansive mappings <strong>with</strong> respect to sequences {k n }, {l n }, {m n }, respectively, such that<br />
F ̸= ∅, k n ≥ 1, l n ≥ 1, m n ≥ 1,<br />
∞<br />
∑<br />
n=1<br />
(k n − 1) < ∞,<br />
∞<br />
∑<br />
n=1<br />
(l n − 1) < ∞ and ∑<br />
∞ (m n − 1) < ∞.<br />
Let {a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } be real sequences in [0, 1] such that a n + δ n ,<br />
b n + c n + σ n and α n + β n + γ n + ρ n are in [0, 1] for all n ≥ 1, and<br />
∞ ∑<br />
n=1<br />
n=1<br />
δ n < ∞,<br />
∞<br />
∑<br />
n=1<br />
□<br />
σ n < ∞,<br />
∞<br />
∑ ρ n < ∞ and let {u n }, {v n }, {w n } be bounded sequences in C. Assume that T 1 , T 2 , T 3 are uniformly<br />
L-Lipschitzian. If one <strong>of</strong> T i (i = 1, 2, 3) is a completely continuous and one <strong>of</strong> the<br />
n=1<br />
following<br />
conditions (C1)-(C5) is satisfied:<br />
(C1) 0 < lim inf a n ≤ lim sup(a n + δ n ) < 1,<br />
n→∞ n→∞<br />
0 < lim inf b n ≤ lim sup(b n + c n + σ n ) < 1, and<br />
n→∞ n→∞<br />
0 < lim inf α n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞<br />
(C2) 0 < lim inf b n , lim inf c n ≤ lim sup(b n + c n + σ n ) < 1, and<br />
n→∞ n→∞ n→∞<br />
0 < lim inf α n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞<br />
(C3) 0 < lim inf b n ≤ lim sup(b n + c n + σ n ) < 1, and<br />
n→∞ n→∞<br />
0 < lim inf α n , lim inf γ n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞ n→∞<br />
(C4) lim inf b n > 0, and 0 < lim inf a n ≤ lim sup(a n + δ n ) < 1, and<br />
n→∞ n→∞ n→∞<br />
0 < lim inf α n , lim inf β n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞ n→∞<br />
(C5) 0 < lim inf α n , lim inf β n , lim inf γ n ≤ lim sup(α n + β n + γ n + ρ n ) < 1.<br />
n→∞ n→∞ n→∞ n→∞<br />
Then the sequences {x n }, {y n }, {z n } defined as in (1) converge strongly to a common fixed point <strong>of</strong> T 1 ,<br />
T 2 and T 3 .<br />
Pro<strong>of</strong>. Suppose one <strong>of</strong> the conditions (C1)-(C5) is satisfied. By Lemma 2.2, we obtain lim<br />
n→∞<br />
‖T i x n −<br />
x n ‖ = 0 for i = 1, 2, 3. Assume one <strong>of</strong> T 1 , T 2 and T 3 says T 1 is completely continuous. Since<br />
{x n } is a bounded sequence in C, there exists a subsequence {x nk } <strong>of</strong> {x n } such that {T 1 x nk }<br />
converges to q ∈ C. Since ‖x nk − q‖ ≤ ‖T 1 x nk − x nk ‖ + ‖T 1 x nk − q‖, we get lim<br />
k→∞<br />
‖x nk − q‖ = 0.<br />
Thus {x nk } converges to q ∈ C. By continuity <strong>of</strong> T i , we have T i x nk → T i q as k → ∞. Since<br />
‖T i q − q‖ ≤ ‖T i x nk − T i q‖ + ‖T i x nk − x nk ‖ + ‖x nk − q‖ → 0 as k → ∞,<br />
we obtain T i q = q (i = 1, 2, 3). Thus q ∈ F. By Lemma 2.1 (i), lim<br />
n→∞<br />
‖x n − q‖ exists. This implies<br />
lim<br />
n→∞ ‖x n − q‖ = 0. By Lemma 2.1, we have<br />
It follows that<br />
‖T 1 (PT 1 ) n−1 x n − x n ‖ → 0 and ‖T 2 (PT 2 ) n−1 z n − x n ‖ → 0 as n → ∞.<br />
‖y n − x n ‖ ≤ b n ‖T 2 (PT 2 ) n−1 z n − x n ‖ + c n ‖T 1 (PT 1 ) n−1 x n − x n ‖ + σ n ‖v n − x n ‖<br />
−→ 0 and ‖z n − x n ‖ ≤ a n ‖T 1 (PT 1 ) n−1 x n − x n ‖ + δ n ‖u n − x n ‖ → 0 as n → ∞. These imply<br />
lim y n = q and lim z n = q.<br />
□<br />
n→∞ n→∞<br />
Remark 2.4. In Theorem 2.3, assume that T 1 , T 2 and T 3 are asymptotically nonexpansive selfmappings<br />
<strong>of</strong> C such that one <strong>of</strong> them is completely continuous and thus T 1 , T 2 , T 3 are uniformly
180 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
L-Lipschitzian and δ n = σ n = ρ n ≡ 0. We obtain the following results.<br />
(1) If one <strong>of</strong> the conditions (C1) − (C5) is satisfied, then the sequences {x n },<br />
{y n }, {z n } defined as in (2) converge strongly to a common fixed<br />
point <strong>of</strong> T 1 , T 2 and T 3 .<br />
(2) If T := T 1 = T 2 = T 3 and one <strong>of</strong> the conditions (C1) − (C5) is satisfied,<br />
then we obtain the results <strong>of</strong> Nilsrakoo and Saejung [8].<br />
(3) If T := T 1 = T 2 = T 3 and one <strong>of</strong> the conditions (C1), (C2), (C4) is<br />
satisfied and γ n ≡ 0, then we obtain the results <strong>of</strong> Suantai [16].<br />
(4) If T := T 1 = T 2 = T 3 <strong>with</strong> condition (C1) is satisfied and c n = β n = γ n<br />
≡ 0, then we obtain the results <strong>of</strong> Xu and Noor [18].<br />
(5) If the condition (C5) is satisfied and a n = b n = c n ≡ 0, then the sequence<br />
{x n } defined as in (3) converges strongly to a common fixed point <strong>of</strong><br />
T 1 , T 2 and T 3 .<br />
The mapping T : C → X <strong>with</strong> F(T) ̸= ∅ is said to satisfy Condition A [14] if there is a<br />
nondecreasing function f : [0, ∞) → [0, ∞) <strong>with</strong> f (0) = 0, f (r) > 0 for all r ∈ (0, ∞) such that<br />
‖x − Tx‖ ≥ f (d(x, F(T))) for all x ∈ C, where d(x, F(T)) = inf { ‖x − q‖ : q ∈ F(T) } . As Tan<br />
and Xu [17] pointed out, the Condition A is weaker than the compactness <strong>of</strong> C.<br />
The following result gives a strong convergence theorem for asymptotically quasi-nonexpansive<br />
nonself-mappings in a uniformly convex Banach space satisfying Condition A.<br />
Theorem 2.5. Let X be a uniformly convex Banach space and C a nonempty closed convex nonexpansive<br />
retract <strong>of</strong> X <strong>with</strong> P as a nonexpansive retraction. Let T 1 , T 2 , T 3 : C → X be asymptotically<br />
quasi-nonexpansive mappings <strong>with</strong> respect to sequences {k n }, {l n }, {m n }, respectively, such that<br />
F ̸= ∅, k n ≥ 1, l n ≥ 1, m n ≥ 1,<br />
∞<br />
∑<br />
n=1<br />
(k n − 1) < ∞,<br />
∞<br />
∑<br />
n=1<br />
(l n − 1) < ∞ and ∑<br />
∞ (m n − 1) < ∞.<br />
Let {a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } be real sequences in [0, 1] such that a n + δ n ,<br />
b n + c n + σ n and α n + β n + γ n + ρ n are in [0, 1] for all n ≥ 1 and<br />
∞<br />
∑<br />
n=1<br />
∞ ∑<br />
n=1<br />
n=1<br />
δ n < ∞,<br />
∞<br />
∑<br />
n=1<br />
σ n < ∞,<br />
ρ n < ∞ and let {u n }, {v n }, {w n } be bounded sequences in C. Suppose T 1 satisfies Condition A<br />
and T 2 , T 3 are uniformly L-Lipschitzian and one <strong>of</strong> the conditions (C1)-(C5) in Theorem 2.3 is satisfied.<br />
Then the sequence {x n } defined as in (1) converges strongly to a common fixed point <strong>of</strong> T 1 , T 2 and T 3 .<br />
Pro<strong>of</strong>. Let q ∈ F. By Lemma 2.1, lim<br />
n→∞<br />
‖x n − q‖ exists. Thus {x n − q} is bounded. Then there<br />
is a constant H such that ‖x n − q‖ ≤ H for all n ≥ 1. This together <strong>with</strong> (6), we have<br />
(21)<br />
‖x n+1 − q‖ ≤ ‖x n − q‖ + D n ,<br />
where D n = KH ( (k n − 1)+ (l n − 1) +(m n − 1) ) + K(δ n + σ n + ρ n ) < ∞ for all n ≥ 1. By<br />
Lemma 2.2, we have lim ‖x n − T i x n ‖ = 0 (i = 1, 2, 3). Since T 1 satisfies Condition A, we obtain<br />
n→∞<br />
lim d(x n, F(T 1 )) = 0. Next, we want to show {x n } is a Cauchy sequence. Since lim d(x n , F(T 1 )) =<br />
n→∞ n→∞<br />
0 and ∞ ∑<br />
n=1<br />
D n < ∞, for any ɛ > 0, there exists a positive integer n 0 such that d(x n , F(T 1 )) < ɛ/4<br />
and ∑<br />
n D k < ɛ/2 for all n ≥ n 0 . Now, let n ∈ N be such that n ≥ n 0 . Then we can find q ∗ ∈ F<br />
k=n 0<br />
such that ‖x n − q ∗ ‖ < ɛ/4. This implies by (21) that for m ≥ 1,<br />
‖x n+m − x n ‖ ≤ ‖x n+m − q ∗ ‖ + ‖x n − q ∗ ‖<br />
≤ 2‖x n − q ∗ n+m−1<br />
‖ +<br />
∑<br />
k=n<br />
= 2‖x n − q ∗ n+m−1<br />
‖ +<br />
∑<br />
k=n 0<br />
D k<br />
D k < 2 ( ɛ ) ɛ +<br />
4 2 = ɛ.
181 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
This shows that {x n } is a Cauchy sequence and so it is convergent. Let lim<br />
n→∞<br />
x n = p. Since<br />
d(x n , F(T 1 )) → 0 as n → ∞, it follows that d(p, F(T 1 )) = 0 and hence p ∈ F(T 1 ). Next, we<br />
want to show p ∈ F(T 2 ) ∩ F(T 3 ). Since T 2 , T 3 are uniformly L-Lipschitzian and by Lemma 2.2,<br />
we obtain<br />
‖T i p − p‖<br />
≤ ‖T i x n − T i p‖ + ‖T i x n − x n ‖ + ‖x n − p‖<br />
≤ L‖x n − p‖ + ‖T i x n − x n ‖ + ‖x n − p‖ → 0 as n → ∞.<br />
Thus T i p = p (i = 2, 3). Therefore p ∈ F.<br />
□<br />
In the next result, we prove weak convergence for the iterative scheme (1) for asymptotically<br />
quasi-nonexpansive nonself-mappings in a uniformly convex Banach space satisfying<br />
Opial’s condition.<br />
Theorem 2.6. Let X be a uniformly convex Banach space which satisfies Opial’s condition and let C be a<br />
nonempty closed convex nonexpansive retract <strong>of</strong> X <strong>with</strong> P as a nonexpansive retraction. Let T 1 , T 2 , T 3 :<br />
C → X be asymptotically quasi-nonexpansive mappings <strong>with</strong> respect to sequences {k n }, {l n }, {m n },<br />
respectively, such that F ̸= ∅, k n ≥ 1, l n ≥ 1, m n ≥ 1,<br />
∞<br />
∑<br />
n=1<br />
(k n − 1) < ∞,<br />
∞<br />
∑ (l n − 1) < ∞ and<br />
n=1<br />
∞<br />
∑ (m n − 1) < ∞. Let {a n }, {b n }, {c n }, {α n }, {β n }, {γ n }, {δ n }, {σ n }, {ρ n } be real sequences in<br />
n=1<br />
[0, 1] such that a n + δ n , b n + c n + σ n and α n + β n + γ n + ρ n are in [0, 1] for all n ≥ 1 and ∞ ∑<br />
∞,<br />
∞<br />
∑<br />
n=1<br />
σ n < ∞,<br />
n=1<br />
δ n <<br />
∞<br />
∑ ρ n < ∞ and let {u n }, {v n }, {w n } be bounded sequences in C. Suppose T 1 , T 2 , T 3<br />
n=1<br />
are uniformly L-Lipschitzian and I − T i (i = 1, 2, 3) is demiclosed at 0. If one <strong>of</strong> the following conditions<br />
(C1)-(C5) in Theorem 2.3 is satisfied, then the sequence {x n } defined as in (1) converges weakly to a<br />
common fixed point <strong>of</strong> T 1 , T 2 and T 3 .<br />
Pro<strong>of</strong>. Assume one <strong>of</strong> the conditions (C1)-(C5) is satisfied. By Lemma 2.1 and Lemma 2.2,<br />
we have lim ‖T i x n − x n ‖ = 0 (i = 1, 2, 3). Since X is uniformly convex and {x n } is bounded,<br />
n→∞<br />
<strong>with</strong>out loss <strong>of</strong> generality, we may assume that x n → u weakly as n → ∞. Since I − T i is<br />
demiclosed at 0, we obtain u ∈ F. Suppose subsequences {x nk } and {x mk } <strong>of</strong> {x n } converge<br />
weakly to u and v, respectively. Also, since I − T i (i = 1, 2, 3) is demiclosed at 0, we have u and<br />
v ∈ F. By Lemma 2.1, we obtain lim ‖x n − u‖ and lim ‖x n − v‖ exist. It follows from Lemma<br />
n→∞ n→∞<br />
1.4 that u = v. Therefore {x n } converges weakly to a common fixed point <strong>of</strong> T 1 , T 2 and T 3 . □<br />
References<br />
[1] A. Bnouhachem, M.A. Noor, Th.M. Rassias, <strong>Three</strong>-steps iterative algorithms for mixed variational inequalities,<br />
Appl. Math. Comput. 183: 436-446 (2006).<br />
[2] H. Fukhar-ud-din, S.H. Khan, Convergence <strong>of</strong> iterates <strong>with</strong> errors <strong>of</strong> asymptotically quasi-nonexpansive mappings<br />
and applications, J. Math. Anal. Appl. 328: 821-829 (2007).<br />
[3] K. Goebel, W.A. Kirk, A fixed point theorem for asymptotically nonexpansive mappings, Proc. Amer. Math.<br />
Soc. 35: 171-174 (1972).<br />
[4] R. Glowinski, P. Le Tallec, Augmented Lagrangian and Operator-Splittting Methods in Nonlinear Mechanics,<br />
SIAM, Philadelphia, 1989.<br />
[5] S.H. Khan, W.Takahashi, Approximating common fixed points <strong>of</strong> two asymptotically nonexpansive mappings,<br />
Sci. Math. Jpn. 53: 133-138 (2001).<br />
[6] S.H. Khan, H. Fukhar-ud-din, Weak and strong convergence <strong>of</strong> a scheme <strong>with</strong> errors for two nonexpansive<br />
mappings, Nonlinear Anal. 8: 1295-1301 (2005).<br />
[7] K. Nammanee, M.A. Noor, S. Suantai, Convergence criteria <strong>of</strong> modified Noor iterations <strong>with</strong> errors for asymptotically<br />
nonexpansive mappings, J. Math. Anal. Appl., in press.<br />
[8] W. Nilsrakoo and S. Saejung, A new three-step fixed point iteration scheme for asymptotically nonexpansive<br />
mappings, J. Appl. Math. Comput. 181: 1026-1034 (2006).<br />
[9] M.A. Noor, <strong>New</strong> approximation schemes for general variational inequalities, J. Math. Anal. Appl. 251: 217-229<br />
(2000).<br />
[10] M.A. Noor, <strong>Three</strong>-step iterative algorithms for multivalued quasi variational inclusions, J. Math. Anal. Appl.<br />
255: 589-604 (2001).<br />
[11] M.A. Noor, Some developments in general variational inequalities, Appl. Math. Comput. 152: 199-277 (2004).
182 U. Inprasit and H. Wattanataweekul / Journal <strong>of</strong> Nonlinear Analysis and Optimization 1 (2010) 169-182<br />
[12] Z. Opial, Weak convergence <strong>of</strong> successive approximations for nonexpansive mappings, Bull. Amer. Math. Soc.<br />
73: 591-597 (1967).<br />
[13] L. Qihou, <strong>Iteration</strong> sequences for asymptotically quasi-nonexpansive mapping <strong>with</strong> an error member <strong>of</strong> uniform<br />
convex Banach space, J. Math. Anal. Appl. 266: 468-471 (2002).<br />
[14] H.F. Senter, W.G. Dotson, Approximating fixed points <strong>of</strong> nonexpansive mappings, Proc. Amer. Math. Soc. 44:<br />
375-380 (1974).<br />
[15] N. Shahzad, A. Udomene, Approximating common fixed points <strong>of</strong> two asymptotically quasi-nonexpansive<br />
mappings in Banach spaces, <strong>Fixed</strong> Point Theory Appl. (2006), article ID 18909, 10 pp.<br />
[16] S. Suantai, Weak and strong convergence criteria <strong>of</strong> Noor iterations for asymptotically nonexpansive mappings,<br />
J. Math. Anal. Appl. 311: 506-517 (2005).<br />
[17] K.K. Tan, H.K. Xu, Approximating fixed points <strong>of</strong> nonexpansive mappings by the Ishikawa iteration process,<br />
J. Math. Anal. Appl. 178: 301-308 (1993).<br />
[18] B.L. Xu, M. Aslam Noor, <strong>Fixed</strong> point iterations for asymptotically nonexpansive mappings in Banach spaces,<br />
J. Math. Anal. Appl. 267: 444-453 (2002).<br />
1,2 DEPARTMENT OF MATHEMATICS, STATISTICS AND COMPUTER, FACULTY OF SCIENCE,, UBON RATCHATHANI<br />
UNIVERSITY, UBON RATCHATHANI 34190, THAILAND.<br />
1,2 CENTRE OF EXCELLENCE IN MATHEMATICS, CHE, SI AYUTTHAYA RD., BANGKOK 10400, THAILAND.<br />
Email address: utith@sci.ubu.ac.th(U. Inprasit) and hathaikarn@sci.ubu.ac.th ( H. Wattanataweekul)