AN OTHER APPROACH FOR THE PROBLEM OF FINDING A

COMMON FIXED POINT OF A FINITE FAMILY OF

NONEXPANSIVE MAPPINGS

TRUONG MINH TUYEN

Address: Department of Mathematics, Thainguyen University - Vietnam

Abstract. The purpose of this paper is to give a Tikhonov regularization method and some regularization iner-
tial proximal point algorithm for the problem of finding a common fixed point of a finite family of nonexpansive
mappings in a uniformly convex and uniformly smooth Banach space E, which admits a weakly sequentially
continuous normalized duality mapping j from E to E ∗ .

Keywords: Accretive operators, uniformly smooth and uniformly convex Banach space, sunny nonexpansive
retraction, weak sequential continuous mapping, regularization.
2010 Mathematics Subject Classification: 47H06, 47H09, 47H10, 47J25.

1. Introduction
Let E be a Banach space. We consider the following problem
Finding an element x∗ ∈ S = ∩N
i=1 F (Ti ), (1.1)
where F (Ti ) is the set of fixed points of nonexpansive mappings Ti : C −→ C and C is a closed
convex nonexpansive retract subset of a uniformly convex and uniformly smooth Banach space
E.
It is well-known that, numerous problems in mathematics and physical sciences can be
recast in terms of a fixed point problem for nonexpansive mappings. For instance, if the
nonexpansive mappings are projections onto some closed and convex sets, then the fixed point
problem becomes the famous convex feasibility problem. Due to the practical importance
of these problems, algorithms for finding fixed points of nonexpansive mappings continue to
be flourishing topic of interest in fixed point theory. This problem has been investigated by
many researchers: see, for instance, Bauschke [7], O’ Hara et al. [22], Jung [16], Chang et
al. [10], Takahashi and Shimoji [27], Ceng et al. [9], Chidume et al. [11, 12], Plubtieng and
Ungchittrakool [23], Kang et al. [17], N. Buong et al. [8] and others.
On the other hand, the problem of finding a fixed point of a nonexpansive mapping T :
E −→ E is equivalent to the problem of finding a zero of m−accretive A = I − T. One of the
methods to solve the problem 0 ∈ A(x), where A is maximal monotone in a Hilbert space H
is proximal point algorithm. This algorithm suggested by Rockafellar [24], starting from any
initial guess x0 ∈ H, this algorithm generates a sequence {xn } given by
xn+1 = JcAn (xn + en ), (1.2)
where JrA = (I + rA)−1 ∀r > 0 is the resolvent of A on the space H. Rockafellar [24]
proved the weak convergence of the algorithm (1.2) provided that the regularization sequence
{cn } remains bounded away from zero and the error sequence {en } satisfies the condition
P ∞ ..
n=0 ken k < ∞. However Guler’s example [15] shows that in infinite dimensional Hilbert
space, proximal point algorithm (1.2) has only weak convergence.

Email address: [email protected] .

1
2 TRUONG MINH TUYEN

Ryazantseva [25] extended the proximal point algorithm (1.2) for the case that A is an
m−accretive mapping in a properly Banach space E and proved the weak convergence of the
sequence generated by (1.2) to a solution of the equation 0 ∈ A(x) which is assumed to be
unique. Then, to obtain the strong convergence for algorithm (1.2), Ryazantseva [26] combined
the proximal algorithm with the regularization, named regularization proximal algorithm, in
the form
cn (A(xn+1 ) + αn xn+1 ) + xn+1 = xn , x0 ∈ E. (1.3)
Under some conditions on cn and αn , the strong convergence of {xn } of (1.3) is guaranteed
only when the dual mapping j is weak sequential continuous and strong continuous, and the
sequence {xn } is bounded.
Attouch and Alvarez [6] considered an extension of the proximal point algorithm (1.2) in
the form
cn A(un+1 ) + un+1 − un = γn (un − un−1 ), u0 , u1 ∈ H, (1.4)
which is called an inertial proximal point algorithm, where {cn } and {γn } are two sequences
of positive numbers. With this algorithm we also only obtained weak convergence of the
sequence {xn } to a solution of problem A(x) 3 0 in Hilbert spaces. Note that this algorithm
was proposed by Alvarez in [2] in the context of convex minimization.
Then, Moudafi [19] applied this algorithm for variational inequalities, Moudafi and Elisabeth
[20] studied this algorithm by using enlargement of a maximal monotone operator, and Moudafi
and Oliny [21] considered convergence of a spliting inertial proximal method. The main result
in these papers is also the weak convergence of the algorithm in Hilbert spaces.
In this paper, we introduced the algorithms in the forms
N
X
Ai (xn ) + αn (xn − y) = 0, (1.5)
i=1
XN
cn ( Ai (un+1 ) + αn (un+1 − y)) + un+1 = QC (un + γn (un − un−1 )), (1.6)
i=1
where y, u0 , u1 ∈ C, and QC : E −→ C is a sunny nonexpansive retraction from E onto C
to solve the problem (1.1).
And also, we give some analogue regularization methods for the more general problems, such
as: the problem of finding a common fixed point of a finite family of nonexpansive nonself
- mapping on a closed and convex subset of E. Finally, the stability of the regularization
algorithms are considered in this paper.

2. Preliminaries
Let E be a Banach space with its dual space E ∗ . For the sake of simplicity, the norms of
E and E ∗ are denoted by the same symbol k.k. We write hx, x∗ i instead of x∗ (x) for x∗ ∈ E ∗
∗
and x ∈ E. We use the symbols *, * and −→ to denote the weak convergence, weak*
convergence and strong convergence, respectively.
Definition 2.1. A Banach space E is said to be uniformly convex if for any ε ∈ (0, 2] the
inequalities kxk ≤ 1, kyk ≤ 1, kx − yk ≥ ε imply there exists a δ = δ(ε) ≥ 0 such that
kx + yk
≤ 1 − δ.
2
The function
δE (ε) = inf{1 − 2−1 kx + yk : kxk = kyk = 1, kx − yk = ε} (2.1)
is called the modulus of convexity of the space E. The function δE (ε) defined on the interval
[0, 2] is continuous, increasing and δE (0) = 0. The space E is uniformly convex if and only if
 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 3

δE (ε) > 0, ∀ε ∈ (0, 2].

The function
ρE (τ ) = sup{2−1 kx + yk + kx − yk − 1 : kxk = 1, kyk = τ },


(2.2)
is called the modulus of smoothness of the space E. The function ρE (τ ) defined on the interval
[0, +∞) is convex, continuous, increasing and ρE (0) = 0.
Definition 2.2. A Banach space E is said to be uniformly smooth, if
ρE (τ )
lim = 0. (2.3)
τ →0 τ
It is well known that every uniformly convex and uniformly smooth Banach space is reflexive.
In what follows, we denote
ρE (τ )
hE (τ ) := . (2.4)
τ
The function hE (τ ) is nondecreasing. In addition, we have the following estimate
hE (Kτ ) ≤ LKhE (τ ), ∀K > 1, τ > 0, (2.5)
where L is the Figiel’s constant [3, 4, 13], 1 < L < 1.7.
Definition 2.3. A mapping j from E onto E ∗ satisfying the condition
J(x) = {f ∈ E ∗ : hx, f i = kxk2 and kf k = kxk} (2.6)
is called the normalized duality mapping of E.
In any smooth Banach space J(x) = 2−1 gradkxk2 and, if E is a Hilbert space, then J = I,
where I is the identity mapping. It is well known that if E ∗ is stricly convex or E is smooth,
then J is single valued. Suppose that J be single valued, then J is said to be weakly sequentially
∗
continuous if for each {xn } ⊂ E with xn * x, J(xn ) * J(x). We denote the single valued
normalized duality mapping by j.
Definition 2.4. An operator A : D(A) ⊆ E ⇒ E is called accretive if for all x, y ∈ D(A)
there exists j(x − y) ∈ J(x − y) such that
hu − v, j(x − y)i ≥ 0, ∀u ∈ A(x), v ∈ A(y). (2.7)
Definition 2.5. A mapping T : C −→ E is said to be nonexpansive on a closed and convex
subset C of Banach space E if
kT (x) − T (y)k ≤ kx − yk, ∀x, y ∈ C. (2.8)
It is clear that, if T : C −→ E is a nonexpansive, then I − T is accretive operator.
Definition 2.6. Let G be a nonempty closed and convex subset of E. A mapping QG : E −→
G is said to be
i) a retraction onto G if Q2G = QG ;
ii) a nonexpansive retraction if it also satisfies the inequality
kQG x − QG yk ≤ kx − yk, ∀x, y ∈ E; (2.9)
iii) a sunny retraction if for all x ∈ E and for all t ∈ [0, +∞),
QG (QG x + t(x − QG x)) = QG x. (2.10)
A closed and convex subset C of E is said to be a nonexpansive retract of E, if there exists
a nonexpansive retraction from E onto C and is said to be a sunny nonexpansive retract of
E, if there exists a sunny nonexpansive retraction from E onto C.
Proposition 2.7. [14] Let C be a nonempty closed convex subset of a smooth Banach E. A
mapping QC : E −→ C is a sunny nonexpansive retraction if and only if
hx − QC x, j(ξ − QC x)i ≤ 0, ∀x ∈ E, ∀ξ ∈ C. (2.11)
4 TRUONG MINH TUYEN

Definition 2.8. Let C1 , C2 be convex subsets of E. The quantity

β(C1 , C2 ) = sup inf ku − vk = sup d(u, C2 )
u∈C1 v∈C2 u∈C1

is said to be semideviation of the set C1 from the set C2 . The function

H(C1 , C2 ) = max{β(C1 , C2 ), β(C2 , C1 )}
is said to be a Hausdorff distance between C1 and C2 .
Lemma 2.9. [5] If E is a uniformly smooth Banach space, C1 and C2 are closed and con-
vex subsets of E such that the Hausdorff H(C1 , C2 ) ≤ δ, and QC1 and QC2 are the sunny
nonexpansive retractions onto the subsets C1 and C2 , respectively, then
16Lδ
kQC1 x − QC2 xk2 ≤ 16R(2r + d)hE ( ), (2.12)
R
where L is Figiel’s constant, r = kxk, d = max{d1 , d2 }, and R = 2(2r + d) + δ. Here
di = dist(θ, Ci ), i = 1, 2, and θ is the origin of the space E.

3. Main results
First, we need the following lemmas in the proof of our results.
Lemma 3.1. [3] Let E be a uniformly convex and uniformly smooth Banach space. If A =
I − T with a nonexpansive mapping T then for all x, y ∈ D(T ), the domain of T ,
 
kAx − Ayk
hAx − Ay, j(x − y)i ≥ L−1 R2 δE , (3.1)
4R
where kxk ≤ R, kyk ≤ R and 1 < L < 1.7 is Figiel constant.
Lemma 3.2 (demiclosedness principle). [1] Let E be a reflexive Banach space having a weakly
sequentially continuous duality mapping, C a nonempty closed convex subset of E, and T :
C −→ E a nonexpansive mapping. Then the mapping I − T is demiclosed on C, where I
is the identity mapping; that is, xn * x in E and (I − T )xn −→ y imply that x ∈ C and
(I − T )x = y.
Lemma 3.3. [28] Let {an } be a sequence of nonnegative real numbers satisfying the following
relation:
an+1 ≤ (1 − αn )an + σn , ∀n ≥ 0,
where {αn } ⊂ (0, 1) for each n ≥ 0 such that (i) limn→∞ αn = 0; (ii) ∞
P
n=1 αn = ∞. Suppose
P∞ σn
either (a) σn = o(αn ), or (b) n=1 |σn | < ∞, or (c) lim sup ≤ 0. Then an → 0 as n → ∞.
αn
Lemma 3.4. [18] Let C be a closed convex subset of a strictly convex Banach space E and let
T : C −→ E be a nonexpansive mapping from C into E. Suppose that C is be sunny nonex-
pansive retract of E. If F (T ) 6= ∅, then F (T ) = F (QC T ), where QC is a sunny nonexpansive
retraction from E onto C.
Theorem 3.5. Let E be a uniformly convex and uniformly smooth Banach space which admits
a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let C be a
nonempty closed convex sunny nonexpansive retract of E and let Ti : C −→ C, i = 1, 2, ..., N
be nonexpansive mappings such that S = ∩N i=1 F (Ti ) 6= ∅. Then
i) For each αn > 0 the equation (1.5) has unique solution xn ;
ii) If the sequence of positive numbers {αn } satisfies limn→∞ αn = 0, then {xn } converges
strongly to QS y, where QS : E −→ S is a sunny nonexpansive retraction from E onto
S.
 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 5

Moreover, we have the following estimate

|αn+1 − αn |
kxn+1 − xn k ≤ R0 ∀n ≥ 0, (3.2)
αn
where R0 = 2ky − QS yk.

Proof. i) For each n ≥ 0, equation (1.5) defines unique sequence {xn } ⊂ E, because for each
n, the element xn is unique fixed point of the contraction mapping T : C −→ C defined by
N
1 X αn
T (x) = Ti (x) + y. (3.3)
N + αn N + αn
i=1

ii) From equation (1.5), we have

XN
h Ai (xn ), j(xn − x∗ )i + αn hxn − y, j(xn − x∗ )i = 0, ∀x∗ ∈ S. (3.4)
i=1
PN
By virtue of the property of i=1 Ai and j, we obtain
N
X
h Ai (xn ), j(xn − x∗ )i ≥ 0, ∀x∗ ∈ S.
i=1

Thus,
hxn − y, j(xn − x∗ )i ≤ 0, ∀x∗ ∈ S. (3.5)
From inequality (3.5), we get
kxn − x∗ k2 ≤ hy − x∗ , j(xn − x∗ )i ≤ ky − x∗ k.kxn − x∗ k, ∀x∗ ∈ S. (3.6)
Therefore
kxn − x∗ k ≤ ky − x∗ k, ∀n ≥ 0, ∀x∗ ∈ S, (3.7)
that implies the boundedness of the sequence {xn }. Every bounded set in a reflexive Banach
space is relatively weakly compact. This means that there exists some subsequence {xnk } ⊆
{xn } which converges weakly to a limit point x. Since C is closed and convex, it is also weakly
closed. Therefore x ∈ C.
We will show that x ∈ S. Indeed, for each i ∈ {1, 2, ..., N }, x∗ ∈ S and R > 0 satisfy
R ≥ max{sup kxn k, kx∗ k}, we have
kAi (xn )k L
δE ( ) ≤ 2 hAi (xn ), j(xn − x∗ )i
4R R
N
L X
≤ 2h Ak (xn ), j(xn − x∗ )i
R
k=1
Lαn
≤ 2 kxn − yk.kxn − x∗ k
R
Lαn
≤ 2 (R + kyk).ky − x∗ k −→ 0, n −→ ∞.
R
Since modulus of convexity δE is continuous and E is the uniformly convex Banach space,
Ai (xn ) −→ 0, n −→ ∞. From Lemma 3.2, it implies that Ai (x) = 0. Since i ∈ {1, 2, ..., N } is
an arbitrary element, we obtain x ∈ S.
In inequality (3.6) replacing xn by xnk and x∗ by x, using the weak continuity of j we obtain
xnk −→ x. From inequality (3.5), we get
hx − y, j(x − x∗ )i ≤ 0, ∀x∗ ∈ S. (3.8)
6 TRUONG MINH TUYEN

Now, we show that the inequality (3.8) has unique solution. Suppose that x1 ∈ S is also its
solution. Then
hx1 − y, j(x1 − x∗ )i ≤ 0, ∀x∗ ∈ S. (3.9)
In inequalities (3.8) and (3.9) replacing x∗ by x1 and x, respectively, we obtain

hx − y, j(x − x1 )i ≤ 0,
hy − x1 , j(x − x1 )i ≤ 0.

Their combination gives kx − x1 k2 ≤ 0, thus x = x1 = QS y and the sequence {xn } converges

weakly to x = QS y, because QS y satisfies the inequality (3.8). Finally, from the first inequality
in (3.6), implies that xn −→ QS y.
Now, we prove the inequality (3.2). In equation (1.5) replacing n by n + 1 we have
N
X
Ai (xn+1 ) + αn+1 (xn+1 − y) = 0. (3.10)
i=1

From (3.10) and (1.5), we get

hαn+1 xn+1 − αn xn , j(xn+1 − xn )i ≤ (αn+1 − αn )hy, j(xn+1 − xn )i. (3.11)

Therefore,

αn kxn+1 − xn k2 ≤ (αn+1 − αn )hy − xn+1 , j(xn+1 − xn )i

≤ |αn+1 − αn |.ky − xn+1 k.kxn+1 − xn k
≤ 2ky − QS yk.|αn+1 − αn |.kxn+1 − xn k.

Thus,
|αn+1 − αn |
kxn+1 − xn k ≤ R0 ∀n ≥ 0,
αn
where R0 = 2ky − QS yk. 

Theorem 3.6. Let E be a uniformly convex and uniformly smooth Banach space which admits
a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let C be a
nonempty closed convex sunny nonexpansive retract of E and let Ti : C −→ C, i = 1, 2, ..., N
be nonexpansive mappings such that S = ∩N i=1 F (Ti ) 6= ∅. If the sequences {cn }, {αn } and
{γn } satisfy
|αn+1 − αn | P∞
i) 0 < c0 < cn , αn > 0, αn −→ 0, −→ 0, n=0 αn = +∞;
αn2
ii) γn ≥ 0, γn αn−1 kun − un−1 k −→ 0,
then the sequence {un } defined by (1.6) converges strongly to QS y, where QS : E −→ S is a
sunny nonexpansive retraction from E onto S.

Proof. First, for each n ≥ 1, equation (1.6) defines unique sequence {un } ⊂ E, because for
each n, the element un+1 is unique fixed point of the contraction mapping f : C −→ C defined
by
N
cn X cn αn 1
f (x) = Ti (x) + y+ z, (3.12)
cn (N + αn ) + 1 cn (N + αn ) + 1 cn (N + αn ) + 1
i=1

where z = QC (un + γn (un − un−1 )) ∈ C.

 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 7

Now, we rewrite equations (1.5) and (1.6) in their equivalent forms

N
X
dn Ai (xn ) + xn − y = βn (xn − y), (3.13)
i=1
XN
dn Ai (un+1 ) + un+1 − y = βn [QC (un + γn (un − un−1 )) − y], (3.14)
i=1

1
where βn = and dn = cn βn .
1 + cn αn
From (3.13), (3.14) and by virtue of the property of N
P
i=1 Ai , we get

kuu+1 − xn k ≤ βn kQC (un + γn (un − un−1 )) − xn k

= βn kQC (un + γn (un − un−1 )) − QC (xn )k
≤ βn kun − xn k + βn γn kun − un−1 k.
Thus,
kun+1 − xn+1 k ≤ kun+1 − xn k + kxn+1 − xn k
|αn+1 − αn | (3.15)
≤ βn kun − xn k + βn γn kun − un−1 k + R,
αn
or equivalent to
cn αn
kun+1 − xn+1 k ≤ (1 − bn )kun − xn k + σn , bn = , (3.16)
1 + cn αn
|αn+1 − αn |
where σn = βn γn kun − un−1 k + R.
αn
By the assumption, we have
σn 1 1 |αn+1 − αn |
= αn−1 γn kun − un−1 k + ( + 1) R
bn cn cn αn2
1 1 |αn+1 − αn |
≤ αn−1 γn kun − un−1 k + ( + 1) R −→ 0.
c0 c0 αn2
Furthermore, since ∞
P P∞
n=0 αn = +∞, n=0 bn = +∞.
By Lemma 3.3, we obtain kun − xn k −→ 0. Since xn −→ QS y, un −→ QS y. 
Corollary 3.7. Let E be a uniformly convex and uniformly smooth Banach space which
admits a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let
Ti : E −→ E, i = 1, 2, ..., N be nonexpansive mappings such that S = ∩Ni=1 F (Ti ) 6= ∅. If the
sequences {cn }, {αn } and {γn } satisfy
|αn+1 − αn | P∞
i) 0 < c0 < cn , αn > 0, αn −→ 0, 2
−→ 0, n=0 αn = +∞;
αn
ii) γn ≥ 0, γn αn−1 kun − un−1 k −→ 0,
then the sequence {un } defined by
XN
cn ( Ai (un+1 ) + αn un+1 ) + un+1 = un + γn (un − un−1 ), u0 , u1 ∈ E
i=1

converges strongly to QS y, where QS : E −→ S is a sunny nonexpansive retraction from E

onto S.

Proof. Applying Theorem 3.6 for C = E and y = θ, we obtain the proof of this corollary. 
8 TRUONG MINH TUYEN

Corollary 3.8. Let E be a uniformly convex and uniformly smooth Banach space which
admits a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let C be
a nonempty closed convex sunny nonexpansive retract of E and let fi : C −→ E, i = 1, 2, ..., N
be nonexpansive mappings such that S = ∩N i=1 F (fi ) 6= ∅. If the sequences {cn }, {αn } and {γn }
satisfy
|αn+1 − αn | P∞
i) 0 < c0 < cn , αn > 0, αn −→ 0, 2
−→ 0, n=0 αn = +∞;
αn
ii) γn ≥ 0, γn αn−1 kun − un−1 k −→ 0,
then the sequence {un } defined by
N
X
cn ( Bi (un+1 ) + αn (un+1 − y)) + un+1 = QC (un + γn (un − un−1 )), (3.17)
i=1
converges strongly to QS y, where Bi = I − QC fi , i = 1, 2, ..., N , QC is a sunny nonexpan-
sive retraction from E onto C, QS is a sunny nonexpansive retraction from E onto S, and
y, u0 , u1 ∈ C.
Proof. By Lemma 3.4, we have S = ∩N i=1 F (QC fi ). Applying Theorem 3.6 for Ti = QC fi , i =
1, 2, ..., N we obtain the proof of this corollary. 
Finally, we study stability of the algorithms (1.5) and (1.6) with respect to perturbations
of both operators Ti and constraint set C satisfying the following conditions:
(P1) Instead of C, there is a sequence of closed convex sunny nonexpasive retract subsets
Cn ⊂ E, n = 1, 2, 3, ... such that the Hausdorff H(Cn , C) ≤ δn , where {δn } is a
sequence of positive numbers with the propertie
δn+1 ≤ δn , ∀n ≥ 1. (3.18)
(P2) On the each set Cn , there is a nonexpansive self-mapping Tin
: Cn −→ Cn , i =
1, 2, ..., N satisfying the conditions: there exists the increasing positive for all t > 0
function g(t) and ξ(t) such that g(0) ≥ 0, ξ(0) = 0 and x ∈ Ck , y ∈ Cm , kx − yk ≤ δ,
then
kTik x − Tim yk ≤ g(max{kxk, kyk})ξ(δ). (3.19)
In this paper, we establish the convergence and stability of the Tikhonov regularization
method (1.5) and the regularization inertial proximal point algorithm (1.6) in the forms
N
X
Ani (zn ) + αn (zn − QCn y) = 0, (3.20)
i=1
XN
cn ( Ani (un+1 ) + αn (un+1 − QCn y)) + un+1 = QCn (un + γn (un − un−1 )), (3.21)
i=1
respectively, where u0 , u1 and y are elements in E, and Ani = I − Tin , i = 1, 2, ..., N ,
with respect to perturbations of the set C, and QCn : E −→ Cn is the sunny nonexpansive
retraction of E onto Cn .
Theorem 3.9. Let E be a uniformly convex and uniformly smooth Banach space which admits
a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let C be a
nonempty closed convex sunny nonexpansive retract of E and let Ti : C −→ C, i = 1, 2, ..., N
be nonexpansive mappings such that S = ∩N i=1 F (Ti ) 6= ∅.
i) For each αn > 0 equation (3.20) has unique solution zn ;
ii) If the conditions (P1) and (P2) are fulfilled and the sequences of positive numbers
{αn }, {δn } satisfy
δn + ξ(δn )
αn −→ 0, −→ 0, as n −→ ∞, (3.22)
αn
 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 9

then {zn } converges strongly to QS (QC y), where QS : E −→ S is a sunny nonexpan-

sive retraction from E onto S.
Moreover, if {αn } is a decreasing sequence, then we have the following estimate
δn + ξ(2δn ) αn − αn+1
kzn+1 − zn k ≤ 4δn + K + R
α αn (3.23)
p np
+ K3 LK4 hE (δn ), ∀n ≥ 0,
where R, K, K3 , K4 are any constants.
Proof. i) For each n ≥ 0, by an argument similar to the proof of Theorem 3.5, it follows that,
the equation (3.20) has a unique solution un+1 .
ii) Since the distance Hausdorff H(Cn , C) ≤ δn , therefore for each solution xn of equation
(1.5) (note that, in the case that the element y in (1.5) is replaced by QC y), there exists an
element un ∈ Cn such that kxn − un k ≤ δn .
From equations (1.5) and (3.20), we have
N
X
(Ani (zn ) − Ani (un )) + αn (zn − xn ) − αn (QCn y − QC y)
i=1
(3.24)
N
X
+ (Ani (un ) − Ai (xn )) = 0.
i=1
PN n
By virtue of the property of i=1 Ai and j, we get
N
X
h (Ani (zn ) − Ani (un )), j(zn − un )i ≥ 0,
i=1
that implies
αn hzn − xn , j(zn − un )i ≤ αn hQCn y − QC y, j(zn − un )i
N
X (3.25)
+ h (Ai (xn ) − Ani (un )), j(zn − un )i.
i=1
Thus,
N
X
αn kzn − un k ≤ αn kxn − un k + αn kQCn y − QC yk + kAi (xn ) − Ani (un )k
i=1
N
X
≤ αn δn + αn kQCn y − QC yk + kAi (xn ) − Ani (un )k.
i=1

Since H(Cn , C) ≤ δn , there exists constants K1 > 0 and K2 > 1 such that the inequalities
p p p
kQCn y − QC yk ≤ K1 hE (K2 δn ) ≤ K1 LK2 hE (δn )
hold.
Next, for each i ∈ {1, 2, ..., N }, we have
kAi (xn ) − Ani (un )k ≤ δn + g(max{kxn k, kun k})ξ(δn )
≤ δn + g(M )ξ(δn ),
where M = max{sup kxn k, sup kun k} < +∞.
Consequently,
p p
αn kzn − un k ≤ αn δn + αn K1 LK2 hE (δn ) + N (δn + g(M )ξ(δn )). (3.26)
10 TRUONG MINH TUYEN

Thus,
kzn − xn k ≤ kzn − un k + kxn − un k
p p δn + g(M )ξ(δn ) (3.27)
≤ 2δn + K1 LK2 hE (δn ) + N .
αn
δn + ξ(δn )
Since αn −→ 0, −→ 0, hence δn −→ 0 and hE (δn ) −→ 0. By inequality (3.27), we
αn
obtain kxn − zn k −→ 0. By Theorem 3.5, it implies that xn −→ QS (QC y), thus the sequence
{zn } also converges strongly to QS (QC y).
Finally, we prove the inequality (3.23). In equation (3.20) replacing n by n + 1, we have
N
X
An+1
i (zn+1 ) + αn (zn+1 − QCn+1 y) = 0. (3.28)
i=1
Since
H(Cn , Cn+1 ) ≤ H(Cn , C) + H(C, Cn+1 ) ≤ 2δn ,
we assert that for any zn+1 ∈ Cn+1 there exists an element vn ∈ Cn such that kzn+1 −vn k ≤ 2δn .
From equations (3.20) and (3.28), we obtain
N
X
(Ani (zn ) − Ani (vn )) + αn (zn − QCn y) − αn+1 (zn+1 − QCn+1 y)
i=1
N
X
+ (Ani (vn ) − An+1
i (zn+1 )) = 0.
i=1
PN n
By virtue of the property of i=1 Ai and j, we get
αn kzn − vn k ≤ αn+1 kvn − zn+1 k + |αn − αn+1 |.kvn − QCn yk
N
X (3.29)
+ αn+1 kQCn y − QCn+1 yk + kAni (vn ) − An+1
i (zn+1 )k
i=1
Since H(Cn , Cn+1 ) ≤ 2δn , there exists constants K3 > 0 and K4 > 1 such that the inequalities
p p p
kQCn y − QCn+1 yk ≤ K3 hE (K4 δn ) ≤ K3 LK4 hE (δn ) (3.30)
hold.
Since vn ∈ Cn , therefore
kvn − QCn yk ≤ kvn − yk ≤ sup kzn k + kyk + 2δ1 := R. (3.31)
Next, for each i ∈ {1, 2, ..., N }, we have
kAni (vn ) − An+1
i (zn+1 )k ≤ 2δn + kTin (vn ) − Tin+1 (zn+1 )k
≤ 2δn + g(max{kvn k, kzn+1 k})ξ(2δn ) (3.32)
0
≤ 2δn + g(M )ξ(2δn ),
where M 0 = max{sup kvn k, sup kzn k} < +∞.
Combining (3.29), (3.30), (3.31) and (3.32), we obtain
p p αn − αn+1 δn + ξ(2δn )
kzn − vn k ≤ 2δn + K3 LK4 hE (δn ) + R +K , (3.33)
αn αn
where K = max{2N, N g(M 0 )}.
Consequently,
δn + ξ(2δn ) αn − αn+1 p p
kzn+1 − zn k ≤ 4δn + K +R + K3 LK4 hE (δn ). (3.34)
αn αn

 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 11

Next, we will prove the strong convergence and stability of regularization inertial proximal
point algorithm (3.21) by the following theorem.

Theorem 3.10. Let E be a uniformly convex and uniformly smooth Banach space which
admits a weakly sequentially continuous normalized duality mapping j from E to E ∗ . Let C be
a nonempty closed convex sunny nonexpansive retract of E and let Ti : C −→ C, i = 1, 2, ..., N
be nonexpansive mappings such that S = ∩N i=1 F (Ti ) 6= ∅. If the conditions (P1) and (P2) are
fulfilled, and the sequences {αn }, {δn }, {c̃n } and {γn } satisfy
αn − αn+1 P∞
i) αn & 0, 2
−→ 0, as n −→ ∞, n=1 αn = +∞,
αn p
δn + ξ(2δn ) hE (δn )
ii) −→ 0, −→ 0, as n −→ ∞,
αn2 αn
iii) 0 < c0 < cn , γn ≥ 0, γn αn−1 kun − un−1 k −→ 0, as n −→ ∞,
then the sequence {un } defined by (3.21) converges strongly to QS (QC y), where QS : E −→ S
is a sunny nonexpansive retraction from E onto S.

Proof. First, for each n by an argument similar to the proof of Theorem 3.6, it follows that,
the equation (3.21) has unique solution un+1 ∈ Cn .
Now, we rewrite equations (3.20) and (3.21) in their equivalent forms
N
X
dn Ani (zn ) + zn − QCn y = βn (zn − QCn y), (3.35)
i=1
XN
dn Ani (un+1 ) + un+1 − QCn y = βn [QCn (un + γn (un − un−1 )) − QCn y], (3.36)
i=1

1
where βn = and dn = cn βn .
1 + cn αn
From (3.35), (3.36) and by virtue of the property of N n
P
i=1 Ai , we have

kuu+1 − zn k ≤ βn kQCn (un + γn (un − un−1 )) − zn k

= βn kQCn (un + γn (un − un−1 )) − QCn (zn )k
≤ βn kun − zn k + βn γn kun − un−1 k.

Consequently,
kun+1 − zn+1 k ≤ kun+1 − zn k + kzn+1 − zn k
δn + ξ(2δn )
≤ βn kun − zn k + βn γn kun − un−1 k + 4δn + K (3.37)
αn
αn − αn+1 p p
+R + K3 LK4 hE (δn ),
αn
or equivalent to
kun+1 − zn+1 k ≤ (1 − bn )kun − zn k + σn , (3.38)
cn αn
where bn = and
1 + c̃n αn
δn + ξ(2δn ) αn − αn+1
σn = βn γn kun − un−1 k + 4δn + K +R
αn αn
p p
+ K3 LK4 hE (δn ).
12 TRUONG MINH TUYEN

By the assumption, we obtain

 
σn 1 −1 1 αn − αn+1 δn δn + ξ(2δn )
= αn γn kun − un−1 k + ( + αn ) R+4 +K
bn cn cn αn2 αn αn2
p
1 p hE (δn )
+ ( + αn )K3 LK4
cn αn
 
1 −1 1 αn − αn+1 δn δn + ξ(2δn )
≤ αn γn kun − un−1 k + ( + αn ) R+4 +K
c0 c0 αn2 αn αn2
p
1 p hE (δn )
+ ( + αn )K3 LK4 −→ 0, n −→ ∞.
c0 αn
Since ∞
P P∞
n=0 αn = +∞, n=0 bn = +∞.
By Lemma 3.3, it implies that kun − zn k −→ 0. Since zn −→ QS (QC y), un −→ QS (QC y).

Acknowledgements The author thanks the reviewer for the valuable comments and sugges-
tions, which improved the presentation of this manuscript.

References
1. R. P. Agarwal, D. O’Regan, D. R. Sahu, Fixed Point Theory for Lipschitzian-type Mappings with Applica-
tions, Springer, 2009.
2. F. Alvarez, On the minimizing property of a second order dissipative system in Hilbert space, SIAM J. of
Control and Optimization, 38(4) (2000), 1102-1119.
3. Y. Alber, On the stability of iterative approximatins to fixed points of nonexpansive mappings, J. Math.
Anal. Appl., 328 (2007), 958-971.
4. Y. Alber, I. Ryazantseva, Nonlinear ill-posed problems of monotone type, Springer, 2006.
5. Y. Alber, S. Reich, J. C. Yao, Iterative methods for solving fixed point problems with nonself-mappings in
Banach spaces, Abstract and Applied Analysis, 4 (2003), 194-216.
6. F. Alvarez and H. Attouch, An inertial proximal method for maximal monotone operators via discretization
of a nonolinear oscillator with damping, Set-Valued Analysis, (2001), 3-11.
7. H. H. Bauschke, The approximation of fixed points of compositions of nonexpansive mappings in Hilbert
spaces, J. Math. Anal. Appl., 202 (1996), 150-159.
8. N. Buong, N. T. Q. Anh, An implicit iteration method for variational inequalities over the set of common
fixed points for a finite family of nonexpansive mappings in Hilbert spaces, Volume 2011 (2011), Article ID
276859, 10 pages.
9. L. C. Ceng, P. Cubiotti, J.-C. Yao, Strong convergence theorems for finitely many nonexpansive mappings
and applications. Nonlinear Anal., 67 (2007), 1464-1473.
10. S.-S. Chang, J.- C. Yao, J. K. Kim, L. Yang, Iterative approximation to convex feasibility problems in
Banach space, Fixed Point Theory and Appl., Volume 2007 (2007), Article ID 46797, 19 pages.
11. C. E. Chidume, B. Ali, Convergence theorems for common fixed points for infinite families of nonexpansive
mappings in reflexive Banach spaces, Nonlinear Anal., 68 (2008), 3410-3418.
12. C. E. Chidume, H. Zegeye, N. Shahzad, Convergence theorems for a common fixed point of finite family of
nonself nonexpansive mappings, Fixed Point Theory and Appl., 2 (2005), 233-241.
13. T. Figiel, On the modunli of convexity and smoothness, Studia Math., 56 (1976), 121-155.
14. K. Goebel, S. Reich, Uniform convexity, hyperbolic geometry and nonexpansive mappings, Marcel Dekker,
New York and Basel,1984.
..
15. O. Guler, On the convergence of the proximal point algorithm for convex minimization, SIAM Journal on
Control and Optimization, 29(2) (1991), 403-419.
16. J. S. Jung, Iterative approaches to common fixed points of nonexpansive mappings in Banach spaces, J.
Math. Anal. Appl., 302 (2005), 509-520.
17. J. I. Kang, Y. J. Cho, H. Zhou, Approximation of common fixed points for a class of finite nonexpansive
mappings in Banach spaces, J. Comp. Anal. Appl., 8(1) (2006), 25-38.
18. S. Matsushita and W. Takahashi, Strong convergence theorem for nonexpansive nonself-mappings without
boundary conditions, Nonlinear Anal., 68 (2008), 412-419.
19. A. Moudafi, A hybrid inertial projection-proximal method for variational inequalities, J. of Inequalities in
Pure and Applied Math., 5(3) (2004), Article 63.
20. A. Moudafi and E. Elizabeth, An approximate inertial proximal mothod using the enlargement of a mono-
tone operator, Intern. J. of Pure and Appl. Math., 5(2) (2003), 283-299.
 AN OTHER APPROACH FOR THE PROBLEM OF FINDING... 13

21. A. Moudafi and M. Oliny, Convergence of a spliting inertial proximal method for monotone operator, J. of
Comp. and Appl. Math., 155 (2003), 447-454.
22. J. G. O’Hara, P. Pilla, H. K. Xu, Iterative approaches to finding nearest common fixed points of nonex-
pansive mappings in Hilbert spaces, Nonlinear Anal., 54 (2003), 1417-1426.
23. S. Plubtieng, K. Ungchittrakool, Weak and strong convergence of finite family with errors of nonexpansive
nonself-mappings, Fixed Point Theory and Appl., 2006 (2006), Article ID 81493, 1-12.
24. R. T. Rockaffelar, Monotone operators and proximal point algorithm, SIAM Journal on Control and Optim.,
5 (1976), 877-898.
25. I. P. Ryazanseva, Regularization for equations with accretive operators by the method of sequential ap-
proximations, Sibir. Math. J., 21(1) (1985), 223-226
26. I. P. Ryazanseva, Regularization proximal algorithm for nonlinear equations of monotone type, Zh. Vychisl.
Mat. i Mat. Fiziki, 42(9) (2002), 1295-1303.
27. W. Takahashi, K. Shimoji, Convergence theorem for nonexpansive mappings and feasibility problems,
Math. Comp. Mod., 32 (2000), 1463-1471.
28. H. K. Xu, Iterative algorithms for nonlinear operators, J. London Math. Soc., 66 (2002), 240-256.