Nonlinear Analysis 67 (2007) 2350–2360

www.elsevier.com/locate/na

Approximation of common fixed points of a countable family of

nonexpansive mappings in a Banach space
Koji Aoyama a,∗ , Yasunori Kimura b , Wataru Takahashi b , Masashi Toyoda c
a Department of Economics, Chiba University, Yayoi-cho, Inage-ku, Chiba-shi, Chiba 263-8522, Japan
b Department of Mathematical and Computing Sciences, Tokyo Institute of Technology, Ookayama, Meguro-ku, Tokyo 152-8552, Japan
c Faculty of Engineering, Tamagawa University, Tamagawa-Gakuen, Machida-shi, Tokyo 194-8610, Japan

Received 3 August 2006; accepted 28 August 2006

Abstract

In this paper, to find a common fixed point of a family of nonexpansive mappings, we introduce a Halpern type iterative
sequence. Then we prove that such a sequence converges strongly to a common fixed point of nonexpansive mappings. Moreover,
we apply our result to the problem of finding a common fixed point of a countable family of nonexpansive mappings and the
problem of finding a zero of an accretive operator.
c 2006 Elsevier Ltd. All rights reserved.

MSC: 47H09; 47H10; 41A65

Keywords: Nonexpansive mapping; Halpern type iteration; Strong convergence

1. Introduction

Let C be a closed convex subset of a Banach space E and T a nonexpansive mapping of C into itself, that is,
kT x − T yk ≤ kx − yk for all x, y ∈ C, where k·k is the norm of E. Halpern [3] introduced the following iterative
scheme for approximating a fixed point of T :
xn+1 = αn x + (1 − αn )T xn (1.1)
for all n ∈ N, where x1 = x ∈ C and {αn } is a sequence of [0, 1]. This iteration process is called a Halpern type
iteration. Strong convergence of this type iterative sequence has been widely studied: Wittmann [15] discussed such
a sequence in a Hilbert space. Shioji and Takahashi [12] extended Wittmann’s result; they proved strong convergence
of {xn } defined by (1.1) in a Banach space; see also Kamimura and Takahashi [6] and Iiduka and Takahashi [5]. On
the other hand, Bauschke [1] used a Halpern type iterative scheme to find a common fixed point of a finite family of
nonexpansive mappings in a Hilbert space. Kimura et al. [7] generalized the result of Shioji and Takahashi [12] and

∗ Corresponding author.
E-mail addresses: [email protected] (K. Aoyama), [email protected] (Y. Kimura), [email protected] (W. Takahashi),
[email protected] (M. Toyoda).

0362-546X/$ - see front matter c 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.na.2006.08.032
 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360 2351

studied strong convergence to a common fixed point of a finite family of nonexpansive mappings in a Banach space;
see also [11,8].
In this paper, motivated by these results related to Halpern type iterative schemes, we introduce the following
iterative sequence: Let x1 = x ∈ C and
xn+1 = αn x + (1 − αn )Tn xn (1.2)
for all n ∈ N, where C is a nonempty closed convex subset of a Banach space, {αn } is a sequence of [0, 1], and {Tn }
is a sequence of nonexpansive mappings with some conditions. Then we prove that {xn } defined by (1.2) converges
strongly to a common fixed point of {Tn } (Theorem 3.4). Further we apply our result to the problem of finding a
common fixed point of a countable family of nonexpansive mappings and the problem of finding a zero of an accretive
operator.

2. Preliminaries

Let E be a real Banach space with norm k·k. The dual of E is denoted by E ∗ , the value of f ∈ E ∗ at x ∈ E by
hx, f i, and the set of all positive integers by N.
∗
The duality mapping J of E into 2 E is defined by
J (x) = {x ∗ ∈ E ∗ : x, x ∗ = kxk2 = kx ∗ k2 }
for all x ∈ E. It is known that
kx + yk2 ≤ kxk2 + 2 hy, f i (2.1)
holds for all x, y ∈ E and f ∈ J (x + y). Let U = {x ∈ E : kxk = 1}. The norm k·k of E is said to be Gâteaux
differentiable if the limit
kx + t yk − kxk
lim (2.2)
t→0 t
exists for all x, y ∈ U . In this case a Banach space E is said to be smooth. The norm of E is said to be uniformly
Gâteaux differentiable if for each y ∈ U , the limit (2.2) is attained uniformly for x ∈ U . We know that the duality
mapping J is single-valued if E is smooth. We also know that J is uniformly norm-to-weak* continuous on each
bounded subset of E if the norm of E is uniformly Gâteaux differentiable; see [13] for more details. A Banach space
E is said to be strictly convex if the following implication holds for x, y ∈ E:
x+y
kxk = kyk = 1 and x 6= y imply < 1.
2
A Banach space E is said to be uniformly convex if for any  > 0, there exists δ > 0 such that
x+y
kxk = kyk = 1 and kx − yk ≥  imply ≤ 1 − δ.
2
It is known that if E is a uniformly convex Banach space, then E is reflexive and strictly convex.
Let C be a subset of a Banach space E. A mapping T of C into itself is said to be nonexpansive if kT x − T yk ≤
kx − yk holds for all x, y ∈ C. The set of all fixed points of T is denoted by F(T ), that is, F(T ) = {x ∈ C : x = T x}.
The following lemma is essentially proved in [12], but we give the proof for the sake of completeness.

Lemma 2.1 (Shioji–Takahashi [12]). Let C be a bounded closed convex subset of a smooth Banach space E. Let
T be a nonexpansive mapping of C into itself and x ∈ C. For each t ∈ (0, 1), let z t be a unique point of C with
z t = t x + (1 − t)T z t . Let {xn } be a sequence of C. If limn→∞ kT xn − xn k = 0, then
t
lim sup hx − z t , J (xn − z t )i ≤ lim sup kz t − xn k2
n→∞ 2 n→∞
for all t ∈ (0, 1), where J is the duality mapping of E.
2352 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360

Proof. From (2.1), we have

kz t − xn k2 = k(1 − t)(T z t − xn ) + t (x − xn )k2
≤ (1 − t)2 kT z t − xn k2 + 2t hx − xn , J (z t − xn )i
≤ (1 − t)2 (kT z t − T xn k + kT xn − xn k)2
+ 2t (hz t − xn , J (z t − xn )i + hx − z t , J (z t − xn )i)
≤ (1 − t)2 (kz t − xn k2 + kT xn − xn k (kT xn − xn k + 2 kz t − xn k))
+ 2t (kz t − xn k2 − hx − z t , J (xn − z t )i)
for all t ∈ (0, 1) and n ∈ N. Therefore,
t (1 − t)2
hx − z t , J (xn − z t )i ≤ kz t − xn k2 + kT xn − xn k (kT xn − xn k + 2 kz t − xn k)
2 2t
for all t ∈ (0, 1) and n ∈ N. Hence we get
t
lim sup hx − z t , J (xn − z t )i ≤ lim sup kz t − xn k2
n→∞ 2 n→∞
for all t ∈ (0, 1). 
Let C be a subset of a Banach space E, D a subset of C, and Q a mapping of C into D. Then Q is said to be sunny if
Q(Qx + t (x − Qx)) = Qx
holds whenever Qx + t (x − Qx) ∈ C for x ∈ C and t ≥ 0. A mapping Q of C into itself is said to be a retraction
if Q 2 = Q. A subset D of C is said to be a sunny nonexpansive retract of C if there exists a sunny nonexpansive
retraction from C onto D. We know the following theorem [14]; see also [10].

Theorem 2.2 (Takahashi–Ueda [14]). Let E be a uniformly convex Banach space whose norm is uniformly Gâteaux
differentiable, C a nonempty closed convex subset of E, and T a nonexpansive mapping of C into itself with
F(T ) 6= ∅. Then F(T ) is a sunny nonexpansive retract of C. Moreover, let x ∈ C and let z t be a unique point
of C which satisfies z t = t x + (1 − t)T z t for t ∈ (0, 1). Then {z t } converges strongly to Qx as t → 0, where Q is the
sunny nonexpansive retraction of C onto F(T ).
Next we prove the following lemma:
P∞
P∞numbers, {αn } a sequence of [0, 1] with n=1 αn = ∞,
Lemma 2.3. Let {sn } be a sequence of nonnegative real
{βn } a sequence of nonnegative real numbers with n=1 βn < ∞, and {γn } a sequence of real numbers with
lim supn→∞ γn ≤ 0. Suppose that
sn+1 ≤ (1 − αn )sn + αn γn + βn
for all n ∈ N. Then limn→∞ sn = 0.
Proof. Let  > 0 be given. There exists N ∈ N such that γm ≤  for all m ≥ N . Thus,
sm+1 ≤ (1 − αm )sm + αm  + βm
for all m ≥ N . By induction on n, it follows that
!
n+m−1
Y n+m−1
Y n+m−1
X
sn+m ≤ (1 − αk )sm + 1 − (1 − αk )  + βk
k=m k=m k=m
for all m ≥ N and n ∈ N. Thus, we have
lim sup sn = lim sup sn+m
n→∞ n→∞
!
∞
Y ∞
Y ∞
X
≤ (1 − αk )sm + 1 − (1 − αk )  + βk
k=m k=m k=m
 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360 2353

∞
X
=+ βk
k=m

for all m ≥ N . Therefore, we have

∞
X
lim sup sn ≤  + lim sup βk = .
n→∞ m→∞ k=m

Since  is arbitrary, we conclude that limn→∞ sn = 0. 

The following theorem was proved by Bruck [2].

Theorem 2.4 (Bruck [2]). Let C be a nonempty closed convex subset of a strictly convex Banach space E. LetP{Sk }
be a sequence
T∞of nonexpansive mappings of C into E and {β such that ∞
Pk }∞a sequence of positive real numbers T k=1
βk = 1. If k=1 F(Sk ) is nonempty, then the mapping T = k=1 βk Sk is well defined and F(T ) = ∞ k=1 F(Sk ).

It is easily seen that the mapping T in this theorem is also nonexpansive. Therefore, we obtain that the set of common
fixed points of a countable family of nonexpansive mappings is a sunny nonexpansive retract if E is uniformly convex
and has a uniformly Gâteaux differentiable norm; see, for example, [13].

3. Main result

In this section, we consider a Halpern type iterative scheme for a family of nonexpansive mappings in a Banach
space. Throughout this section, let E be a uniformly convex Banach space whose norm is uniformly Gâteaux
T C a nonempty closed convex subset of E, and {Tn } a sequence of nonexpansive mappings of C into
differentiable,
itself with ∞ n=1 F(Tn ) 6= ∅. We consider an iterative sequence {x n } of C defined by x 1 = x ∈ C and

xn+1 = αn x + (1 − αn )Tn xn
for each n ∈ N, where {αn } is a sequence of [0, 1].
To obtain our result, we need some lemmas.
P∞
Lemma 3.1. Let {αn } be a sequence of [0, 1] such that n=1 αn = ∞. Suppose that C is bounded and
∞
X
sup{kTn+1 z − Tn zk : z ∈ C} < ∞.
n=1

If either
(i) ∞ n=1 |αn+1 − αn | < ∞ or
P
(ii) αn ∈ (0, 1] for every n ∈ N and limn→∞ αn /αn+1 = 1,
then limn→∞ kxn+1 − xn k = 0.
Proof. Since C is bounded, {x − Tn+1 xn } is bounded. Let M = sup{kx − Tn+1 xn k : n ∈ N}. From the definition of
{xn }, it is easily seen that
kxn+2 − xn+1 k = kαn+1 x + (1 − αn+1 )Tn+1 xn+1 − αn x − (1 − αn )Tn xn k
≤ k(1 − αn+1 )(Tn+1 xn+1 − Tn+1 xn )k
+ k(αn+1 − αn )(x − Tn+1 xn )k + k(1 − αn )(Tn+1 xn − Tn xn )k
≤ (1 − αn+1 ) kxn+1 − xn k
+ |αn+1 − αn | kx − Tn+1 xn k + (1 − αn ) kTn+1 xn − Tn xn k
≤ (1 − αn+1 ) kxn+1 − xn k + |αn+1 − αn | M + kTn+1 xn − Tn xn k (3.1)
for each n ∈ N. Assume that (i) holds. Putting
βn = |αn+1 − αn | M + kTn+1 xn − Tn xn k ,
2354 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360

we have
X∞ ∞
X
βn = (|αn+1 − αn | M + kTn+1 xn − Tn xn k)
n=1 n=1
∞
X ∞
X
≤ M |αn+1 − αn | + sup{kTn+1 z − Tn zk : z ∈ C} < ∞.
n=1 n=1

Therefore it follows from Lemma 2.3 that limn→∞ kxn+1 − xn k = 0. If (ii) holds, then, from (3.1), we have
αn
kxn+2 − xn+1 k ≤ (1 − αn+1 ) kxn+1 − xn k + αn+1 1 − M + kTn+1 xn − Tn xn k
αn+1
for every n ∈ N. Since limn→∞ |1 − αn /αn+1 | = 0, it follows from Lemma 2.3 that limn→∞ kxn+1 − xn k = 0. 
P∞
Lemma 3.2. Suppose that n=1 sup{kTn+1 z − Tn zk : z ∈ C} < ∞. Then, for each y ∈ C, {Tn y} converges strongly
to some point of C. Moreover, let T be a mapping of C into itself defined by
T y = lim Tn y for all y ∈ C.
n→∞

Then limn→∞ sup{kT z − Tn zk : z ∈ C} = 0.

Proof. The assumption guarantees that {Tn y} is a Cauchy sequence for each y ∈ C. Indeed, let k, l ∈ N with k > l.
We have that
kTk y − Tl yk ≤ sup{kTk z − Tl zk : z ∈ C}
≤ sup{kTk z − Tk−1 zk : z ∈ C} + sup{kTk−1 z − Tl zk : z ∈ C}
≤ ···
k−1
X
≤ sup{kTn+1 z − Tn zk : z ∈ C}
n=l
∞
X
≤ sup{kTn+1 z − Tn zk : z ∈ C}. (3.2)
n=l

Taking the limit l → ∞, we get that ∞

P
n=l sup{kTn+1 z − Tn zk : z ∈ C} converges to 0. Therefore we have that {Tn y}
is a Cauchy sequence and hence it converges strongly to some point of C. Moreover, (3.2) implies that
∞
X
kT y − Tl yk = lim kTk y − Tl yk ≤ sup{kTn+1 z − Tn zk : z ∈ C}
k→∞
n=l

for all y ∈ C and l ∈ N. So,

∞
X
sup{kT y − Tl yk : y ∈ C} ≤ sup{kTn+1 z − Tn zk : z ∈ C}
n=l

holds for all l ∈ N. Therefore we conclude that liml→∞ sup{kT y − Tl yk : y ∈ C} = 0. This completes the proof. 

Lemma 3.3. Let {αn } be a sequence of [0, 1] such that limn→∞ αn = 0. Suppose that C is bounded and
∞
n=1 sup{kTn+1 z − Tn zk : z ∈ C} < ∞. Let T be a mapping of C into itself defined by T z = limn→∞ Tn z for
P
all z ∈ C. Then F(T ) is a sunny nonexpansive retract of C. Moreover, if limn→∞ kxn+1 − xn k = 0, then
lim sup hx − Qx, J (xn − Qx)i ≤ 0,
n→∞

where Q is the sunny nonexpansive retraction of C onto F(T ).

Proof. It is clear that T is a nonexpansive mapping C into itself and ∞ n=1 F(Tn ) ⊂ F(T ). So, we have F(T ) 6= ∅.
T
For each t ∈ (0, 1), let z t be a unique point of C such that z t = t x + (1 − t)T z t . It follows from Theorem 2.2 that
 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360 2355

F(T ) is a sunny nonexpansive retract of C and z t converges strongly to Qx ∈ F(T ) as t → 0, where Q is a sunny
nonexpansive retraction of C onto F(T ).
Let  > 0 be given. From the fact that limt→0 z t = Qx and C is bounded, there exists t1 ∈ (0, 1) such that

hx − Qx, J (xn − z t )i − hx − z t , J (xn − z t )i ≤ kz t − Qxk kxn − z t k <
3
for all t ∈ (0, t1 ) and n ∈ N. Since the norm of E is uniformly Gâteaux differentiable, J is uniformly norm-to-weak*
continuous on C. Hence there exists t2 ∈ (0, 1) such that

hx − Qx, J (xn − Qx)i − hx − Qx, J (xn − z t )i <
3
for all t ∈ (0, t2 ) and n ∈ N.
The definitions of T and xn imply that
kT xn − xn k ≤ kT xn − Tn xn k + kTn xn − xn+1 k + kxn+1 − xn k
≤ sup{kT z − Tn zk : z ∈ C} + αn kx − Tn xn k + kxn+1 − xn k
for each n ∈ N. Since limn→∞ sup{kTn z − T zk : z ∈ C} = 0, limn→∞ αn = 0, limn→∞ kxn+1 − xn k = 0, and C is
bounded, we get limn→∞ kT xn − xn k = 0. Therefore it follows from Lemma 2.1 that
t
lim sup hx − z t , J (xn − z t )i ≤ lim sup kz t − xn k2
n→∞ 2 n→∞
for all t ∈ (0, 1). So, there exists t3 ∈ (0, 1) such that

lim sup hx − z t , J (xn − z t )i <
n→∞ 3
for all t ∈ (0, t3 ).
On the other hand, we have
hx − Qx, J (xn − Qx)i = hx − Qx, J (xn − Qx)i − hx − Qx, J (xn − z t )i
+ hx − Qx, J (xn − z t )i − hx − z t , J (xn − z t )i + hx − z t , J (xn − z t )i
for all t ∈ (0, 1) and n ∈ N. Hence we have
  
lim sup hx − Qx, J (xn − Qx)i < + + = 
n→∞ 3 3 3
for all t ∈ (0, min{t1 , t2 , t3 }). Since  is arbitrary, we deduce that
lim sup hx − Qx, J (xn − Qx)i ≤ 0.
n→∞

This completes the proof. 

Now we prove the following theorem which is the main result of this paper.

Theorem 3.4. Let E be a uniformly convex Banach space whose norm is uniformly Gâteaux differentiable and C a
nonempty
T∞ closed convex subset of E. Let {Tn } be a sequence of nonexpansive mappingsP∞of C into itself such that
n=1 F(T n ) is nonempty and {α n } a sequence of [0, 1] such that lim α
n→∞ n = 0 and n=1 αn = ∞. Let {x n } be a
sequence of C defined as follows: x1 = x ∈ C and
xn+1 = αn x + (1 − αn )Tn xn
for every n ∈ N. Suppose that ∞ n=1 sup{kTn+1 z − Tn zk : z ∈ B} < ∞ for any bounded subset
P
T B of C. Let T be a
mapping of C into itself defined by T z = limn→∞ Tn z for all z ∈ C and suppose that F(T ) = ∞ n=1 F(Tn ). If either
P∞
(i) n=1 |αn+1 − αn | < ∞ or
(ii) αn ∈ (0, 1] for every n ∈ N and limn→∞ αn /αn+1 = 1,
then {xn } converges strongly to Qx, where Q is the sunny nonexpansive retraction of E onto F(T ) = ∞n=1 F(Tn ).
T
2356 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360
T∞
Proof. Let u ∈ n=1 F(Tn ). By the definition of xn , we have
kx2 − uk ≤ α1 kx − uk + (1 − α1 ) kT1 x1 − uk
= α1 kx − uk + (1 − α1 ) kT1 x1 − T1 uk
≤ α1 kx − uk + (1 − α1 ) kx1 − uk
= kx − uk.
By induction on n, we obtain kxn − uk ≤ kx − uk for every n ∈ T N. So, {xn } is bounded and, without loss of generality,
∞
we
T∞ may assume that C is bounded. By the assumption F(T ) = n=1 F(Tn ) 6= ∅, Lemmas 3.1 and 3.3, we know that
n=1 F(T n ) is a sunny nonexpansive retract of C and
lim sup hx − Qx, J (xn − Qx)i ≤ 0, (3.3)
n→∞
T∞
where Q is a sunny nonexpansive retraction of C onto n=1 F(Tn ). Since Tn Qx = Qx for every n ∈ N, by (2.1), we
have
kxn+1 − Qxk2 = k(1 − αn )(Tn xn − Qx) + αn (x − Qx)k2
≤ (1 − αn )2 kTn xn − Tn Qxk2 + 2αn hx − Qx, J (xn+1 − Qx)i
≤ (1 − αn )2 kxn − Qxk2 + 2αn hx − Qx, J (xn+1 − Qx)i
≤ (1 − αn ) kxn − Qxk2 + 2αn hx − Qx, J (xn+1 − Qx)i
for all n ∈ N. By using Lemma 2.3 combined with (3.3), we have limn→∞ kxn − Qxk2 = 0. Consequently, we
deduce that {xn } converges strongly to Qx, which completes the proof. 

4. Applications

In our main theorem, we assume that ∞ n=1 sup{kTn+1 z − Tn zk : z ∈ B} < ∞ for any bounded subset B of C. In
P
general, one cannot apply this result for a sequence of nonexpansive mappings. However, we can generate a sequence
{Tn } of nonexpansive mappings satisfying this condition by using convex combination of a general sequence {Sk } of
nonexpansive mappings with a common fixed point.
Let {βnk } be a family of nonnegative numbers with indices n, k ∈ N with k ≤ n. Suppose that the following
conditions hold:
(i) nk=1 βnk = 1 for every n ∈ N;
P

Pβn n > 0k for every

(ii) lim k k ∈ N;
P∞n→∞
(iii) n=1 k=1 βn+1 − βnk < ∞.
Let C be a nonempty closed convex subset of a uniformly convex Banach space E. For a sequence {Sk } of
nonexpansive mappings of C into itself with a common fixed point, we define a sequence {Tn } of mappings on C
by
n
X
Tn x = βnk Sk x
k=1

for x ∈ C andT n ∈ N. ThenTfrom the condition (i), each Tn is also a nonexpansive mapping of C into itself. It is
obvious that ∞ k=1 F(Sk ) ⊂
∞
n=1 F(Tn ). Further, by (ii) we have that for every k ∈ N, there
T exists n 0 ∈ N such that
βnk0 > 0. Thus we have that F(Tn 0 ) ⊂ F(Sk ) for k ∈ N by Theorem 2.4, which implies that ∞ n=1 F(Tn ) ⊂ F(Sk ) for
all k ∈ N. Therefore we obtain that
∞
\ \∞
F(Tn ) = F(Sk ) 6= ∅.
n=1 k=1

Let B be a bounded subset of C. Then, since ∞

k=1 F(Sk ) is not empty, {Sk x : x ∈ B, k ∈ N} is also bounded. Let
T
M = sup{kSk xk : x ∈ B, k ∈ N}. For x ∈ B and n ∈ N, we have that
n+1
X n
X
kTn+1 x − Tn xk = βn+1
k
Sk x − βnk Sk x
k=1 k=1
 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360 2357

n
X
≤ βn+1
k
− βnk kSk xk + βn+1
n+1
kSn+1 xk
k=1
!
n
X n
X
≤ βn+1
k
− βnk M + 1 − βn+1
k
M
k=1 k=1
!
n
X n
X n
X
= βn+1
k
− βnk M+ βnk − βn+1
k
M
k=1 k=1 k=1
n
X
≤ 2M βn+1
k
− βnk .
k=1
Pn
Thus we have that sup{kTn+1 x − Tn xk : x ∈ B} ≤ 2M k=1 βn+1
k − βnk and hence, using (iii), we have that
∞
X ∞ X
X n
sup{kTn+1 x − Tn xk : x ∈ B} ≤ 2M βn+1
k
− βnk < ∞.
n=1 n=1 k=1

For l, m ∈ N with l > m, we have that

l−1
X
βlk − βmk ≤ βn+1
k
− βnk
n=m

for all k ∈ N with k ≤ m. Summing up for k = 1, 2, . . . , m, we get that

m
X m X
X l−1
βlk − βmk ≤ βn+1
k
− βnk
k=1 k=1 n=m
l−1 X
X m
= βn+1
k
− βnk
n=m k=1
l−1 X
X n
≤ βn+1
k
− βnk .
n=m k=1

Taking the limit l → ∞, we have that

m
X ∞ X
X n
β k − βmk ≤ βn+1
k
− βnk < ∞,
k=1 n=m k=1

where β k = limn→∞ βnk for k ∈ N. Taking the limit m → ∞, we obtain that

m
X
lim β k − βmk = 0.
m→∞
k=1

This implies that

∞
X m
X
0≤ β k − 1 = lim βk − 1
m→∞
k=1 k=1
Xm m
X
= lim βk − βmk
m→∞
k=1 k=1
m
X
≤ lim β k − βmk = 0
m→∞
k=1
2358 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360
P∞
k=1 β
and thus k = 1. Now we define a mapping T by
∞
X
Tx = β k Sk x
k=1

for x ∈ C. Then Theorem 2.4 implies that T is a nonexpansive mapping of C into itself satisfying that
∞
\ ∞
\
F(T ) = F(Sk ) = F(Tn ),
k=1 n=1

since β k > 0 for every k ∈ N. We also have that

m
X ∞
X
kTm x − T xk = βmk Sk x − β k Sk x
k=1 k=1
m
X ∞
X
≤ (βmk − β k )Sk x + β k Sk x
k=1 k=m+1
m
X ∞
X
≤ βmk − β k M + k
β M.
k=1 k=m+1

Taking the limit m → ∞, we have that

lim Tm x = T x.
m→∞

Hence we obtain that {Tn } satisfies all the conditions of our main theorem. Namely, we get the following result:

Theorem 4.1. Let E be a uniformly convex Banach space whose norm is uniformly Gâteaux differentiable and C a
nonempty closed convex subset of E. Let {αn } be a sequence of [0, 1] such that limn αn = 0, ∞n=1 αn = ∞, and
P
either
(i) ∞ n=1 |αn+1 − αn | < ∞ or
P
(ii) αn ∈ (0, 1] for every n ∈ N and limn→∞ αn /αn+1 = 1.
Let {βnk } be a family of nonnegative real numbers with indices n, k ∈ N with k ≤ n such that
(iii) nk=1 βnk = 1 for every n ∈ N,
P

Pβn n > 0k for every

(iv) lim k k ∈ N, and
P∞n→∞
(v) n=1 k=1 βn+1 − βnk < ∞.
Let {Sk } be a sequence of nonexpansive mappings of C into itself with a common fixed point. Then a sequence {xn }
defined by
n
X
xn+1 = αn x + (1 − αn ) βnk Sk xn for all n ∈ N,
k=1
T∞
converges strongly to Qx, where Q is the sunny nonexpansive retraction of E onto k=1 F(Sk ).
One example of {βnk }
satisfying these conditions is:

(k < n)
 −k
2
βnk = 1−k
2 (k = n),
for n, k ∈ N with k ≤ n. In this case, the sequence {Tn } of mappings generated by {Sk } is defined as follows: For
x ∈ C,
T1 x = S1 x,
1 1
T2 x = S1 x + S2 x,
2 2
 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360 2359

1 1 1
T3 x = S1 x + S2 x + S3 x,
2 4 4
1 1 1 1
T4 x = S1 x + S2 x + S3 x + S4 x,
2 4 8 8
···
1 1 1 1 1 1
Tn x = S1 x + S2 x + S3 x + S4 x + · · · + n−1 Sn−1 x + n−1 Sn x.
2 4 8 16 2 2
We note that this theorem is still valid even if we omit the first few mappings of {Tn }. Namely, for fixed m ∈ N, we
may let
n+m−1
X
Tn x = βn+m−1
k
Sk x
k=1

for x ∈ C and n ∈ N. This {Tn } satisfies all the conditions of our main theorem so that we may get strong convergence.
From this fact, we can obtain the result proved by Kimura et al. [7] by taking Sm = Sm+1 = Sm+2 = · · · for fixed
m ∈ N.
Finally, we consider the problem of finding a zero of an accretive operator. An operator A ⊂ E × E is said to be
accretive if for each (x1 , y1 ) and (x2 , y2 ) ∈ A, there exists j ∈ J (x1 − x2 ) such that hy1 − y2 , ji ≥ 0. An accretive
operator A is said to satisfy the range condition if D(A) ⊂ R(I +λA) for all λ > 0, where D(A) is the domain of A, I
is the identity mapping on E, R(I + λA) is the range of I + λA, and D(A) is the closure of D(A). If A is an accretive
operator which satisfies the range condition, then we can define, for each λ > 0, a mapping Jλ : R(I + λA) → D(A)
by Jλ = (I + λA)−1 , which is called the resolvent of A. We know that Jλ is nonexpansive and F(Jλ ) = A−1 0 for all
λ > 0. We also know the following [4]: For each λ, µ > 0 and x ∈ R(I + λA) ∩ R(I + µA), it holds that
|λ − µ|
Jλ x − Jµ x ≤ kx − Jλ xk. (4.1)
λ
Nakajo [9] proved the following.

Theorem 4.2. Let E be a uniformly convex Banach space whose norm is uniformly Gâteaux differentiable and C
a nonempty closed convex subset of E. Let A ⊂ E × E be an accretive operator such that A−1 0 6= ∅ and
D(A) ⊂ C ⊂ λ>0 R(I + λA). Let {xn } be a sequence defined by x1 = x ∈ C and
T

xn+1 = αn x + (1 − αn )Jλn xn (4.2)

{λn } is a sequence of (0, ∞). If limn αn = 0, n αn =
P
Pn∞ ∈ N, where {αn } is a sequence of [0, 1] andP
for all
∞
∞, n=1 |αn+1 − αn | < ∞, inf{λn : n ∈ N} > 0 and n=1 |λn+1 − λn | < ∞, then {x n } converges strongly to Qx,
where Q is the sunny nonexpansive retraction of C onto A−1 0.
By using Theorem 3.4, we may obtain the following improvement:

Theorem 4.3. Let E be a uniformly convex Banach space whose norm is uniformly Gâteaux differentiable and C
−1
a nonempty closed T convex subset of E. Let A ⊂ E × E be an accretive operator such thatPA 0 6= ∅ and
D(A) ⊂ C ⊂ λ>0 R(I + λA). Let {αn } be a sequence of [0, 1] such that P limn αn = 0 and n αn = ∞. Let
{λn } be a sequence of positive real numbers such that inf{λn : n ∈ N} > 0 and ∞ n=1 |λn+1 − λn | < ∞. Let {x n } be
a sequence defined by x1 = x ∈ C and (4.2) for all n ∈ N. If either
(i) ∞ n=1 |αn+1 − αn | < ∞ or
P
(ii) αn ∈ (0, 1] for every n ∈ N and limn→∞ αn /αn+1 = 1,
then {xn } converges strongly to Qx, where Q is the sunny nonexpansive retraction of C onto A−1 0.
Proof. We first verify that ∞ n=1 sup{kJλn+1 z − Jλn zk : z ∈ B} < ∞ for every bounded subset B of C. Let B be a
P

bounded subset of C. Since F(Jλn ) = A−1 0 6= ∅ for every n ∈ N, {Jλn z : z ∈ B, n ∈ N} is bounded. From (4.1) and
the boundedness of B, we have that
|λn+1 − λn |
Jλn+1 z − Jλn z ≤ z − Jλn+1 z
λn+1
≤ M |λn+1 − λn |
2360 K. Aoyama et al. / Nonlinear Analysis 67 (2007) 2350–2360

for all z ∈ B and n ∈ N, where

sup{ z − Jλn+1 z : z ∈ B, n ∈ N}
M= .
inf{λn : n ∈ N}
Hence we get
∞
X ∞
X
sup{ Jλn+1 z − Jλn z : z ∈ B} ≤ M |λn+1 − λn | < ∞.
n=1 n=1
By the assumptions for {λn }, we know that λn → λ for some λ > 0. Since
|λ − λn |
Jλ z − Jλn z ≤ kz − Jλ zk,
λ
T∞
we obtain limn→∞ Jλn z = Jλ z for all z ∈ C. Since F(Jµ ) = A−1 0 for all µ > 0, we have F(Jλ ) = n=1 F(Jλn ) =
A−1 0 6= ∅. Therefore, by Theorem 3.4, {xn } converges strongly to Qx. 

References

[1] H.H. Bauschke, The approximation of fixed points of compositions of nonexpansive mappings in Hilbert space, J. Math. Anal. Appl. 202
(1996) 150–159.
[2] R.E. Bruck Jr., Properties of fixed-point sets of nonexpansive mappings in Banach spaces, Trans. Amer. Math. Soc. 179 (1973) 251–262.
[3] B. Halpern, Fixed points of nonexpanding maps, Bull. Amer. Math. Soc. 73 (1967) 957–961.
[4] K. Eshita, W. Takahashi, Approximating zero points of accretive operators in general Banach spaces (in press).
[5] H. Iiduka, W. Takahashi, Strong convergence theorems for nonexpansive mappings and inverse-strongly monotone mappings, Nonlinear Anal.
61 (2005) 341–350.
[6] S. Kamimura, W. Takahashi, Weak and strong convergence of solutions to accretive operator inclusions and applications, Set-Valued Anal. 8
(2000) 361–374.
[7] Y. Kimura, W. Takahashi, M. Toyoda, Convergence to common fixed points of a finite family of nonexpansive mappings, Arch. Math. (Basel)
84 (2005) 350–363.
[8] J.G. O’Hara, P. Pillay, H.-K. Xu, Iterative approaches to finding nearest common fixed points of nonexpansive mappings in Hilbert spaces,
Nonlinear Anal. 54 (2003) 1417–1426.
[9] K. Nakajo, Strong convergence to zeros of accretive operators in Banach spaces, J. Nonlinear Convex Anal. 7 (2006) 71–81.
[10] S. Reich, Strong convergence theorems for resolvents of accretive operators in Banach spaces, J. Math. Anal. Appl. 75 (1980) 287–292.
[11] T. Shimizu, W. Takahashi, Strong convergence to common fixed points of families of nonexpansive mappings, J. Math. Anal. Appl. 211 (1997)
71–83.
[12] N. Shioji, W. Takahashi, Strong convergence of approximated sequences for nonexpansive mappings in Banach spaces, Proc. Amer. Math.
Soc. 125 (1997) 3641–3645.
[13] W. Takahashi, Nonlinear Functional Analysis, Yokohama Publishers, Yokohama, 2000.
[14] W. Takahashi, Y. Ueda, On Reich’s strong convergence theorems for resolvents of accretive operators, J. Math. Anal. Appl. 104 (1984)
546–553.
[15] R. Wittmann, Approximation of fixed points of nonexpansive mappings, Arch. Math. (Basel) 58 (1992) 486–491.