1 Introduction

Chidume [1] established that the Mann iteration sequence converges strongly to the unique fixed point of T in case T is a Lipschitz strongly pseudo-contractive mapping from a bounded closed convex subset of L p (or l p ) into itself. Schu [2] generalized the result in [1] to both uniformly continuous strongly pseudo-contractive mappings and real smooth Banach spaces. Park [3] extended the result in [1] to both strongly pseudocontractive mappings and certain smooth Banach spaces. Rhoades [4] proved that the Mann and Ishikawa iteration methods may exhibit different behavior for different classes of nonlinear mappings. Harder and Hicks [5, 6] revealed the importance of investigating the stability of various iteration procedures for various classes of nonlinear mappings. Harder [7] established applications of stability results to first-order differential equations. Afterwords, several generalizations have been made in various directions (see, for example, [2, 4, 821].

Let K be a nonempty closed bounded convex subset of an arbitrary smooth Banach space X and T:KK be a continuous strictly hemicontractive mapping. Under some conditions, we obtain that the Mann iteration method with error term converges strongly to a unique fixed point of T and is almost T-stable on K. As an application, we shall also establish strong convergence of a multi-step iteration process. The results presented here generalize the corresponding results in [24, 10, 11, 22].

2 Preliminaries

Let K be a nonempty subset of an arbitrary Banach space X and X be its dual space. The symbols D(T), R(T) and F(T) stand for the domain, the range and the set of fixed points of T:XX respectively (x is called a fixed point of T iff T(x)=x). We denote by J the normalized duality mapping from X to 2 X defined by

J(x)= { f X : x , f = x 2 = f 2 } .

Let T be a self-mapping of K.

Definition 1 The mapping T is called Lipshitzian if there exists L>0 such that

TxTyLxy

for all x,yK. If L=1, then T is called non-expansive and if 0L<1, T is called contraction.

Definition 2 [10, 22]

  1. 1.

    The mapping T is said to be pseudocontractive if the inequality

    xy x y + t [ ( I T ) x ( I T ) y ]
    (2.1)

holds for each x,yK and for all t>0.

  1. 2.

    T is said to be strongly pseudocontractive if there exists t>1 such that

    xy ( 1 + r ) ( x y ) r t ( T x T y )
    (2.2)

for all x,yD(T) and r>0.

  1. 3.

    T is said to be local strongly pseudocontractive if for each xD(T), there exists t x >1 such that

    xy ( 1 + r ) ( x y ) r t x ( T x T y )
    (2.3)

for all yD(T) and r>0.

  1. 4.

    T is said to be strictly hemicontractive if F(T) and if there exists t>1 such that

    xq ( 1 + r ) ( x q ) r t ( T x q )
    (2.4)

for all xD(T), qF(T) and r>0.

Clearly, each strongly pseudocontractive operator is local strongly pseudocontractive.

Definition 3 [57]

Let K be a nonempty convex subset of X and T:KK be an operator. Assume that x o K and x n + 1 =f(T, x n ) defines an iteration scheme which produces a sequence { x n } n = 0 K. Suppose, furthermore, that { x n } n = 0 converges strongly to qF(T). Let { y n } n = 0 be any bounded sequence in K and put ε n = y n + 1 f(T, y n ).

  1. (1)

    The iteration scheme { x n } n = 0 defined by x n + 1 =f(T, x n ) is said to be T-stable on K if lim n ε n =0 implies that lim n y n =q.

  2. (2)

    The iteration scheme { x n } n = 0 defined by x n + 1 =f(T, x n ) is said to be almost T-stable on K if n = 0 ε n < implies that lim n y n =q.

It is easy to verify that an iteration scheme { x n } n = 0 which is T-stable on K is almost T-stable on K.

Lemma 4 [3]

Let X be a smooth Banach space. Suppose one of the following holds:

  1. (1)

    J is uniformly continuous on any bounded subsets of X,

  2. (2)

    xy,j(x)j(y) x y 2 for all x, y in X,

  3. (3)

    for any bounded subset D of X, there is a c:[0,)[0,) such that

    Re x y , j ( x ) j ( y ) c ( x y ) ,

for all x,yD, where c satisfies

lim t 0 + c ( t ) t =0.
(2.5)

Then for any ϵ>0 and any bounded subset K, there exists δ>0 such that

s x + ( 1 s ) y 2 (12s) y 2 +2sRe x , j ( y ) +2sϵ
(2.6)

for all x,yK and s[0,δ].

Lemma 5 [10]

Let T:D(T)XX be an operator with F(T)φ. Then T is strictly hemicontractive if and only if there exists t>1 such that for all xD(T) and qF(T), there exists j(xq)J(xq) satisfying

Re x T x , j ( x q ) ( 1 1 t ) x q 2 .
(2.7)

Lemma 6 [4]

Let X be an arbitrary normed linear space and T:D(T)XX be an operator.

  1. (1)

    If T is a local strongly pseudocontractive operator and F(T), then F(T) is a singleton and T is strictly hemicontractive.

  2. (2)

    If T is strictly hemicontractive, then F(T)is a singleton.

3 Main results

We now prove our main results.

Lemma 7 Let { α n } n = 0 , { β n } n = 0 and { γ n } n = 0 be nonnegative real sequences, and let ϵ >0 be a constant satisfying

β n + 1 ( 1 α n l ) β n + ϵ α n + γ n ;l1,n0,

where n = 0 α n l =, α n 1 for all n0 and n = 0 γ n <. Then, lim n sup β n ϵ .

Proof By a straightforward argument, for nk0,

β n + 1 β k j = k n ( 1 α j l ) + ϵ j = k n α j i = j + 1 n ( 1 α i l ) + j = k n γ j i = j + 1 n ( 1 α i l ) ,
(3.1)

where we put i = n + 1 n (1 α i l )=1. Note that j = k n α j i = j + 1 n (1 α i l )1. It follows from (3.1) that

β n + 1 exp ( j = k n α j l ) β k + ϵ + j = k n γ j .
(3.2)

For a given δ>0, there exists a positive integer k such that j = k γ j <δ. Thus (3.2) ensures that

lim n sup β n ϵ +δ.

Letting δ 0 + yields lim n sup β n ϵ . □

Remark 8

  1. (i)

    If γ n =0 for each n0, then Lemma 7 reduces to Lemma 1 of Park [3].

  2. (ii)

    If l=1, then Lemma 7 reduces to Lemma 2.1 of Liu et al. [4].

Theorem 9 Let Xbe a smooth Banach space satisfying any one of the Axioms (1)-(3) of Lemma 4. Let K be a nonempty closed bounded convex subset of X and T:KK be a continuous strictly hemicontractive mapping. Suppose that { u n } n = 0 is an arbitrary sequence in K and { a n } n = 0 , { b n } n = 0 and { c n } n = 0 are any sequences in [0,1] satisfying conditions (i) a n + b n + c n =1, (ii) c n =o( b n ), (iii) lim n b n =0 and (iv) n = 0 b n =.

For a sequence { v n } n = 0 in K, suppose that { x n } n = 0 is the sequence generated from an arbitrary x 0 K by

x n + 1 = a n x n + b n T v n + c n u n ,n0,
(3.3)

and satisfying lim n v n x n =0.

Let { y n } n = 0 be any sequence in K and define { ε n } n = 0 by

ε n = y n + 1 p n ,n0,
(3.4)

where p n = a n y n + b n T v n + c n u n , such that lim n v n y n =0.

Then

  1. (a)

    the sequence { x n } n = 0 converges strongly to a unique fixed point q of T,

  2. (b)

    n = 0 ε n < implies that lim n y n =q, so that { x n } n = 0 is almost T-stable on K,

  3. (c)

    lim n y n =q implies that lim n ε n =0.

Proof From (ii), we have c n = t n b n , where t n 0 as n.

It follows from Lemma 6 that F(T) is a singleton. That is, F(T)={q} for some qK.

Set M=1+diamK. For all n0, it is easy to verify that

M = sup n 0 x n q + sup n 0 T v n q + sup n 0 u n q + sup n 0 { p n q } + sup n 0 y n q .
(3.5)

For given any ϵ>0 and the bounded subset K, there exists a δ>0 satisfying (2.6). Note that (ii), (iii), lim n v n x n =0 and the continuity of T ensure that there exists an N such that

b n <min { δ , 1 2 ( 1 k ) } , t n ϵ 16 M 2 ,T v n T x n ϵ 4 M ,nN,
(3.6)

where k= 1 t and t satisfies (2.7). Using (3.3) and Lemma 4, we infer that

x n + 1 q 2 = ( 1 b n ) ( x n q ) + b n ( T v n q ) + c n ( u n x n ) 2 ( ( 1 b n ) ( x n q ) + b n ( T v n q ) + 2 M c n ) 2 ( 1 b n ) ( x n q ) + b n ( T v n q ) 2 + 8 M 2 c n ( 1 2 b n ) x n q 2 + 2 b n Re ( T v n q , j ( x n q ) ) + 2 ϵ b n + 8 M 2 c n = ( 1 2 b n ) x n q 2 + 2 b n Re ( T x n q , j ( x n q ) ) + 2 b n Re ( T v n T x n , j ( x n q ) ) + 2 ϵ b n + 8 M 2 c n ( 1 2 b n ) x n q 2 + 2 k b n x n q 2 + 2 b n T v n T x n x n q + 2 ϵ b n + 8 M 2 c n ( 1 2 ( 1 k ) b n ) x n q 2 + 2 M b n T v n T x n + 2 ϵ b n + 8 M 2 c n ( 1 2 ( 1 k ) b n ) x n q 2 + 3 ϵ b n ,
(3.7)

for all nN.

Put

we have from (3.7)

β n + 1 (1 α n ) β n + ϵ α n + γ n ,n0.

Observe that n = 0 α n =, α n <1 for all n0. It follows from Lemma 7 that

lim n sup x n q 2 ϵ .

Letting ϵ 0 + , we obtain that lim n sup x n q 2 =0, which implies that x n q as n.

On the same lines, we obtain

p n q 2 = ( 1 b n ) ( y n q ) + b n ( T v n q ) + c n ( u n y n ) 2 ( ( 1 b n ) ( y n q ) + b n ( T v n q ) + 2 M c n ) 2 ( 1 b n ) ( y n q ) + b n ( T v n q ) 2 + 8 M 2 c n ( 1 2 b n ) y n q 2 + 2 b n Re ( T v n q , j ( y n q ) ) + 2 ϵ b n + 8 M 2 c n = ( 1 2 b n ) y n q 2 + 2 b n Re ( T y n q , j ( y n q ) ) + 2 b n Re ( T v n T y n , j ( y n q ) ) + 2 ϵ b n + 8 M 2 c n ( 1 2 b n ) y n q 2 + 2 k b n y n q 2 + 2 b n T v n T y n y n q + 2 ϵ b n + 8 M 2 c n ( 1 2 ( 1 k ) b n ) y n q 2 + 2 M b n T v n T y n + 2 ϵ b n + 8 M 2 c n ( 1 2 ( 1 k ) b n ) y n q 2 + 3 ϵ b n ,
(3.8)

for all nN.

Suppose that n = 0 ε n <. In view of (3.4) and (3.8), we infer that

y n + 1 q 2 ( y n + 1 p n + p n q ) 2 p n q 2 + 3 M ε n [ 1 2 b n ( 1 k ) ] y n q 2 + 3 ϵ b n + 3 M ε n ,
(3.9)

for all nN.

Now, put

and we have from (3.9)

β n + 1 (1 α n ) β n + ϵ α n + γ n ,n0.

Observe that n = 0 α n =, α n <1 and n = 0 γ n < for all n0. It follows from Lemma 7 that

lim n sup y n q 2 ϵ .

Letting ϵ 0 + , we obtain that lim n sup y n q 2 =0, which implies that y n q as n.

Conversely, suppose that lim n y n =q, then (iii) and (3.8) imply that

ε n y n + 1 q + p n q y n + 1 q + [ [ 1 2 ( 1 k ) b n ] y n q 2 + 3 ϵ b n ] 1 2 0 ,

as n, that is, ε n 0 as n. □

Using the methods of the proof of Theorem 9, we can easily prove the following.

Theorem 10 Let X, K, T and { u n } n = 0 , be as in Theorem 9. Suppose that { a n } n = 0 , { b n } n = 0 and { c n } n = 0 are sequences in [0,1] satisfying conditions (i), (iii)-(iv) and

n = 0 c n <.

If { x n } n = 0 , { v n } n = 0 , { y n } n = 0 , { p n } n = 0 and { ε n } n = 0 are as in Theorem 9, then the conclusions of Theorem 9 hold.

Corollary 11 Let X be a smooth Banach space satisfying any one of the Axioms (1)-(3) of Lemma 4. Let K be a nonempty closed bounded convex subset of X and T:KK be a Lipschitz strictly hemicontractive mapping. Suppose that { u n } n = 0 is an arbitrary sequence in K and { a n } n = 0 , { b n } n = 0 and { c n } n = 0 are any sequences in [0,1] satisfying conditions (i) a n + b n + c n =1, (ii) c n =o( b n ), (iii) lim n b n =0 and (iv) n = 0 b n =.

For a sequence { v n } n = 0 in K, suppose that { x n } n = 0 is the sequence generated from an arbitrary x 0 K by

x n + 1 = a n x n + b n T v n + c n u n ,n0,

and satisfying lim n v n x n =0.

Let { y n } n = 0 be any sequence in K and define { ε n } n = 0 by

ε n = y n + 1 p n ,n0,

where p n = a n y n + b n T v n + c n u n , such that lim n v n y n =0.

Then

  1. (a)

    the sequence { x n } n = 0 converges strongly to a unique fixed point q of T,

  2. (b)

    n = 0 ε n < implies that lim n y n =q, so that { x n } n = 0 is almost T-stable on K,

  3. (c)

    lim n y n =q implies that lim n ε n =0.

Corollary 12 Let X, K, T and { u n } n = 0 be as in Corollary 11. Suppose that { a n } n = 0 , { b n } n = 0 and { c n } n = 0 are sequences in [0,1] satisfying conditions (i), (iii)-(iv) and

n = 0 c n <.

If { x n } n = 0 , { v n } n = 0 , { y n } n = 0 , { p n } n = 0 and { ε n } n = 0 are as in Corollary 11, then the conclusions of Corollary 11 hold.

Corollary 13 Let X be a smooth Banach space satisfying any one of the Axioms (1)-(3) of Lemma 4. Let K be a nonempty closed bounded convex subset of X and T:KK be a continuous strictly hemicontractive mapping. Suppose that { α n } n = 0 is a sequence in [0,1] satisfying conditions (i) lim n α n =0 and (ii) n = 0 α n =.

For a sequence { v n } n = 0 in K, suppose that { x n } n = 0 is the sequence generated from an arbitrary x 0 K by

x n + 1 = α n x n +(1 α n )T v n ,n0,

and satisfying lim n v n x n =0.

Let { y n } n = 0 be any sequence in K and define { ε n } n = 0 by

ε n = y n + 1 p n ,n0,

where p n = α n y n +(1 α n )T v n , such that lim n v n y n =0.

Then

  1. (a)

    the sequence { x n } n = 0 converges strongly to a unique fixed point q of T,

  2. (b)

    n = 0 ε n < implies that lim n y n =q, so that { x n } n = 0 is almost T-stable on K,

  3. (c)

    lim n y n =q implies that lim n ε n =0.

Corollary 14 Let X be a smooth Banach space satisfying any one of the Axioms (1)-(3) of Lemma 4. Let K be a nonempty closed bounded convex subset of X and T:KK be a Lipschitz strictly hemicontractive mapping. Suppose that { α n } n = 0 is a sequence in [0,1] satisfying conditions (i) lim n α n =0 and (ii) n = 0 α n =.

For a sequence { v n } n = 0 in K, suppose that { x n } n = 0 is the sequence generated from an arbitrary x 0 K by

x n + 1 = α n x n +(1 α n )T v n ,n0,

and satisfying lim n v n x n =0.

Let { y n } n = 0 be any sequence in K and define { ε n } n = 0 by

ε n = y n + 1 p n ,n0,

where p n = α n y n +(1 α n )T v n , such that lim n v n y n =0.

Then

  1. (a)

    the sequence { x n } n = 0 converges strongly to a unique fixed point q of T,

  2. (b)

    n = 0 ε n < implies that lim n y n =q, so that { x n } n = 0 is almost T-stable on K,

  3. (c)

    lim n y n =q implies that lim n ε n =0.

4 Applications to a multi-step iteration process

Khan et al. [23] have introduced and studied a multi-step iteration process for a finite family of selfmappings. We now introduce a modified multi-step process as follows:

Let K be a nonempty closed convex subset of a real normed space E and T 1 , T 2 ,, T p :KK (p2) be a family of selfmappings.

Algorithm 1 For a given x 0 K, compute the sequence { x n } n 0 by the iteration process of arbitrary fixed order p2,

(4.1)

which is called the modified multi-step iteration process, where { α n } n 0 , { β n i } n 0 [0,1], i=1,2,,p1.

For p=3, we obtain the following three-step iteration process:

Algorithm 2 For a given x 0 K, compute the sequence { x n } n 0 by the iteration process:

(4.2)

where { α n } n 0 , { β n 1 } n 0 and { β n 2 } n 0 are three real sequences in [0,1].

For p=2, we obtain the Ishikawa [24] iteration process:

Algorithm 3 For a given x 0 K, compute the sequence { x n } n 0 by the iteration process

x n + 1 = ( 1 α n ) x n + α n T 1 y n 1 , y n 1 = ( 1 β n 1 ) x n + β n 1 T 2 x n , n 0 ,
(4.3)

where { α n } n 0 and { β n 1 } n 0 are two real sequences in [0,1].

If T 1 =T, T 2 =I, β n 1 =0 in (4.3), we obtain the Mann iteration process [14]:

Algorithm 4 For any given x 0 K, compute the sequence { x n } n 0 by the iteration process

x n + 1 =(1 α n ) x n + α n T x n ,n0,
(4.4)

where { α n } is a real sequence in [0,1].

Theorem 15 Let K be a nonempty closed bounded convex subset of a smooth Banach space X and T 1 , T 2 ,, T p (p2) be selfmappings of K. Let T 1 be a continuous strictly hemicontractive mapping. Let { α n } n 0 , { β n i } n 0 [0,1], i=1,2,,p1 be real sequences in [0,1] satisfying n 0 α n =, lim n α n =0 and lim n β n 1 =0. For arbitrary x 0 K, define the sequence { x n } n 0 by (4.1). Then { x n } n 0 converges strongly to a point in i = 1 p F( T i ).

Proof By applying Corollary 13 under assumption that T 1 is continuous strictly hemicontractive mapping, we obtain Theorem 15 which proves strong convergence of the iteration process defined by (4.1). We will check only the condition lim n v n x n =0 by taking T 1 =T and v n = y n 1 ,

v n x n = y n 1 x n = ( 1 β n 1 ) x n + β n 1 T 2 y n 2 x n = β n 1 T 2 y n 2 x n 2 M β n 1 .

Now, from the condition lim n β n 1 =0, it can be easily seen that lim n v n x n =0. □

Corollary 16 Let K be a nonempty closed bounded convex subset of a smooth Banach space X and T 1 , T 2 ,, T p (p2) be selfmappings of K. Let T 1 be a Lipschitz strictly hemicontractive mapping. Let { α n } n 0 , { β n i } n 0 [0,1], i=1,2,,p1 be real sequences in [0,1] satisfying n 0 α n =, lim n α n =0 and lim n β n 1 =0. For arbitrary x 0 K, define the sequence { x n } n 0 by (4.1). Then { x n } n 0 converges strongly to a point in i = 1 p F( T i ).

Remark 17 Similar results can be found for the iteration processes with error terms, we omit the details.