Open Access

Convergence of splitting algorithms for the sum of two accretive operators with applications

Fixed Point Theory and Applications20142014:166

https://doi.org/10.1186/1687-1812-2014-166

Received: 15 March 2014

Accepted: 25 June 2014

Published: 15 August 2014

Abstract

We study a splitting algorithm for problems involving the sum of two accretive operators. We prove the strong convergence of the algorithm. Applications to variational inequality, fixed point, equilibrium, and minimization problems are provided.

Keywords

accretive operatorfixed pointnonexpansive mappingresolventzero point

1 Introduction

Many important real world problems have reformulations which require finding zero points of some nonlinear operators, for instance, evolution equations, complementarity problems, mini-max problems, variational inequalities and optimization problems; see [111] and the references therein. It is well known that minimizing a convex function f can be reduced to finding zero points of the subdifferential mapping ∂f. Forward-backward splitting algorithms were proposed by Lions and Mercier [12], by Passty [13], and, in a dual form for convex programming, by Han and Lou [14]. The algorithms, which provide a range of approaches to solving large-scale optimization problems and variational inequalities, have recently received much attention due to the fact that many nonlinear problems arising in applied areas such as image recovery, signal processing, and machine learning are mathematically modeled as a nonlinear operator equation, and this operator is decomposed as the sum of two nonlinear operators. This paper concerns a forward-backward splitting algorithm with computational errors designed to find zeros of the sum of two accretive operators A and B.

The paper is organized in the following way. In Section 2, we present the preliminaries that are needed in our work. In Section 3, we present a splitting algorithm for finding zeros of the sum of two accretive operators A and B. Convergence analysis of the algorithms is investigated. In Section 4, applications of our main results are provided.

2 Preliminaries

Let E be a real Banach space with the dual E . Given a continuous strictly increasing function φ : R + R + , where R + denotes the set of nonnegative real numbers, such that φ ( 0 ) = 0 and lim r φ ( r ) = , we associate with it a (possibly multivalued) generalized duality map J φ : E 2 E , defined as J φ ( x ) = { x E : x ( x ) = x φ ( x ) , x = φ ( x ) } , x E . In this paper, we use the generalized duality map associated with the gauge function φ ( t ) = t q 1 for q > 1 . Let ρ E : [ 0 , ) [ 0 , ) be the modulus of smoothness of E by ρ E ( t ) = sup { x + y x y 2 1 , x U E , y t } . A Banach space E is said to be uniformly smooth if ρ E ( t ) t 0 as t 0 . Let q > 1 . E is said to be q-uniformly smooth if there exists a fixed constant c > 0 such that ρ E ( t ) c t q . The modulus of convexity of E is the function δ E ( ϵ ) : ( 0 , 2 ] [ 0 , 1 ] defined by δ E ( ϵ ) = inf { 1 x + t 2 : x = y = 1 , x y ϵ } . Recall that E is said to be uniformly convex if δ E ( ϵ ) > 0 for any ϵ ( 0 , 2 ] . Let E be a smooth Banach space, and let C be a nonempty subset of E. Let Proj C : E C be a retraction and be the normalized duality mapping on E. Then the following are equivalent [15]: (1) Proj C is sunny and nonexpansive; (2) x Proj C x , J ( y Proj C x ) 0 , x E , y C .

Let I denote the identity operator on E. An operator A E × E with domain D ( A ) = { z E : A z } and range R ( A ) = { A z : z D ( A ) } is said to be accretive iff, for t > 0 and x , y D ( A ) , x y x y + t ( u v ) , u A x , v A y . It follows from Kato [16] that A is accretive iff, for x , y D ( A ) , there exists j q ( x 1 x 2 ) such that u v , j q ( x y ) 0 . An accretive operator A is said to be m-accretive iff R ( I + r A ) = E for all r > 0 . In this paper, we use A 1 ( 0 ) to denote the set of zeros of A. For an accretive operator A, we can define a nonexpansive single-valued mapping J r : R ( I + r A ) D ( A ) by J r = ( I + r A ) 1 for each r > 0 , which is called the resolvent of A. Recall that a single-valued operator A : C E is said to be α-inverse strongly accretive if there exists a constant α > 0 and some j q ( x y ) J q ( x y ) such that A x A y , j q ( x y ) α A x A y q , x , y C .

Let T : C C be a mapping. Recall that T is said to be κ-contractive iff there exists a constant κ ( 0 , 1 ) such that T x T y κ x , y , x , y C . T is said to be nonexpansive iff κ = 1 . T is said to be κ-strictly pseudocontractive iff there exists a constant κ ( 0 , 1 ) such that
T x T y , j q ( x y ) x y q κ ( x T x ) ( y T y ) q , x , y C

for some j q ( x y ) J q ( x y ) . T is said to be pseudocontractive iff T x T y , j q ( x y ) x y q , x , y C for some j q ( x y ) J q ( x y ) .

In order to obtain our main results, we also need the following lemmas.

Lemma 2.1 [17]

Let E be a real q-uniformly smooth Banach space. Then the following inequality holds: x + y q x q + q y , J q ( x + y ) and x + y q x q + q y , J q ( x ) + K q y q , x , y E , where K q is some fixed positive constant.

Lemma 2.2 [18]

Let E be a real Banach space, and let C be a nonempty closed and convex subset of E. Let A : C E be a single-valued operator, and let B : E 2 E be an m-accretive operator. Then F ( J a ( I a A ) ) = ( A + B ) 1 ( 0 ) , where J a ( I a A ) is the resolvent of B for a > 0 .

Lemma 2.3 [19]

Let E be a real Banach space, and A be an m-accretive operator. For λ > 0 , μ > 0 , and x E , we have J λ x = J μ ( μ λ x + ( 1 μ λ ) J λ x ) , where J λ = ( I + λ A ) 1 and J μ = ( I + μ A ) 1 .

Lemma 2.4 [20]

Let { a n } be a sequence of nonnegative real numbers such that a n + 1 ( 1 t n ) a n + b n + c n , where { c n } is a sequence of nonnegative real numbers, { t n } ( 0 , 1 ) and { b n } is a number sequence. Assume that n = 0 t n = , lim sup n b n t n 0 , and n = 0 c n < . Then lim n a n = 0 .

Lemma 2.5 [21]

Let q > 1 . Then the following inequality holds: a b a q q + q 1 q b q q 1 , for arbitrary positive real numbers a and b.

Lemma 2.6 [22]

Let C be a nonempty closed convex subset of a real uniformly smooth Banach space E. Let f : C C be a contractive mapping, and let T : C C be a nonexpansive mapping. For each t ( 0 , 1 ) , let x t be the unique solution of the equation x = t f ( x ) + ( 1 t ) T x . Then { x t } converges strongly to a fixed point x ¯ = Q F ( T ) f ( x ¯ ) .

3 Main results

Theorem 3.1 Let E be a real q-uniformly smooth Banach space with the constant K q . Let B : E 2 E be an m-accretive operator such that D ( B ) ¯ is convex. Let A : D ( B ) ¯ E be an α-inverse strongly accretive operator. Assume that ( A + B ) 1 ( 0 ) . Let f : D ( B ) ¯ D ( B ) ¯ be a fixed κ-contraction. Let { r n } , { α n } , { β n } , and { γ n } be positive real number sequences, where { α n } , { β n } , and { γ n } are in ( 0 , 1 ) . Let { x n } be a sequence generated in the following iterative process:
x 0 C , x n + 1 = α n f ( x n ) + β n J r n ( x n r n A x n + e n ) + γ n f n , n 0 ,
where J r n = ( I + r n B ) 1 , { e n } is a sequence in E, and { f n } is a bounded sequence in D ( B ) ¯ . Assume that the sequences { α n } , { β n } , { γ n } , { e n } , and { r n } satisfy the following restrictions:
  1. (1)

    α n + β n + γ n = 1 ;

     
  2. (2)

    lim n α n = 0 , n = 0 α n = ;

     
  3. (3)

    n = 1 | β n β n 1 | < ;

     
  4. (4)

    lim inf n r n > 0 , r n ( q α K q ) 1 q 1 , n = 1 | r n r n 1 | < ;

     
  5. (5)

    n = 0 e n < , n = 0 γ n < .

     

Then the sequence { x n } converges strongly to x = Proj ( A + B ) 1 ( 0 ) f ( x ) , where Proj ( A + B ) 1 ( 0 ) is the unique sunny nonexpansive retraction of C onto ( A + B ) 1 ( 0 ) .

Proof First, we show that { x n } is bounded. In view of Lemma 2.1, we find that
( I r n A ) x ( I r n A ) y q x y q q r n A x A y , J q ( x y ) + K q r n q A x A y q x y q q r n α A x A y q + K q r n q A x A y q = x y q ( α q K q r n q 1 ) r n A x A y q .
From restriction (4), we find that I r n A is nonexpansive. Fixing p ( A + B ) 1 ( 0 ) , we find from Lemma 2.2 that p = J r n ( x n r n A x n ) p . It follows from restriction (5) that
x n + 1 p α n f ( x n ) p + β n J r n ( x n r n A x n + e n ) p + γ n f n p α n κ x n p + α n f ( p ) p + β n ( x n r n A x n + e n ) ( I r n A ) p + γ n f n p ( 1 α n ( 1 κ ) ) x n p + α n f ( p ) p + e n + γ n f n p max { x n p , f ( p ) p 1 κ } + e n + γ n f n p max { x n 1 p , f ( p ) p 1 κ } + e n 1 + e n + γ n 1 f n 1 p + γ n f n p max { x 0 p , f ( p ) p 1 κ } + i = 0 n e i + i = 0 n γ i M max { x 0 p , f ( p ) p 1 κ } + i = 0 e i + i = 0 γ i M < ,
where M = sup n 0 { f n p } . This shows that { x n } is bounded. Set y n = ( I r n A ) x n + e n . It follows that
y n 1 y n x n x n 1 + | r n r n 1 | A x n 1 + e n + e n 1 .
Using Lemma 2.3, we find that
J r n 1 y n 1 J r n y n = J r n 1 ( r n 1 r n y n + ( 1 r n 1 r n ) J r n y n ) J r n 1 y n 1 r n 1 r n ( y n y n 1 ) + ( 1 r n 1 r n ) ( J r n y n y n 1 ) y n y n 1 + | r n r n 1 | r n J r n y n y n x n x n 1 + | r n r n 1 | ( A x n 1 + J r n y n y n r n ) + e n + e n 1 .
(3.1)
On the other hand, we have
x n + 1 x n α n κ x n x n 1 + | α n α n 1 | f ( x n 1 ) + γ n f n f n 1 + | γ n γ n 1 | f n 1 + β n J r n 1 y n 1 J r n y n + | β n β n 1 | J r n 1 y n 1 .
(3.2)
Using (3.1) and (3.2), we find
x n + 1 x n ( 1 α n ( 1 κ ) ) x n x n 1 + | α n α n 1 | f ( x n 1 ) + γ n f n f n 1 + | γ n γ n 1 | f n 1 + | r n r n 1 | ( A x n 1 + J r n y n y n r n ) + e n + e n 1 + | β n β n 1 | J r n 1 y n 1 .
Using restrictions (3), (4), and (5), we obtain from Lemma 2.4 that
lim n x n + 1 x n = 0 .
(3.3)
On the other hand, we have
J r n y n x n x n + 1 x n β n + α n β n x n f ( x n ) + γ n β n x n f n .
Using restrictions (2) and (5), we obtain from (3.3) that
lim n J r n y n x n = 0 .
(3.4)
Since
J r n ( x n r n A x n ) x n J r n ( x n r n A x n ) J r n y n + J r n y n x n e n + J r n y n x n ,
we find from (3.4) and restriction (5) that
lim n J r n ( x n r n A x n ) x n = 0 .
(3.5)
Without loss of generality, let us assume that there exists a real number r such that r n r > 0 . Since B is accretive, we have
x n J r ( I r A ) x n r x n J r n ( I r n A ) x n r n , J q ( J r ( I r A ) x n J r n ( I r n A ) x n ) 0 .
It follows that
J r ( I r A ) x n J r n ( I r n A ) x n q r n r r n x n J r n ( I r n A ) x n , J q ( J r ( I r A ) x n J r n ( I r n A ) x n ) x n J r n ( I r n A ) x n J r ( I r A ) x n J r n ( I r n A ) x n q 1 .
This implies from (3.5) that
lim n J r ( x n r A x n ) x n = 0 .
(3.6)

Since J r ( I r A ) is nonexpansive and f is contractive, we find that the mapping t f + ( 1 t ) J r ( I r A ) is contractive. Let x t be the unique fixed point of the mapping t f + ( 1 t ) J r ( I r A ) , that is, x t = t f ( x t ) + ( 1 t ) J r ( I r A ) x t , t ( 0 , 1 ) . Setting x = lim t 0 x t , we have x = Proj ( A + B ) 1 ( 0 ) f ( x ) , where Proj ( A + B ) 1 ( 0 ) is the unique sunny nonexpansive retraction from C onto ( A + B ) 1 ( 0 ) .

Next, we show that
lim sup n f ( x ) x , J q ( x n x ) 0 .
Since
x t x n q t f ( x t ) x n , J q ( x t x n ) + ( 1 t ) J r ( I r A ) x t x n , J q ( x t x n ) t f ( x t ) x t , J q ( x t x n ) + t x t x n , J q ( x t x n ) + ( 1 t ) J r ( I r A ) x t J r ( I r A ) x n , J q ( x t x n ) + ( 1 t ) J r ( I r A ) x n x n , J q ( x t x n ) t f ( x t ) x t , J q ( x t x n ) + x t x n q + J r ( I r A ) x n x n x t x n q 1 ,
we have
f ( x t ) x t , J q ( x n x t ) 1 t J r ( I r A ) x n x n x t x n q 1 .
Fix t and let n . It follows from (3.6) that
lim sup n f ( x t ) x t , J q ( x n x t ) 0 .
(3.7)
Since the duality map J q is single-valued and strong-weak uniformly continuous on bounded sets of a Banach space E with a uniformly Gâteaux differentiable norm, one has
| f ( x t ) x t , J q ( x n x t ) f ( x ) x , J q ( x n x ) | = | f ( x ) x , J q ( x n x ) J q ( x n x t ) + f ( x ) x ( f ( x t ) x t ) , J q ( x n x t ) | | f ( x ) x , J q ( x n x ) J q ( x n x t ) | + f ( x ) x ( f ( x t ) x t ) x n x t q 1 .
Hence, ϵ > 0 , δ > 0 such that t ( 0 , δ ) , one has
f ( x ) x , J q ( x n x ) f ( x t ) x t , J q ( x n x t ) + ϵ .
Using (3.7), we see that
lim sup n f ( x ) x , J q ( x n x ¯ ) 0 .
(3.8)
Using Lemma 2.5, one has
x n + 1 x q α n f ( x n ) f ( x ) , J q ( x n + 1 x ) + α n f ( x ) x , J q ( x n + 1 x ¯ ) + β n J r n ( x n r n A x n + e n ) x x n + 1 x q 1 + γ n f n x x n + 1 x q 1 ( 1 α n ( 1 κ ) ) x n x x n + 1 x q 1 + α n f ( x ) x , J q ( x n + 1 x ) + e n x n + 1 x q 1 + γ n f n x x n + 1 x q 1 ( 1 α n ( 1 κ ) ) ( 1 q x n x q + q 1 q x n + 1 x q ) + α n f ( x ) x , J q ( x n + 1 x ) + e n x n + 1 x q 1 + γ n f n x x n + 1 x q 1 .
It follows that
x n + 1 x q ( 1 α n ( 1 κ ) ) x n x q + q α n f ( x ) x , J q ( x n + 1 x ) + q e n x n + 1 x q 1 + q γ n f n x x n + 1 x q 1 .

Using restrictions (2) and (5), we see from (3.8) that { x n } converges strongly to x. This completes the proof. □

Remark 3.2 The framework of the space in Theorem 3.1 can be applicable to L p , where p > 1 .

Corollary 3.3 Let E be a real q-uniformly smooth Banach space with the constant K q . Let B : E 2 E be an m-accretive operator such that D ( B ) ¯ is convex. Assume that B 1 ( 0 ) . Let f : D ( B ) ¯ D ( B ) ¯ be a fixed κ-contraction. Let { r n } and { α n } be positive real number sequences, where { α n } is in ( 0 , 1 ) . Let { x n } be a sequence generated in the following iterative process:
x 0 C , x n + 1 = α n f ( x n ) + ( 1 α n ) J r n x n , n 0 ,
where J r n = ( I + r n B ) 1 . Assume that the sequences { α n } and { r n } satisfy the following restrictions:
  1. (1)

    lim n α n = 0 , n = 0 α n = ;

     
  2. (2)

    n = 1 | α n α n 1 | < ;

     
  3. (3)

    lim inf n r n > 0 , n = 1 | r n r n 1 | < .

     

Then the sequence { x n } converges strongly to x = Proj B 1 ( 0 ) f ( x ) , where Proj B 1 ( 0 ) is the unique sunny nonexpansive retraction of C onto B 1 ( 0 ) .

Corollary 3.4 Let E be a real q-uniformly smooth Banach space with the constant K q , and let C be a nonempty closed and convex subset of E. Let f : C C be a fixed κ-contraction. Let T : C C be an α-strictly pseudocontractive mapping with a nonempty fixed point set. Let { r n } , { α n } , { β n } , and { γ n } be positive number sequences, where { α n } , { β n } , and { γ n } in ( 0 , 1 ) . Let { x n } be a sequence generated in the following process:
x 0 C , x n + 1 = α n f ( x n ) + β n ( 1 r n ) x n + r n β n T x n + γ n f n , n 0 ,
where { f n } is a bounded sequence in C. Assume that the sequences { α n } , { β n } , { γ n } , and { r n } satisfy the following restrictions:
  1. (1)

    α n + β n + γ n = 1 ;

     
  2. (2)

    lim n α n = 0 , n = 0 α n = ;

     
  3. (3)

    n = 1 | β n β n 1 | < ;

     
  4. (4)

    lim inf n r n > 0 , r n ( q α K q ) 1 q 1 , n = 1 | r n r n 1 | < ;

     
  5. (5)

    n = 0 γ n < .

     

Then the sequence { x n } converges strongly to x = Proj F ( T ) f ( x ) , where Proj F ( T ) is the unique sunny nonexpansive retraction of C onto F ( T ) .

Proof Putting A = I T , we find that A is α-inverse strongly accretive and F ( T ) = A 1 ( 0 ) . Notice that
x n + 1 = α n f ( x n ) + β n ( 1 r n ) x n + r n β n T x n + γ n f n = α n f ( x n ) + β n ( ( 1 r n ) x n + r n T x n ) + γ n f n = α n f ( x n ) + β n ( x n r n ( I T ) x n ) + γ n f n = α n f ( x n ) + β n ( x n r n A x n ) + γ n f n .

Using Theorem 3.1, we find the desired conclusion immediately. □

4 Applications

In this section, we give some applications of our main results in the framework of Hilbert spaces.

From now on, we always assume that C is a nonempty closed and convex subset of a real Hilbert space H and P C stands for the metric projection from H onto C. Let A : C H be a monotone operator. Recall that the classical variational inequality is to find x C such that
A x , y x 0 , y C .
(4.1)

The solution set of the variational inequality is denoted by VI ( C , A ) .

Let i C be a function defined by
i C ( x ) = { 0 , x C , , x C .
It is easy to see that i C is a proper lower and semicontinuous convex function on H, and the subdifferential i C of i C is maximal monotone. Define the resolvent J r : = ( I + r i C ) 1 of the subdifferential operator i C . Letting x = J r y , we find that
y x + r i C x y x + r N C x y x , v x 0 , v C x = P C y ,

where N C x : = { e H : e , v x , v C } .

Putting B = i C in Theorem 3.1, we find that J r n = P C . Hence, the following result can be obtained immediately.

Theorem 4.1 Let f : C C be a fixed κ-contraction. Let A : C E be an α-inverse strongly monotone operator with VI ( C , A ) . Let { r n } be a positive number sequence. Let { α n } , { β n } , and { γ n } be real number sequences in ( 0 , 1 ) . Let { x n } be a sequence generated in the following iterative process:
x 0 C , x n + 1 = α n f ( x n ) + β n P C ( x n r n A x n + e n ) + γ n f n , n 0 ,
where { e n } is a sequence in H and { f n } is a bounded sequence in C. Assume that the sequences { α n } , { β n } , { γ n } , { e n } , and { r n } satisfy the following restrictions:
  1. (1)

    α n + β n + γ n = 1 ;

     
  2. (2)

    lim n α n = 0 , n = 0 α n = ;

     
  3. (3)

    n = 1 | β n β n 1 | < ;

     
  4. (4)

    lim inf n r n > 0 , r n 2 α , n = 1 | r n r n 1 | < ;

     
  5. (5)

    n = 0 e n < , n = 0 γ n < .

     

Then the sequence { x n } converges strongly to x = P VI ( C , A ) f ( x ) , where P VI ( C , A ) is the unique metric projection from C onto VI ( C , A ) .

Next, we consider the problem of finding a solution of a Ky Fan inequality [23], which is known as an equilibrium problem in the terminology of Blum and Oettli; see [24] and the references therein.

Let F be a bifunction of C × C into , where denotes the set of real numbers. Recall the following equilibrium problem:
Find  x C  such that  F ( x , y ) 0 , y C .
(4.2)

The solution set of the problem is denoted by EP ( F ) in this section.

To study the equilibrium problem (4.2), we may assume that F satisfies the following restrictions:

(A1) F ( x , x ) = 0 for all x C ;

(A2) F is monotone, i.e., F ( x , y ) + F ( y , x ) 0 for all x , y C ;

(A3) for each x , y , z C , lim sup t 0 F ( t z + ( 1 t ) x , y ) F ( x , y ) ;

(A4) for each x C , y F ( x , y ) is convex and lower semicontinuous.

The following lemma can be found in [24].

Lemma 4.2 Let F : C × C R be a bifunction satisfying (A1)-(A4). Then, for any r > 0 and x H , there exists z C such that F ( z , y ) + 1 r y z , z x 0 , y C . Further, define
T r x = { z C : F ( z , y ) + 1 r y z , z x 0 , y C }
(4.3)

for all r > 0 and x H . Then (1) T r is single-valued and firmly nonexpansive; (2) F ( T r ) = EP ( F ) is closed and convex.

Lemma 4.3 Let F be a bifunction from C × C to which satisfies (A1)-(A4), and let A F be a multivalued mapping of H into itself defined by
A F x = { { z H : F ( x , y ) y x , z , y C } , x C , , x C .
(4.4)

Then A F is a maximal monotone operator with the domain D ( A F ) C , EP ( F ) = A F 1 ( 0 ) , and T r x = ( I + r A F ) 1 x , x H , r > 0 , where T r is defined as in (4.3).

Based on Lemmas 4.2 and 4.3, we find from Theorem 3.1 the following immediately.

Theorem 4.4 Let F : C × C R be a bifunction satisfying (A1)-(A4) such that EP ( F ) is not empty. Let f : C C be a fixed κ-contraction. Let { r n } be a positive number sequence. Let { α n } , { β n } , and { γ n } be real number sequences in ( 0 , 1 ) . Let { x n } be a sequence generated in the following iterative process:
x 0 C , x n + 1 = α n f ( x n ) + β n T r n ( x n r n A x n + e n ) + γ n f n , n 0 ,
where J r n = ( I + r n A F ) 1 , { e n } is a sequence in H and { f n } is a bounded sequence in C. Assume that the sequences { α n } , { β n } , { γ n } , { e n } , and { r n } satisfy the following restrictions:
  1. (1)

    α n + β n + γ n = 1 ;

     
  2. (2)

    lim n α n = 0 , n = 0 α n = ;

     
  3. (3)

    n = 1 | β n β n 1 | < ;

     
  4. (4)

    lim inf n r n > 0 , r n 2 α , n = 1 | r n r n 1 | < ;

     
  5. (5)

    n = 0 e n < , n = 0 γ n < .

     

Then the sequence { x n } converges strongly to some point in EP ( F ) .

For any matrix D R m × n , we denote its transpose by D T and its operator norm by D = max x R n : x = 1 D x .

Consider the inclusion problem [10] 0 A ( x 1 , x 2 , y ) + B ( x 1 , x 2 , y ) , where A ( x 1 , x 2 , y ) = ( D T y , E T y , D x 1 E x 2 ) , B ( x 1 , x 2 , y ) = T 1 x 1 × T 2 x 2 × { b } and T 1 and T 2 are maximal monotone mappings on R n 1 and R n 2 , respectively, and D R m × n 1 , E R m × n 2 , b R m . Then, A and B are maximal monotone and A is Lipschitz continuous on R m + n 1 + n 2 with the constant
η = D T 2 + E T 2 + D 2 + E 2 .
The special case where T 1 = f 1 , T 2 = f 2 yields the following convex program:
{ minimize  f 1 ( x 1 ) + f 2 ( x 2 ) subject to  D x 1 + E x 2 = b ,

where f 1 and f 2 are closed proper convex functions on, respectively, R n 1 and R n 2 . The special case where n 1 = n 2 , D = E = I and b = 0 yields the inclusion 0 T 1 x + T x 2 .

Finally, we consider finding minimizers of proper lower semicontinuous convex functions.

For a proper lower semicontinuous convex function h : H ( , ] , the subdifferential mapping ∂h of h is defined by
h ( x ) = { x H : h ( x ) + y x , x h ( y ) , y H } , x H .

Rockafellar [25] proved that ∂h is a maximal monotone operator. It is easy to verify that 0 h ( v ) if and only if h ( v ) = min x H h ( x ) .

Theorem 4.5 Let f : C C be a fixed κ-contraction. Let h : H ( , + ] be a proper convex lower semicontinuous function such that ( h ) 1 ( 0 ) is not empty. Let { r n } be a positive number sequence. Let { α n } , { β n } , and { γ n } be real number sequences in ( 0 , 1 ) . Let { x n } be a sequence generated in the following iterative process:
x 0 C , x n + 1 = α n f ( x n ) + β n arg min z H { h ( z ) + z x n e n 2 2 r n } + γ n f n , n 0 ,
where { e n } is a sequence in H and { f n } is a bounded sequence in C. Assume that the sequences { α n } , { β n } , { γ n } , { e n } , and { r n } satisfy the following restrictions:
  1. (1)

    α n + β n + γ n = 1 ;

     
  2. (2)

    lim n α n = 0 , n = 0 α n = ;

     
  3. (3)

    n = 1 | β n β n 1 | < ;

     
  4. (4)

    lim inf n r n > 0 , r n 2 α , n = 1 | r n r n 1 | < ;

     
  5. (5)

    n = 0 e n < , n = 0 γ n < .

     

Then the sequence { x n } converges strongly to some minimizer of h.

Proof Since h : H ( , ] is a proper convex and lower semicontinuous function, we see that the subdifferential ∂h of h is maximal monotone. Putting A = 0 and y n = J r n ( x n + e n ) , we see that
y n = arg min z H { h ( z ) + z x n e n 2 2 r n }
is equivalent to
0 h ( y n ) + 1 r n ( y n x n e n ) .
It follows that
x n + e n y n + r n h ( y n ) .

By using Theorem 3.1, we draw the desired conclusion immediately. □

Declarations

Acknowledgements

The authors are grateful to the anonymous referees for useful suggestions, which improved the contents of the article.

Authors’ Affiliations

(1)
Department of Mathematics, Hangzhou Normal University
(2)
Department of Mathematics, Gyeongsang National University
(3)
College of Statistics and Mathematics, Yunnan University of Finance and Economics

References

  1. Combettes PL, Wajs VR: Single recovery by proximal forward-backward splitting. Multiscale Model. Simul. 2005, 4: 1168–1200. 10.1137/050626090View ArticleMathSciNetGoogle Scholar
  2. Chen GHG, Rockafellar RT: Convergence rates in forward-backward splitting. SIAM J. Optim. 1997, 7: 421–444. 10.1137/S1052623495290179View ArticleMathSciNetGoogle Scholar
  3. Rockafellar RT: Monotone operators and the proximal point algorithm. SIAM J. Control Optim. 1976, 14: 877–898. 10.1137/0314056View ArticleMathSciNetGoogle Scholar
  4. He RH: Coincidence theorem and existence theorems of solutions for a system of Ky Fan type minimax inequalities in FC-spaces. Adv. Fixed Point Theory 2012, 2: 47–57.Google Scholar
  5. Peaceman DH, Rachford HH: The numerical solutions of parabolic and elliptic differential equations. J. Soc. Ind. Appl. Math. 1955, 3: 28–41. 10.1137/0103003View ArticleMathSciNetGoogle Scholar
  6. Cho SY, Qin X, Wang L: Strong convergence of a splitting algorithm for treating monotone operators. Fixed Point Theory Appl. 2014., 2014: Article ID 94Google Scholar
  7. Moudafi A, Thera M: Finding a zero of the sum of two maximal monotone operators. J. Optim. Theory Appl. 1997, 94: 425–448. 10.1023/A:1022643914538View ArticleMathSciNetGoogle Scholar
  8. Bauschke HH: A note on the paper by Eckstein and Svaiter on general projective splitting methods for sums of maximal monotone operators. SIAM J. Control Optim. 2009, 48: 2513–2515. 10.1137/090759690View ArticleMathSciNetGoogle Scholar
  9. Tseng P: Applications of a splitting algorithm to decomposition in convex programming and variational inequalities. SIAM J. Control Optim. 1991, 29: 119–138. 10.1137/0329006View ArticleMathSciNetGoogle Scholar
  10. Tseng P: A modified forward-backward splitting method for maximal monotone mappings. SIAM J. Control Optim. 2000, 38: 431–446. 10.1137/S0363012998338806View ArticleMathSciNetGoogle Scholar
  11. Cho SY, Qin X: On the strong convergence of an iterative process for asymptotically strict pseudocontractions and equilibrium problems. Appl. Math. Comput. 2014, 235: 430–438.View ArticleMathSciNetGoogle Scholar
  12. Lions PL, Mercier B: Splitting algorithms for the sum of two nonlinear operators. SIAM J. Numer. Anal. 1979, 16: 964–979. 10.1137/0716071View ArticleMathSciNetGoogle Scholar
  13. Passty GB: Ergodic convergence to a zero of the sum of monotone operators in Hilbert space. J. Math. Anal. Appl. 1979, 72: 383–390. 10.1016/0022-247X(79)90234-8View ArticleMathSciNetGoogle Scholar
  14. Han SP, Lou G: A parallel algorithm for a class of convex programs. SIAM J. Control Optim. 1988, 26: 345–355. 10.1137/0326019View ArticleMathSciNetGoogle Scholar
  15. Reich S: Asymptotic behavior of contractions in Banach spaces. J. Math. Anal. Appl. 1973, 44: 57–70. 10.1016/0022-247X(73)90024-3View ArticleMathSciNetGoogle Scholar
  16. Kato T: Nonlinear semigroups and evolution equations. J. Math. Soc. Jpn. 1967, 19: 508–520. 10.2969/jmsj/01940508View ArticleGoogle Scholar
  17. Xu HK: Inequalities in Banach spaces with applications. Nonlinear Anal. 1991, 16: 1127–1138. 10.1016/0362-546X(91)90200-KView ArticleMathSciNetGoogle Scholar
  18. Aoyama K, Kimura Y, Takahashi W, Toyoda M: On a strongly nonexpansive sequence in Hilbert spaces. J. Nonlinear Convex Anal. 2007, 8: 471–489.MathSciNetGoogle Scholar
  19. Barbu V: Nonlinear Semigroups and Differential Equations in Banach Space. Noordhoff, Groningen; 1976.View ArticleGoogle Scholar
  20. Liu LS: Ishikawa and Mann iterative process with errors for nonlinear strongly accretive mappings in Banach spaces. J. Math. Anal. Appl. 1995, 194: 114–125. 10.1006/jmaa.1995.1289View ArticleMathSciNetGoogle Scholar
  21. Mitrinovic DS: Analytic Inequalities. Springer, New York; 1970.View ArticleGoogle Scholar
  22. Qin X, Cho SY, Wang L: Iterative algorithms with errors for zero points of m -accretive operators. Fixed Point Theory Appl. 2013., 2013: Article ID 148Google Scholar
  23. Fan K: A minimax inequality and applications. In Inequalities III. Edited by: Shisha O. Academic Press, New York; 1972:103–113.Google Scholar
  24. Blum E, Oettli W: From optimization and variational inequalities to equilibriums problems. Math. Stud. 1994, 63: 123–145.MathSciNetGoogle Scholar
  25. Rockafellar RT: Monotone operators and the proximal point algorithm. SIAM J. Control Optim. 1976, 14: 877–898. 10.1137/0314056View ArticleMathSciNetGoogle Scholar

Copyright

© Qin et al.; licensee Springer. 2014

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.