Skip to content

Advertisement

  • Research
  • Open Access

Proximal algorithms for a class of mixed equilibrium problems

Fixed Point Theory and Applications20122012:166

https://doi.org/10.1186/1687-1812-2012-166

  • Received: 21 February 2012
  • Accepted: 17 September 2012
  • Published:

Abstract

We present two proximal algorithms for solving the mixed equilibrium problems. Under some simpler framework, the strong and weak convergence of the sequences defined by two general algorithms is respectively obtained. In particular, we deal with several iterative schemes in a united way and apply our algorithms for solving the classical equilibrium problem, the minimization problem, the classical variational inequality problem and the generalized variational inequality problem. Our results properly include some corresponding results in this field as a special case.

MSC:47H06, 47J05, 47J25, 47H10, 90C33, 90C25, 49M45, 65C25, 49J40, 65J15, 47H09.

Keywords

  • mixed equilibrium problem
  • proximal algorithm
  • variational inequality

1 Introduction

Throughout the paper, H is a real Hilbert space with inner product , and induced norm . Let K be a nonempty closed convex subset of H, F : K × K R be a bifunction and φ : K R { + } be a proper generalized real valued function, where R is the set of real numbers. Our interest is in finding a solution to the following problem which is referred to as the mixed equilibrium problem (for short, MEP) for F, φ
find  x K  such that  F ( x , y ) + φ ( y ) φ ( x ) 0 , y K .
(1.1)
We denote the set of solutions for MEP by
MEP ( F , φ ) = { x K ; F ( x , y ) + φ ( y ) φ ( x ) 0 , y K } .
Obviously, MEP (1.1) is a classical equilibrium problem (for short, EP) for F when φ 0
find  x K  such that  F ( x , y ) 0 , y K .
(1.2)

The set MEP ( F , 0 ) of solutions of EP (1.2) is denoted by EP ( F ) .

If F 0 , then MEP (1.1) becomes the minimization problem (for short, MP) for a function φ
find  x K  such that  φ ( y ) φ ( x ) , y K ,
(1.3)

and MEP ( 0 , φ ) is denoted by Argmin ( φ ) .

Given a mapping T : K H , let F ( x , y ) = T x , y x for all x , y K . Then z EP ( F ) if and only if T z , y z 0 for all y K , i.e., EP (1.2) turns into a classical variational inequality problem (for short, VIP) for T
find  x K  such that  T x , y x 0 , y K .
(1.4)
At the same time, MEP (1.1) also reduces a generalized variational inequality problem (for short, GVIP) for a mapping T and a function φ
find  x K  such that  T x , y x + φ ( y ) φ ( x ) 0 for all  y K ,
(1.5)
and denote
GVIP ( T , φ , K ) = { x K ; T x , y x + φ ( y ) φ ( x ) 0  for all  y K }

and VIP ( T , K ) = GVIP ( T , 0 , K ) .

The mixed equilibrium problem (MEP) is very interesting because it covers mathematical programs and optimization problems over equilibrium constraints, hierarchical minimization problems, variational inequality, complementarity problems, monotone inclusion problems, saddle point problems, Nash equilibria in noncooperative games as well as certain fixed point problems. In other words, the mixed equilibrium problem unifies several problems arising from engineering, physics, statistics, computer science, optimization theory, operation research, economics and others. The interest of this problem is that it unites all these particular problems in a convenient way. Moreover, many methods devoted to solving one of these problems can be extended, with suitable modifications, to solving the mixed equilibrium problem (MEP).

In the last 20 years or so, many mathematical works have been devoted to studying the method for finding an approximate solution of EP ( F ) with various types of additional conditions. Moudafi [1] extended the proximal method to monotone equilibrium problems and Konnov [2] used the proximal method to solve equilibrium problems with weakly monotone bifunctions. In 2005, Combettes and Hirstoaga [3] introduced an iterative scheme of finding the best approximation to the initial data when EP ( F ) is nonempty and they also proved a strong convergence theorem under the condition lim n r n = + . Song and Zheng [4] and Song, Kang and Cho [5] considered the convergence of a Halpern-type iteration for EP ( F ) . The bundle methods and extragradient methods were extended to equilibrium problems in [6] and [7]. In 2007, Takahashi and Takahashi [8] introduced the viscosity approximation method for finding a common element of the set of solutions of an equilibrium problem EP ( F ) and the set of fixed points of a nonexpansive mapping T under the conditions lim inf n r n = + and n = 0 + r n r n + 1 < + . Subsequently, the above viscosity approximation method was studied by different mathematicians using varying environmental conditions; see references [913] for details. Recently, Moudafi [14] showed weak convergence of proximal methods for a class of bilevel monotone equilibrium problems. For other solution methods regarding the equilibrium problems see [2, 1522].

In this paper, our main objective is to show the mixed equilibrium problem (1.1) can be solved by two very simple proximal methods under simpler conditions, where a bifunction F and a function φ satisfy the following standard assumptions.

Condition 1.1 The function φ : K R { + } is a proper lower semicontinuous convex function and the bifunction F : K × K R satisfies the following:

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

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

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

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

To this end, we introduce two proximal algorithms.

Algorithm 1 For any anchor u K and initialization x 0 K , if x n ( n 0 ) is the current iterate and α n , r n are the current parameters, then x n + 1 K is defined iteratively by
{ F ( u n , y ) + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n u + ( 1 α n ) u n , y K , n 0 .
(1.6)
We will show { x n } strongly converges to some element x of MEP ( F , φ ) if only { α n } ( 0 , 1 ) and { r n } ( 0 , + ) satisfy the following simple conditions:
( C 1 ) lim n α n = 0 , ( C 2 ) n = 1 + α n = + , ( C 3 ) lim inf n r n > 0 .
Algorithm 2 For any initialization x 0 K , if x n ( n 0 ) is the current iterate and α n , r n are the current parameters, then x n + 1 K is given iteratively by
{ F ( u n , y ) + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n x n + ( 1 α n ) u n , y K , n 0 .
(1.7)
We will prove { x n } weakly converges to some element x of MEP ( F , φ ) whenever { r n } ( 0 , + ) and { α n } ( 0 , 1 ) only satisfy the simpler conditions
( C 3 ) lim inf n r n > 0 and ( C 4 ) lim sup n α n < 1 .

Some other proximal methods and the related strong and weak convergence results can be derived from our main theorem by particularizing the bifunction F, the function H and the parameters α n , r n . Our results treat several iterative schemes in a united way and properly include some results of [35, 8, 20, 21, 23, 24] as a special case.

The rest of the paper is organized as follows. In Section 2, we introduce some necessary lemmas and results, and show our algorithms are well defined. In Section 3, we study the strong convergence of a sequence iteratively given by Algorithm 1. In Section 4, we deal with the weak convergence of an iterative scheme defined by Algorithm 2. Some other proximal methods and the related results derived from our main theorems and some concluding remarks can be found in Section 5.

2 Preliminaries and basic results

Let { x n } be a sequence in H. We write x n x to indicate that the sequence { x n } weakly converges to x; as usual, x n x will symbolize strong convergence. Let F ( T ) = { x H ; T x = x } be the set of all fixed points for a mapping T. In a real Hilbert space H, we have
t x + ( 1 t ) y 2 = t x 2 + ( 1 t ) y 2 t ( 1 t ) x y 2
(2.1)
for all x , y H and t [ 0 , 1 ] . Let K be a nonempty closed convex subset of H. Then for any x H , there exists a unique nearest point in K, denoted by P K ( x ) , such that
x P K ( x ) x y for all  y K .
Such a P K is called the metric projection of H onto K. We know that P K is firmly nonexpansive, i.e., for any x , y H ,
P K ( x ) P K ( y ) 2 P K ( x ) P K ( y ) , x y .
Further, for x H and z K ,
z = P K ( x ) x z , y z 0 for all  y K .
(2.2)

For a bifunction Ψ : K × K R , we have the following lemmas which were also given in [17].

Lemma 2.1 (Blum-Oettli [[17], Corollary 1])

Let K be a nonempty closed convex subset of H and Ψ be a bifunction of K × K into R satisfying (A1)-(A4) in Condition 1.1. Let r > 0 and x H . Then there exists z K such that
Ψ ( z , y ) + 1 r y z , z x 0 , y K .

Using a similar proof technique of Combettes-Hirstoaga [[3], Lemma 2.12], also see Ceng and Yao [[15], Lemma 3.1], Peng, Liou and Yao [[16], Lemma 2.2] obtained the following lemma which guarantees that our algorithms are well defined.

Lemma 2.2 ([[16], Lemma 2.2])

Let K be a nonempty closed convex subset of H. Suppose that the bifunction F : K × K R and the function φ : K R { + } satisfy Condition  1.1. For r > 0 and x H , define a mapping T r : H 2 K as follows:
T r ( x ) = { z K : F ( z , y ) + φ ( y ) φ ( z ) + 1 r y z , z x 0 , y K } .
Then the following hold:
  1. (1)

    the domain of T r is H.

     
  2. (2)
    T r is single-valued and firmly nonexpansive, i.e., for any x , y H ,
    T r ( x ) T r ( y ) 2 T r ( x ) T r ( y ) , x y
     
or equivalently,
T r ( x ) T r ( y ) 2 x y 2 ( x T r ( x ) ) ( y T r ( y ) ) 2 .
  1. (3)

    F ( T r ) = MEP ( F , φ ) .

     
  2. (4)

    MEP ( F , φ ) is closed and convex.

     

Lemma 2.3 was showed and used by several authors. For detailed proofs, see Liu [25] and Xu [26, 27]. Furthermore, a variant of Lemma 2.3 has already been used by Reich in [[28], Theorem 1].

Lemma 2.3 Let { a n } be a sequence of nonnegative real numbers satisfying the property
a n + 1 ( 1 t n ) a n + t n c n , n 0 ,

where { t n } and { c n } satisfy the restrictions n = 0 t n = and lim sup n c n 0 . Then { a n } converges to zero as n .

3 Strong convergence of Algorithm 1

In this section, we deal with an iterative scheme given by Algorithm 1 for finding an element of the set of solutions of the MEP (1.1) in a Hilbert space.

Theorem 3.1 Let K be a nonempty closed convex subset of a real Hilbert space H. Assume that a bifunction F and a function φ satisfy Condition 1.1 and MEP ( F , φ ) . For any anchor u K and initialization x 0 K , let { x n } and { u n } be two sequences generated iteratively by
{ F ( u n , y ) + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n u + ( 1 α n ) u n , y K , n 0 .
(3.1)
If only { α n } ( 0 , 1 ) and { r n } ( 0 , + ) satisfy
( C 1 ) lim n α n = 0 , ( C 2 ) n = 1 + α n = + and ( C 3 ) lim inf n r n > 0 ,

then { x n } converge strongly to some element x = P MEP ( F , φ ) ( u ) of MEP ( F , φ ) .

Proof At first, we show that { x n } and { u n } are bounded. Taking p MEP ( F , φ ) , it follows from Lemma 2.2 that u n = T r n ( x n ) and
x n + 1 p ( 1 α n ) T r n ( x n ) p + α n u p ( 1 α n ) x n p + α n u p max { x n p , u p } max { x 0 p , u p } .

Thus, { x n } is bounded, and hence so is { u n } .

With the help of the geometric properties of a Hilbert space, from the closed convexity of MEP ( F , φ ) , for u K , there exists a unique nearest point x MEP ( F , φ ) , that is, x = P MEP ( F , φ ) u . By equation (2.1) and Lemma 2.2(2), (3), for each fixed n, we also have
x n + 1 x 2 = α n ( u x ) + ( 1 α n ) ( u n x ) 2 α n u x 2 + ( 1 α n ) T r n ( x n ) x 2 T r n ( x n ) x 2 + α n u x 2 x n x 2 x n T r n ( x n ) 2 + α n u x 2 ,
which can be rewritten as
x n u n 2 α n M + x n x 2 x n + 1 x 2 ,
(3.2)

where M = u x 2 . Following the proof technique in Mainge [[29], Lemma 3.2, Theorem 3.1], the proof may be divided into two cases.

Case 1. If there exists N 0 such that the sequence { x n x 2 } is nonincreasing for n N 0 , then the limit lim n x n x 2 exists, and hence lim n ( x n x 2 x n + 1 x 2 ) = 0 . So, by the condition (C1) and inequality (3.2), we have
lim n x n u n = 0 .
(3.3)
Next, we show
lim sup n u x , u n x 0 .
(3.4)
Indeed, we can take a subsequence { u n k } of { u n } such that
lim sup n u x , u n x = lim k u x , u n k x .
Without loss of generality, we may assume that u n k x by the boundedness of { u n } . Let us show x MEP ( F , φ ) . In fact, the first inequality of (3.1) yields
F ( u n , y ) + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , y K , n 0 .
From the property (A2) of Condition 1.1, we also have
1 r n y u n , u n x n F ( u n , y ) φ ( y ) + φ ( u n ) F ( y , u n ) + φ ( u n ) φ ( y )
and hence by replacing n by n k , we obtain
y u n k , u n k x n k r n k F ( y , u n k ) + φ ( u n k ) φ ( y ) , y K .
By passing to the limit in this inequality and by taking into account the condition (C3) and the fact that lim k u n k x n k r n k = 0 , u n k x and that the bifunction F ( y , ) and the function φ ( ) are weak lower-semicontinuous, we deduce that
0 F ( y , x ) + φ ( x ) φ ( y ) , y K .
Now, for t with 0 < t 1 and y K , let y t = t y + ( 1 t ) x . Since y K and x K , we have y t K , and hence 0 F ( y t , x ) + φ ( x ) φ ( y t ) . So, by virtue of (A1), (A4) of the bifunction F and convexity of the function φ, we get
0 = F ( y t , y t ) + φ ( y t ) φ ( y t ) t F ( y t , y ) + ( 1 t ) F ( y t , x ) + t φ ( y ) + ( 1 t ) φ ( x ) φ ( y t ) = t F ( y t , y ) + t ( φ ( y ) φ ( y t ) ) + ( 1 t ) ( F ( y t , x ) + φ ( x ) φ ( y t ) ) t ( F ( y t , y ) + φ ( y ) φ ( y t ) ) .
Consequently, we deduce
F ( y t , y ) + φ ( y ) φ ( y t ) 0 , y K .
In the light of (A3) of the bifunction F and lower-semicontinuity of the function φ, we have
F ( x , y ) + φ ( y ) φ ( x ) 0 , y K ,
thus x MEP ( F , φ ) . Since x = P MEP ( F , φ ) u , then by (2.2), we have
lim sup n u x , u n x = lim k u x , u n k x = u x , x x 0 .
Now, we show that x n x ( n ). In fact, from the second equality of (3.1) and x F ( T r ) = MEP ( F , φ ) for each r > 0 together with Lemma 2.2(2), it follows
x n + 1 x 2 = α n ( u x ) + ( 1 α n ) ( u n x ) 2 = α n 2 u x 2 + ( 1 α n ) 2 u n x 2 + 2 α n ( 1 α n ) u x , u n x ( 1 α n ) T r n ( x n ) T r n ( x ) 2 + α n β n ( 1 α n ) x n x 2 + α n β n ,
which can be rewritten as
x n + 1 x 2 ( 1 α n ) x n x 2 + α n β n ,
(3.5)
where β n = α n u x 2 + 2 ( 1 α n ) u x , u n x . From the condition (C1) and (3.4), we obtain
lim sup n β n = lim sup n ( α n u x 2 + 2 ( 1 α n ) u x , u n x ) 0 .

So, an application of Lemma 2.3 onto (3.5) yields x n x .

Case 2. Assume that there exists a subsequence { x n k x 2 } of { x n x 2 } such that x n k x 2 < x n k + 1 x 2 for all k 0 . Let
Γ n = x n x 2 and τ ( n ) = max { k n ; Γ k < Γ k + 1 } .
It follows from Mainge [[29], Lemma 3.1] that τ ( n ) is a nondecreasing sequence verifying lim n τ ( n ) = + , and for n large enough,
Γ τ ( n ) Γ τ ( n ) + 1 , Γ n = x n x 2 Γ τ ( n ) + 1 .
(3.6)
In light of equation (3.2), we have x τ ( n ) u τ ( n ) 2 α τ ( n ) M , and so by the condition (C1), we have
lim n x τ ( n ) u τ ( n ) = 0 .
(3.7)
Using the similar proof techniques as in Case 1, the only modification being that n ( n k ) is replaced by τ ( n ) ( τ ( n k ) , respectively), we have
lim sup n u x , u τ ( n ) x 0
(3.8)
and
x τ ( n ) + 1 x 2 ( 1 α τ ( n ) ) x τ ( n ) x 2 + α τ ( n ) β τ ( n ) ,
(3.9)
where β τ ( n ) = α τ ( n ) u x 2 + 2 ( 1 α τ ( n ) ) u x , u τ ( n ) x , and so
lim sup n β τ ( n ) 0 .
By (3.6), we have
Γ τ ( n ) Γ τ ( n ) + 1 ( 1 α τ ( n ) ) Γ τ ( n ) + α τ ( n ) β τ ( n ) ,
and hence
Γ τ ( n ) β τ ( n ) .
So, we obtain
lim n Γ τ ( n ) = lim n x τ ( n ) x = 0 .
Together with (3.9), we have
lim n x τ ( n ) + 1 x = lim n x τ ( n ) x = 0 .

Now, it follows from (3.6) that Γ n = x n x 0 . The proof is completed. □

4 Weak convergence of Algorithm 2

In this section, we show a weak convergence theorem which solves the MEP (1.1) in a Hilbert space.

Theorem 4.1 Let K be a nonempty closed convex subset of a real Hilbert space H. Assume that a bifunction F and a function φ satisfy Condition 1.1 and MEP ( F , φ ) . For any initialization x 0 K , if x n ( n 0 ) is the current iterate and positive real numbers α n , r n are the current parameters, then x n + 1 K is given iteratively by
{ F ( u n , y ) + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n x n + ( 1 α n ) u n , y K , n 0 .
(4.1)
Assume that { α n } ( 0 , 1 ) satisfies (C4) lim sup n α n < 1 . Then
n = 0 + u n x n 2 < + .

If, in addition, { r n } ( 0 , + ) satisfies (C3) lim inf n r n > 0 , then { x n } and { u n } converge weakly to some element x of MEP ( F , φ ) .

Proof Take p MEP ( F , φ ) . It follows from Lemma 2.2 that u n = T r n ( x n ) and
x n + 1 p ( 1 α n ) T r n ( x n ) p + α n x n p ( 1 α n ) x n p + α n x n p x n p x 0 p .

So, { x n } and { u n } are bounded. Moreover, the limit lim n x n p exists for each p MEP ( F , φ ) .

It follows from equation (2.1) and Lemma 2.2(2), (3) that for each fixed n,
x n + 1 p 2 = α n ( x n p ) + ( 1 α n ) ( u n p ) 2 α n x n p 2 + ( 1 α n ) T r n ( x n ) p 2 α n x n p 2 + ( 1 α n ) ( x n p 2 x n T r n ( x n ) 2 ) x n p 2 ( 1 α n ) x n T r n ( x n ) 2 ,
which can be rewritten as
( 1 α n ) x n u n 2 x n p 2 x n + 1 p 2 .
(4.2)
Therefore, we obtain
n = 0 m ( 1 α n ) x n u n 2 x 0 p 2 x m p 2 x 0 p 2 ,
and hence
n = 0 + ( 1 α n ) x n u n 2 < + .
Since { α n } ( 0 , 1 ) satisfies (C4) lim sup n α n < 1 , then there exists a ( 0 , 1 ) and N sufficiently large such that 0 < α n a < 1 for all n > N , which implies ( 1 a ) x n u n 2 ( 1 α n ) x n u n 2 . Consequently,
n = 0 + x n u n 2 < + ,
and so
lim n x n u n = 0 .
(4.3)

Therefore, the limit lim n u n p exists for each p MEP ( F , φ ) since x n p x n u n u n p x n p .

By properties of Hilbert spaces, the boundedness of { u n } means that the sequence { u n } is weak compact in H, and hence there exists a subsequence { u n k } of { u n } such that u n k x . Then using the same argument as in the proof of Theorem 3.1 by means of (4.3) and the condition (C3) together with the properties of the bifunction F and the function φ, we must have x MEP ( F , φ ) .

Next, we show that there is a unique weak cluster point x of the sequence { u n } , that is, u n x . This fact can be reached from the Opial property of H and the existence of lim n u n p for each p MEP ( F , φ ) . The proof is presented here for completeness. Indeed, let z be another weak cluster point of { u n } , write as u n i z . Then we also have z MEP ( F , φ ) . Let us show that x = z . For x, z, we have
u n i x 2 = u n i z 2 + z x 2 + 2 u n i z , z x
and
u n k z 2 = u n k x 2 + x z 2 + 2 u n k x , x z .
Since both lim n u n x and lim n u n z exist, lim i u n i z , z x = 0 and lim k u n k x , x z = 0 , then
lim n u n x 2 = lim n u n z 2 + z x 2
and
lim n u n z 2 = lim n u n x 2 + x z 2 .

Adding the above equations, we must have 2 x z 2 = 0 , and so x = z . This implies u n x MEP ( F , φ ) . By (4.3), it is obvious that { x n } also converges weakly to x. The desired results are reached. □

5 Deduced results and some remarks

Let K be a nonempty closed convex subset of a real Hilbert space H. We recall that a mapping T : K H is said to be monotone if
T x T y , x y 0 for all  x , y K .

A continuous monotone mapping has the following properties which were given in Nilsrakoo and Saejung [23] and Combettes and Hirstoaga [3] independently.

Lemma 5.1 (Combettes and Hirstoaga [[3], Lemma 2.15], Nilsrakoo and Saejung [[23], Lemmas 19, 20])

Let K be a nonempty closed convex subset of a real Hilbert space H. Let T be a continuous monotone mapping of K into H. Define a bifunction F : K × K R as follows:
F ( x , y ) : = T x , y x for all x , y K .
Then the following hold:
  1. (i)

    F satisfies (A1)-(A4) in Condition 1.1 and VIP ( K , T ) = EP ( F ) ;

     
  2. (ii)
    for x H , z K and r > 0 ,
    F ( z , y ) + 1 r y z , z x 0 , y K z = P K ( x r T z ) .
     

Using Theorems 3.1 and 4.1 along with Lemma 5.1, we have the following theorems which solve GVIP (1.5) and VIP (1.4) under the simpler framework.

Theorem 5.2 Let K be a nonempty closed convex subset of H and T be a continuous monotone mapping of K such that VIP ( K , T ) . For any anchor u K and initialization x 0 K ,
  1. (1)
    let { x n } and { u n } be sequences generated iteratively by
    { u n = P K ( x n r n T u n ) , x n + 1 = α n u + ( 1 α n ) u n .
    (5.1)
     
Assume that α n ( 0 , 1 ) and r n ( 0 , + ) satisfy
( C 1 ) lim n α n = 0 , ( C 2 ) n = 1 + α n = + and ( C 3 ) lim inf n r n > 0 .
Then { x n } converges strongly to x VIP ( K , T ) , which is the nearest point from K to VIP ( K , T ) ;
  1. (2)
    let { x n } and { u n } be sequences generated iteratively by
    { u n = P K ( x n r n T u n ) , x n + 1 = α n x n + ( 1 α n ) u n ,
    (5.2)
     

where r n ( 0 , + ) and α n ( 0 , 1 ) satisfy (C3) and (C4) lim sup n α n < 1 . Then { x n } converges weakly to some x VIP ( K , T ) .

Proof Let F ( x , y ) = T x , y x for all x , y K . It follows from Lemma 5.1 that two iteration schemes (5.1) and (5.2) respectively turn into
{ F ( u n , y ) + 1 r n y u n , u n x n 0 , x n + 1 = α n u + ( 1 α n ) u n , y K , n 0 ,
and
{ F ( u n , y ) + 1 r n y u n , u n x n 0 , x n + 1 = α n x n + ( 1 α n ) u n , y K , n 0 .

Following Theorems 3.1 and 4.1 ( φ 0 ), the desired results are proved. □

Theorem 5.3 Let K be a nonempty closed convex subset of H, φ be a convex and lower semicontinuous function from K to R , and T be a continuous monotone mapping of K such that GVIP ( K , φ , T ) . For any anchor u K and initialization x 0 K ,
  1. (1)
    let { x n } and { u n } be sequences generated iteratively by
    { T u n , y u n + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n u + ( 1 α n ) u n , y K , n 0 ,
    (5.3)
     
where α n ( 0 , 1 ) and r n ( 0 , + ) satisfy (C1), (C2) and (C3). Then { x n } converges strongly to x GVIP ( K , φ , T ) , which is the nearest point from K to GVIP ( K , φ , T ) ;
  1. (2)
    let { x n } and { u n } be sequences generated iteratively by
    { T u n , y u n + φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , x n + 1 = α n x n + ( 1 α n ) u n , y K , n 0 ,
    (5.4)
     

where r n ( 0 , + ) and α n ( 0 , 1 ) satisfy (C3) and (C4). Then { x n } converges weakly to some x GVIP ( K , φ , T ) .

Proof Let F ( x , y ) = T x , y x for all x , y K . It follows from Lemma 5.1, Theorems 3.1 and 4.1, the desired results are obtained. □

For the classical equilibrium problem (1.2), the following is obvious ( φ 0 in Theorems 3.1 and 4.1).

Theorem 5.4 Let K be a nonempty closed convex subset of H and F be a bifunction from K × K to R such that (A1)-(A4) of Condition 1.1 and EP ( F ) . For any anchor u K and initialization x 0 K ,
  1. (1)
    let { x n } and { u n } be sequences generated iteratively by
    { F ( u n , y ) + 1 r n y u n , u n x n 0 , y K , x n + 1 = α n u + ( 1 α n ) u n , n 0 ,
    (5.5)
     
where α n ( 0 , 1 ) and r n ( 0 , + ) satisfy (C1), (C2) and (C3). Then { x n } converges strongly to x EP ( F ) , which is the nearest point from K to EP ( F ) ;
  1. (2)
    let { x n } and { u n } be sequences generated iteratively by
    { F ( u n , y ) + 1 r n y u n , u n x n 0 , y K , x n + 1 = α n x n + ( 1 α n ) u n , n 0 ,
    (5.6)
     

where r n ( 0 , + ) and α n ( 0 , 1 ) satisfy (C3) and (C4). Then { x n } converges weakly to some x EP ( F ) .

When F 0 in Theorems 3.1 and 4.1, we also have better approximated algorithm about solving the minimization problem (1.3) of a function φ.

Theorem 5.5 Let K be a nonempty closed convex subset of H and φ be a convex and lower-continuous function from K to R such that Argmin ( φ ) . For any anchor u K and initialization x 0 K ,
  1. (1)
    let { x n } and { u n } be sequences generated iteratively by
    { φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , y K , x n + 1 = α n u + ( 1 α n ) u n , n 0 ,
    (5.7)
     
where α n ( 0 , 1 ) and r n ( 0 , + ) satisfy (C1), (C2) and (C3). Then { x n } converges strongly to x Argmin ( φ ) , which is the nearest point from K to Argmin ( φ ) ;
  1. (2)
    let { x n } and { u n } be sequences generated iteratively by
    { φ ( y ) φ ( u n ) + 1 r n y u n , u n x n 0 , y K , x n + 1 = α n x n + ( 1 α n ) u n , n 0 ,
    (5.8)
     

where r n ( 0 , + ) and α n ( 0 , 1 ) satisfy (C3) and (C4). Then { x n } converges weakly to some x Argmin ( φ ) .

We also observe the condition (C4) that lim sup n α n < 1 includes the case that α n = 0 as a special case. Therefore, our iteration scheme contains several proximal point algorithms that compute the solution x n + 1 K of some regularized problem.

Theorem 5.6 Let K be a nonempty closed convex subset of a real Hilbert space H. Assume that a bifunction F and a function φ satisfy Condition 1.1 and MEP ( F , φ ) . Let T be a continuous monotone mapping of K such that GVIP ( K , φ , T ) . For any initialization x 0 K , if x n ( n 0 ) is the current iterate and a positive real number r n is the current parameter, then x n + 1 is defined by computing the solution of the regularized problem
F ( x n + 1 , y ) + φ ( y ) φ ( x n + 1 ) + 1 r n y x n + 1 , x n + 1 x n 0 , y K
(5.9)
or
T x n + 1 , y x n + 1 + φ ( y ) φ ( x n + 1 ) + 1 r n y x n + 1 , x n + 1 x n 0 , y K .
(5.10)
Then
n = 0 + x n + 1 x n 2 < + .
If, in addition, { r n } ( 0 , + ) satisfies (C3) lim inf n r n > 0 , then
  1. (i)

    { x n } , given by (5.9) converge weakly to some element x of MEP ( F , φ ) ;

     
  2. (ii)

    { x n } , given by (5.10) converge weakly to some element x of GVIP ( K , φ , T ) .

     

Proof It follows from Lemma 2.2 or Lemma 5.1 together with (5.9) or (5.10) that x n + 1 = T r n ( x n ) . Then by Theorem 4.1, the desired results are obtained. □

Remark 5.7 (1) By particularizing the bifunction F, the function φ and the parameters α n , r n , some other methods and related results can be derived from our main theorems. For example, taking φ 0 in (5.9) and lim n r n = + in (5.5), we obtain several main results in Combettes and Hirstoaga [3], and in the case where (C3), n = 0 + r n r n + 1 < + and n = 0 + α n ( 1 α n ) = + in (5.6), we have the main result in Nilsrakoo and Saejung [23], and when taking α n [ a , b ] ( 0 , 1 ) in (5.6), a result is proved by Tada and Takahashi [21], and if α n 0 in (5.2), then we obtain Algorithm 2 and Theorem 5 in Solodov and Svaiter [24]; also see [4, 5, 8, 20] and others for varying versions.
  1. (2)

    All strong convergence theorems of this paper remain true if one replaces the anchor point u by a contraction f (that is, so-called viscosity approximation methods with a contraction f) since Song [30] has showed their equivalency. So our main results could recover or develop some viscosity approximation results such as ones in references [8, 10, 11, 13] as well as others not referenced here.

     
  2. (3)

    The framework of holding our conclusions is more general and our results treat several iterative schemes and corresponding results in a united way. Consequently, our main results could be considered as recovering, developing and improving some known related convergence results in this field.

     

Declarations

Acknowledgements

The authors would like to thank the editor and the anonymous referee for useful comments and valuable suggestions on the language and structure of our manuscript. This work is supported by the National Natural Science Foundation of P.R. China (11071279, 11171094, 11271112).

Authors’ Affiliations

(1)
College of Mathematics and Information Science, Henan Normal University, Xinxiang, Henan, 453007, P.R. China
(2)
Department of Applied Mathematics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, P.R. China
(3)
College of Mathematics and Information Science, North China University of Water Resources and Electric Power, Zhengzhou, 450011, P.R. China

References

  1. Moudafi A: Proximal point algorithm extended for equilibrium problems. J. Nat. Geom. 1999, 15: 91–100.MathSciNetGoogle Scholar
  2. Konnov IV: Application of the proximal point method to nonmonotone equilibrium problems. J. Optim. Theory Appl. 2003, 119: 317–333.MathSciNetView ArticleGoogle Scholar
  3. Combettes PL, Hirstoaga SA: Equilibrium programming in Hilbert spaces. J. Nonlinear Convex Anal. 2005, 6: 117–136.MathSciNetGoogle Scholar
  4. Song Y, Zheng Y: Strong convergence of iteration algorithms for a countable family of nonexpansive mappings. Nonlinear Anal. 2009, 71(7–8):3072–3082. 10.1016/j.na.2009.01.219MathSciNetView ArticleGoogle Scholar
  5. Song Y, Kang JI, Cho YJ: On iterations methods for zeros of accretive operators in Banach spaces. Appl. Math. Comput. 2010, 216: 1007–1017. 10.1016/j.amc.2010.01.124MathSciNetView ArticleGoogle Scholar
  6. Tran DQ, Muu LD, Nguyen VH: Extragradient algorithms extended to solving equilibrium problems. Optimization 2008, 57(6):749–776. 10.1080/02331930601122876MathSciNetView ArticleGoogle Scholar
  7. Van NTT, Strodiot JJ, Nguyen VH: A bundle method for solving equilibrium problems. Math. Program., Ser. B 2009, 116(1–2):529–552. 10.1007/s10107-007-0112-xView ArticleGoogle Scholar
  8. Takahashi S, Takahashi W: Viscosity approximation methods for equilibrium problems and fixed point problems in Hilbert spaces. J. Math. Anal. Appl. 2007, 331(1):506–515. 10.1016/j.jmaa.2006.08.036MathSciNetView ArticleGoogle Scholar
  9. Plubtieng S, Thammathiwat T: A viscosity approximation method for equilibrium problems, fixed point problems of nonexpansive mappings and a general system of variational inequalities. J. Glob. Optim. 2010, 46: 447–464. 10.1007/s10898-009-9448-5MathSciNetView ArticleGoogle Scholar
  10. Qin X, Cho YJ, Kang SM: Viscosity approximation methods for generalized equilibrium problems and fixed point problems with applications. Nonlinear Anal. 2010, 72: 99–112. 10.1016/j.na.2009.06.042MathSciNetView ArticleGoogle Scholar
  11. Qin X, Cho YJ, Kang SM: Convergence theorems of common elements for equilibrium problems and fixed point problems in Banach spaces. J. Comput. Appl. Math. 2009, 225: 20–30. 10.1016/j.cam.2008.06.011MathSciNetView ArticleGoogle Scholar
  12. Qin X, Shang M, Su Y: Strong convergence of a general iterative algorithm for equilibrium problems and variational inequality problems. Math. Comput. Model. 2008, 48: 1033–1046. 10.1016/j.mcm.2007.12.008MathSciNetView ArticleGoogle Scholar
  13. Qin X, Shang M, Su Y: A general iterative method for equilibrium problems and fixed point problems in Hilbert spaces. Nonlinear Anal. 2008, 69: 3897–3909. 10.1016/j.na.2007.10.025MathSciNetView ArticleGoogle Scholar
  14. Moudafi A: Proximal methods for a class of bilevel monotone equilibrium problems. J. Glob. Optim. 2010, 47: 287–292. 10.1007/s10898-009-9476-1MathSciNetView ArticleGoogle Scholar
  15. Ceng LC, Yao JC: A hybrid iterative scheme for mixed equilibrium problems and fixed point problems. J. Comput. Appl. Math. 2008, 214: 186–201. 10.1016/j.cam.2007.02.022MathSciNetView ArticleGoogle Scholar
  16. Peng JW, Liou YC, Yao JC: An iterative algorithm combining viscosity method with parallel method for a generalized equilibrium problem and strict pseudocontractions. Fixed Point Theory Appl. 2009., 2009: Article ID 794178Google Scholar
  17. Blum E, Oettli W: From optimization and variational inequalities to equilibriums problems. Math. Stud. 1994, 63: 123–145.MathSciNetGoogle Scholar
  18. Chadli O, Chbani Z, Riahi H: Equilibrium problems with generalized monotone bifunctions and applications to variational inequalities. J. Optim. Theory Appl. 2000, 105(2):299–323. 10.1023/A:1004657817758MathSciNetView ArticleGoogle Scholar
  19. Cabot A: Proximal point algorithm controlled by a slowly vanishing term: applications to hierarchical minimization. SIAM J. Optim. 2005, 15(2):555–572. 10.1137/S105262340343467XMathSciNetView ArticleGoogle Scholar
  20. Iusem AN, Sosa W: Iterative algorithms for equilibrium problems. Optimization 2003, 52: 301–316. 10.1080/0233193031000120039MathSciNetView ArticleGoogle Scholar
  21. Tada A, Takahashi W: Weak and strong convergence theorems for a nonexpansive mapping and an equilibrium problem. J. Optim. Theory Appl. 2007, 133: 359–370. 10.1007/s10957-007-9187-zMathSciNetView ArticleGoogle Scholar
  22. Yamada I, Ogura N: Hybrid steepest descent method for the variational inequality problem over the fixed point set of certain quasi-nonexpansive mappings. Numer. Funct. Anal. Optim. 2004, 25(7–8):619–655.MathSciNetView ArticleGoogle Scholar
  23. Nilsrakoo W, Saejung S: Weak and strong convergence theorems for countable Lipschitzian mappings and its applications. Nonlinear Anal. 2008, 69: 2695–2708. 10.1016/j.na.2007.08.044MathSciNetView ArticleGoogle Scholar
  24. Solodov MV, Svaiter BF: Error bounds for proximal point subproblems and associated inexact proximal point algorithms. Math. Program., Ser. B 2000, 88: 371–389. 10.1007/s101070050022MathSciNetView ArticleGoogle Scholar
  25. Liu LS: Ishikawa and Mann iteration process with errors for nonlinear strongly accretive mappings in Banach spaces. J. Math. Anal. Appl. 1995, 194: 114–125. 10.1006/jmaa.1995.1289MathSciNetView ArticleGoogle Scholar
  26. Xu HK: An iterative approach to quadratic optimization. J. Optim. Theory Appl. 2003, 116: 659–678. 10.1023/A:1023073621589MathSciNetView ArticleGoogle Scholar
  27. Xu HK: Iterative algorithms for nonlinear operators. J. Lond. Math. Soc. 2002, 66: 240–256. 10.1112/S0024610702003332View ArticleGoogle Scholar
  28. Reich S: Constructive techniques for accretive and monotone operators. In Applied Nonlinear Analysis. Academic Press, New York; 1979:335–345.Google Scholar
  29. Mainge P-E: Strong convergence of projected subgradient methods for nonsmooth and nonstrictly convex minimization. Set-Valued Anal. 2008, 16: 899–912. 10.1007/s11228-008-0102-zMathSciNetView ArticleGoogle Scholar
  30. Song Y: Equivalent theorems of the convergence between proximal-type algorithms. Nonlinear Anal. 2009, 71: 293–300. 10.1016/j.na.2008.10.067MathSciNetView ArticleGoogle Scholar

Copyright

© Song and Zhang; licensee Springer 2012

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/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Advertisement