Motivated and inspired by Korpelevich's and Noor's extragradient methods, we suggest an extragradient method by using the sunny nonexpansive
retraction which has strong convergence for solving the generalized variational inequalities in Banach spaces.
1. Introduction
In the present paper, we focus on the following generalized variational inequality:Findx*∈Csuchthat〈Ax*,J(x-x*)〉≥0,∀x∈C,
where C is a nonempty closed convex subset of a real Banach space E, A:C→E is a nonlinear mapping, and J:E→2E* is the normalized duality mapping defined byJ(x)={f∈E*:〈x,f〉=‖x‖2,‖f‖=‖x‖},∀x∈E.
We use S(C,A) to denote the solution set of (1.1). It is clear that (1.1) is reduced to the following variational inequality in Hilbert spaces:Findx*∈Csuchthat〈Ax*,x-x*〉≥0,∀x∈C,
which was introduced and studied by Stampacchia [1]. Variational inequalities are being used as mathematical programming tools and models to study a wide class of unrelated problems arising in mathematical, physical, regional, engineering, and nonlinear optimization sciences. See, for instance, [2–23]. In order to solve (1.3), especially, Korpelevich [24] introduced the following well-known extragradient method:yn=PC(I-λA)xn,xn+1=PC(xn-λAyn),n≥0,
where PC is the metric projection from Rn onto its subset C, λ∈(0,1/k) and A:C→Rn is a monotone operator. He showed that the sequence {xn} converges to some solution of the above variational inequality (1.3). Noor [10] further suggested and analyzed the following new iterative methods:xn+1=PC(yn-λAyn),yn=PC[xn-λAxn],n≥0,
which is known as the modified Noor's extragradient method. We would like to point out that this algorithm (1.5) is quite different from the method of Koperlevich. However, these two algorithms fail, in general, to converge strongly in the setting of infinite-dimensional Hilbert spaces.
The generalized variational inequality (1.1) was introduced by Aoyama et al. [25] which is connected with the fixed point problem for nonlinear mapping. For solving the aforementioned generalized variational inequality (1.1), Aoyama et al. [25] introduced an iterative algorithm:xn+1=αnxn+(1-αn)QC[xn-λnAxn],n≥0,
where QC is a sunny nonexpansive retraction from E onto C, and {αn}⊂(0,1) and {λn}⊂(0,∞) are two real number sequences. Aoyama et al. [25] obtained on the aforementioned method (1.6) for solving variational inequality (1.1). Motivated by (1.6), Yao and Maruster [26] presented a modification of (1.5):xn+1=βnxn+(1-βn)QC[(1-αn)(xn-λAxn)],n≥0.
Yao and Maruster [26] proved that (1.7) converges strongly to the solution of the generalized variational inequality (1.1). Yao et al. [27] further considered the following extended extragradient method for solving (1.1):yn=QC[xn-λnAxn],xn+1=αnu+βnxn+γnQC[yn-λAyn],n≥0.
In this paper, motivated and inspired by Korpelevich's and Noor's extragradient methods, (1.7) and (1.8), we suggest a modified Noor's extragradient method via the sunny nonexpansive retraction for solving the variational inequalities (1.1) in Banach spaces.
2. Preliminaries
Let C be a nonempty closed convex subset of a real Banach space E. Recall that a mapping A of C into E is said to be accretive if there exists j(x-y)∈J(x-y) such that〈Ax-Ay,j(x-y)〉≥0,
for all x,y∈C. A mapping A of C into E is said to be α-strongly accretive if, for α>0,〈Ax-Ay,j(x-y)〉≥α‖x-y‖2,
for all x,y∈C. A mapping A of C into E is said to be α-inverse-strongly accretive if, for α>0,〈Ax-Ay,j(x-y)〉≥α‖Ax-Ay‖2,
for all x,y∈C.
Let U={x∈E:∥x∥=1}. A Banach space E is said to uniformly convex if, for each ϵ∈(0,2], there exists δ>0 such that for any x,y∈U,‖x-y‖≥ϵimplies‖x+y2‖≤1-δ.
It is known that a uniformly convex Banach space is reflexive and strictly convex. A Banach space E is said to be smooth if the limitlimt→0‖x+ty‖-‖x‖t
exists for all x,y∈U. It is also said to be uniformly smooth if the limit (2.5) is attained uniformly for x,y∈U. The norm of E is said to be Fréchet differentiable if, for each x∈U, the limit (2.5) is attained uniformly for y∈U. And we define a function ρ:[0,∞)→[0,∞) called the modulus of smoothness of E as follows:ρ(τ)=sup{12(‖x+y‖+‖x-y‖)-1:x,y∈X,‖x‖=1,‖y‖=τ}.
It is known that E is uniformly smooth if and only if limτ→0(ρ(τ)/τ)=0. Let q be a fixed real number with 1<q≤2. Then a Banach space E is said to be q-uniformly smooth if there exists a constant c>0 such that ρ(τ)≤cτq for all τ>0.
We need the following lemmas for proof of our main results.
Lemma 2.1 (see [28]).
Let q be a given real number with 1<q≤2 and let E be a q-uniformly smooth Banach space. Then
‖x+y‖q≤‖x‖q+q〈y,Jq(x)〉+2‖Ky‖q
for all x,y∈E, where K is the q-uniformly smoothness constant of E and Jq is the generalized duality mapping from E into 2E* defined by
Jq(x)={f∈E*:〈x,f〉=‖x‖q,‖f‖=‖x‖q-1},∀x∈E.
Let D be a subset of C and let Q be a mapping of C into D. Then Q is said to be sunny ifQ(Qx+t(x-Qx))=Qx,
whenever Qx+t(x-Qx)∈C for x∈C and t≥0. A mapping Q of C into itself is called a retraction if Q2=Q. If a mapping Q of C into itself is a retraction, then Qz=z for every z∈R(Q), where R(Q) is the range of Q. A subset D of C is called a sunny nonexpansive retract of C if there exists a sunny nonexpansive retraction from C onto D. One knows the following lemma concerning sunny nonexpansive retraction.
Lemma 2.2 (see [29]).
Let C be a closed convex subset of a smooth Banach space E, let D be a nonempty subset of C, and let Q be a retraction from C onto D. Then Q is sunny and nonexpansive if and only if
〈u-Qu,j(y-Qu)〉≤0
for all u∈C and y∈D.
Remark 2.3.
(1) It is well known that if E is a Hilbert space, then a sunny nonexpansive retraction QC is coincident with the metric projection from E onto C.
(2) Let C be a nonempty closed convex subset of a uniformly convex and uniformly smooth Banach space E and let T be a nonexpansive mapping of C into itself with F(T)≠∅. Then the set F(T) is a sunny nonexpansive retract of C.
The following lemma characterized the set of solution of (1.1) by using sunny nonexpansive retractions.
Lemma 2.4 (see [25]).
Let C be a nonempty closed convex subset of a smooth Banach space X. Let QC be a sunny nonexpansive retraction from X onto C and let A be an accretive operator of C into X. Then for all λ>0,
S(C,A)=F(QC(I-λA)),
where S(C,A)={x*∈C:〈Ax*,J(x-x*)〉≥0,∀x∈C}.
Lemma 2.5 (see [30]).
Let C be a nonempty bounded closed convex subset of a uniformly convex Banach space E and let T be nonexpansive mapping of C into itself. If {xn} is a sequence of C such that xn→x weakly and xn-Txn→0 strongly, then x is a fixed point of T.
Lemma 2.6 (see [31]).
Assume that {an} is a sequence of nonnegative real numbers such that
an+1≤(1-γn)an+δn,n≥0,
where {γn} is a sequence in (0,1) and {δn} is a sequence in R such that
∑n=0∞γn=∞;
limsupn→∞(δn/γn)≤0 or ∑n=0∞|δn|<∞.
Then limn→∞an=0.
3. Main Results
In this section, we will state and prove our main result.
Theorem 3.1.
Let E be a uniformly convex and 2-uniformly smooth Banach space and let C be a nonempty closed convex subset of E. Let QC be a sunny nonexpansive retraction from E onto C and let A:C→E be an α-strongly accretive and L-Lipschitz continuous mapping with S(C,A)≠∅. For given x0∈C, let the sequence {xn} be generated iteratively by
yn=QC[xn-λAxn],xn+1=QC[(1-αn)(yn-λAyn)],n≥0,
where {αn} and {βn} are two sequences in (0,1) and λ is a constant in [a,b] for some a,b with 0<a<b<α/K2L2. Assume that the following conditions hold:
limn→∞αn=0 and ∑n=1∞αn=∞;
limn→∞(αn+1/αn)=1.
Then {xn} defined by (3.1) converges strongly to Q′(0), where Q′ is a sunny nonexpansive retraction of E onto S(C,A).
Proof.
First, we note that A must be α/L2-inverse-strongly accretive mapping. Take p∈S(C,A). By using Lemmas 2.1 and 2.4, we easily obtain the following facts.
p=QC[p-λAp] for all λ>0; in particular,
p=QC[p-λ(1-αn)Ap]=QC[αnp+(1-αn)(p-λAp)],n≥0.
If λ∈(0,α/K2L2], then I-λA is nonexpansive and for all x,y∈C‖(I-λA)x-(I-λA)y‖2≤‖x-y‖2+2λ(K2λ-αL2)‖Ax-Ay‖2.
Indeed, from Lemma 2.1, we have
‖(I-λA)x-(I-λA)y‖2=‖(x-y)-λ(Ax-Ay)‖2≤‖x-y‖2-2λ〈Ax-Ay,j(x-y)〉+2K2λ2‖Ax-Ay‖2≤‖x-y‖2-2λαL2‖Ax-Ay‖2+2K2λ2‖Ax-Ay‖2=‖x-y‖2+2λ(K2λ-αL2)‖Ax-Ay‖2.
From (3.1), we have
‖yn-p‖=‖QC(xn-λAxn)-QC(p-λAp)‖≤‖(xn-λAxn)-(p-λAp)‖≤‖xn-p‖.
By (3.1) and (3.5), we have
‖xn+1-p‖=‖QC[(1-αn)(yn-λAyn)]-QC[αnp+(1-αn)(p-λAp)]‖≤‖[(1-αn)(yn-λAyn)]-[αnp+(1-αn)(p-λAp)]‖≤αn‖p‖+(1-αn)‖(yn-λAyn)-(p-λAp)‖≤αn‖p‖+(1-αn)‖yn-p‖≤αn‖p‖+(1-αn)‖xn-p‖≤max{‖p‖,‖x0-p‖}.
Therefore, {xn} is bounded. We observe that
‖yn-yn-1‖=‖QC[xn-λAxn]-QC[xn-1-λAxn-1]‖≤‖(xn-λAxn)-(xn-1-λAxn-1)‖≤‖xn-xn-1‖,
and hence
‖xn+1-xn‖=‖QC[(1-αn)(yn-λAyn)]-QC[(1-αn-1)(yn-1-λAyn-1)]‖≤‖[(1-αn)(yn-λAyn)]-[(1-αn-1)(yn-1-λAyn-1)]‖=‖(1-αn)[(yn-λAyn)-(yn-1-λAyn-1)]+(αn-1-αn)(yn-1-λAyn-1)‖≤(1-αn)‖(yn-λAyn)-(yn-1-λAyn-1)‖+|αn-αn-1|‖yn-1-λAyn-1‖≤(1-αn)‖yn-yn-1‖+|αn-αn-1|‖yn-1-λAyn-1‖≤(1-αn)‖xn-xn-1‖+|αn-αn-1|‖yn-1-λAyn-1‖.
By Lemma 2.6, we obtain
limn→∞‖xn+1-xn‖=0.
From (3.1), we also have
‖yn-p‖2=‖QC[xn-λAxn]-QC[p-λAp]‖2≤‖(xn-λAxn)-(p-λAp)‖2≤‖xn-p‖2+2λ(K2λ-αL2)‖Axn-Ap‖2.
By (3.1) and (3.10), we obtain
‖xn+1-p‖2≤‖αn(-p)+(1-αn)[(yn-λAyn)-(p-λAp)]‖2≤αn‖p‖2+(1-αn)‖(yn-λAyn)-(p-λAp)‖2≤αn‖p‖2+(1-αn)[‖yn-p‖2+2λ(K2λ-αL2)‖Ayn-Ap‖2]≤αn‖p‖2+(1-αn)[‖xn-p‖2+2λ(K2λ-αL2)‖Axn-Ap‖2]+2(1-αn)λ(K2λ-αL2)‖Ayn-Ap‖2≤αn‖p‖2+‖xn-p‖2+2(1-αn)λ(K2λ-αL2)‖Axn-Ap‖2+2(1-αn)λ(K2λ-αL2)‖Ayn-Ap‖2.
Therefore, we have
0≤-2(1-αn)λ(K2λ-αL2)‖Axn-Ap‖2-2(1-αn)λ(K2λ-αL2)‖Ayn-Ap‖2≤αn‖p‖2+‖xn-p‖2-‖xn+1-p‖2=αn‖p‖2+(‖xn-p‖+‖xn+1-p‖)(‖xn-p‖-‖xn+1-p‖)≤αn‖p‖2+(‖xn-p‖+‖xn+1-p‖)‖xn-xn+1‖.
Since liminfn→∞2(1-αn)λ(K2λ-α/L2)>0, αn→0 and ∥xn-xn+1∥→0, we obtain
limn→∞‖Axn-Ap‖=limn→∞‖Ayn-Ap‖=0.
It follows that
limn→∞‖Ayn-Axn‖=0.
Since A is α-strongly accretive, we deduce
‖Ayn-Axn‖≥α‖yn-xn‖,
which implies that
limn→∞‖yn-xn‖=0,
that is,
limn→∞‖QC(xn-λAxn)-xn‖=0.
Next, we show that
limsupn→∞〈Q′(0),j(xn-Q′(0))〉≥0.
To show (3.18), since {xn} is bounded, we can choose that a sequence {xni} of {xn} converges weakly to z such that
limsupn→∞〈Q′(0),j(xn-Q′(0))〉=limsupi→∞〈Q′(0),j(xni-Q′(0))〉.
We first prove z∈S(C,A). It follows that
limi→∞‖QC(I-λA)xni-xni‖=0.
By Lemma 2.5 and (3.20), we have z∈F(QC(I-λA)); it follows from Lemma 2.4 that z∈S(C,A).
Now, from (3.19) and Lemma 2.2, we have
limsupn→∞〈Q′(0),j(xn-Q′(0))〉=limsupi→∞〈Q′(0),j(xni-Q′(0))〉=〈Q′(0),j(z-Q′(0))〉≥0.
Since xn+1=QC[(1-αn)(yn-λAyn)] and x*=QC[αnx*+(1-αn)(x*-λAx*)] for all n≥0, we can deduce from Lemma 2.2 that
〈QC[(1-αn)(yn-λnAyn)]-[(1-αn)(yn-λnAyn)],j(xn+1-x*)〉≤0,〈[αnx*+(1-αn)(x*-λnAx*)]-QC[αnx*+(1-αn)(x*-λnAx*)],j(xn+1-x*)〉≤0.
Therefore, we have
‖xn+1-x*‖2=〈QC[(1-αn)(yn-λnAyn)]-QC[αnx*+(1-αn)(x*-λnAx*)],j(xn+1-x*)〉=〈QC[(1-αn)(yn-λnAyn)]-[(1-αn)(yn-λnAyn)],j(xn+1-x*)〉+〈[(1-αn)(yn-λnAyn)]-[αnx*+(1-αn)(x*-λnAx*)],j(xn+1-x*)〉+〈[αnx*+(1-αn)(x*-λnAx*)]-QC[αnx*+(1-αn)(x*-λnAx*)],j(xn+1-x*)〉≤〈(1-αn)(yn-λnAyn)-(1-αn)(x*-λnAx*)-αnx*,j(xn+1-x*)〉≤(1-αn)‖yn-x*‖‖xn+1-x*‖-αn〈x*,j(xn+1-x*)〉≤(1-αn)‖xn-x*‖‖xn+1-x*‖-αn〈x*,j(xn+1-x*)〉≤1-αn2(‖xn-x*‖2+‖xn+1-x*‖2)-αn〈x*,j(xn+1-x*)〉,
which implies that
‖xn+1-z‖2≤(1-αn)‖xn-z‖2+2αn〈-z,j(xn+1-z)〉.
Finally, by Lemma 2.6 and (3.24), we conclude that xn converges strongly to Q′(0). This completes the proof.
4. Conclusion
Variational inequality theory provides a simple, natural, and unified framework for a general treatment of unrelated problems. These activities have motivated to generalize and extend the variational inequalities and related optimization problems in several directions using new and novel techniques. A well-known method to solve the VI is the following gradient method:xn+1=PC(xn-αnA(xn)),n≥0,
This method requires some monotonicity properties of A. However, we remark that there is no chance of relaxing the assumption on A to plain monotonicity. To overcome this weakness of the method, Korpelevich proposed a so-called Korpelevich's method which has been extensively extended and studied. Noor [10] especially, suggested another method referred as Noor's method which is different from Korpelevich's method. However, these two algorithms fail, in general, to converge strongly in the setting of infinite-dimensional Hilbert spaces. In the present paper, we suggested a modified Noor's method which has strong convergence in Banach spaces. We hope that the ideas and technique of this paper may stimulate further research in this field.
Acknowledgment
The authors thank the referees for useful comments and suggestions.
StampacchiaG.Formes bilineaires coercitives sur les ensembles convexes1964258441344160166591ZBL0124.06401CensorY.IusemA. N.ZeniosS. A.An interior point method with Bregman functions for the variational inequality problem with paramonotone operators199881337340010.1007/BF015800891617732ZBL0919.90123GlowinskiR.1984New York, NY, USASpringer-Verlag737005IusemA. N.An iterative algorithm for the variational inequality problem19941321031141361588ZBL0811.65049JailletP.LambertonD.LapeyreB.Variational inequalities and the pricing of American options199021326328910.1007/BF000472111096582ZBL0714.90004NoorM. A.1975London, UKBrunel UniversityNoorM. A.General variational inequalities19881211912210.1016/0893-9659(88)90054-7953368ZBL0655.49005NoorM. A.Wiener-Hopf equations and variational inequalities199379119720610.1007/BF00941894124650310.1007/BF00941894ZBL0799.49010NoorM. A.New approximation schemes for general variational inequalities20002511217229179040610.1006/jmaa.2000.7042ZBL0964.49007NoorM. A.A class of new iterative methods for general mixed variational inequalities20003113111910.1016/S0895-7177(00)00108-41773134ZBL0953.49016NoorM. A.Some developments in general variational inequalities20041521199277205006310.1016/S0096-3003(03)00558-7ZBL1134.49304NoorM. A.Extended general variational inequalities200922218218610.1016/j.aml.2008.03.0072482273ZBL1163.49303NoorM. A.Some aspects of extended general variational inequalities201220121610.1155/2012/303569303569NoorM. A.Al-SaidE. A.Wiener-Hopf equations technique for quasimonotone variational inequalities19991033705714172724610.1023/A:1021796326831ZBL0953.65050Aslam NoorM.HuangZ.Wiener-Hopf equation technique for variational inequalities and nonexpansive mappings20071912504510238555210.1016/j.amc.2007.02.117ZBL1193.49009NoorM. A.NoorK. I.Al-SaidE.Iterative methods for solving nonconvex equilibrium variational inequalities2012616569XuH. K.KimT. H.Convergence of hybrid steepest-descent methods for variational inequalities20031191185201202844510.1023/B:JOTA.0000005048.79379.b6ZBL1045.49018YamadaI.ButnariuD.CensorY.ReichS.The hybrid steepest descent method for the variational inequality problem over the intersection of fixed point sets of nonexpansive mappings20018Amsterdam, HollandNorth-Holland4735041853237ZBL1013.49005YaoJ. C.Variational inequalities with generalized monotone operators1994193691705128889410.1287/moor.19.3.691ZBL0813.49010YaoY.LiouY.-C.KangS. M.Two-step projection methods for a system of variational inequality problems in Banach spacesJournal of Global Optimization. In press10.1007/s10898-011-9804-0YaoY.NoorM. A.LiouY.-CStrong convergence of a modified extra-gradient method to the minimum-norm solution of variational inequalities20122012910.1155/2012/817436817436YaoY.NoorM. A.LiouY.-CIterative algorithms for general multi-valued variational inequalities201220121010.1155/2012/768272768272ZeidlerE.1985New York, NY, USASpringer-Verlag768749KorpelevichG. M.An extragradient method for finding saddle points and for other problems19761247477560451121AoyamaK.IidukaH.TakahashiW.Weak convergence of an iterative sequence for accretive operators in Banach spaces200620061310.1155/FPTA/2006/35390353902235489ZBL1128.47056YaoY.MarusterS.Strong convergence of an iterative algorithm for variational inequalities in Banach spaces2011541-2325329280188910.1016/j.mcm.2011.02.016ZBL1225.65067YaoY.LiouY. C.LiC. L.LinH. T.Extended extragradient methods for generalized variational inequalities201220121410.1155/2012/237083237083XuH. K.Inequalities in Banach spaces with applications199116121127113810.1016/0362-546X(91)90200-K1111623ZBL0757.46033BruckR. E.Jr.Nonexpansive retracts of Banach spaces197076384386025613510.1090/S0002-9904-1970-12486-7ZBL0224.47034BrowderF. E.Nonlinear operators and nonlinear equations of evolution in Banach spaces1976Providence, RI, USAAmerican Mathematical Society13080405188ZBL0327.47022XuH. K.Iterative algorithms for nonlinear operators2002661240256191187210.1112/S0024610702003332ZBL1013.47032