Volume 2011, Article ID 973028,23pages doi:10.1155/2011/973028
Research Article
Hybrid Proximal-Type Algorithms for Generalized Equilibrium Problems, Maximal Monotone
Operators, and Relatively Nonexpansive Mappings
Lu-Chuan Zeng,
1, 2Q. H. Ansari,
3and S. Al-Homidan
31Department of Mathematics, Shanghai Normal University, Shanghai 200234, China
2Scientific Computing Key Laboratory of Shanghai Universities, Shanghai 200234, China
3Department of Mathematics and Statistics, King Fahd University of Petroleum & Minerals (KFUPM), P.O. Box 119, Dhahran 31261, Saudi Arabia
Correspondence should be addressed to S. Al-Homidan,homidan@kfupm.edu.sa Received 24 September 2010; Accepted 18 October 2010
Academic Editor: Jen Chih Yao
Copyrightq2011 Lu-Chuan Zeng et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The purpose of this paper is to introduce and consider new hybrid proximal-type algorithms for finding a common element of the set EP of solutions of a generalized equilibrium problem, the setFSof fixed points of a relatively nonexpansive mappingS, and the setT−10 of zeros of a maximal monotone operatorTin a uniformly smooth and uniformly convex Banach space. Strong convergence theorems for these hybrid proximal-type algorithms are established; that is, under appropriate conditions, the sequences generated by these various algorithms converge strongly to the same point in EP∩FS∩T−10. These new results represent the improvement, generalization, and development of the previously known ones in the literature.
1. Introduction
LetEbe a real Banach space with the dualE∗andCbe a nonempty closed convex subset of E. We denote byNandRthe sets of positive integers and real numbers, respectively. Also, we denote byJthe normalized duality mapping fromEto 2E∗defined by
Jx
x∗∈E∗:x, x∗x2x∗2
, ∀x∈E, 1.1
where ·,· denotes the generalized duality pairing. Recall that if E is smooth, then J is single valued and ifEis uniformly smooth, thenJ is uniformly norm-to-norm continuous on bounded subsets ofE. We will still denote byJ the single valued duality mapping. Let
f : C×C → Rbe a bifunction andA: C → E∗be a nonlinear mapping. We consider the following generalized equilibrium problem:
findu∈Csuch thatf u, y
Au, y−u
≥0, ∀y∈C. 1.2
The set of suchu∈Cis denoted by EP, that is, EP
u∈C:f u, y
Au, y−u
≥0, ∀y∈C . 1.3 WheneverEHa Hilbert space, problem1.2was introduced and studied by S. Takahashi and W. Takahashi 1. Similar problems have been studied extensively recently. See, for example,2–11. In the case ofA ≡ 0,EP is denoted by EPf. In the case of f ≡ 0, EP is also denoted by VIC, A. The problem1.2is very general in the sense that it includes, as special cases, optimization problems, variational inequalities, minimax problems, the Nash equilibrium problem in noncooperative games and others; see, for example, 12–14. A mappingS:C → Eis called nonexpansive ifSx−Sy ≤ x−yfor allx, y∈C. Denote by FSthe set of fixed points ofS, that is,FS {x∈C:Sxx}. A mappingA:C → E∗is calledα-inverse-strongly monotone, if there exists anα >0 such that
Ax−Ay, x−y
≥αAx−Ay2, ∀x, y∈C. 1.4 It is easy to see that ifA : C → E∗is anα-inverse-strongly monotone mapping, then it is 1/α-Lipschitzian.
LetEbe a real Banach space with the dualE∗. A multivalued operatorT :E → 2E∗ with domainDT {z ∈ E : Tz /∅}is called monotone ifx1−x2, y1 −y2 ≥0 for each xi ∈ DT and yi ∈ Txi, i 1,2. A monotone operator T is called maximal if its graph GT {x, y : y ∈ Tx} is not properly contained in the graph of any other monotone operator. A method for solving the inclusion 0∈Txis the proximal point algorithm. Denote byIthe identity operator onEHa Hilbert space. The proximal point algorithm generates, for any initial pointx0x∈H, a sequence{xn}inH, by the iterative scheme
xn1 IrnT−1xn, n0,1,2, . . . , 1.5 where{rn}is a sequence in the interval0,∞. Note that this iteration is equivalent to
0∈Txn1 1
rnxn1−xn, n0,1,2, . . . . 1.6 This algorithm was first introduced by Martinet12and generally studied by Rockafellar 15in the framework of a Hilbert space. Later many authors studied its convergence in a Hilbert space or a Banach space. See, for instance,16–21and the references therein.
LetEbe a reflexive, strictly convex, and smooth Banach space with the dualE∗andC be a nonempty closed convex subset ofE. LetT :E → 2E∗ be a maximal monotone operator with domainDT CandS :C → Cbe a relatively nonexpansive mapping. LetA:C → X∗be anα-inverse-strongly monotone mapping andf:C×C → Rbe a bifunction satisfying
A1–A4:A1 fx, x 0, ∀x ∈ C; A2 f is monotone, that is, fx, y fy, x ≤ 0,
∀x, y ∈ C;A3limsupt↓0fxtz−x, y ≤ fx, y,∀x, y, z ∈ C;A4 the functiony → fx, yis convex and lower semicontinuous. The purpose of this paper is to introduce and investigate two new hybrid proximal-type Algorithms1.1and1.2for finding an element of EP∩ FS∩ T−10.
Algorithm 1.1.
x0∈Carbitrarily chosen, 0vn 1
rnJxn−Jxn, vn∈Txn, znJ−1
βnJxn 1−βn
JSxn , ynJ−1αnJxn 1−αnJSzn,
un∈Csuch that f
un, y
Aun, y−un 1
rn
y−un, Jun−Jyn
≥0, ∀y∈C, Hn
v∈C:φv, un≤αnφv,xn 1−αnφv, zn, v−xn, vn ≤0 , Wn{v∈C:v−xn, Jx0−Jxn ≤0},
xn1 ΠHn∩Wnx0, n0,1,2, . . . ,
1.7
where{rn}∞n0is a sequence in0,∞and{αn}∞n0,{βn}∞n0are sequences in0,1.
Algorithm 1.2.
x0∈Carbitrarily chosen, 0vn 1
rnJxn−Jxn, vn∈Txn, ynJ−1αnJx0 1−αnJSxn,
un∈Csuch that f
un, y
Aun, y−un 1
rn
y−un, Jun−Jyn
≥0, ∀y∈C, Hn
v∈C:φv, un≤αnφv, x0 1−αnφv,xn, v−xn, vn ≤0 , Wn{v∈C:v−xn, Jx0−Jxn ≤0},
xn1 ΠHn∩Wnx0, n0,1,2, . . . ,
1.8
where{rn}∞n0is a sequence in0,∞and{αn}∞n0is a sequence in0,1.
In this paper, strong convergence results on these two hybrid proximal-type algorithms are established; that is, under appropriate conditions, the sequence{xn}generated byAlgorithm 1.1and the sequence{xn}generated byAlgorithm 1.2, converge strongly to the same pointΠEP∩FS∩T−10x0. These new results represent the improvement, generalization and development of the previously known ones in the literature including Solodov and Svaiter 22, Kamimura and Takahashi23, Qin and Su24, and Ceng et al.25.
Throughout this paper the symbolstands for weak convergence and → stands for strong convergence.
2. Preliminaries
Let E be a real Banach space with the dual E∗. We denote by J the normalized duality mapping fromEto 2E∗defined by
Jx
x∗∈E∗:x, x∗x2x∗2
, ∀x∈X, 2.1
where·,·denotes the generalized duality pairing. A Banach spaceEis called strictly convex ifxy/2 < 1 for allx, y ∈ Ewithx y 1 andx /y. It is said to be uniformly convex ifxn−yn → 0 for any two sequences{xn},{yn} ⊂Esuch thatxn yn 1 and limn→ ∞xnyn/21. LetU{x∈E:x1}be a unit sphere ofE, then the Banach spaceEis called smooth if
limt→0
xty− x
t 2.2
exists for each x, y ∈ U. IfEis smooth, thenJ is single valued. We still denote the single valued duality mapping byJ.
It is also said to be uniformly smooth if the limit is attained uniformly forx, y ∈ U.
Recall also that ifEis uniformly smooth, then Jis uniformly norm-to-norm continuous on bounded subsets ofE. A Banach spaceEis said to have the Kadec-Klee property if for any sequence{xn} ⊂E, wheneverxn x∈Eandxn → x, we havexn → x. It is known that ifEis uniformly convex, thenEhas the Kadec-Klee property; see26,27for more details.
LetCbe a nonempty closed convex subset of a real Hilbert spaceHandPC :H → C be the metric projection ofHonto C, thenPCis nonexpansive. This fact actually characterizes Hilbert spaces and hence, it is not available in more general Banach spaces. Nevertheless, Alber 28 recently introduced a generalized projection operator ΠC in a Banach space E which is an analogue of the metric projection in Hilbert spaces.
Next, we assume thatEis a smooth Banach space. Consider the functional defined as in28,29by
φ x, y
x2−2x, Jyy2, ∀x, y∈E. 2.3
It is clear that in a Hilbert spaceH,2.3reduces toφx, y x−y2, for allx, y∈H.
The generalized projectionΠC:E → Cis a mapping that assigns to an arbitrary point x∈Ethe minimum point of the functionalφy, x; that is,ΠCxx, wherexis the solution to the minimization problem
φx, x min
y∈C φ y, x
. 2.4
The existence and uniqueness of the operatorΠCfollows from the properties of the functional φx, yand strict monotonicity of the mappingJsee, e.g.,30. In a Hilbert spaceH,ΠC PC. From28, in uniformly smooth and uniformly convex Banach spaces, we have
x −y2≤φ x, y
≤
xy2, ∀x, y∈E. 2.5 LetCbe a nonempty closed convex subset ofE, and letSbe a mapping fromCinto itself.
A pointp∈Cis called an asymptotically fixed point ofS31ifCcontains a sequence{xn} which converges weakly topsuch thatSxn−xn → 0. The set of asymptotical fixed points of Swill be denoted byFS. A mapping SfromCinto itself is called relatively nonexpansive 32–34ifFS FSandφp, Sx≤φp, xfor allx∈Candp∈FS.
We remark that ifEis a reflexive, strictly convex and smooth Banach space, then for anyx, y ∈E, φx, y 0 if and only ifx y. It is sufficient to show that ifφx, y 0 then x y. From2.5, we havex y. This implies that x, Jy x2 y2. From the definition ofJ, we haveJxJy. Therefore, we havexy; see26,27for more details.
We need the following lemmas for the proof of our main results.
Lemma 2.1see23. LetEbe a smooth and uniformly convex Banach space and let{xn}and{yn} be two sequences ofE. Ifφxn, yn → 0 and either{xn}or{yn}is bounded, thenxn−yn → 0.
Lemma 2.2see23,28. LetCbe a nonempty closed convex subset of a smooth, strictly convex and reflexive Banach spaceE, letx∈Eand letz∈C, then
z ΠCx⇐⇒
y−z, Jx−Jz
≤0, ∀y∈C. 2.6
Lemma 2.3see23,28. LetCbe a nonempty closed convex subset of a smooth, strictly convex and reflexive Banach spaceE, then
φ
x,ΠCy φ
ΠCy, y
≤φ x, y
, ∀x∈C, y∈E. 2.7
Lemma 2.4see35. LetCbe a nonempty closed convex subset of a reflexive, strictly convex and smooth Banach spaceE, and letS:C → Cbe a relatively nonexpansive mapping, thenFSis closed and convex.
The following result is according to Blum and Oettli36.
Lemma 2.5see36. LetCbe a nonempty closed convex subset of a smooth, strictly convex and reflexive Banach spaceE, letf be a bifunction fromC×CtoRsatisfying (A1)–(A4), and letr >0 andx∈E, then, there existsz∈Csuch that
f z, y
1 r
y−z, Jz−Jx
≥0, ∀y∈C. 2.8
Motivated by Combettes and Hirstoaga 37 in a Hilbert space, Takahashi and Zembayashi38established the following lemma.
Lemma 2.6 see38. LetCbe a nonempty closed convex subset of a uniformly smooth, strictly convex and reflexive Banach spaceE, and letfbe a bifunction fromC×CtoRsatisfying (A1)–(A4).
Forr >0 andx∈E, define a mappingTr :E → Cas follows:
Trx
z∈C:f z, y
1 r
y−z, Jz−Jx
≥0, ∀y∈C
2.9
for allx∈E, then, the following hold:
iTr is single valued;
iiTr is a firmly nonexpansive-type mapping, that is, for allx, y∈E,
Trx−Try, JTrx−JTry
≤
Trx−Try, Jx−Jy
; 2.10
iiiFTr FT r EPf;
ivEPfis closed and convex.
UsingLemma 2.6, one has the following result.
Lemma 2.7see38. LetCbe a nonempty closed convex subset of a smooth, strictly convex and reflexive Banach spaceE, letfbe a bifunction fromC×CtoRsatisfying (A1)–(A4), and letr >0, then, forx∈Eandq∈FTr,
φ q, Trx
φTrx, x≤φ q, x
. 2.11
Utilizing Lemmas2.5,2.6and2.7as above, Chang39derived the following result.
Proposition 2.8 see39, Lemma 2.5. LetEbe a smooth, strictly convex and reflexive Banach space and Cbe a nonempty closed convex subset ofE. LetA : C → E∗ be anα-inverse-strongly monotone mapping, letf be a bifunction fromC×CtoRsatisfying (A1)–(A4), and letr >0, then there hold the following:
Iforx∈E, there existsu∈Csuch that
f u, y
Au, y−u 1
r
y−u, Ju−Jx
≥0, ∀y∈C; 2.12
IIifEis additionally uniformly smooth andKr :E → Cis defined as
Krx
u∈C:f u, y
Au, y−u 1
r
y−u, Ju−Jx
≥0, ∀y∈C
, ∀x∈E, 2.13
then the mappingKrhas the following properties:
iKris single valued,
iiKris a firmly nonexpansive-type mapping, that is, Krx−Kry, JKrx−JKry
≤
Krx−Kry, Jx−Jy
, ∀x, y∈E, 2.14
iiiFKr FK r EP,
ivEP is a closed convex subset ofC,
vφp, Krx φKrx, x≤φp, x, for allp∈FKr. Proof. Define a bifunctionF:C×C → Ras follows:
F x, y
f x, y
Ax, y−x
, ∀x, y∈C. 2.15
Then it is easy to verify thatFsatisfies the conditionsA1–A4. Therefore, The conclusions IandIIofProposition 2.8follow immediately from Lemmas2.5,2.6and2.7.
Lemma 2.9see13,14. LetE be a reflexive, strictly convex and smooth Banach space, and let T :E → 2E∗be a maximal monotone operator withT−10/∅, then,
φz, Jrx φJrx, x≤φz, x, ∀r >0, z∈T−10, x∈E. 2.16
3. Main Results
Throughout this section, unless otherwise stated, we assume thatT :E → 2E∗is a maximal monotone operator with domainDT C,S:C → Cis a relatively nonexpansive mapping, A:C → E∗is anα-inverse-strongly monotone mapping andf :C×C → Ris a bifunction satisfying A1–A4, where C is a nonempty closed convex subset of a reflexive, strictly convex, and smooth Banach spaceE. In this section, we study the following algorithm.
Algorithm 3.1.
x0∈Carbitrarily chosen, 0vn 1
rnJxn−Jxn, vn∈Txn, znJ−1
βnJxn 1−βn
JSxn , ynJ−1αnJxn 1−αnJSzn,
un∈Csuch that f
un, y
Aun, y−un 1
rn
y−un, Jun−Jyn
≥0, ∀y∈C, Hn
v∈C:φv, un≤αnφv,xn 1−αnφv, zn, v−xn, vn ≤0 , Wn{v∈C:v−xn, Jx0−Jxn ≤0},
xn1 ΠHn∩Wnx0, n0,1,2, . . . ,
3.1
where{rn}∞n0is a sequence in0,∞and{αn}∞n0,{βn}∞n0are sequences in0,1.
First we investigate the condition under which the Algorithm 3.1 is well defined.
Rockafellar40proved the following result.
Lemma 3.2Rockafellar40. LetEbe a reflexive, strictly convex, and smooth Banach space and letT:E → 2E∗be a multivalued operator, then there hold the following:
iT−10 is closed and convex ifTis maximal monotone such thatT−10/∅;
iiT is maximal monotone if and only ifTis monotone withRJrT E∗for allr >0.
Utilizing this result, we can show the following lemma.
Lemma 3.3. LetEbe a reflexive, strictly convex, and smooth Banach space. If EP∩FS∩T−10/∅, then the sequence{xn}generated byAlgorithm 3.1is well defined.
Proof. For eachn≥0, define two setsCnandDnas follows:
Cn
v∈C:φv, un≤αnφv,xn 1−αnφv, zn ,
Dn{v∈C:v−xn, vn ≤0}. 3.2
It is obvious thatCnis closed andDn, Wnare closed convex sets for eachn≥0. Let us show thatCnis convex. Forv1, v2∈Cnandt∈0,1, putvtv1 1−tv2. It is sufficient to show thatv∈Cn. Indeed, observe that
φv, un≤αnφv,xn 1−αnφv, zn 3.3
is equivalent to
2αnv, Jxn21−αnv, Jzn −2v, Jun ≤αnxn2 1−αnzn2− un2. 3.4
Note that there hold the following:
φv, un v2−2v, Junun2, φv,xn v2−2v, Jxnxn2, φv, zn v2−2v, Jznzn2,
3.5
Thus we have
2αnv, Jxn21−αnv, Jzn −2v, Jun
2αntv1 1−tv2, Jxn21−αntv1 1−tv2, Jzn
−2tv1 1−tv2, Jun
2tαnv1, Jxn21−tαnv2, Jxn21−αntv1, Jzn 21−αn1−tv2, Jzn −2tv1, Jun −21−tv2, Jun
≤αnxn2 1−αnzn2− un2.
3.6
This implies thatv ∈ Cn. Therefore, Cn is convex and henceHn Cn ∩ Dn is closed and convex.
On the other hand, letw ∈EP∩ FS∩ T−10 be arbitrarily chosen, thenw ∈EP, w∈ FSandw∈T−10. FromAlgorithm 3.1, it follows that
φw, un φ
w, Krnyn
≤φ w, yn φ
w, J−1αnJxn 1−αnJSzn
w2−2w, αnJxn 1−αnJSznαnJxn 1−αnJSzn2
≤ w2−2αnw, Jxn −21−αnw, JSznαnxn2 1−αnSzn2
≤αnφw,xn 1−αnφw, Szn
≤αnφw,xn 1−αnφw, zn.
3.7
Sow∈Cnfor alln≥0. Now, fromLemma 3.2it follows that there existsx0, v0∈E×E∗such that 0v01/r0Jx0−Jx0andv0∈Tx0. SinceTis monotone, it follows thatx0−w, v0 ≥0, which implies thatw∈D0and hencew∈H0. Furthermore, it is clear thatw∈W0C, then w ∈H0∩ W0, and thereforex1 ΠH0∩W0x0is well defined. Suppose thatw ∈Hn−1∩ Wn−1
andxnis well defined for somen≥1. Again byLemma 3.2, we deduce thatxn, vn∈E×E∗ such that 0 vn 1/rnJxn−Jxnand vn ∈ Txn, then from the monotonicity of T we
conclude thatxn−w, vn ≥0, which implies thatw∈Dnand hencew∈Hn. It follows from Lemma 2.4that
w−xn, Jx0−Jxnw−ΠHn−1∩Wn−1x0, Jx0−JΠHn−1∩Wn−1x0 ≤0, 3.8
which implies thatw∈Wn. Consequently,w∈Hn∩Wnand so EP∩FS∩T−10⊂Hn∩Wn. Thereforexn1 ΠHn∩Wnx0 is well defined, then, by induction, the sequence{xn}generated byAlgorithm 3.1, is well defined for each integern≥0.
Remark 3.4. From the above proof, we obtain that
EP∩ FS∩ T−10⊂Hn∩ Wn 3.9
for each integern≥0.
We are now in a position to prove the main theorem.
Theorem 3.5. LetEbe a uniformly smooth and uniformly convex Banach space. Let{rn}∞n0 be a sequence in0,∞and{αn}∞n0,{βn}∞n0be sequences in0,1such that
lim inf
n→ ∞ rn>0, lim sup
n→ ∞ αn<1, lim
n→ ∞βn1. 3.10
Let EP∩ FS∩ T−10/∅. If S is uniformly continuous, then the sequence {xn} generated by Algorithm 3.1converges strongly toΠEP∩FS∩T−10x0.
Proof. First of all, if follows from the definition of Wn that xn ΠWnx0. Since xn1 ΠHn∩Wnx0∈Wn, we have
φxn, x0≤φxn1, x0, ∀n≥0. 3.11
Thus{φxn, x0}is nondecreasing. Also fromxn ΠWnx0andLemma 2.3, we have that φxn, x0 φΠWnx0, x0≤φw, x0−φw, xn≤φw, x0 3.12
for eachw∈EP∩FS∩T−10⊂Wnand for eachn≥0. Consequently,{φxn, x0}is bounded.
Moreover, according to the inequality
xn − x02≤φxn, x0≤xnx02, 3.13
we conclude that {xn} is bounded. Thus, we have that limn→ ∞φxn, x0 exists. From Lemma 2.3, we derive the following:
φxn1, xn φxn1,ΠWnx0
≤φxn1, x0−φΠWnx0, x0 φxn1, x0−φxn, x0,
3.14
for alln≥0. This implies thatφxn1, xn → 0. So it follows fromLemma 2.1thatxn1−xn → 0. Sincexn1 ΠHn∩Wnx0∈Hn, from the definition ofHn, we also have
φxn1, un≤αnφxn1,xn 1−αnφxn1, zn, xn1−xn, vn ≤0. 3.15
Observe that φxn1, zn φ
xn1, J−1
βnJxn 1−βn
JSxn xn12−2
xn1, βnJxn 1−βn
JSxn
βnJxn 1−βn
JSxn2
≤ xn12−2βnxn1, Jxn −2 1−βn
xn1, JSxnβnxn2 1−βn
Sxn2 βnφxn1,xn
1−βn
φxn1, Sxn.
3.16
At the same time,
φΠHnxn, xn−φxn, xn ΠHnxn2− xn22xn−ΠHnxn, Jxn
≥2ΠHnxn−xn, Jxn2xn−ΠHnxn, Jxn 2xn−ΠHnxn, Jxn−Jxn.
3.17
SinceΠHnxn∈Hnandvn 1/rnJxn−Jxn, it follows that
xn−ΠHnxn, Jxn−Jxnrnxn−ΠHnxn, vn ≥0 3.18
and hence that φΠHnxn, xn ≥ φxn, xn. Further, from xn1 ∈ Hn, we have φxn1, xn ≥ φΠHnxn, xn, which yields
φxn1, xn≥φΠHnxn, xn≥φxn, xn. 3.19
Then it follows fromφxn1, xn → 0 thatφxn, xn → 0. Hence it follows fromLemma 2.1 thatxn−xn → 0. Since from3.15we derive
φxn1,xn−φxn, xn
xn12−2xn1, Jxnxn2−
xn2−2xn, Jxnxn2 xn12− xn2−2xn1, Jxn2xn, Jxn
xn12− xn2−2xn1−xn, Jxn−Jxn
−2xn1−xn, Jxn2xn, Jxn−Jxn
xn1 − xnxn1xn 2rnxn1−xn, vn −2xn1−xn, Jxn 2xn, Jxn−Jxn
≤ xn1−xnxn1xn 2xn1−xnxn2xnJxn−Jxn
≤ xn1−xnxn1xn 2xn1−xnxn−xnxn2xnJxn−Jxn, 3.20
we have
φxn1,xn≤φxn, xn xn1−xnxn1xn
2xn1−xnxn−xnxn2xnJxn−Jxn.
3.21
Thus, fromφxn, xn → 0,xn−xn → 0, andxn1−xn → 0, we know thatφxn1,xn → 0.
Consequently from3.16,φxn1,xn → 0, andβn → 1 it follows that
φxn1, zn−→0. 3.22
So it follows from3.15,φxn1,xn → 0, andφxn1, zn → 0 thatφxn1, un → 0. Utilizing Lemma 2.1we deduce that
nlim→ ∞xn1−un lim
n→ ∞xn1−xn lim
n→ ∞xn1−zn0. 3.23
Furthermore, foru∈EP∩FS∩ T−10 arbitrarily fixed, it follows fromProposition 2.8that φ
un, yn φ
Krnyn, yn
≤φ u, yn
−φ
u, Krnyn φ
u, J−1αnJxn 1−αnJSzn
−φu, un
u2−2u, αnJxn 1−αnJSznαnJxn 1−αnJSzn2−φu, un
≤ u2−2αnu, Jxn −21−αnu, JSznαnxn2 1−αnSzn2−φu, un αnφu,xn 1−αnφu, Szn−φu, un
≤1−αnφu, zn αnφu,xn−φu, un 1−αnφ
u, J−1
βnJxn 1−βn
JSxn
αnφu,xn−φu, un 1−αn
u2−2
u, βnJxn 1−βn
JSxn
βnJxn 1−βn
JSxn2 αnφu,xn−φu, un
≤1−αn
u2−2βnu, Jxn −2 1−βn
u, JSxnβnxn2 1−βn
Sxn2 αnφu,xn−φu, un
1−αn
βnφu,xn 1−βn
φu, Sxn
αnφu,xn−φu, un
≤1−αn
βnφu,xn 1−βn
φu,xn
αnφu,xn−φu, un 1−αnφu,xn αnφu,xn−φu, un
φu,xn−φu, un
φu,xn−φu, xn1 φu, xn1−φu, un
xn2− xn122u, Jxn1−Jxnxn12− un22u, Jun−Jxn1
≤ xn−xn1xnxn1 2uJxn1−Jxn xn1−unxn1un 2uJun−Jxn1.
3.24 SinceJis uniformly norm-to-norm continuous on bounded subsets ofE, it follows from3.23 thatJxn1−Jxn → 0 andJun−Jxn1 → 0, which hence yieldφun, yn → 0. Utilizing Lemma 2.1, we getun−yn → 0. Observe that
xn1−yn≤ xn1−unun−yn−→0, 3.25 due to3.23. Since J is uniformly norm-to-norm continuous on bounded subsets ofE, we have that
nlim→ ∞Jxn1−Jyn lim
n→ ∞Jxn1−Jxn0. 3.26
On the other hand, we have
xn−zn ≤ xn−xn1xn1−zn −→0. 3.27 Noticing that
Jxn1−JynJxn1−αnJxn 1−αnJSzn
αnJxn1−Jxn 1−αnJxn1−JSzn 1−αnJxn1−JSzn−αnJxn−Jxn1
≥1−αnJxn1−JSzn −αnJxn−Jxn1,
3.28
we have
Jxn1−JSzn ≤ 1
1−αnJxn1−JynαnJxn−Jxn1
. 3.29
From3.26and lim supn→ ∞αn<1, we obtain
nlim→ ∞Jxn1−JSzn0. 3.30
SinceJ−1is also uniformly norm-to-norm continuous on bounded subsets ofE∗, we obtain
nlim→ ∞xn1−Szn0. 3.31
Observe that
xn−Sxn ≤ xn−xn1xn1−SznSzn−Sxn. 3.32
SinceS is uniformly continuous, it follows from3.27,3.31andxn1−xn → 0 thatxn− Sxn → 0.
Now let us show thatωw{xn}⊂EP∩FS∩ T−10, where ωw{xn}:
x∈C:xnk xfor some subsequence{nk} ⊂ {n}withnk↑ ∞ . 3.33 Indeed, since {xn} is bounded and X is reflexive, we know that ωw{xn}/∅. Take x ∈ ωw{xn}arbitrarily, then there exists a subsequence{xnk}of{xn}such thatxnk x. Hence
x∈FS. Let us show thatx∈T−10. Sincexn−xn → 0, we have thatxnk x. Moreover, since Jis uniformly norm-to-norm continuous on bounded subsets ofEand lim infn→ ∞rn>0, we obtain
vn 1
rnJxn−Jxn−→0. 3.34
It follows fromvn∈Txnand the monotonicity ofTthat z−xn, z−vn
≥0 3.35
for allz∈DTandz∈Tz. This implies that z−x, z
≥0 3.36
for allz∈DTandz∈Tz. Thus from the maximality ofT, we infer thatx∈T−10. Therefore,
x∈FS∩ T−10. Further, let us show thatx∈EP. Sinceun−yn → 0 andxn−un → 0, from xnk xwe obtain thatynk xandunk x.
SinceJ is uniformly norm-to-norm continuous on bounded subsets ofE, fromun − yn → 0 we derive
nlim→ ∞Jun−Jyn0. 3.37
From lim infn→ ∞rn>0, it follows that
nlim→ ∞
Jun−Jyn
rn 0. 3.38
By the definition ofun :Krnyn, we have F
un, y 1
rn
y−un, Jun−Jyn
≥0, ∀y∈C, 3.39
where
F un, y
f un, y
Aun, y−un
. 3.40
Replacingnbynk, we have fromA2that 1
rnk
y−unk, Junk −Jynk
≥ −F unk, y
≥F y, unk
, ∀y∈C. 3.41
Sincey→fx, y Ax, y−xis convex and lower semicontinuous, it is also weakly lower semicontinuous. Lettingnk → ∞in the last inequality, from3.38andA4we have
F y,x
≤0, ∀y∈C. 3.42 Fort, with 0< t≤1, andy∈C, letytty 1−tx. Since y∈Candx∈C, thenyt∈Cand henceFyt,x ≤0. So, fromA1we have
0F yt, yt
≤tF yt, y
1−tF yt,x
≤tF yt, y
. 3.43
Dividing byt, we have
F yt, y
≥0, ∀y∈C. 3.44
Lettingt↓0, fromA3it follows that F x, y
≥0, ∀y∈C. 3.45
So,x∈EP. Therefore, we obtain thatωw{xn}⊂EP∩ FS∩ T−10 by the arbitrariness ofx.
Next, let us show thatωw{xn} {ΠEP∩FS∩T−10x0}andxn → ΠEP∩FS∩T−10x0. Indeed, putx ΠEP∩FS∩T−10x0. Fromxn1 ΠHn∩Wnx0andx∈EP∩ FS∩ T−10 ⊂ Hn∩Wn, we haveφxn1, x0≤φx, x0. Now from weakly lower semicontinuity of the norm, we derive for eachx∈ωw{xn}
φx, x 0 x 2−2x, x 0x02
≤lim inf
k→ ∞
xnk2−2xnk, x0x02
lim inf
k→ ∞ φxnk, x0
≤lim sup
k→ ∞ φxnk, x0
≤φx, x0.
3.46
It follows from the definition ofΠEP∩FS∩T−10x0thatxxand hence
klim→ ∞φxnk, x0 φx, x0. 3.47
So we have limk→ ∞xnk x. Utilizing the Kadec-Klee property ofE, we conclude that {xnk} converges strongly toΠEP∩FS∩T−10x0. Since{xnk} is an arbitrary weakly convergent subsequence of {xn}, we know that {xn} converges strongly to ΠEP∩FS∩T−10x0. This completes the proof.
Theorem 3.5 covers 25, Theorem 3.1 by taking C E, f ≡ 0 and A ≡ 0. Also Theorem 3.5covers24, Theorem 2.1by takingf ≡0,A≡0 andT≡0.
Theorem 3.6. LetCbe a nonempty closed convex subset of a uniformly smooth and uniformly convex Banach spaceE. LetT :E → 2E∗be a maximal monotone operator with domainDT C,S:C → Cbe a relatively nonexpansive mapping,A :C → E∗be anα-inverse-strongly monotone mapping andf:C×C → Rbe a bifunction satisfying (A1)–(A4). Assume that{rn}∞n0is a sequence in0,∞ satisfying lim infn→ ∞rn >0 and that{αn}∞n0is a sequences in0,1satisfying limn→ ∞αn0.
Define a sequence{xn}by the following algorithm.
Algorithm 3.7.
x0∈Carbitrarily chosen, 0vn 1
rnJxn−Jxn, vn∈Txn, ynJ−1αnJx0 1−αnJSxn,
un∈Csuch that f
un, y
Aun, y−un 1
rn
y−un, Jun−Jyn
≥0, ∀y∈C, Hn
v∈C:φv, un≤αnφv, x0 1−αnφv,xn, v−xn, vn ≤0 , Wn{v∈C:v−xn, Jx0−Jxn ≤0},
xn1 ΠHn∩Wnx0, n0,1,2, . . . ,
3.48
whereJis the single valued duality mapping onE. Let EP∩FS∩T−10/∅. IfSis uniformly continuous, then{xn}converges strongly toΠEP∩FS∩T−10x0.
Proof. For eachn≥0, define two setsCnandDnas follows:
Cn
v∈C:φv, un≤αnφv, x0 1−αnφv,xn , Dn{v∈C:v−xn, vn ≤0}.
3.49
It is obvious that Cn is closed and Dn, Wn are closed convex sets for each n ≥ 0. Let us show thatCn is convex and so Hn Cn∩ Dn is closed and convex. Similarly to the proof ofLemma 3.3, since
φv, un≤αnφv, x0 1−αnφv,xn 3.50
is equivalent to
2αnv, Jx021−αnv, Jxn −2v, Jun ≤αnx02 1−αnxn2− un2, 3.51