OF SEMIGROUPS OF NONEXPANSIVE MAPPINGS AND SOLVING SYSTEMS OF VARIATIONAL INEQUALITIES
HOSSEIN PIRI
Communicated by the former editorial board
In this paper, we introduce a new general iterative process for approximating a common fixed point of a semigroup of nonexpansive mappings, with respect to strongly left regular sequence of means defined on an appropriate space of bounded real-valued functions of the semigroups and the set of solutions a system of variational inequality for finite family of inverse strongly monotone mappings in a real Hilbert space. Our result extends and improves the result announced by H. Piri and A.H. Badali [13] and some others.
AMS 2010 Subject Classification: 47H09, 47H10, 43A07, 47J25.
Key words: projection, common fixed point, amenable semigroup, iterative pro- cess, strong convergence, system of variational inequality.
1. INTRODUCTION
Let H be a real Hilbert space with inner product h., .i and norm are denoted by k . k, and let C be a nonempty closed convex subset of H. A mapping T:C → C is called nonexpansive if k T x−T y k≤k x−y k, for all x, y ∈ C. By F ix(T), we denote the set of fixed points of T (i.e., F ix(T) = {x∈H:T x=x}). It is well known thatF ix(T) is closed and convex. Recall that a self-mappingf:C→C is a contraction onC if there exists a constant α∈[0,1) such thatkf(x)−f(y)k≤αkx−yk for allx, y∈C.
Let B:C → H be a mapping. The variational inequality problem, de- noted byV I(C, B), is to findx∈C such that
hBx, y−xi ≥0, (1.1)
for ally ∈C. The variational inequality problem has been extensively studied in literature. See, for example, [19, 20] and the references therein.
Definition 1.1.Let B:C→H be a mapping. Then B is called
(i) strongly positive with constantγ > 0, if there is a constantγ >0 such that
hBx, xi ≥γ kxk2, ∀x∈C.
MATH. REPORTS16(66),2(2014), 295–317
(ii) η−strongly monotone if there exists a positive constantη such that hBx−By, x−yi ≥ηkx−yk2, ∀x, y∈C,
(iii) k−Lipschitzian if there exists a positive constant ksuch that kBx−Byk≤kkx−y k, ∀x, y∈C,
(iv) β−inverse strongly monotone if there exists a positive real numberβ >0 such that
hBx−By, x−yi ≥β kBx−Byk2, ∀x, y∈C.
It is obvious that anyβ−inverse strongly monotone mappingBis 1β−Lip- schitzian.
In 2001 Yamada [18] introduced the following hybrid iterative method for solving the variational inequality
xn+1=T xn−µαnF(T xn), n∈N, (1.2)
whereF isk−Lipschitzian andη−strongly monotone operator withk >0,η >
0, 0< µ < 2ηk2, then he proved that if{αn} satisfying appropriate conditions, then {xn}generated by (1.2) converges strongly to the unique solution of the variational inequality:
hF x∗, x−x∗i ≥0, ∀x∈F ix(T).
In 2006, Marino and Xu [12] consider the following iterative method:
xn+1 = (I −αnA)T xn+αnγf(xn), n≥0.
(1.3)
It is proved that if the sequence{αn}of parameters satisfies the following conditions:
(C1) αn→0, (C2)
∞
P
n=0
αn=∞, (C3) either
∞
P
n=0
|αn+1−αn|<∞or lim
n→∞
αn+1
αn = 1.
Then, the sequence{xn}generated by (1.3) converges strongly, asn→ ∞, to the unique solution of the variational inequality:
h(γf−A)x∗, x−x∗i ≤0, ∀x∈F ix(T), which is the optimality condition for minimization problem
x∈F ix(Tmin )
1
2hAx, xi −h(x),
whereh is a potential function for γf i.e.,h0(x) =γf(x), for all x∈H. Some people also study the application of the iterative method (1.3) [11, 14].
In 2010, Tian [16] combined the iterative (1.3) with the iterative method (1.2) and considered the iterative methods:
xn+1 = (I−µαnF)T xn+αnγf(xn), n≥0, (1.4)
and he proved that if the sequence{αn}of parameters satisfies the conditions (C1),(C2) and (C3), then the sequences {xn} generated by (1.4) converges strongly to the unique solution x∗ ∈F ix(T) of the variational inequality
h(µF −γf)x∗, x−x∗i ≥0, ∀x∈F ix(T).
Motivated and inspired by Atsushiba and Takahashi [2], Ceng and Yao [4], Kim [8], Lau et al. [9], Lauet al. [10], Marino and Xu [12], Tian [16], Xu [17] and Yamada [18], Piri and Badali [13] introduced the following iterative algorithm: Letx0∈C and
yn=βnxn+ (1−βn)PC(xn−δnBxn),
xn+1 =αnγf(xn) + (I−αnµF)Tµnyn, n≥0.
(1.5)
wherePCis a metric projection of H onto C, B isβ−inverse strongly monotone, ϕ = {Tt : t ∈ S} is a nonexpansive semigroup on H such that the set F = F ix(ϕ)∩V I(C, B) 6= ∅, X is a subspace of B(S) such that 1 ∈ X and the mapping t → hTtx, yi is an element of X for each x, y ∈ H, and {µn} is a sequence of means on X. They prove that if {µn} is left regular and the sequences {αn}, {βn} and {δn} of parameters satisfy appropriate conditions, then the sequences{xn} and {yn} generated by (1.5) converge strongly to the unique solutionx∗∈ F of the variational inequalities:
h(µF −γf)x∗, x−x∗i ≥0, ∀x∈ F, hBx∗, y−x∗i ≥0 ∀y∈C,
In this paper, motivated and inspired by Atsushiba and Takahashi [2], Ceng and Yao [4], Kim [8], Lau et al. [9], Lauet al. [10], Marino and Xu [12], Katchang Kumam [7], Tian [16], Xu [17], Yamada [18] and Piri and Badali [13], we introduce the following general iterative algorithm: Let x0 ∈C and
ym+1,n=xn,
yi,n=γi,nPC(I−δi,nAi)yi+1,n
+ (1−γi,n)PC(I−δi+1,nAi+1)yi+1,n, i= 1,2,· · ·, m, xn+1=αnγf(TµnPC(I−δ1,nA1)y1,n) +βnxn
+ ((1−βn)I−αnµF)TµnPC(I−δ2,nA2)y1,n, (1.6)
where PC is a metric projection of H onto C, for i = 1,2,· · ·, m+ 1, Ai be inverse strongly monotone, ϕ ={Tt :t ∈ S} is a nonexpansive semigroup on
C such that the set F = Tm+1
i=1 V I(C, Ai)∩F ix(ϕ) 6= ∅, X is a subspace of B(S) such that 1 ∈ X and the mapping t → hTtx, yi is an element of X for each x, y ∈ H, and {µn} is a sequence of means on X. Our purpose in this paper is to introduce this general iterative algorithm for approximating a com- mon element of the set of common fixed point of a semigroup of nonexpansive mappings and the set of solutions of system of variational inequalities for finite family of inverse strongly monotone mapping. We will prove that if {µn} is left regular sequence of means and the sequences{αn},{βn},{γi,n}m,i=1,n=1∞ and {δi,n}m,i=1,n=1∞ of parameters satisfy appropriate conditions, then the sequences {xn} and {yi,n}m,i=1,n=1∞ generated by (1.6) converge strongly to the unique so- lutionx∗∈ F of the system of the variational inequalities:
h(µF −γf)x∗, x−x∗i ≥0, ∀x∈ F,
hAix∗, y−x∗i ≥0, ∀y∈C, i= 1,2,· · ·, m+ 1.
2.PRELIMINARIES
This section collects some lemmas which will be used in the proofs of the main results in the next section.
Lemma 2.1 ([6]). For a givenx∈H, y∈C,
y=PCx⇔ hy−x, z−yi ≥0, ∀z∈C.
It is well known that PC is a firmly nonexpansive mapping of H onto C and satisfies
kPCx−PCyk2≤ hPCx−PCy, x−yi, ∀x, y∈H (2.1)
Moreover, PC is characterized by the following properties: PCx∈C and for all x∈H, y∈C,
hx−PCx, y−PCxi ≤0.
(2.2)
It is easy to see that (2.2) is equivalent to the following inequality kx−yk2≥kx−PCxk2 +ky−PCxk2.
(2.3)
Using Lemma 2.1, one can see that the variational inequality (1.1) is equivalent to a fixed point problem.
It is easy to see that the following is true:
u∈V I(C, A)⇔u=PC(u−λAu), λ >0.
(2.4)
A set-valued mapping U:H → 2H is called monotone if for all x, y ∈ H, f ∈ U x and g ∈ U y imply hx −y, f −gi ≥ 0. A monotone mapping U:H → 2H is maximal if the graph of G(U) of U is not properly contained
in the graph of any other monotone mapping. It is known that a monotone mappingU is maximal if and only if for (x, f)∈H×H,hx−y, f−gi ≥0 for every (y, g)∈G(U) implies that f ∈U x. LetA be a monotone mapping ofC into H and letNCx be the normal cone to C atx ∈C, that is, NCx ={y ∈ H :hz−x, yi ≤0,∀z∈C}and define
U x=
Ax+NCx, x∈C,
∅ x /∈C.
(2.5)
ThenU is the maximal monotone and 0∈U xif and only ifx∈V I(C, A);
see [15].
Let C be a nonempty subset of a Hilbert space H and T: C → H a mapping. Then T is said to be demiclosed at v∈H if, for any sequence {xn} inC, the following implication holds:
xn* u∈C and T xn→v imply T u=v, where→ (resp. * ) denotes strong (resp. weak) convergence.
Lemma 2.2 ([1]). Let C be a nonempty closed convex subset of a Hilbert space H and suppose that T:C → H is nonexpansive. Then, the mapping I−T is demiclosed at zero.
Lemma 2.3 ([1]). Let H be a real Hilbert space. Then, for allx, y∈H (i) kx−yk2≤kxk2 +2hy, x+yi,
(ii) kx−yk2≥kxk2 +2hy, xi.
Lemma 2.4 ([18]). Let {xn} and {yn} be bounded sequences in a Banach spaceE and let{αn}be a sequence in[0,1]with0<lim inf
n→∞ αn≤lim sup
n→∞
αn<1.
Suppose xn+1=αnxn+ (1−αn)yn for all integers n≥0 and lim sup
n→∞
(kyn+1−ynk − kxn+1−xnk)≤0.
Then, lim
n→∞kyn−xnk= 0.
Lemma 2.5 ([17]). Let {an} be a sequence of nonnegative real numbers such that
an+1≤(1−bn)an+bncn, n≥0,
where {bn} and {cn} are sequences of real numbers satisfying the following conditions:
(i) {bn} ⊂(0,1),
∞
P
n=0
bn=∞, (ii) either lim sup
n→∞ cn≤0 or
∞
P
n=0
|bncn|<∞.
Then, lim
n→∞an= 0.
Lemma2.6 ([13]). LetHbe a real Hilbert space andF be ak−Lipschitzian and η−strongly monotone operator with k > 0, η > 0. Let 0 < µ < 2ηk2 and τ = µ(η − µk22). Then for t ∈ (0,min{1,1τ}), I −tµF is contraction with constant 1−tτ.
LetS be a semigroup and letB(S) be the space of all bounded real valued functions defined on S with supremum norm. For s ∈ S and f ∈ B(S), we define elementslsf and rsf inB(S) by
(lsf)(t) =f(st), (rsf)(t) =f(ts), ∀t∈S.
Let X be a subspace ofB(S) containing 1 and let X∗ be its topological dual. An element µ of X∗ is said to be a mean on X ifk µk=µ(1) = 1. We often write µt(f(t)) instead of µ(f) for µ ∈ X∗ and f ∈ X. Let X be left invariant (resp. right invariant), i.e., ls(X) ⊂X (resp. rs(X) ⊂ X) for each s ∈ S. A mean µ on X is said to be left invariant (resp. right invariant) if µ(lsf) = µ(f) (resp. µ(rsf) = µ(f)) for eachs∈ S and f ∈X. X is said to be left (resp. right) amenable if X has a left (resp. right) invariant mean. X is amenable if X is both left and right amenable. As is well known, B(S) is amenable when S is a commutative semigroup, see [10]. A net {µα} of means on X is said to be strongly left regular if
limα kls∗
µα−µα k= 0, for each s∈S, wherel∗s is the adjoint operator ofls.
LetS be a semigroup and letC be a nonempty closed and convex subset of a reflexive Banach spaceE. A familyϕ={Tt :t∈S} of mapping fromC into itself is said to be a nonexpansive semigroup on C if Tt is nonexpansive andTts=TtTsfor eacht, s∈S. ByF ix(ϕ) we denote the set of common fixed points of ϕ,i.e.
F ix(ϕ) = \
t∈S
{x∈C :Ttx=x}.
Lemma 2.7 ([10]). Let S be a semigroup and C be a nonempty closed convex subset of a reflexive Banach space E. Let ϕ = {Tt : t ∈ S} be a nonexpansive semigroup onH such that {Ttx:t∈S}is bounded for somex∈ C, letX be a subspace ofB(S) such that1∈X and the mappingt→ hTtx, y∗i is an element of X for eachx∈C and y∗ ∈E∗, andµis a mean on X. If we write Tµx instead of R
Ttxdµ(t), then the following hold.
(i) Tµ is nonexpansive mapping from C into C.
(ii) Tµx=x for each x∈F ix(ϕ).
(iii) Tµx∈co{Ttx:t∈S} for each x∈C.
Notation. The open ball of radiusr centered at 0 is denoted byBr and for a subsetDofH, bycoD, we denote the closed convex hull ofD. For >0 and a mappingT:D→H, we letF(T;D) be the set of−approximate fixed points sets of T,i.e. F(T;D) ={x∈D:kx−T xk≤}.
3. MAIN RESULTS
Theorem 3.1. Let C be a nonempty closed convex subset of real Hilbert space H, F:C → C be a k−Lipschitzian and η−strongly monotone operator withk >0,η >0,f:C→Cbe a contraction with coefficient0< α <1,γ be a number in(0,ατ), whereτ =µ(η−µk22)and0< µ < 2ηk2. Fori= 1,2,· · ·, m+1, letAi:C→Hbeδi−inverse strongly monotone mapping. LetSbe a semigroup and ϕ= {Tt : t ∈S} be a nonexpansive semigroup of C into itself such that F =Tm+1
i=1 V I(C, Ai)∩F ix(ϕ)6=∅. LetX be a left invariant subspace ofB(S) such that 1 ∈ X, and the function t → hTtx, yi is an element of X for each x∈C andy∈H. Let{µn}be a left regular sequence of means onX such that P∞
n=1 kµn+1−µnk<∞. Let {αn}, {γi,n}m,i=1,n=1∞ be sequences in (0,1), {βn} and {δi,n}m+1,i=1,n=1∞ be a sequence in [0,1) satisfy the following conditions:
(B1) 0 < lim infn→∞γi,n ≤ lim supn→∞γi,n < 1 and {δi,n} ⊂ (0,2δi) i = 1,2,· · · , m+ 1,
(B2) limn→∞αn= 0, P∞
n=1αn=∞ and limn→∞βn= 0.
(B3) P∞
n=1 |αn+1−αn|<∞, P∞
n=1 |βn+1−βn|<∞ P∞
n=1 |δi,n+1−δi,n|<∞,P∞
n=1 |γi,n+1−γi,n|<∞ i= 1,2,· · · , m+ 1.
If x0 ∈C and {xn} and {yi,n}m,i=1,n=1∞ be generated by the iteration algo- rithm:
ym+1,n=xn,
yi,n=γi,nPC(I−δi,nAi)yi+1,n
+ (1−γi,n)PC(I−δi+1,nAi+1)yi+1,n, i= 1,2,· · · , m, xn+1=αnγf(TµnPC(I−δ1,nA1)y1,n) +βnxn
+ ((1−βn)I−αnµF)TµnPC(I−δ2,nA2)y1,n,
then, {xn} and {yi,n}m,i=1,n=1∞ converge strongly, as n → ∞, to x∗ ∈ F, which is a unique solution of system of the variational inequalities.
h(µF −γf)x∗, x−x∗i ≥0, ∀x∈ F,
hAix∗, y−x∗i ≥0, ∀y∈C, i= 1,2,· · ·, m+ 1.
Proof. Since {δi,n}m+1,i=1,n=1∞ satisfy in condition (B1) and Ai is δi−inverse strongly monotone mapping, for any x, y∈C, we have
k(I−δi,nAi)x−(I−δi,nAi)yk2
=k(x−y)−δi,n(Aix−Aiy)k2
=kx−yk2 −2δi,nhx−y, Aix−Aiyi+δi,n2 kAix−Aiyk2
≤kx−yk2 −2δi,nδikAix−Aiyk2 +δi,n2 kAix−Aiyk2
=kx−yk2 +δi,n(δi,n−2δi)kAix−Aiy k2
≤kx−yk2 It follows that
k(I−δi,nAi)x−(I −δi,nAi)yk≤kx−y k, i= 1,2,· · · , m+ 1.
(3.1)
Let p ∈ F, in the context of the variational inequality problem the characterization of projection (2.4) implies that p = PC(I −δi,nAi)p, i = 1,2,· · · , m+ 1.Using (3.1), we get
kyi,n−pk2
=kγi,nPC(I −δi,nAi)yi+1,n+ (1−γi,n)PC(I−δi+1,nAi+1)yi+1,n−pk2
≤γi,nkPC(I−δi,nAi)yi+1,n−pk2+(1−γi,n)kPC(I−δi+1,nAi+1)yi+1,n−pk2
≤γi,nkyi+1,n−pk2+(1−γi,n)kyi+1,n−pk2=kyi+1,n−pk2 Which implies that
ky1,n−pk≤ky2,n−pk≤ · · · ≤kym,n−pk≤kym+1,n−pk=kxn−pk (3.2)
We claim that {xn}is bounded. Let p∈ F, using Lemma 2.6, we have kxn+1−pk=kαnγf(TµnPC(I−δ1,nA1)y1,n) +βnxn
+ ((1−βn)I−αnµF)TµnPC(I−δ2,nA2)y1,n−pk
=k((1−βn)I−αnµF)TµnPC(I−δ2,nA2)y1,n
−((1−βn)I−αnµF)TµnPC(I−δ2,nA2)pk
+βnkxn−pk+αnkγf(TµnPC(I −δ1,nA1)y1,n)−µF pk
= (1−βn−αnτ)kTµnPC(I−δ2,nA2)y1,n−TµnPC(I−δ2,nA2)pk +βnkxn−pk+αnkγf(TµnPC(I −δ1,nA1)y1,n)
−γf(p)k+kγf(p)−µF pk
≤(1−βn−αnτ)ky1,n−pk+βnkxn−pk+αnγαky1,n−pk +αnkγf(p)−µF pk
≤(1−αn(τ −γα))kxn−pk+αnkγf(p)−µF pk
By induction, kxn−pk≤max
(τ−γα)−1 kγf(p)−µF pk,kx0−pk =M0. Hence,{xn}is bounded and also{yi,n}m,i=1,n=1∞ and{f(PC(I−δ1,nA1)y1,n)}
are bounded. Set D = {y ∈ H :k y −p k≤ M0}. We remark that D is ϕ-invariant bounded closed convex set and {xn}∞n=1,{yi,n}m,i=1,n=1∞ ,{PC(I− δi,nAi)yi+1,n}m,i=1,n=1∞ ⊂D. Now, we claim that
(3.3) lim sup
n→∞ sup
y∈D
kTµny−TtTµnyk= 0, ∀t∈S.
This assertion is proved in [13]. We now show that
n→∞lim kxn+1−xnk= 0 (3.4)
From definition of {xn}, we have kxn+1−xnk
=kαnγf(TµnPC(I−δ1,nA1)y1,n) +βnxn + ((1−βn)I −αnµF)TµnPC(I−δ2,nA2)y1,n
−αn−1γf(Tµn−1PC(I−δ1,n−1A1)y1,n−1) +βn−1xn−1
+ ((1−βn−1)I−αn−1µF)Tµn−1PC(I−δ2,n−1A2)y1,n−1 k
=kαnγ[f(TµnPC(I−δ1,nA1)y1,n)−f(Tµn−1PC(I−δ1,n−1A1)y1,n−1)]
+ (αn−αn−1)γf(Tµn−1PC(I−δ1,n−1A1)y1,n−1) +βn[xn−xn−1] + (βn−βn−1)xn−1+ [((1−βn)I−αnµF)TµnPC(I−δ2,nA2)y1,n
−((1−βn)I −αnµF)Tµn−1PC(I−δ2,n−1A2)y1,n−1] + [((1−βn)I−αnµF)Tµn−1PC(I−δ2,n−1A2)y1,n−1
−((1−βn−1)I−αn−1µF)Tµn−1PC(I−δ2,n−1A2)y1,n−1]k
≤αnγαkTµnPC(I−δ1,nA1)y1,n−Tµn−1PC(I−δ1,n−1A1)y1,n−1 k +|αn−αn−1 |kγf(Tµn−1PC(I−δ1,n−1A1)y1,n−1)k
+βnkxn−xn−1 k+|βn−βn−1 |kxn−1 k
+ (1−βn−αnτ)kTµnPC(I−δ2,nA2)y1,n−Tµn−1PC(I−δ2,n−1A2)y1,n−1k +|βn−βn−1 |kTµn−1PC(I−δ2,n−1A2)y1,n−1k
+|αn−αn−1 |kµF Tµn−1PC(I−δ2,n−1A2)y1,n−1k From (2.4) for p∈ F, we have
kTµn−1PC(I−δi,n−1Ai)y1,n−1−TµnPC(I−δi,nAi)y1,nk
≤kTµn−1PC(I−δi,n−1Ai)y1,n+1−Tµn−1PC(I−δi,n+1Ai)y1,n k +kTµn+1PC(I−δi,n−1Ai)y1,n−Tµn−1PC(I−δi,nAi)y1,n k
+kTµn−1PC(I−δi,nAi)y1,n−TµnPC(I−δi,nAi)y1,nk
≤ky1,n−1−y1,nk+|δi,n−1−δi,n|kAiy1,n k +kµn−1−µnk[ky1,n−pk+kpk]
By taking
ui,n=PC(I−δi,nAi)yi+1,n,
vi,n=PC(I−δi+1,nAi+1)yi+1,n, i= 1,2,· · · , m.
From definition of {yi,n}m+1,i=1,n=1∞, we have kyi,n+1−yi,nk
=kγi,n+1ui,n+1+ (1−γi,n+1)vi,n+1−γi,nui,n−(1−γi,n)vi,nk
=kγi,n+1(ui,n+1−ui,n) + (γi,n+1−γi,n)ui,n+ (1−γi,n+1)(vi,n+1−vi,n) + (γi,n−γi,n+1)vi,nk
=γi,n+1 kui,n+1−ui,nk+(1−γi,n+1)kvi,n+1−vi,nk +|γi,n+1−γi,n|[kui,nk+kvi,nk]
=γi,n+1 kPC(I−δi,n+1Ai)yi+1,n+1−PC(I−δi,n+1Ai)yi+1,n
+PC(I −δi,n+1Ai)yi+1,n−PC(I−δi,nAi)yi+1,nk
+ (1−γi,n+1)kPC(I−δi+1,n+1Ai+1)yi+1,n+1−PC(I−δi+1,n+1Ai+1)yi+1,n
+PC(I −δi+1,n+1Ai+1)yi+1,n−PC(I−δi+1,nAi+1)yi+1,nk +|γi,n+1−γi,n|[kui,nk+kvi,nk]
≤γi,n+1 kyi+1,n+1−yi+1,n k+γi,n+1 |δi,n+1−δi,n|kAiyi+1,nk + (1−γi,n+1)kyi+1,n+1−yi+1,nk
+ (1−γi,n+1)|δi+1,n+1−δi+1,n |kAi+1yi+1,n k +|γi,n+1−γi,n|[kui,nk+kvi,nk]
=kyi+1,n+1−yi+1,nk+γi,n+1|δi,n+1−δi,n+1 |kAiyi+1,nk + (1−γi,n+1)|δi+1,n+1−δi+1,n kAi+1yi+1,nk
+|γi,n+1−γi,n|[kui,nk+kvi,nk]
which implies that for some approximate constant Mi >0, kyi,n+1−yi,n≤kyi+1,n+1−yi+1,n k+[|δi,n+1−δi,n|
+|δi+1,n+1−δi+1,n |+|γi,n+1−γi,n|]Mi. From this and (3.2), we get
kyi,n+1−yi,nk
≤kxn+1−xnk+
m
X
j=i
[|δj,n+1−δj,n |
+|δj+1,n+1−δj+1,n |+|γj,n+1−γj,n|]Mj. (3.5)
Therefore, kxn+1−xnk
≤(1−βn−αn(τ−γα))[kxn−xn−1 k+
m
X
j=1
[|δj,n−δj,n−1 |
+|δj+1,n−δj+1,n−1|+|γj,n−γj,n−1 |]Mj+|δ2,n−1−δ2,n |kA2y1,nk +|δ1,n−1−δ1,n |kA1y1,n k+kµn−1−2µnk(ky1,n−pk+kpk)]
+|αn−αn−1 |kγf(Tµn−1PC(I−δ1,n−1A1)y1,n−1)k +βnkxn−xn−1 k+|βn−βn−1|kxn−1 k
+|βn−βn−1|kTµn−1PC(I−δ2,n−1A2)y1,n−1 k +|αn−αn−1 |kµF Tµn−1PC(I−δ2,n−1A2)y1,n−1 k
≤(1−αn(τ −γα))kxn−xn−1k+[
m
X
j=1
[|δj,n−δj,n−1 |
+|δj+1,n−δj+1,n−1|+|γj,n−γj,n−1 |]Mj+|δ1,n−1−δ2,n |kA2y1,nk +|δ2,n−1−δ2,n |kA2y1,n k+2kµn−1−µnk(ky1,n−pk+kpk)]
+|αn−αn−1 |[kγf(Tµn−1PC(I−δ1,n−1A1)y1,n−1)k +kµF Tµn−1PC(I−δ2,n−1A2)y1,n−1 k]
+|βn−βn−1|[kxn−1 k+kTµn−1PC(I−δ2,n−1A2)y1,n−1k]
(3.6)
Thus, it follows from (3.6), condition (B3) and Lemma 2.4 that
n→∞lim (kxn+1−xnk) = 0.
In this stage, we will show
n→∞lim kxn−Ttxnk= 0, ∀t∈S.
(3.7)
Let t∈S and >0. By [3, Theorem 1.2], there existsδ >0 such that coFδ(Tt;D) +Bδ⊂F(Tt;D), ∀t∈S.
(3.8)
From condition B1 and (3.3), there exists N1∈Nsuch that
αn≤ δ
2(τ +µk)M0, βn≤ δ
4M0, TµnPC(I−δ2,nA2)y1,n∈Fδ(Tt;D), n≥N1. By Lemma 2.6 and (2.4), forp∈ F we have
αnkγf(PC(I−δ1,nA1)y1,n)−µF TµnPC(I−δ2,nA2)y1,n k
≤αn[γ kf(PC(I −δ1,nA1)y1,n)−f(p)k+kγf(p)−µF pk +kµF p−µF TµnPC(I−δ2,nA2)y1,n k]
≤αn(γαky1,n−pk+kγf(p)−µF pk+µkky1,n−pk)
≤αn(γαM0+ (τ −γα)M0+µkM0)
≤αn(τ +µk)M0 ≤ δ 2, and
βnkxn−TµnPC(I−δ2,nA2)y1,nk
≤βn[kxn−pk+kTµnPC(I−δ2,nA2)y1,n−pk]
≤βn[kxn−pk+ky1,n−pk]
≤2βnkxn−pk
≤2βnM0 ≤ δ 2. Therefore, we have
xn+1 =TµnPC(I−δ2,nA2)y1,n+αn[γf(PC(I−δ1,nA1)y1,n)
−µF TµnPC(I −δ2,nA2)y1,n] +βn[xn−TµnPC(I−δ2,nA2)y1,n]
∈Fδ(Tt;D) +Bδ 2 +Bδ
2
⊂F(Tt;D), n≥N1. This shows that
kxn−Ttxnk≤, ∀n≥N1. Since >0 is arbitrary, we get (3.7).
Let Q=PF. Then Q(I−µF +γf) is a contraction of H into itself. In fact, we see that
kQ(I−µF +γf)x−Q(I−µF +γf)yk
≤k(I−µF +γf)x−(I−µF +γf)y k
≤k(I−µF)x−(I−µF)yk+γ kf(x)−f(y)k
≤(1−τ)kx−y k+γαkx−y k
and hence, Q(I−µF +γf) is a contraction due to (1−(τ −γα))∈(0,1).
Therefore, by Banach’s contraction principle, PF(I −µF +γf) has a unique fixed point x∗. Then using (2.4), x∗ is the unique solution of the variational inequality:
h(γf −µF)x∗, x−x∗i ≤0, ∀x∈ F. (3.9)
We show that
lim sup
n→∞
hγf(x∗)−µF x∗, xn−x∗i ≤0.
(3.10)
We can choose a subsequence {xnj} of {xn}such that lim sup
n→∞
hγf(x∗)−µF x∗, xn−x∗i= lim
j→∞hγf(x∗)−µF x∗, xnj−x∗i.
(3.11)
Because {xnj} is bounded, therefore {xnj} has subsequence{xnjk} such that xnjk * z. With no loss of generality, we may assume that xnj * z. It follows from (3.7) and Lemma 2.2 that z∈F ix(T).
In this stage, we will show that
n→∞lim kui,n−yi+1,n k= 0
n→∞lim kvi,n−yi+1,nk= 0, i= 1,2,· · ·, m.
(3.12)
Let p∈ F, from (3.2), Lemma 2.3 and definition of {xn}, we have kxn+1−pk2
=kαnγf(T PC(I−δ1,nA1)y1,n) +βnxn
+ ((1−βn)I−αnµF)T PC(I−δ2,nA2)y1,n−pk2
=k[((1−βn)I−αnF)T PC(I−δ2,nA2)y1,n−((1−βn)I−αnF)p]
+βn[xn−p] +αn[γf(T PC(I−δ1,nA1)y1,n)−F p]k2
≤k((1−βn)I −αnF)T PC(I−δ2,nA2)y1,n−((1−βn)I−αnF)pk2 + 2βnhxn−p, xn+1−pi
+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤(1−βn−αnτ )2 ky1,n−pk2 +2βnhxn−p, xn+1−pi + 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤ky1,n−pk2+2βnhxn−p, xn+1−pi
+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤kyi,n−pk2+2βnhxn−p, xn+1−pi
+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi.
(3.13)
From (2.4), (3.2), (3.13) and Lemma 2.3 , we have (for i= 1,2,· · · , m) kxn+1−pk2
≤kγi,nPC(I−δi,nAi)yi+1,n+ (1−γi,n)PC(I−δi+1,nAi+1)yi+1,n−pk2 + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤γi,nkPC(I−δi,nAi)yi+1,n−pk2 +(1−γi,n)kPC(I−δi+1,nAi+1)yi+1,n−pk2 + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤γi,nk(yi+1,n−p)−δi,n(Aiyi+1,n−Aip)k2 +(1−γi,n)kyi+1,n−pk2
+ 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤γi,n[kyi+1,n−pk2+δi,n2 kAiyi+1,n−Aipk2−2δi,nhAiyi+1,n−Aip, yi+1,n−pi]
+ (1−γi,n)kyi+1,n−pk2 +2βnhxn−p, xn+1−pi + 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤kyi+1,n−pk2 +δ2i,nkAiyi+1,n−Aipk2 −2δi,nδikAiyi+1,n−Aipk2 + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤kxn−pk2 +δi,n(δi,n−2δi)kAiyi+1,n−Aipk2
+ 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi which implies that
−δi,n(δi,n−2δi)kAiyi+1,n−Aipk2
≤kxn−pk2− kxn+1−pk2 +2βnhxn−p, xn+1−pi + 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi
≤[kxn−pk+kxn+1−pk]kxn+1−xnk + 2βnhxn−p, xn+1−pi
+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi (3.14)
Using (3.4), (3.14) and condition B2, we get
limn→∞ kAiyi+1,n−Aipk= 0, i= 1,2,· · · , m.
(3.15)
Repeating the same argument as above, we conclude that limn→∞ kAi+1yi+1,n−Ai+1pk= 0, i= 1,2,· · ·, m.
(3.16)
From (2.1), we have kui,n−pk2
=kPC(I−δi,nAi)yi+1,n−PC(I−δi,nAip)k2
≤ h(I−δi,nAi)yi+1,n−(I−δi,nAi)p, ui,n−pi
= 1
2[k(I−δi,nAi)yi+1,n−(I−δi,nAi)pk2 +kui,n−pk2
− k(I−δi,nAi)yi+1,n−(I−δi,nAi)p−(ui,n−p)k2]
≤ 1
2[kyi+1,n−pk2+kui,n−pk2
− k(I−βi,nAi)yi+1,n−(I−δi,nAi)p−(ui,n−p)k2]
= 1
2[kyi+1,n−pk2+kui,n−pk2 − kyi+1,n−ui,nk2
+ 2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi −δi,n2 kAiyi+1,n−Aipk2].
So, we obtain
kui,n−pk2 ≤kyi+1,n−pk2− kyi+1,n−ui,nk2 + 2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi
−δ2i,nkAiyi+1,n−Aipk2, i= 1,2,· · · , m.
(3.17)
By using same method as (3.17), we have kvi,n−pk2 ≤kyi+1,n−pk2− kyi+1,n−vi,nk2
+ 2δi+1,nhyi+1,n−vi,n, Ai+1yi+1,n−Ai+1pi
−δ2i+1,n kAi+1yi+1,n−Ai+1pk2, i= 1,2,· · ·, m.
(3.18)
From (3.13), (3.17), (3.18) and Lemma 2.3, we have kxn+1−pk2
≤kyi,n−pk2+2βnhxn−p, xn+1−pi
+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi
≤kγi,nPC(I−δi,nAi)yi+1,n+ (1−γi,n)PC(I−δi+1,nAi+1)yi+1,n−pk2 + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi
≤γi,nkui,n−pk2 +(1−γi,n)kvi,n−pk2
+ 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi
≤γi,n[kyi+1,n−pk2− kyi+1,n−ui,nk2+2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi
−δi,n2 kAiyi+1,n−Aipk2] + (1−γi,n)[kyi+1,n−pk2 − kyi+1,n−vi,nk2 + 2δi+1,nhyi+1,n−vi,n, Ai+1yi+1,n−Ai+1pi−δ2i+1,nkAi+1yi+1,n−Ai+1pk2] + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi
=kyi+1,n−pk2+γi,n[− kyi+1,n−ui,nk2+2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi
−δi,n2 kAiyi+1,n−Aipk2] + (1−γi,n)[− kyi+1,n−vi,nk2
+ 2δi+1,nhyi+1,n−vi,n, Ai+1yi+1,n−Ai+1pi−δ2i+1,nkAi+1yi+1,n−Ai+1pk2] + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi
≤kxn−pk2 +γi,n[− kyi+1,n−ui,nk2 +2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi
−δi,n2 kAiyi+1,n−Aipk2] + (1−γi,n)[− kyi+1,n−vi,nk2
+ 2δi+1,nhyi+1,n−vi,n, Ai+1yi+1,n−Ai+1pi−δ2i+1,n kAi+1yi+1,n−Ai+1pk2] + 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I −δ1,nA1)y1,n)−F p, xn+1−pi Which implies that
γi,nkyi+1,n−ui,nk2
≤[kxn−pk+kxn+1−pk]kxn+1−xnk
+γi,n[2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi
−δi,n2 kAiyi+1,n−Aipk2] + (1−γi,n)[+2δi+1,nhyi+1,n−vi,n, Ai+1yi+1,n−Ai+1pi −δi+1,n2 kAi+1yi+1,n−Ai+1pk2]
+ 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi and
(1−γi,n)kyi+1,n−vi,nk2
≤[kxn−pk+kxn+1−pk]kxn+1−xnk+γi,n[2δi,nhyi+1,n−ui,n, Aiyi+1,n−Aipi+δi,n2 kAiyi+1,n−Aipk2] + (1−γi,n)[2δi+1,nhyi+1,n
−vi,n, Ai+1yi+1,n−Ai+1pi −δi+1,n2 kAi+1yi+1,n−Ai+1pk2]
+ 2βnhxn−p, xn+1−pi+ 2αnhγf(T PC(I−δ1,nA1)y1,n)−F p, xn+1−pi Therefore, using condition B2, (3.4), (3.15) and (3.16) we get (3.12).
Simple calculation shows that, kyi,n−xnk≤
m
X
j=i
kyj,n−yj+1,nk, , i= 1,2,· · ·, m.
(3.19)
From (3.12) and (3.19), we get
n→∞lim kyi,n−xnk= 0, i= 1,2,· · · , m.
(3.20)
Now, let us show that fori= 1,2,· · ·, m+1,z∈V I(C, Ai). LetUi:H → 2H be a set-valued mapping is defined by
Uix=
Aix+NCx, x∈C,
∅, x /∈C.
where NCx is the normal cone to C at x ∈ C. Since Ai is monotone. Thus, Ui is maximal monotone see [15]. Let (x, y) ∈ G(Ui), hence, y−Aix ∈ NCx and sinceui,n=PC(I−δi,nAi)yi+1,n therefore,hx−ui,n, y−Aixi ≥0. On the other hand from (2.3), we have
hx−ui,n, ui,n−(yi+1,n−δi,nAiyi+1,n)i ≥0, that is
hx−ui,n,ui,n−yi+1,n
δi,n +Aiyi+1,ni ≥0.
Therefore, we have hx−ui,nj, yi
≥ hx−ui,nj, Aixi
≥ hx−ui,nj, Aixi − hx−ui,nj,ui,nj−yi+1,nj δi,nj
+Aiyi+1,nji
=hx−ui,nj, Aix−ui,nj−yi+1,nj δi,nj
−Aiyi+1,nji
=hx−ui,nj, Aix−Aiui,nji+hx−ui,nj, Aiui,nj−Aiyi+1,nji
− hx−ui,nj,ui,nj−yi+1,nj
δi,nj i
≥ hx−ui,nj, Aiui,nj−Aiyi+1,nji − hx−ui,nj,ui,nj−yi+1,nj
δi,nj i
≥ hx−ui,nj, Aiui,nj−Aiyi+1,nji− kx−unj kk ui,nj−yi+1,nj δi,nj
k. (3.21)
Since xnj * z Ai is δ1
i−Lipschitzian, from (3.12), (3.20) and (3.21) we obtain
hx−z, yi ≥0.
Since Ui is maximal monotone, we havez∈Ui−10, and hence, z∈V I(C, Ai), i= 1,2,· · ·, m.
Repeating the same argument as above for vi,n, we conclude that z∈V I(C, Ai), i= 2,3,· · ·, m+ 1.
Therefore z∈ F and applying (3.9) and (3.11), we have lim sup
n→∞
h(γf −F)x∗, xn−x∗i ≤0.
Finally, we prove that xn → x∗ as n → ∞. Using (2.4), (3.2), and Lemma 2.3, we have
kxn+1−x∗k2
=kαnγf(T PC(I−δ1,nA1)y1,n) +βnxn
+ ((1−βn)I−αnF)T PC(I−δ2,nA2)y1,n−x∗ k2
=k[((1−βn)I−αnF)T PC(I−δ2,nA2)y1,n
−((1−βn)I−αnF)T PC(I−δ3,nA3)x∗]
+βn[xn−x∗] +αn[γf(T PC(I−δ1,nA1)y1,n)−F x∗]k2
≤k[((1−βn)I−αnF)T PC(I−δ2,nA2)y1,n
−((1−βn)I−αnF)T PC(I−δ2,nA2)x∗]
+βn[xn−x∗]k2+2αnhγf(T PC(I−δ1,nA1)y1,n)−F x∗, xn+1−x∗i
≤[(1−βn−αnτ)ky1,n−x∗ k+βnkxn−x∗k]2 + 2αnhγf(PC(I−δ1,nA1)T yn)−F x∗, xn+1−x∗i
≤k[(1−βn−αnτ)ky1,n−x∗k+βnkxn−x∗k]2