We explore the local dynamical characteristics, chaos and bifurcations of a two-dimensional discrete laser model in R2+. It is shown that for all a, b, c and p, model has boundary fixed point P0y(0,pc), and the unique positive fixed point P+xy(ap−bcab,ba) if p>bca. Further, local dynamical characteristics with topological classifications for the fixed points P0y(0,pc) and P+xy(ap−bcab,ba) have explored by stability theory. It is investigated that flip bifurcation exists for the boundary fixed point P0y(0,pc), and also there exists a flip bifurcation if parameters vary in a small neighborhood of the unique positive fixed point P+xy(ap−bcab,ba). Moreover, it is also explored that for the fixed point P+xy(ap−bcab,ba), laser model undergoes a Neimark-Sacker bifurcation, and in the meantime stable invariant curve appears. Numerical simulations are implemented to verify not only obtain results but also exhibit complex dynamics of period −2, −3, −4, −5, −8 and −9. Further, maximum lyapunov exponents along with fractal dimension are computed numerically to validate chaotic behavior of the laser model. Lastly, feedback control method is utilized to stabilize chaos exists in the model.
Citation: Abdul Qadeer Khan, Mohammed Bakheet Almatrafi. Two-dimensional discrete-time laser model with chaos and bifurcations[J]. AIMS Mathematics, 2023, 8(3): 6804-6828. doi: 10.3934/math.2023346
[1] | Lu-Chuan Ceng, Shih-Hsin Chen, Yeong-Cheng Liou, Tzu-Chien Yin . Modified inertial subgradient extragradient algorithms for generalized equilibria systems with constraints of variational inequalities and fixed points. AIMS Mathematics, 2024, 9(6): 13819-13842. doi: 10.3934/math.2024672 |
[2] | Francis Akutsah, Akindele Adebayo Mebawondu, Austine Efut Ofem, Reny George, Hossam A. Nabwey, Ojen Kumar Narain . Modified mildly inertial subgradient extragradient method for solving pseudomonotone equilibrium problems and nonexpansive fixed point problems. AIMS Mathematics, 2024, 9(7): 17276-17290. doi: 10.3934/math.2024839 |
[3] | Fei Ma, Jun Yang, Min Yin . A strong convergence theorem for solving pseudo-monotone variational inequalities and fixed point problems using subgradient extragradient method in Banach spaces. AIMS Mathematics, 2022, 7(4): 5015-5028. doi: 10.3934/math.2022279 |
[4] | Habib ur Rehman, Poom Kumam, Kanokwan Sitthithakerngkiet . Viscosity-type method for solving pseudomonotone equilibrium problems in a real Hilbert space with applications. AIMS Mathematics, 2021, 6(2): 1538-1560. doi: 10.3934/math.2021093 |
[5] | Habib ur Rehman, Wiyada Kumam, Poom Kumam, Meshal Shutaywi . A new weak convergence non-monotonic self-adaptive iterative scheme for solving equilibrium problems. AIMS Mathematics, 2021, 6(6): 5612-5638. doi: 10.3934/math.2021332 |
[6] | Rose Maluleka, Godwin Chidi Ugwunnadi, Maggie Aphane . Inertial subgradient extragradient with projection method for solving variational inequality and fixed point problems. AIMS Mathematics, 2023, 8(12): 30102-30119. doi: 10.3934/math.20231539 |
[7] | Yuanheng Wang, Chenjing Wu, Yekini Shehu, Bin Huang . Self adaptive alternated inertial algorithm for solving variational inequality and fixed point problems. AIMS Mathematics, 2024, 9(4): 9705-9720. doi: 10.3934/math.2024475 |
[8] | Ziqi Zhu, Kaiye Zheng, Shenghua Wang . A new double inertial subgradient extragradient method for solving a non-monotone variational inequality problem in Hilbert space. AIMS Mathematics, 2024, 9(8): 20956-20975. doi: 10.3934/math.20241020 |
[9] | Wenlong Sun, Gang Lu, Yuanfeng Jin, Zufeng Peng . Strong convergence theorems for split variational inequality problems in Hilbert spaces. AIMS Mathematics, 2023, 8(11): 27291-27308. doi: 10.3934/math.20231396 |
[10] | Bancha Panyanak, Chainarong Khunpanuk, Nattawut Pholasa, Nuttapol Pakkaranang . A novel class of forward-backward explicit iterative algorithms using inertial techniques to solve variational inequality problems with quasi-monotone operators. AIMS Mathematics, 2023, 8(4): 9692-9715. doi: 10.3934/math.2023489 |
We explore the local dynamical characteristics, chaos and bifurcations of a two-dimensional discrete laser model in R2+. It is shown that for all a, b, c and p, model has boundary fixed point P0y(0,pc), and the unique positive fixed point P+xy(ap−bcab,ba) if p>bca. Further, local dynamical characteristics with topological classifications for the fixed points P0y(0,pc) and P+xy(ap−bcab,ba) have explored by stability theory. It is investigated that flip bifurcation exists for the boundary fixed point P0y(0,pc), and also there exists a flip bifurcation if parameters vary in a small neighborhood of the unique positive fixed point P+xy(ap−bcab,ba). Moreover, it is also explored that for the fixed point P+xy(ap−bcab,ba), laser model undergoes a Neimark-Sacker bifurcation, and in the meantime stable invariant curve appears. Numerical simulations are implemented to verify not only obtain results but also exhibit complex dynamics of period −2, −3, −4, −5, −8 and −9. Further, maximum lyapunov exponents along with fractal dimension are computed numerically to validate chaotic behavior of the laser model. Lastly, feedback control method is utilized to stabilize chaos exists in the model.
Let H be a real Hilbert space with its inner product ⟨⋅,⋅⟩ and induced norm ‖⋅‖. Let ∅≠C⊂H be a closed convex set, and PC be the metric projection (or nearest point) from H onto C. Let S:C→H be a nonlinear mapping. Let the Fix(S) and R indicate the fixed-point set of S and the real-number set, respectively. We denote by the → and ⇀ the strong convergence and weak convergence in H, respectively. A mapping S:C→C is referred to as being asymptotically nonexpansive if ∃{θr}∞r=1⊂[0,+∞) s.t. limr→∞θr=0 and θr‖u−v‖+‖u−v‖≥‖Sru−Srv‖,∀r≥1,u,v∈C. In particular, if θr=0,∀r≥1, then S is known as being nonexpansive.
Let Θ:C×C→R be a bifunction. The equilibrium problem (EP) for Θ is to determine its equilibrium points, that is, the set EP(Θ)={u∈C:Θ(u,v)≥0,∀v∈C}. Under the theory framework of equilibrium problems, there is a unified way to investigate a wide number of problems arising in the physics, optimization, structural analysis, transportation, finance and economics. In order to find an element in EP(Θ), one assumes that the following hold:
(H1) Θ(u,u)=0,∀u∈C;
(H2) Θ is monotone, that is, Θ(u,v)+Θ(v,u)≤0,∀u,v∈C;
(H3) limλ→0+Θ((1−λ)u+λw,v)≤Θ(u,v),∀u,v,w∈C;
(H4) v↦Θ(u,v) is convex and lower semicontinuous (l.s.c.) for each u∈C.
In 1994, Blum and Oettli [34] gave the following lemma, which plays a key role in solving the equilibrium problems.
Lemma 1.1 ([34]). Let Θ:C×C→R satisfy the hypotheses (H1)–(H4). For any u∈C and ℓ>0, let Sℓ:H→C be the mapping formulated below:
TΘℓ(u):={w∈C:Θ(w,v)+1ℓ⟨v−w,w−u⟩≥0,∀v∈C}. |
Then TΘℓ is well defined and the following hold: (i) TΘℓ is single-valued, and firmly nonexpansive, that is, ‖TΘℓu−TΘℓv‖2≤⟨TΘℓu−TΘℓv,u−v⟩,∀u,v∈H; and (ii) Fix(TΘℓ)=EP(Θ), and EP(Θ) is convex and closed.
It is worth pointing out that the variational inequality problem (VIP) is a special case of the EP. In particular, if Θ(u,v)=⟨Au,v−u⟩,∀u,v∈C, then the EP reduces to the classical VIP of finding u∈C s.t. ⟨Au,v−u⟩≥0,∀v∈C, where A is a self-mapping on H. The solution set of the VIP is denoted by VI(C,A). It is well known that, one of the most popular techniques for solving the VIP is the extragradient one put forth by Korpelevich [26] in 1976, that is, for any starting point u0∈C, let {up} be the sequence constructed below
{vp=PC(up−μAup),up+1=PC(up−μAvp),∀p≥0, |
where μ∈(0,1L) and L is Lipschitz constant of A. Whenever VI(C,A)≠∅, the sequence {up} converges weakly to a point in VI(C,A). Till now, the vast literature on the Korpelevich extragradient technique shows that many authors have paid great attention to it and enhanced it in different manners; see e.g., [1,2,3,4,5,6,7,9,10,12,13,14,15,16,17,18,20,21,22,23,24,25,27,28,29,30,31,36,37,38,39,40,41] and references therein.
Let Θ1,Θ2:C×C→R be two bifunctions and let B1,B2:C→H be two nonlinear mappings. In 2010, Ceng and Yao [35] considered the following problem of finding (u∗,v∗)∈C×C such that
{Θ1(u∗,u)+⟨B1v∗,u−u∗⟩+1μ1⟨u∗−v∗,u−u∗⟩≥0,∀u∈C,Θ2(v∗,v)+⟨B2u∗,v−v∗⟩+1μ2⟨v∗−u∗,v−v∗⟩≥0,∀v∈C, | (1.1) |
with μ1,μ2>0, which is called a system of generalized equilibrium problems (SGEP). In particular, if Θ1=Θ2=0, then the SGEP reduces to the following general system of variational inequalities (GSVI) considered in [6]: Find (u∗,v∗)∈C×C such that
{⟨μ1B1v∗+u∗−v∗,u−u∗⟩≥0,∀u∈C,⟨μ2B2u∗+v∗−u∗,v−v∗⟩≥0,∀v∈C. | (1.2) |
Note that SGEP (1.1) can be transformed into the fixed-point problem.
Lemma 1.2 ([35]). Let Θ1,Θ2:C×C→R be two bifunctions satisfying the hypotheses (H1)–(H4) and let the mappings B1,B2:C→H be α-inverse-strongly monotone and β-inverse-strongly monotone, respectively. Let μ1∈(0,2α) and μ2∈(0,2β), respectively. Then, for given u∗,v∗∈C, (u∗,v∗) is a solution of SGEP (1.1) if and only if x∗∈Fix(G), where Fix(G) is the fixed point set of the mapping G:=TΘ1μ1(I−μ1B1)TΘ2μ2(I−μ2B2), and v∗=TΘ2μ2(I−μ2B2)u∗.
On the other hand, suppose that the mappings B1,B2:C→H are α-inverse-strongly monotone and β-inverse-strongly monotone, respectively. Let f:C→C be contractive with constant δ∈[0,1) and F:C→H be κ-Lipschitzian and η-strongly monotone with constants κ,η>0 such that δ<ζ:=1−√1−μ(2η−μκ2)∈(0,1] for μ∈(0,2ηκ2). Let S:C→C be an asymptotically nonexpansive mapping with a sequence {θr} such that Ω:=Fix(S)∩Fix(G)≠∅, where Fix(G) is the fixed-point set of the mapping G:=PC(I−μ1B1)PC(I−μ2B2) for μ1∈(0,2α) and μ2∈(0,2β). Recently, Cai, Shehu and Iyiola [13] proposed the modified viscosity implicit rule for finding an element of Ω, that is, for any starting point x1∈C, let {xp} be the sequence constructed below
{up=βpxp+(1−βp)yp,vp=PC(up−μ2B2up),yp=PC(vp−μ1B1vp),xp+1=PC[αpf(xp)+(I−αpμF)Tpyp],∀p≥1, | (1.3) |
where {αp},{βp}⊂(0,1] s.t. (ⅰ) ∑∞p=1|αp+1−αp|<∞,∑∞p=1αp=∞; (ⅱ) limp→∞αp=0,limp→∞θpαp=0; (ⅲ) 0<ε≤βp≤1,∑∞p=1|βp+1−βp|<∞; and (ⅳ) ∑∞p=1‖Tp+1yp−Tpyp‖<∞. It was proved in [13] that the sequence {xp} converges strongly to an element u∗∈Ω, which is a unique solution of the hierarchical variational inequality (HVI): ⟨(μF−f)u∗,u−u∗⟩≥0,∀u∈Ω. Very recently, Reich et al. [29] suggested the modified projection-type method for solving the pseudomonotone VIP with uniform continuity mapping A. Let {αp}⊂(0,1) and f:C→C be contractive with constant δ∈[0,1). Given any initial x1∈C.
Algorithm 1.3 ([29]). Initial step: Let ν>0,ℓ∈(0,1),λ∈(0,1ν).
Iterations: Given the current iterate xp, calculate xp+1 as follows:
Step 1. Compute yp=PC(xp−λAxp) and Rλ(xp):=xp−yp. If Rλ(xp)=0, then stop; xp is a solution of VI(C,A). Otherwise,
Step 2. Compute wp=xp−τpRλ(xp), where τp:=ℓjp and jp is the smallest nonnegative integer j s.t. ν2‖Rλ(xp)‖2≥⟨Axp−A(xp−ℓjRλ(xp)),Rλ(xp)⟩.
Step 3. Compute xp+1=αpf(xp)+(1−αp)PCp(xp), where Cp:={x∈C:ℏp(xp)≤0} and ℏp(x)=⟨Awp,x−xp⟩+τp2λ‖Rλ(xp)‖2. Again set p:=p+1 and go to Step 1.
It was proven in [29] that under mild conditions, {xp} converges strongly to an element of VI(C,A). In a real Hilbert space H, we always assume that the SGEP, VIP, HVI and FPP represent a system of generalized equilibrium problems, a pseudomonotone variational inequality problem, a hierarchical variational inequality and a fixed-point problem of an asymptotically nonexpansive mapping, respectively. We introduce the modified Mann-like subgradient-like extragradient implicit rules with linear-search process for finding a common solution of the SGEP, VIP and FPP. The proposed algorithms are based on the subgradient extragradient rule with linear-search process, Mann implicit iteration approach, and hybrid deepest-descent technique. Under mild restrictions, we demonstrate the strong convergence of the proposed algorithms to a common solution of the SGEP, VIP and FPP, which is a unique solution of a certain HVI defined on their common solution set. In addition, an illustrated example is provided to illustrate the feasibility and implementability of our suggested rules.
The architecture of this article is constituted below: In Section 2, we present some concepts and basic tools for further use. Section 3 treats the convergence analysis of the suggested algorithms. Last, Section 4 applies our main results to solve the SGEP, VIP and FPP in an illustrated example. Our results improve and extend the ones associated with very recent literature, e.g., [13,17,29].
Let H be a real Hilbert space and ∅≠C⊂H be a convex and closed set. Given a sequence {uk}⊂H. We denote by the uk→u∗ (resp., uk⇀u∗) the strong (resp., weak) convergence of {uk} to u∗. For all u,v∈C, an operator Ψ:C→H is referred to as being
(a) L-Lipschitzian (or L-Lipschitz continuous) if ∃L>0 s.t. ‖Ψu−Ψv‖≤L‖u−v‖;
(b) pseudomonotone if ⟨Ψu,v−u⟩≥0⇒⟨Ψv,v−u⟩≥0;
(c) monotone if ⟨Ψu−Ψv,u−v⟩≥0;
(d) α-strongly monotone if ∃α>0 s.t. ⟨Ψu−Ψv,u−v⟩≥α‖u−v‖2;
(e) β-inverse-strongly monotone if ∃β>0 s.t. ⟨Ψu−Ψv,u−v⟩≥β‖Ψu−Ψv‖2;
(f) sequentially weakly continuous if ∀{vk}⊂C, the relation holds: vk⇀v⇒Ψvk⇀Ψv. It is clear that each monotone mapping is pseudomonotone but the converse is not true. Also, ∀v∈H, ∃| (nearest point) PC(v)∈C s.t. ‖v−PC(v)‖≤‖v−w‖∀w∈C. PC is called a nearest point (or metric) projection of H onto C. The following conclusions hold (see [19]):
(a) ⟨v−w,PC(v)−PC(w)⟩≥‖PC(v)−PC(w)‖2,∀v,w∈H;
(b) w=PC(v)⇔⟨v−w,u−w⟩≤0,∀v∈H,u∈C;
(c) ‖v−w‖2≥‖v−PC(v)‖2+‖w−PC(v)‖2,∀v∈H,w∈C;
(d) ‖v−w‖2=‖v‖2−‖w‖2−2⟨v−w,w⟩,∀v,w∈H;
(e) ‖sv+(1−s)w‖2=s‖v‖2+(1−s)‖w‖2−s(1−s)‖v−w‖2,∀v,w∈H,s∈[0,1].
The following inequality is an immediate consequence of the subdifferential inequality of the function 12‖⋅‖2:
‖u+v‖2≤‖u‖2+2⟨v,u+v⟩,∀u,v∈H. |
The following lemmas will be used for demonstrating our main results in the sequel.
Lemma 2.1. Let the mapping B:C→H be γ-inverse-strongly monotone. Then, for a given λ≥0,
‖(I−λB)u−(I−λB)v‖2≤‖u−v‖2−λ(2γ−λ)‖Bu−Bv‖2. |
In particular, if 0≤λ≤2γ, then I−λB is nonexpansive.
Using Lemma 1.1 and Lemma 2.1, we immediately derive the following lemma.
Lemma 2.2 ([35]). Let Θ1,Θ2:C×C→R be two bifunctions satisfying the hypotheses (H1)–(H4), and the mappings B1,B2:C→H be α-inverse-strongly monotone and β-inverse-strongly monotone, respectively. Let the mapping G:C→C be defined as G:=TΘ1μ1(I−μ1B1)TΘ2μ2(I−μ2B2). Then G:C→C is a nonexpansive mapping provided 0<μ1≤2α and 0<μ2≤2β.
In particular, if Θ1=Θ2=0, using Lemma 1.1 we deduce that TΘ1μ1=TΘ2μ2=PC. Thus, from Lemma 2.2 we obtain the corollary below.
Corollary 2.3 ([6]). Let the mappings B1,B2:C→H be α-inverse-strongly monotone and β-inverse-strongly monotone, respectively. Let the mapping G:C→C be defined as G:=PC(I−μ1B1)PC(I−μ2B2). If 0<μ1≤2α and 0<μ2≤2β, then G:C→C is nonexpansive.
Lemma 2.4 ([6]). Let A:C→H be pseudomonotone and continuous. Then u∈C is a solution to the VIP ⟨Au,υ−u⟩≥0,∀υ∈C, if and only if ⟨Aυ,υ−u⟩≥0,∀υ∈C.
Lemma 2.5 ([8]). Let {al} be a sequence of nonnegative numbers satisfying the conditions: al+1≤(1−λl)al+λlγl∀l≥1, where {λl} and {γl} are sequences of real numbers such that (i) {λl}⊂[0,1] and ∑∞l=1λl=∞, and (ii) lim supl→∞γl≤0 or ∑∞l=1|λlγl|<∞. Then liml→∞al=0.
Later on, we will make use of the following lemmas to demonstrate our main results.
Lemma 2.6 ([32]). Let H1 and H2 be two real Hilbert spaces. Suppose that A:H1→H2 is uniformly continuous on bounded subsets of H1 and M is a bounded subset of H1. Then, A(M) is bounded.
Lemma 2.7 ([33]). Let h be a real-valued function on H and define K:={x∈C:h(x)≤0}. If K is nonempty and h is Lipschitz continuous on C with modulus θ>0, then dist(x,K)≥θ−1max{h(x),0}∀x∈C, where dist(x,K) denotes the distance of x to K.
Lemma 2.8 ([11]). Let X be a Banach space which admits a weakly continuous duality mapping, C be a nonempty closed convex subset of X, and T:C→C be an asymptotically nonexpansive mapping with Fix(T)≠∅. Then I−T is demiclosed at zero, i.e., if {uk} is a sequence in C such that uk⇀u∈C and (I−T)uk→0, then (I−T)u=0, where I is the identity mapping of X.
The following lemmas are very crucial to the convergence analysis of the proposed algorithms.
Lemma 2.9 ([30]). Let {Λm} be a sequence of real numbers that does not decrease at infinity in the sense that, ∃{Λmk}⊂{Λm} s.t. Λmk<Λmk+1∀k≥1. Let the sequence {ϕ(m)}m≥m0 of integers be formulated below:
ϕ(m)=max{k≤m:Λk<Λk+1}, |
with integer m0≥1 satisfying {k≤m0:Λk<Λk+1}≠∅. Then there hold the statements below:
(i) ϕ(m0)≤ϕ(m0+1)≤⋯ and ϕ(m)→∞;
(ii) Λϕ(m)≤Λϕ(m)+1 and Λm≤Λϕ(m)+1,∀m≥m0.
Lemma 2.10 ([8]). Let λ∈(0,1] and Let S:C→C be a nonexpansive mapping. Let Sλ:C→H be the mapping formulated by Sλu:=(I−λμF)Su∀u∈C with F:C→H being κ-Lipschitzian and η-strongly monotone. Then Sλ is a contraction provided 0<μ<2ηκ2, i.e., ‖Sλu−Sλv‖≤(1−λτ)‖u−v‖,∀u,v∈C, where τ=1−√1−μ(2η−μκ2)∈(0,1].
In this section, let the feasible set C be a nonempty closed convex subset of a real Hilbert space H, and assume always that the following conditions hold:
(1) S:C→C is an asymptotically nonexpansive mapping with a sequence {θn}, and A:H→H is pseudomonotone and uniformly continuous on C, s.t. ‖Az‖≤lim infn→∞ ‖Aun‖ for each {un}⊂C with un⇀z.
(2) Θ1,Θ2:C×C→R are two bifunctions satisfying the hypotheses (H1)–(H4), and B1,B2:C→H are α-inverse-strongly monotone and β-inverse-strongly monotone, respectively.
(3) Ω=Fix(S)∩Fix(G)∩VI(C,A)≠∅ where G:=TΘ1μ1(I−μ1B1)TΘ2μ2(I−μ2B2) for μ1∈(0,2α) and μ2∈(0,2β).
(4) f:C→H is a contraction with constant δ∈[0,1), and F:C→H is η-strongly monotone and κ-Lipschitzian such that δ<τ:=1−√1−μ(2η−μκ2) for μ∈(0,2ηκ2).
(5) {σn},{αn}⊂(0,1] and {βn}⊂[0,1] are three real number sequences satisfying
(ⅰ) ∑∞n=1αn=∞,limn→∞αn=0 and limn→∞θnαn=0;
(ⅱ) 0<lim infn→∞βn≤lim supn→∞βn<1;
(ⅲ) lim supn→∞σn<1.
Algorithm 3.1. Initial step: Given ν>0,ℓ∈(0,1),λ∈(0,1ν). Let x1∈C be arbitrary.
Iterations: Given the current iterate xn, calculate xn+1 below:
Step 1. Calculate wn=σnxn+(1−σn)un with
vn=TΘ2μ2(wn−μ2B2wn), |
un=TΘ1μ1(vn−μ1B1vn). |
Step 2. Calculate yn=PC(wn−λAwn) and Rλ(wn):=wn−yn.
Step 3. Calculate tn=wn−τnRλ(wn), where τn:=ℓjn and jn is the smallest nonnegative integer j satisfying
⟨Awn−A(wn−ℓjRλ(wn)),wn−yn⟩≤ν2‖Rλ(wn)‖2. | (3.1) |
Step 4. Compute zn=PCn(wn) and xn+1=βnxn+(1−βn)PC[αnf(xn)+(I−αnμF)Snzn], where Cn:={u∈C:ℏn(u)≤0} and
ℏn(u)=⟨Atn,u−wn⟩+τn2λ‖Rλ(wn)‖2. | (3.2) |
Again put n:=n+1 and return to Step 1.
Lemma 3.2. The Armijo-type search approach (3.1) is well formulated, and the relation holds: λ−1‖Rλ(wn)‖2≤⟨Rλ(wn),Awn⟩.
Proof. Since ℓ∈(0,1) and A is of uniform continuity on C, it is clear that limj→∞⟨Awn−A(wn−ℓjRλ(wn)),Rλ(wn)⟩=0. If Rλ(wn)=0, one gets jn=0. Otherwise, from Rλ(wn)≠0, it follows that ∃ (integer) jn≥0 fulfilling (3.1). It is readily known that the firm nonexpansivity of PC implies ⟨u−PCv,u−v⟩≥‖u−PCv‖2,∀u∈C,v∈H. Setting v=wn−λAwn and u=wn, one has λ⟨wn−PC(wn−λAwn),Awn⟩≥‖wn−PC(wn−λAwn)‖2. Hence the relation holds.
Lemma 3.3. Suppose that ℏn is the function formulated in (3.2). Then, ℏn(υ)≤0∀υ∈Ω. In addition, when Rλ(wn)≠0, one has ℏn(wn)>0.
Proof. It suffices to show the former claim of Lemma 3.3 because the latter claim is clear. In fact, pick an arbitrary υ∈Ω. By Lemma 2.4 one gets ⟨Atn,tn−υ⟩≥0. Thus, one has
ℏn(υ)=⟨Atn,υ−wn⟩+τn2λ‖Rλ(wn)‖2=⟨Atn,tn−wn⟩+⟨Atn,υ−tn⟩+τn2λ‖Rλ(wn)‖2≤−τn⟨Atn,Rλ(wn)⟩+τn2λ‖Rλ(wn)‖2. | (3.3) |
Meanwhile, from (3.1) it follows that ν2‖Rλ(wn)‖2≥⟨Awn−Atn,Rλ(wn)⟩. So, from Lemma 3.2 one gets
⟨Atn,Rλ(wn)⟩≥−ν2‖Rλ(wn)‖2+⟨Rλ(wn),Awn⟩≥(−ν2+1λ)‖Rλ(wn)‖2. | (3.4) |
This along with (3.3), arrives at
ℏn(υ)≤−τn2(1λ−ν)‖Rλ(wn)‖2. | (3.5) |
Therefore, we derive the desired result.
Lemma 3.4. Let {wn},{xn},{yn},{zn} be the bounded sequences constructed in Algorithm 3.1. Assume that xn−xn+1→0,xn−Gwn→0,wn−yn→0 and xn−zn→0. If Snxn−Sn+1xn→0 and ∃{xnk}⊂{xn} such that xnk⇀z∈C, then z∈Ω.
Proof. From Algorithm 3.1, we get wn−xn=(1−σn)(un−xn)∀n≥1, and hence ‖wn−xn‖=(1−σn)‖un−xn‖≤‖un−xn‖. Utilizing the assumption un−xn→0, we have
limn→∞‖wn−xn‖=0. | (3.6) |
Putting qn:=αnf(xn)+(I−αnμF)Snzn, by Algorithm 3.1 we know that xn+1=βnxn+(1−βn)PC(qn) and qn−Snzn=αnf(xn)−αnμFSnzn. Hence one gets
‖xn−Snzn‖≤‖xn−xn+1‖+‖xn+1−Snzn‖≤‖xn−xn+1‖+βn‖xn−Snzn‖+(1−βn)‖qn−Snzn‖≤‖xn−xn+1‖+βn‖xn−Snzn‖+αn‖f(xn)‖+αn‖μFSnzn‖. |
This immediately ensures that
(1−βn)‖xn−Snzn‖≤‖xn−xn+1‖+αn‖f(xn)‖+αn‖μFSnzn‖. |
Since xn−xn+1→0,αn→0 and lim infn→∞(1−βn)>0, by the boundedness of {xn},{zn} we obtain
limn→∞‖xn−Snzn‖=0. |
We claim that limn→∞‖xn−Sxn‖=0. In fact, using the asymptotical nonexpansivity of S, one deduces that
‖xn−Sxn‖≤‖xn−Snzn‖+‖Snzn−Snxn‖+‖Snxn−Sn+1xn‖+‖Sn+1xn−Sn+1zn‖+‖Sn+1zn−Sxn‖≤‖xn−Snzn‖+(1+θn)‖zn−xn‖+‖Snxn−Sn+1xn‖+(1+θn+1)‖xn−zn‖+(1+θ1)‖Snzn−xn‖=(2+θ1)‖xn−Snzn‖+(2+θn+θn+1)‖zn−xn‖+‖Snxn−Sn+1xn‖. |
Since xn−zn→0, xn−Snzn→0 and Snxn−Sn+1xn→0, we obtain
limn→∞‖xn−Sxn‖=0. | (3.7) |
Also, let us show that limn→∞‖xn−Gxn‖=0. In fact, by Lemma 2.2 we know that G:C→C is nonexpansive for μ1∈(0,2α) and μ2∈(0,2β). Again from Algorithm 3.1, we have un=Gwn. Since
‖Gxn−xn‖≤‖Gxn−Gwn‖+‖Gwn−xn‖≤‖xn−wn‖+‖un−xn‖, |
Noticing un−xn→0 and xn−wn→0 (due to (3.6)), we obtain
limn→∞‖Gxn−xn‖=0. | (3.8) |
Next, let us show z∈VI(C,A). Indeed, noticing xn−wn→0 and xnk⇀z, we know that wnk⇀z. Since C is convex and closed, from {wn}⊂C and wnk⇀z we get z∈C. In what follows, we consider two cases. In the case of Az=0, it is clear that z∈VI(C,A) because ⟨Az,y−z⟩≥0,∀y∈C. In the case of Az≠0, it follows from wn−xn→0 and xnk⇀z that wnk⇀z as k→∞. Utilizing the assumption on A, instead of the sequentially weak continuity of A, we get 0<‖Az‖≤lim infk→∞‖Awnk‖. So, we might assume that ‖Awnk‖≠0∀k≥1. On the other hand, from yn=PC(wn−λAwn), one has ⟨wn−λAwn−yn,x−yn⟩≤0,∀x∈C, and hence
1λ⟨wn−yn,x−yn⟩+⟨Awn,yn−wn⟩≤⟨Awn,x−wn⟩,∀x∈C. | (3.9) |
In the light of the uniform continuity of A on C, one knows that {Awn} is bounded (due to Lemma 2.6). Note that {yn} is bounded as well. Thus, from (3.9) we get lim infk→∞⟨Awnk,x−wnk⟩≥0∀x∈C.
To show that z∈VI(C,A), we now choose a sequence {γk}⊂(0,1) satisfying γk↓0 as k→∞. For each k≥1, we denote by lk the smallest positive integer such that
⟨Awnj,x−wnj⟩+γk≥0,∀j≥lk. | (3.10) |
Because {γk} is decreasing, it is readily known that {lk} is increasing. Note that Awlk≠0∀k≥1 (due to {Awlk}⊂{Awnk}). Then one puts υlk=Awlk‖Awlk‖2, one gets ⟨Awlk,υlk⟩=1,∀k≥1. So, using (3.10) one has ⟨Awlk,x+γkυlk−wlk⟩≥0,∀k≥1. Again from the pseudo-monotonicity of A one has ⟨A(x+γkυlk),x+γkυlk−wlk⟩≥0,∀k≥1. This immediately arrives at
⟨Ax,x−wlk⟩≥⟨Ax−A(x+γkυlk),x+γkυlk−wlk⟩−γk⟨Ax,υlk⟩,∀k≥1. | (3.11) |
We claim that limk→∞γkυlk=0. In fact, from xnk⇀z∈C and wn−xn→0, we obtain wnk⇀z. Note that {wlk}⊂{wnk} and γk↓0 as k→∞. So it follows that
0≤lim supk→∞‖γkυlk‖=lim supk→∞γk‖Awlk‖≤lim supk→∞γklim infk→∞‖Awnk‖=0. |
Hence one gets γkυlk→0 as k→∞. Thus, letting k→∞, we deduce that the right-hand side of (3.11) tends to zero by the uniform continuity of A, the boundedness of {wlk},{υlk} and the limit limk→∞γkυlk=0. Therefore, ⟨Ax,x−z⟩=lim infk→∞⟨Ax,x−wlk⟩≥0∀x∈C. Using Lemma 2.4 one has z∈VI(C,A).
Last, we claim that z∈Ω. In fact, because (3.7) yields xnk−Sxnk→0. By Lemma 2.8 one knows that I−S is demiclosed at zero. So, from xnk⇀z it follows that (I−S)z=0, i.e., z∈Fix(S). Besides, let us claim that z∈Fix(G). Actually, by Lemma 2.8 we deduce that I−G is demiclosed at zero. Thus, from (3.8) and xnk⇀z one has (I−G)z=0, i.e., z∈Fix(G). Accordingly, z∈Fix(S)∩Fix(G)∩VI(C,A)=Ω. This completes the proof.
Lemma 3.5. Let {wn} be the sequence constructed in Algorithm 3.1. Then,
limn→∞τn‖Rλ(wn)‖2=0⇒limn→∞‖Rλ(wn)‖=0. | (3.12) |
Proof. We claim that lim supn→∞‖Rλ(wn)‖=0. Conversely, suppose that lim supn→∞ ‖Rλ(wn)‖=d>0. Then, ∃{np}⊂{n} s.t. limp→∞‖Rλ(wnp)‖=d>0. Note that limp→∞τnp‖Rλ(wnp)‖2=0. First, if lim infp→∞τnp>0, we might assume that ∃ξ>0 s.t. τnp≥ξ>0,∀p≥1. So it follows that
‖Rλ(wnp)‖2=1τnpτnp‖Rλ(wnp)‖2≤1ξ⋅τnp‖Rλ(wnp)‖2=1ξ⋅τnp‖Rλ(wnp)‖2, | (3.13) |
which immediately leads to
0<d2=limp→∞‖Rλ(wnp)‖2≤limp→∞{1ξ⋅τnp‖Rλ(wnp)‖2}=0. |
So, this reaches at a contradiction.
If lim infp→∞τnp=0, there exists a subsequence of {τnp}, still denoted by {τnp}, s.t. limp→∞τnp=0. We now set
υnp:=1ℓτnpynp+(1−1ℓτnp)wnp=wnp−1ℓτnp(wnp−ynp). |
Then, from limp→∞τnp‖Rλ(wnp)‖2=0 we infer that
limp→∞‖υnp−wnp‖2=limp→∞1ℓ2τnp⋅τnp‖Rλ(wnp)‖2=0. | (3.14) |
Using the stepsize rule (3.1), one gets ⟨Awnp−Aυnp,Rλ(wnp)⟩>ν2‖Rλ(wnp)‖2. Since A is uniformly continuous on bounded subsets of C, (3.14) guarantees that
limp→∞‖Awnp−Aυnp‖=0, | (3.15) |
which hence attains limp→∞‖Rλ(wnp)‖=0. So, this reaches a contradiction. Therefore, Rλ(wn)→0 as n→∞. This completes the proof.
Theorem 3.6. Suppose that {xn} is the sequence constructed in Algorithm 3.1. Then xn→x∗∈Ω provided Snxn−Sn+1xn→0, with x∗∈Ω being only a solution to the HVI
⟨(μF−f)x∗,y−x∗⟩≥0,∀y∈Ω. |
Proof. First of all, noticing 0<lim infn→∞βn≤lim supn→∞βn<1 and limn→∞θnαn=0, we may assume, without loss of generality, that {σn}⊂[a,b]⊂(0,1) and θn≤αn(τ−δ)2,∀n≥1. We claim that PΩ(I−μF+f):C→C is contractive map. In fact, using Lemma 2.10, one has
‖PΩ(I−μF+f)u−PΩ(I−μF+f)v‖≤[1−(τ−δ)]‖u−v‖,∀u,v∈C. |
This ensures that PΩ(I−μF+f) is contractive. Banach's Contraction Mapping Principle guarantees that there exists a unique fixed point of PΩ(I−μF+f) in C. Say x∗∈C s.t. x∗=PΩ(I−μF+f)x∗. That is, ∃∣ (solution) x∗∈Ω=Fix(S)∩Fix(G)∩VI(C,A) of the HVI
⟨(μF−f)x∗,y−x∗⟩≥0,∀y∈Ω. | (3.16) |
Next we demonstrate the conclusion of the theorem. To the goal, we divide the remainder of the proof into several aspects.
Aspect 1. We assert that {xn} is of boundedness. Indeed, for x∗∈Ω=Fix(S)∩Fix(G)∩VI(C,A) we have Sx∗=x∗,Gx∗=x∗ and PC(x∗−λAx∗)=x∗. We observe that
‖zn−x∗‖2=‖PCn(wn)−x∗‖2≤‖wn−x∗‖2−‖wn−PCn(wn)‖2=‖wn−x∗‖2−dist2(wn,Cn), | (3.17) |
which hence leads to
‖zn−x∗‖≤‖wn−x∗‖,∀n≥1. | (3.18) |
Using Lemma 2.2, one knows that G=Tθ1μ1(I−μ1B1)Tθ2μ2(I−μ2B2) is nonexpansive for μ1∈(0,2α) and μ2∈(0,2β). Thus, by the definition of wn, one gets
‖wn−x∗‖≤σn‖xn−x∗‖+(1−σn)‖Gwn−x∗‖≤σn‖xn−x∗‖+(1−σn)‖wn−x∗‖, |
which immediately yields
‖wn−x∗‖≤‖xn−x∗‖,∀n≥1. |
This together with (3.18), yields
‖zn−x∗‖≤‖wn−x∗‖≤‖xn−x∗‖,∀n≥1. | (3.19) |
Thus, using (3.19), from Lemma 2.10 we obtain
‖xn+1−x∗‖≤βn‖xn−x∗‖+(1−βn)‖αnf(xn)+(I−αnμF)Snzn−x∗‖=βn‖xn−x∗‖+(1−βn)‖αn(f(xn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗+αn(f−μF)x∗‖≤βn‖xn−x∗‖+(1−βn){αnδ‖xn−x∗‖+(1−αnτ)(1+θn)‖zn−x∗‖+αn‖(f−μF)x∗‖}≤βn‖xn−x∗‖+(1−βn){[αnδ+(1−αnτ)+θn]‖xn−x∗‖+αn‖(f−μF)x∗‖}≤βn‖xn−x∗‖+(1−βn){[1−αn(τ−δ)+αn(τ−δ)2]‖xn−x∗‖+αn‖(f−μF)x∗‖}=βn‖xn−x∗‖+(1−βn){[1−αn(τ−δ)2]‖xn−x∗‖+αn‖(f−μF)x∗‖}=[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖+αn(1−βn)(τ−δ)2⋅2‖(f−μF)x∗‖τ−δ≤max{‖xn−x∗‖,2‖(f−μF)x∗‖τ−δ}. |
By induction, we get
‖xn−x∗‖≤max{‖x1−x∗‖,2‖(f−μF)x∗‖τ−δ},∀n≥1. |
Thus, {xn} is bounded, and so are the sequences {wn},{yn},{zn},{f(xn)},{Atn},{Gwn},{Snzn}.
Aspect 2. We assert that
(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖qn−PC(qn)‖2}+βn(1−βn)‖xn−PC(qn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1, |
for some M1>0. In fact, noticing zn=PCn(wn) and wn=σnxn+(1−σn)un, we obtain
‖zn−x∗‖2≤‖wn−x∗‖2−‖wn−zn‖2≤σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2. |
Since xn+1=βnxn+(1−βn)PC(qn) where qn=αnf(xn)+(I−αnμF)Snzn, by Lemma 2.10 and the convexity of the function h(s)=s2,∀s∈R we deduce that
‖xn+1−x∗‖2=βn‖xn−x∗‖2+(1−βn)‖PC(qn)−x∗‖2−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){‖qn−x∗‖2−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2=βn‖xn−x∗‖2+(1−βn){‖αn(f(xn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗+αn(f−μF)x∗‖2−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){‖αn(f(xn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){[αn‖f(xn)−f(x∗)‖+‖(I−αnμF)Snzn−(I−αnμF)x∗‖]2+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){[αnδ‖xn−x∗‖+(1−αnτ)(1+θn)‖zn−x∗‖]2+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){[αnδ‖xn−x∗‖+[(1−αnτ)+θn]‖zn−x∗‖]2+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]‖zn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn][σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2]+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2, | (3.20) |
(due to αnδ+(1−αnτ)+θn≤1−αn(τ−δ)+αn(τ−δ)2=1−αn(τ−δ)2≤1), which together with un=Gwn and (3.19), ensures that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩+[(1−αnτ)+αn(τ−δ)2][σn‖xn−x∗‖2+(1−σn)‖xn−x∗‖2−‖wn−zn‖2]−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn){‖qn−PC(qn)‖2+[1−αn(τ+δ)2]‖wn−zn‖2}−βn(1−βn)‖xn−PC(qn)‖2+2αn(1−βn)⟨(f−μF)x∗,qn−x∗⟩≤‖xn−x∗‖2−(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2+αnM1, | (3.21) |
where supn≥12‖(f−μF)x∗‖‖qn−x∗‖≤M1 for some M1>0. This attains the desired assertion.
Aspect 3. We assert that
(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1. |
In fact, we claim that for some ˉL>0,
‖zn−x∗‖2≤‖wn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2. | (3.22) |
Noticing the boundedness of {Atn}, one knows that ∃ˉL>0 s.t. ‖Atn‖≤ˉL,∀n≥1. This implies that
|ℏn(u)−ℏn(v)|=|⟨Atn,u−v⟩|≤‖Atn‖‖u−v‖≤ˉL‖u−v‖,∀u,v∈Cn, |
which hence guarantees that ℏn(⋅) is ˉL-Lipschitz continuous on Cn. By Lemmas 2.7 and 3.3, we have
dist(wn,Cn)≥1ˉLℏn(wn)=τn2λˉL‖Rλ(wn)‖2. | (3.23) |
Combining (3.17) and (3.23) yields
‖zn−x∗‖2≤‖wn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2. |
From (3.20), (3.19) and (3.22) it follows that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]‖zn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn][‖wn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2]+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+αn(τ−δ)2][‖xn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2]+2αn⟨(f−μF)x∗,qn−x∗⟩}=[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2+2αn(1−βn)⟨(f−μF)x∗,qn−x∗⟩≤‖xn−x∗‖2−(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2+αnM1. |
This hence leads to
(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1. |
Aspect 4. We assert that
‖xn+1−x∗‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,qn−x∗⟩τ−δ+θnαn⋅Mτ−δ] | (3.24) |
for some M>0. In fact, from Lemma 2.10 and (3.19), one obtains
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn)‖qn−x∗‖2=βn‖xn−x∗‖2+(1−βn)‖αn(f(xn)−f(x∗))+αn(f−μF)x∗+(I−αnμF)Snzn−(I−αnμF)x∗‖2≤(1−βn){‖αn(f(xn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗‖2+βn‖xn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤(1−βn){[αn‖f(xn)−f(x∗)‖+‖(I−αnμF)Snzn−(I−αnμF)x∗‖]2+βn‖xn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){[αnδ‖xn−x∗‖+(1−αnτ)(1+θn)‖zn−x∗‖]2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){[αnδ‖xn−x∗‖+(1−αnτ+θn)‖zn−x∗‖]2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+(1−αnτ+θn)‖zn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+(1−βn){θn‖xn−x∗‖2+2αn⟨(f−μF)x∗,qn−x∗⟩}≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,qn−x∗⟩τ−δ+θnαn⋅Mτ−δ], |
where supn≥1‖xn−x∗‖2≤M for some M>0.
Aspect 5. We assert that xn→x∗∈Ω, which is only a solution of the HVI (3.16).
In fact, from (3.24), we have
‖xn+1−x∗‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,qn−x∗⟩τ−δ+θnαn⋅Mτ−δ]. | (3.25) |
Setting Λn=‖xn−x∗‖2, we demonstrate the convergence of {Λn} to zero by the following two situations.
Situation 1. ∃(integer)n0≥1 s.t. {Λn} is nonincreasing. It is clear that the limit limn→∞Λn=k<+∞ and limn→∞(Λn−Λn+1)=0. From Aspect 2 and {βn}⊂[a,b]⊂(0,1) we obtain
(1−b){[1−αn(τ+δ)2]‖wn−zn‖2+‖qn−PC(qn)‖2}+a(1−b)‖xn−PC(qn)‖2≤(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖qn−PC(qn)‖2}+βn(1−βn)‖xn−PC(qn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
Thanks to the facts that αn→0 and Λn−Λn+1→0, from τ+δ2∈(0,1) one deduces that
limn→∞‖wn−zn‖=limn→∞‖qn−PC(qn)‖=limn→∞‖xn−PC(qn)‖=0. | (3.26) |
Hence it is readily known that
‖xn−qn‖≤‖xn−PC(qn)‖+‖PC(qn)−qn‖→0(n→∞), |
‖xn+1−xn‖=(1−βn)‖PC(qn)−xn‖≤‖qn−xn‖→0(n→∞), |
and
‖Snzn−xn‖=‖qn−xn−αn(f(xn)−μFSnzn)‖≤‖qn−xn‖+αn(‖f(xn)‖+μ‖FSnzn‖)→0(n→∞). |
Next, we show that ‖xn−un‖→0 as n→∞. Indeed, note that y∗=TΘ2μ2(x∗−μ2B2x∗), vn=TΘ2μ2(wn−μ2B2wn) and un=TΘ1μ1(vn−μ1B1vn). Then un=Gwn. By Lemma 2.1 we have
‖vn−y∗‖2≤‖wn−x∗‖2−μ2(2β−μ2)‖B2wn−B2x∗‖2 |
and
‖un−x∗‖2≤‖vn−y∗‖2−μ1(2α−μ1)‖B1vn−B1y∗‖2. |
Combining the last two inequalities, from (3.19) we obtain
‖un−x∗‖2≤‖xn−x∗‖2−μ2(2β−μ2)‖B2wn−B2x∗‖2−μ1(2α−μ1)‖B1vn−B1y∗‖2. |
This together with (3.20), implies that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn][σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2]+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+αn(τ−δ)2]×[σn‖xn−x∗‖2+(1−σn)(‖xn−x∗‖2−μ2(2β−μ2)‖B2wn−B2x∗‖2−μ1(2α−μ1)‖B1vn−B1y∗‖2)]+αnM1}≤[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]×{μ2(2β−μ2)‖B2wn−B2x∗‖2+μ1(2α−μ1)‖B1vn−B1y∗‖2}+αnM1≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]{μ2(2β−μ2)‖B2wn−B2x∗‖2+μ1(2α−μ1)‖B1vn−B1y∗‖2}+αnM1, |
which immediately arrives at
(1−βn)(1−σn)[1−αn(τ+δ)2]{μ2(2β−μ2)‖B2wn−B2x∗‖2+μ1(2α−μ1)‖B1vn−B1y∗‖2}≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
Since βn≤b<1,μ1∈(0,2α),μ2∈(0,2β),limn→∞αn=0 and lim supn→∞σn<1, we obtain from Λn−Λn+1→0 that
limn→∞‖B2wn−B2x∗‖=0andlimn→∞‖B1vn−B1y∗‖=0. | (3.27) |
On the other hand, by Lemma 1.1 one has
‖un−x∗‖2≤⟨vn−y∗,un−x∗⟩+μ1⟨B1y∗−B1vn,un−x∗⟩≤12[‖vn−y∗‖2+‖un−x∗‖2−‖vn−un+x∗−y∗‖2]+μ1‖B1y∗−B1vn‖‖un−x∗‖, |
which hence leads to
‖un−x∗‖2≤‖vn−y∗‖2−‖vn−un+x∗−y∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖. |
Similarly, one gets
‖vn−y∗‖2≤‖wn−x∗‖2−‖wn−vn+y∗−x∗‖2+2μ2‖B2x∗−B2wn‖‖vn−y∗‖. |
Combining the last two inequalities, from (3.19) we get
‖un−x∗‖2≤‖xn−x∗‖2−‖wn−vn+y∗−x∗‖2−‖vn−un+x∗−y∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖. |
This together with (3.20), ensures that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]×[σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2]+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+αn(τ−δ)2][σn‖xn−x∗‖2+(1−σn)(‖xn−x∗‖2−‖wn−vn+y∗−x∗‖2−‖vn−un+x∗−y∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖)]+αnM1}≤[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]{‖wn−vn+y∗−x∗‖2+‖vn−un+x∗−y∗‖2}+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]{‖wn−vn+y∗−x∗‖2+‖vn−un+x∗−y∗‖2}+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1, |
which immediately leads to
(1−βn)(1−σn)[1−αn(τ+δ)2]{‖wn−vn+y∗−x∗‖2+‖vn−un+x∗−y∗‖2}≤‖xn−x∗‖2−‖xn+1−x∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1=Λn−Λn+1+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1. |
Since βn≤b<1,limn→∞αn=0, lim supn→∞σn<1 and lim supn→∞(Λn−Λn+1)=0, we deduce from (3.27) and the boundedness of {un},{vn} that
limn→∞‖wn−vn+y∗−x∗‖=0andlimn→∞‖vn−un+x∗−y∗‖=0. |
Therefore,
‖wn−Gwn‖=‖wn−un‖≤‖wn−vn+y∗−x∗‖+‖vn−un+x∗−y∗‖→0(n→∞). | (3.28) |
Noticing wn=σnxn+(1−σn)un, we get
‖wn−x∗‖2=σn⟨xn−x∗,wn−x∗⟩+(1−σn)⟨un−x∗,wn−x∗⟩≤σn⟨xn−x∗,wn−x∗⟩+(1−σn)‖wn−x∗‖2, |
which immediately yields
‖wn−x∗‖2≤⟨xn−x∗,wn−x∗⟩≤12[‖xn−x∗‖2+‖wn−x∗‖2−‖xn−wn‖2]. |
So it follows that
‖wn−x∗‖2≤‖xn−x∗‖2−‖xn−wn‖2, |
which together with (3.20), leads to
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]×[σn‖xn−x∗‖2+(1−σn)‖wn−x∗‖2]+2αn⟨(f−μF)x∗,qn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+αn(τ−δ)2]×[σn‖xn−x∗‖2+(1−σn)(‖xn−x∗‖2−‖xn−wn‖2)]+αnM1}≤[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]×‖xn−wn‖2+αnM1≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]‖xn−wn‖2+αnM1. |
This hence arrives at
(1−βn)(1−σn)[1−αn(τ+δ)2]‖xn−wn‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
Since βn≤b<1,limn→∞αn=0, lim supn→∞σn<1 and lim supn→∞(Λn−Λn+1)=0, we deduce from τ+δ2∈(0,1) that
limn→∞‖xn−wn‖=0. |
So it follows from (3.26) and (3.28) that
‖xn−zn‖≤‖xn−wn‖+‖wn−zn‖→0(n→∞), | (3.29) |
and
‖xn−Gwn‖≤‖xn−wn‖+‖wn−Gwn‖→0(n→∞). | (3.30) |
Meanwhile, from Aspect 3 we obtain
(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
Noticing βn≤b<1,αn→0 and Λn−Λn+1→0, one gets
limn→∞[τn2λˉL‖Rλ(wn)‖2]2=0, |
which together with Lemma 3.5, yields
limn→∞‖wn−yn‖=0. | (3.31) |
By the boundedness of {xn}, we know that ∃subsequence{xnp}⊂{xn} s.t.
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=limp→∞⟨(f−μF)x∗,xnp−x∗⟩. | (3.32) |
Since H is reflexive and {xn} is bounded, we might assume that xnp⇀ˉx. Thus, from (3.32) one has
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=limp→∞⟨(f−μF)x∗,xnp−x∗⟩=⟨(f−μF)x∗,ˉx−x∗⟩. | (3.33) |
Since xn−xn+1→0,xn−Gwn→0,wn−yn→0,xn−zn→0 and xnp⇀ˉx, by Lemma 3.4 we infer that ˉx∈Ω. Thus, using (3.16) and (3.33) one has
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=⟨(f−μF)x∗,ˉx−x∗⟩≤0, | (3.34) |
which together with (3.26), arrives at
lim supn→∞⟨(f−μF)x∗,qn−x∗⟩=lim supn→∞[⟨(f−μF)x∗,qn−PC(qn)+PC(qn)−xn⟩+⟨(f−μF)x∗,xn−x∗⟩]≤lim supn→∞[‖(f−μF)x∗‖(‖qn−PC(qn)‖+‖PC(qn)−xn‖)+⟨(f−μF)x∗,xn−x∗⟩]≤0. | (3.35) |
Note that {αn(1−βn)(τ−δ)}⊂[0,1],∑∞n=1αn(1−βn)(τ−δ)=∞, and
lim supn→∞[2⟨(f−μF)x∗,qn−x∗⟩τ−δ+θnαn⋅Mτ−δ]≤0. |
Consequently, applying Lemma 2.5 to (3.25), one has limn→∞‖xn−x∗‖2=0.
Situation 2. ∃{Λnp}⊂{Λn} s.t. Λnp<Λnp+1∀p∈N, with N being the set of all natural numbers. Let ϕ:N→N be formulated as
ϕ(n):=max{p≤n:Λp<Λp+1}. |
Using Lemma 2.9, we have
Λϕ(n)≤Λϕ(n)+1andΛn≤Λϕ(n)+1. |
By Aspect 2 one gets
(1−b){[1−αϕ(n)(τ+δ)2]‖wϕ(n)−zϕ(n)‖2+‖qϕ(n)−PC(qϕ(n))‖2}+a(1−b)‖xϕ(n)−PC(qϕ(n))‖2≤(1−βϕ(n)){[1−αϕ(n)(τ+δ)2]‖wϕ(n)−zϕ(n)‖2+‖qϕ(n)−PC(qϕ(n))‖2}+βϕ(n)(1−βϕ(n))‖xϕ(n)−PC(qϕ(n))‖2≤‖xϕ(n)−x∗‖2−‖xϕ(n)+1−x∗‖2+αϕ(n)M1=Λϕ(n)−Λϕ(n)+1+αϕ(n)M1, | (3.36) |
which immediately ensures that
limn→∞‖wϕ(n)−zϕ(n)‖=limn→∞‖qϕ(n)−PC(qϕ(n))‖=limn→∞‖xϕ(n)−PC(qϕ(n))‖=0. |
By Aspect 3 we have
(1−βϕ(n))[1−αϕ(n)(τ+δ)2][τϕ(n)2λˉL‖Rλ(wϕ(n))‖2]2≤‖xϕ(n)−x∗‖2−‖xϕ(n)+1−x∗‖2+αϕ(n)M1=Λϕ(n)−Λϕ(n)+1+αϕ(n)M1, |
which hence leads to
limn→∞[τϕ(n)2λˉL‖Rλ(wϕ(n))‖2]2=0. |
Using the similar arguments to those of Situation 1, we infer that limn→∞‖xϕ(n)+1−xϕ(n)‖=0,
limn→∞‖xϕ(n)−Gwϕ(n)‖=limn→∞‖wϕ(n)−yϕ(n)‖=limn→∞‖xϕ(n)−zϕ(n)‖=0, |
and
lim supn→∞⟨(f−μF)x∗,qϕ(n)−x∗⟩≤0. | (3.37) |
On the other hand, from (3.25) we obtain
αϕ(n)(1−βϕ(n))(τ−δ)Λϕ(n)≤Λϕ(n)−Λϕ(n)+1+αϕ(n)(1−βϕ(n))(τ−δ)[2⟨(f−μF)x∗,qϕ(n)−x∗⟩τ−δ+θϕ(n)αϕ(n)⋅Mτ−δ]≤αϕ(n)(1−βϕ(n))(τ−δ)[2⟨(f−μF)x∗,qϕ(n)−x∗⟩τ−δ+θϕ(n)αϕ(n)⋅Mτ−δ], |
which immediately attains
lim supn→∞Λϕ(n)≤lim supn→∞[2⟨(f−μF)x∗,qϕ(n)−x∗⟩τ−δ+θϕ(n)αϕ(n)⋅Mτ−δ]≤0. |
Thus, limn→∞‖xϕ(n)−x∗‖2=0. In addition, observe that
‖xϕ(n)+1−x∗‖2−‖xϕ(n)−x∗‖2=2⟨xϕ(n)+1−xϕ(n),xϕ(n)−x∗⟩+‖xϕ(n)+1−xϕ(n)‖2≤2‖xϕ(n)+1−xϕ(n)‖‖xϕ(n)−x∗‖+‖xϕ(n)+1−xϕ(n)‖2. | (3.38) |
Thanks to Λn≤Λϕ(n)+1, one gets
‖xn−x∗‖2≤‖xϕ(n)−x∗‖2+2‖xϕ(n)+1−xϕ(n)‖‖xϕ(n)−x∗‖+‖xϕ(n)+1−xϕ(n)‖2→0(n→∞). |
That is, xn→x∗ as n→∞. This completes the proof.
Theorem 3.7. If S:C→C is nonexpansive and {xn} is the sequence constructed in the modified version of Algorithm 3.1, that is, for any initial x1∈C,
{wn=σnxn+(1−σn)un,vn=TΘ2μ2(wn−μ2B2wn),un=TΘ1μ1(vn−μ1B1vn),yn=PC(wn−λAwn),tn=(1−τn)wn+τnyn,zn=PCn(wn),xn+1=βnxn+(1−βn)PC[αnf(xn)+(I−αnμF)Szn]∀n≥1, | (3.39) |
where for each n≥1, Cn and τn are picked as in Algorithm 3.1, then xn→x∗∈Ω with x∗∈Ω being only a solution to the HVI: ⟨(μF−f)x∗,y−x∗⟩≥0,∀y∈Ω.
Proof. We divide the proof of the theorem into several aspects.
Aspect 1. We assert the boundedness of {xn}. Indeed, using the same reasonings as in Aspect 1 of the proof of Theorem 3.6, one derives the desired assertion.
Aspect 2. We assert that
(1−βn){(1−αnτ)‖wn−zn‖2+‖qn−PC(qn)‖2}+βn(1−βn)‖xn−PC(qn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1, |
for some M1>0. In fact, putting θn=0, from (3.20) we get
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+(1−αnτ)[σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2]+2αn⟨(f−μF)x∗,qn−x∗⟩−‖qn−PC(qn)‖2}−βn(1−βn)‖xn−PC(qn)‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2−(1−βn){(1−αnτ)‖wn−zn‖2+‖qn−PC(qn)‖2}+αnM1−βn(1−βn)‖xn−PC(qn)‖2≤‖xn−x∗‖2−(1−βn){(1−αnτ)‖wn−zn‖2+‖qn−PC(qn)‖2}+αnM1−βn(1−βn)‖xn−PC(qn)‖2, |
where supn≥12‖(f−μF)x∗‖‖qn−x∗‖≤M1 for some M1>0. This attains the desired assertion.
Aspect 3. We assert that
(1−βn)(1−αnτ)[τn2λˉL‖Rλ(wn)‖2]2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1. |
Indeed, using the same reasonings as in Aspect 3 of the proof of Theorem 3.6, one deduces the desired assertion.
Aspect 4. We assert that
‖xn+1−x∗‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)⋅2⟨(f−μF)x∗,qn−x∗⟩τ−δ. | (3.40) |
Indeed, using the same reasonings as in Aspect 4 of the proof of Theorem 3.6, one obtains the desired assertion.
Aspect 5. We assert that {xn} converges strongly to the unique solution x∗∈Ω of the HVI (3.16). Indeed, using the same reasonings as in Aspect 5 of the proof of Theorem 3.6, one gets the desired assertion. On the other hand, we put forth another modification of Mann-like subgradient-like extragradient implicit rule with linear-search process.
Algorithm 3.8. Initial step: Given ν>0,ℓ∈(0,1),λ∈(0,1ν). Let x1∈C be arbitrary.
Iterations: Given the current iterate xn, calculate xn+1 below:
Step 1. Calculate wn=σnxn+(1−σn)un with
vn=TΘ2μ2(wn−μ2B2wn), |
un=TΘ1μ1(vn−μ1B1vn). |
Step 2. Calculate yn=PC(wn−λAwn) and Rλ(wn):=wn−yn.
Step 3. Calculate tn=wn−τnRλ(wn), where τn:=ℓjn and jn is the smallest nonnegative integer j satisfying
⟨Awn−A(wn−ℓjRλ(wn)),wn−yn⟩≤ν2‖Rλ(wn)‖2. |
Step 4. Compute zn=PCn(wn) and xn+1=βnwn+(1−βn)PC[αnf(zn)+(I−αnμF)Snzn], where Cn:={u∈C:ℏn(u)≤0} and
ℏn(u)=⟨Atn,u−wn⟩+τn2λ‖Rλ(wn)‖2. |
Again put n:=n+1 and return to Step 1.
It is worth mentioning that (3.16)–(3.19) and Lemmas 3.2–3.5 remain true for Algorithm 3.8.
Theorem 3.9. Suppose that {xn} is the sequence constructed in Algorithm 3.8. Then xn→x∗∈Ω provided Snxn−Sn+1xn→0, with x∗∈Ω being only a solution to the HVI: ⟨(μF−f)x∗,y−x∗⟩≥0,∀y∈Ω.
Proof. In what follows, under the assumption Snxn−Sn+1xn→0, one divides the proof into several aspects.
Aspect 1. We assert that {xn} is of boundedness. Indeed, for x∗∈Ω=Fix(S)∩Fix(G)∩VI(C,A) we have Sx∗=x∗,Gx∗=x∗ and PC(x∗−λAx∗)=x∗. Using (3.19), from Lemma 2.10 we obtain
‖xn+1−x∗‖≤βn‖wn−x∗‖+(1−βn){αnδ‖zn−x∗‖+(1−αnτ)(1+θn)‖zn−x∗‖+αn‖(f−μF)x∗‖}≤βn‖xn−x∗‖+(1−βn){[αnδ+(1−αnτ)+θn]‖xn−x∗‖+αn‖(f−μF)x∗‖}≤βn‖xn−x∗‖+(1−βn){[1−αn(τ−δ)+αn(τ−δ)2]‖xn−x∗‖+αn‖(f−μF)x∗‖}=[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖+αn(1−βn)(τ−δ)2⋅2‖(f−μF)x∗‖τ−δ≤max{‖xn−x∗‖,2‖(f−μF)x∗‖τ−δ}. |
By induction, we get ‖xn−x∗‖≤max{‖x1−x∗‖,2‖(f−μF)x∗‖τ−δ}∀n≥1. Thus, {xn} is bounded, and so are the sequences {wn},{yn},{zn},{f(zn)},{Atn},{Gwn},{Snzn}.
Aspect 2. We assert that
(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+βn(1−βn)‖wn−PC(ˉqn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1, |
for some M1>0. In fact, it is clear that
‖zn−x∗‖2≤‖wn−x∗‖2−‖wn−zn‖2≤σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2. |
Since xn+1=βnwn+(1−βn)PC(ˉqn) where ˉqn=αnf(zn)+(I−αnμF)Snzn, Using Lemma 2.10 and the convexity of the function h(s)=s2∀s∈R, from (3.19) we obtain that
‖xn+1−x∗‖2=βn‖wn−x∗‖2+(1−βn)‖PC(ˉqn)−x∗‖2−βn(1−βn)‖wn−PC(ˉqn)‖2≤βn‖wn−x∗‖2+(1−βn){‖ˉqn−x∗‖2−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2=βn‖wn−x∗‖2+(1−βn){‖αn(f(zn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗+αn(f−μF)x∗‖2−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤βn‖wn−x∗‖2+(1−βn){‖αn(f(zn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗‖2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤βn‖wn−x∗‖2+(1−βn){[αn‖f(zn)−f(x∗)‖+‖(I−αnμF)Snzn−(I−αnμF)x∗‖]2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]‖zn−x∗‖2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn][σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2 | (3.41) |
(due to αnδ+(1−αnτ)+θn≤1−αn(τ−δ)+αn(τ−δ)2=1−αn(τ−δ)2≤1), which together with un=Gwn, guarantees that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+αn(τ−δ)2][σn‖xn−x∗‖2+(1−σn)‖xn−x∗‖2−‖wn−zn‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤‖xn−x∗‖2−(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2+αnM1, | (3.42) |
where supn≥12‖(f−μF)x∗‖‖ˉqn−x∗‖≤M1 for some M1>0. This attains the desired assertion.
Aspect 3. We assert that
(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1. |
In fact, using the similar arguments to those of (3.22) in the proof of Theorem 3.56, we can deduce that for some ˉL>0,
‖zn−x∗‖2≤‖wn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2. | (3.43) |
From (3.41), (3.19) and (3.43) it follows that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]‖zn−x∗‖2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn][‖wn−x∗‖2−[τn2λˉL‖Rλ(wn)‖2]2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤[1−αn(1−βn)(τ−δ)2]‖xn−x∗‖2−(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2+2αn(1−βn)⟨(f−μF)x∗,ˉqn−x∗⟩≤‖xn−x∗‖2−(1−βn)[1−αn(τ+δ)2][τn2λˉL‖Rλ(wn)‖2]2+αnM1, |
which hence yields the desired assertion.
Aspect 4. We assert that
‖xn+1−x∗‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,ˉqn−x∗⟩τ−δ+θnαn⋅Mτ−δ] | (3.44) |
for some M>0. In fact, from Lemma 2.10 and (3.19), one obtains
‖xn+1−x∗‖2≤βn‖wn−x∗‖2+(1−βn)‖ˉqn−x∗‖2=βn‖wn−x∗‖2+(1−βn)‖αn(f(zn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗+αn(f−μF)x∗‖2≤βn‖wn−x∗‖2+(1−βn){‖αn(f(zn)−f(x∗))+(I−αnμF)Snzn−(I−αnμF)x∗‖2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤βn‖wn−x∗‖2+(1−βn){[αnδ‖zn−x∗‖+(1−αnτ)(1+θn)‖zn−x∗‖]2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+(1−αnτ+θn)‖zn−x∗‖2+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,ˉqn−x∗⟩τ−δ+θnαn⋅Mτ−δ], |
where supn≥1‖xn−x∗‖2≤M for some M>0.
Aspect 5. We assert that xn→x∗∈Ω, which is only a solution of the HVI (3.16).
In fact, from (3.44), we have
‖xn+1−x∗‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2+αn(1−βn)(τ−δ)×[2⟨(f−μF)x∗,ˉqn−x∗⟩τ−δ+θnαn⋅Mτ−δ]. | (3.45) |
Setting Λn=‖xn−x∗‖2, we demonstrate the convergence of {Λn} to zero by the following two situations.
Situation 1. ∃(integer)n0≥1 s.t. {Λn} is nonincreasing. It is clear that the limit limn→∞Λn=k<+∞ and limn→∞(Λn−Λn+1)=0. From Aspect 2 and {βn}⊂[a,b]⊂(0,1) we obtain
(1−b){[1−αn(τ+δ)2]‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+a(1−b)‖wn−PC(ˉqn)‖2≤(1−βn){[1−αn(τ+δ)2]‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+βn(1−βn)‖wn−PC(ˉqn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
Owing to the facts that αn→0 and Λn−Λn+1→0, from τ+δ2∈(0,1) one deduces that
limn→∞‖wn−zn‖=limn→∞‖ˉqn−PC(ˉqn)‖=limn→∞‖wn−PC(ˉqn)‖=0. | (3.46) |
Hence it is readily known that
‖wn−ˉqn‖≤‖wn−PC(ˉqn)‖+‖PC(ˉqn)−ˉqn‖→0(n→∞),‖xn+1−wn‖=(1−βn)‖PC(ˉqn)−wn‖≤‖ˉqn−wn‖→0(n→∞), | (3.47) |
and
‖Snzn−wn‖=‖ˉqn−wn−αn(f(zn)−μFSnzn)‖≤‖ˉqn−wn‖+αn(‖f(zn)‖+μ‖FSnzn‖)→0(n→∞). |
Next, we show that ‖xn−un‖→0 and ‖xn−xn+1‖→0 as n→∞. Indeed, note that y∗=TΘ2μ2(x∗−μ2B2x∗), vn=TΘ2μ2(wn−μ2B2wn) and un=TΘ1μ1(vn−μ1B1vn). Then un=Gwn. Using Lemma 2.1, from (3.19) we have
‖un−x∗‖2≤‖xn−x∗‖2−μ2(2β−μ2)‖B2wn−B2x∗‖2−μ1(2α−μ1)‖B1vn−B1y∗‖2. |
This together with (3.41), implies that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]×[σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]{μ2(2β−μ2)‖B2wn−B2x∗‖2+μ1(2α−μ1)‖B1vn−B1y∗‖2}+αnM1, |
which immediately arrives at
(1−βn)(1−σn)[1−αn(τ+δ)2]{μ2(2β−μ2)‖B2wn−B2x∗‖2+μ1(2α−μ1)‖B1vn−B1y∗‖2}≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1. |
This hence ensures that
limn→∞‖B2wn−B2x∗‖=0andlimn→∞‖B1vn−B1y∗‖=0. |
On the other hand, using Lemma 1.1, from (3.19) we get
‖un−x∗‖2≤‖xn−x∗‖2−‖wn−vn+y∗−x∗‖2−‖vn−un+x∗−y∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖. |
This together with (3.41), implies that
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]×[σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]{‖wn−vn+y∗−x∗‖2+‖vn−un+x∗−y∗‖2}+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1, |
which immediately leads to
(1−βn)(1−σn)[1−αn(τ+δ)2]{‖wn−vn+y∗−x∗‖2+‖vn−un+x∗−y∗‖2}≤‖xn−x∗‖2−‖xn+1−x∗‖2+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1=Λn−Λn+1+2μ1‖B1y∗−B1vn‖‖un−x∗‖+2μ2‖B2x∗−B2wn‖‖vn−y∗‖+αnM1. |
This hence ensures that
limn→∞‖wn−vn+y∗−x∗‖=0andlimn→∞‖vn−un+x∗−y∗‖=0. |
Therefore,
‖wn−Gwn‖=‖wn−un‖≤‖wn−vn+y∗−x∗‖+‖vn−un+x∗−y∗‖→0(n→∞). | (3.48) |
Noticing wn=σnxn+(1−σn)un, we get
‖wn−x∗‖2=σn⟨xn−x∗,wn−x∗⟩+(1−σn)⟨un−x∗,wn−x∗⟩≤σn⟨xn−x∗,wn−x∗⟩+(1−σn)‖wn−x∗‖2=12σn[‖xn−x∗‖2+‖wn−x∗‖2−‖xn−wn‖2]+(1−σn)‖wn−x∗‖2, |
which immediately yields
‖wn−x∗‖2≤‖xn−x∗‖2−‖xn−wn‖2. |
This together with (3.41), arrives at
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+[(1−αnτ)+θn]×[σn‖xn−x∗‖2+(1−σn)‖wn−x∗‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩}≤‖xn−x∗‖2−(1−βn)(1−σn)[1−αn(τ+δ)2]‖xn−wn‖2+αnM1. |
So it follows that
(1−βn)(1−σn)[1−αn(τ+δ)2]‖xn−wn‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1=Λn−Λn+1+αnM1, |
which immediately yields
limn→∞‖xn−wn‖=0. |
So it follows from (3.46)–(3.48) that
‖xn−zn‖≤‖xn−wn‖+‖wn−zn‖→0(n→∞), | (3.49) |
‖xn−xn+1‖≤‖xn−wn‖+‖wn−xn+1‖→0(n→∞), | (3.50) |
and
‖xn−Gwn‖≤‖xn−wn‖+‖wn−Gwn‖→0(n→∞). | (3.51) |
Also, using the similar arguments to those of (3.31) in the proof of Theorem 3.6, we can obtain that
limn→∞‖wn−yn‖=0. | (3.52) |
By the boundedness of {xn}, we know that ∃subsequence{xnp}⊂{xn} s.t.
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=limp→∞⟨(f−μF)x∗,xnp−x∗⟩. | (3.53) |
Since H is reflexive and {xn} is bounded, we might assume that xnp⇀ˉx. Thus, from (3.53) one has
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=limp→∞⟨(f−μF)x∗,xnp−x∗⟩=⟨(f−μF)x∗,ˉx−x∗⟩. | (3.54) |
Since xn−xn+1→0,xn−Gwn→0,wn−yn→0,xn−zn→0 and xnp⇀ˉx, by Lemma 3.4 we infer that ˉx∈Ω. Thus, using (3.16) and (3.54) one has
lim supn→∞⟨(f−μF)x∗,xn−x∗⟩=⟨(f−μF)x∗,ˉx−x∗⟩≤0, | (3.55) |
which together with (3.46), arrives at
lim supn→∞⟨(f−μF)x∗,ˉqn−x∗⟩=lim supn→∞[⟨(f−μF)x∗,ˉqn−PC(ˉqn)+PC(ˉqn)−wn+wn−xn⟩+⟨(f−μF)x∗,xn−x∗⟩]≤lim supn→∞[‖(f−μF)x∗‖(‖ˉqn−PC(ˉqn)‖+‖PC(ˉqn)−wn‖+‖wn−xn‖)+⟨(f−μF)x∗,xn−x∗⟩]≤0. | (3.56) |
Note that {αn(1−βn)(τ−δ)}⊂[0,1],∑∞n=1αn(1−βn)(τ−δ)=∞, and
lim supn→∞[2⟨(f−μF)x∗,ˉqn−x∗⟩τ−δ+θnαn⋅Mτ−δ]≤0. |
Consequently, applying Lemma 2.5 to (3.45), one has limn→∞‖xn−x∗‖2=0.
Situation 2. ∃{Λnp}⊂{Λn} s.t. Λnp<Λnp+1∀p∈N, with N being the set of all natural numbers. Let ϕ:N→N be formulated as
ϕ(n):=max{p≤n:Λp<Λp+1}. |
Using Lemma 2.9, we get
Λϕ(n)≤Λϕ(n)+1andΛn≤Λϕ(n)+1. |
In the remainder of the proof, using the same reasonings as in Situation 2 of Aspect 5 in the proof of Theorem 3.6, we obtain the desired assertion.
Theorem 3.10. If S:C→C is nonexpansive and {xn} is the sequence constructed in the modified version of Algorithm 3.8, that is, for any initial x1∈C,
{wn=σnxn+(1−σn)un,vn=TΘ2μ2(wn−μ2B2wn),un=TΘ1μ1(vn−μ1B1vn),yn=PC(wn−λAwn),tn=(1−τn)wn+τnyn,zn=PCn(wn),xn+1=βnwn+(1−βn)PC[αnf(zn)+(I−αnμF)Szn],∀n≥1, | (3.57) |
where for each n≥1, Cn and τn are picked as in Algorithm 3.8, then xn→x∗∈Ω with x∗∈Ω being only a solution to the HVI: ⟨(μF−f)x∗,y−x∗⟩≥0,∀y∈Ω.
Proof. We divide the proof of the theorem into several aspects.
Aspect 1. We assert the boundedness of {xn}. Indeed, using the same reasonings as in Aspect 1 of the proof of Theorem 3.9, one derives the desired assertion.
Aspect 2. We assert that
(1−βn){(1−αnτ)‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+βn(1−βn)‖wn−PC(ˉqn)‖2≤‖xn−x∗‖2−‖xn+1−x∗‖2+αnM1, |
for some M1>0. In fact, putting θn=0, from (3.41) we get
‖xn+1−x∗‖2≤βn‖xn−x∗‖2+(1−βn){αnδ‖xn−x∗‖2+(1−αnτ)[σn‖xn−x∗‖2+(1−σn)‖un−x∗‖2−‖wn−zn‖2]+2αn⟨(f−μF)x∗,ˉqn−x∗⟩−‖ˉqn−PC(ˉqn)‖2}−βn(1−βn)‖wn−PC(ˉqn)‖2≤[1−αn(1−βn)(τ−δ)]‖xn−x∗‖2−(1−βn){(1−αnτ)‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+αnM1−βn(1−βn)‖wn−PC(ˉqn)‖2≤‖xn−x∗‖2−(1−βn){(1−αnτ)‖wn−zn‖2+‖ˉqn−PC(ˉqn)‖2}+αnM1−βn(1−βn)‖wn−PC(ˉqn)‖2, |
where \sup_{n\geq1}2\|(f-\mu F)x^*\|\|\bar q_n-x^*\|\leq M_1 for some M_1 > 0 . This attains the desired assertion.
Aspect 3. We assert that
(1-\beta_n)(1-\alpha_n\tau)[\frac{\tau_n}{2\lambda\bar L}\|R_\lambda(w_n)\|^2]^2\leq\|x_n-x^*\|^2-\|x_{n+1}-x^*\|^2 +\alpha_nM_1. |
Indeed, using the same reasonings as in Aspect 3 of the proof of Theorem 3.9, one deduces the desired assertion.
Aspect 4. We assert that
\begin{eqnarray} \begin{aligned} \|x_{n+1}-x^*\|^2&\leq[1-\alpha_n(1-\beta_n)(\tau-\delta)]\|x_n-x^*\|^2\\ &\quad+\alpha_n(1-\beta_n)(\tau-\delta) \cdot\frac{2\langle(f-\mu F)x^*, \bar q_n-x^*\rangle}{\tau-\delta}. \end{aligned} \end{eqnarray} | (3.58) |
Indeed, using the same reasonings as in Aspect 4 of the proof of Theorem 3.9, one obtains the desired assertion.
Aspect 5. We assert that \{x_n\} converges strongly to the unique solution x^*\in\Omega of the HVI (3.16). Indeed, using the same reasonings as in Aspect 5 of the proof of Theorem 3.9, one gets the desired assertion.
Remark 3.11. Compared with the corresponding results in Cai, Shehu and Iyiola [13], Thong and Hieu [17] and Reich et al. [29], our results improve and extend them in the following aspects.
(ⅰ) The problem of finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G}) (with {\mathcal G} = P_C(I-\mu_1B_1) P_C(I-\mu_2B_2) ) in [13] is extended to develop our problem of finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) where {\mathcal G} = T^{\Theta_1}_{\mu_1}(I-\mu_1B_1)T^{\Theta_2}_{\mu_2} (I-\mu_2B_2) and S is asymptotically nonexpansive mapping. The modified viscosity implicit rule for finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G}) in [13] is extended to develop our modified Mann-like subgradient-like extragradient implicit rules with linear-search process for finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) , which is on the basis of the subgradient extragradient rule with linear-search process, Mann implicit iteration approach, and hybrid deepest-descent technique.
(ⅱ) The problem of finding an element of {\rm Fix}(S)\cap{\rm VI}(C, A) with quasi-nonexpansive mapping S in [17] is extended to develop our problem of finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) with asymptotically nonexpansive mapping S . The inertial subgradient extragradient method with linear-search process for finding an element of {\rm Fix}(S)\cap{\rm VI}(C, A) in [17] is extended to develop our modified Mann-like subgradient-like extragradient implicit rules with linear-search process for finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) , which is on the basis of the subgradient extragradient rule with linear-search process, Mann implicit iteration approach, and hybrid deepest-descent technique.
(ⅲ) The problem of finding an element of {\rm VI}(C, A) with pseudomonotone uniform continuity mapping A is extended to develop our problem of finding an element of {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) with both asymptotically nonexpansive mapping S and nonexpansive mapping {\mathcal G} . The modified projection-type method with linear-search process in [29] is extended to develop our modified Mann-like subgradient-like extragradient implicit rule with linear-search process, e.g., the original projection step y_n = P_C(x_n-\lambda Ax_n) in [29] is developed into the modified Mann-like implicit projection step w_n = (1-\sigma_n)x_n+\sigma_n{\mathcal G}w_n and y_n = P_C(w_n-\lambda Aw_n) ; meantime, the original viscosity step x_{n+1} = \alpha_nf(x_n)+(1-\alpha_n)P_{C_n}(x_n) is developed into the composite viscosity iterative step x_{n+1} = P_C[\alpha_nf(x_n)+(I-\alpha_n\mu F)S^nP_{C_n}(w_n)] .
In what follows, we provide an illustrated instance to show the feasibility and implementability of suggested rules. Put \Theta_1 = \Theta_2 = 0 , \mu = 2 , \mu_1 = \mu_2 = \frac{1}{3}, \; \nu = 1, \; \lambda = \ell = \frac{1}{2}, \; \sigma_n = \beta_n = \frac{2}{3} and \alpha_n = \frac{1}{3(n+1)} . We first provide an example of two inverse-strongly monotone mappings B_1, B_2:C\to H , Lipschitz continuous and pseudomonotone mapping A and asymptotically nonexpansive mapping S with \Omega = {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A)\neq\emptyset , where {\mathcal G}: = T^{\Theta_1}_{\mu_1}(I-\mu_1B_1) T^{\Theta_2}_{\mu_2}(I-\mu_2B_2) = P_C(I-\mu_1B_1)P_C(I-\mu_2B_2) . We set H = {\bf R} and use the \langle a, b\rangle = ab and \|\cdot\| = |\cdot| to denote its inner product and induced norm, respectively. Moreover, we put C = [-2, 3] . The starting point x_1 is arbitrarily chosen in C . Let f(x) = F(x) = \frac{1}{2}x\; \forall x\in C with
\delta = \frac{1}{2} < \tau = 1-\sqrt{1-\mu(2\eta-\mu\kappa^2)} = 1-\sqrt{1-2(2\cdot\frac{1}{2}-2(\frac{1}{2})^2)} = 1. |
We let B_1x = B_2x: = Bx = x-\frac{1}{2}\sin x, \; \forall x\in C . Let A:H\to H and S:C\to C be formulated as Ax: = \frac{1}{1+|\sin x|}-\frac{1}{1+|x|} and Sx: = \frac{5}{6}\sin x . We now assert that B is \frac{2} {9} -inverse-strongly monotone. In fact, since B is \frac{1}{2} -strongly monotone and \frac{3}{2} -Lipschitz continuous, we know that B is \frac{2}{9} -inverse-strongly monotone with \alpha = \beta = \frac{2}{9} . Let us assert that A is pseudomonotone and Lipschitz continuous. In fact, for each a, b\in H one has
\begin{eqnarray*} \begin{aligned} \|Aa-Ab\| &\leq|\frac{\|b\|-\|a\|}{(1+\|b\|)(1+\|a\|)}|+|\frac{\|\sin b\|-\|\sin a\|}{(1+\|\sin b\|)(1+\|\sin a\|)}|\\ &\leq\frac{\|a-b\|}{(1+\|a\|)(1+\|b\|)}+\frac{\|\sin a-\sin b\|}{(1+\|\sin a\|)(1+\|\sin b\|)}\\ &\leq\|a-b\|+\|\sin a-\sin b\|\leq2\|a-b\|. \end{aligned} \end{eqnarray*} |
This means that A is Lipschitz continuous with L = 2 . Next, we assert that A is pseudomonotone. For each a, b\in H , it is readily known that
\begin{eqnarray*} \begin{aligned} &\langle Aa, b-a\rangle = (\frac{1}{1+|\sin a|}-\frac{1}{1+|a|})(b-a)\geq0\\ &\quad\Rightarrow\; \langle Ab, b-a\rangle = (\frac{1}{1+|\sin b|}-\frac{1}{1+|b|})(b-a)\geq0. \end{aligned} \end{eqnarray*} |
Moreover, it is easy to check that S is asymptotically nonexpansive with \theta_n = (\frac{5}{6})^n, \; \forall n\geq1 , such that \|S^{n+1}x_n-S^nx_n\|\to0 as n\to\infty . In fact, note that
\|S^na-S^nb\|\leq{\frac{5}{6}}\|S^{n-1}a-S^{n-1}b\|\leq\cdots\leq(\frac{5}{6})^n\|a-b\|\leq(1+\theta_n)\|a-b\|, |
and
\|S^{n+1}x_n-S^nx_n\|\leq(\frac{5}{6})^{n-1}\|S^2x_n-Sx_n\| = (\frac{5}{6})^{n-1}\|\frac{5}{6}\sin(Sx_n) -\frac{5}{6}\sin x_n\|\leq2(\frac{5}{6})^n\to0. |
It is obvious that {\rm Fix}(S) = \{0\} and
\lim\limits_{n\to\infty}\frac{\theta_n}{\alpha_n} = \lim\limits_{n\to\infty}\frac{(5/6)^n}{1/3(n+1)} = 0. |
Accordingly, \Omega = {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) = \{0\}\neq\emptyset . In this case, noticing
{\mathcal G} = P_C(I-\mu_1B_1)P_C(I-\mu_2B_2) = [P_C(I-\frac{1}{3}B)]^2, |
we rewrite Algorithm 3.1 as follows:
\begin{eqnarray} \begin{cases} w_n = \frac{2}{3}x_n+\frac{1}{3}u_n, \\ v_n = P_C(I-\frac{1}{3}B)w_n, \\ u_n = P_C(I-\frac{1}{3}B)v_n, \\ y_n = P_C(w_n-\frac{1}{2}Aw_n), \\ t_n = (1-\tau_n)w_n+\tau_ny_n, \\ z_n = P_{C_n}(w_n), \\ x_{n+1} = \frac{2}{3}x_n+\frac{1}{3}P_C[\frac{1}{3(n+1)}\cdot\frac{1}{2}x_n+(1-\frac{1}{3(n+1)})S^nz_n], \quad\forall n\geq1, \end{cases} \end{eqnarray} | (4.1) |
where for each n\geq1 , C_n and \tau_n are chosen as in Algorithm 3.1. Then, by Theorem 3.6, we know that \{x_n\} converges to 0\in\Omega = {\rm Fix}(S)\cap\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) .
In particular, since Sx: = \frac{5}{6}\sin x is also nonexpansive, we consider the modified version of Algorithm 3.1, that is,
\begin{eqnarray} \begin{cases} w_n = \frac{2}{3}x_n+\frac{1}{3}u_n, \\ v_n = P_C(I-\frac{1}{3}B)w_n, \\ u_n = P_C(I-\frac{1}{3}B)v_n, \\ y_n = P_C(w_n-\frac{1}{2}Aw_n), \\ t_n = (1-\tau_n)w_n+\tau_ny_n, \\ z_n = P_{C_n}(w_n), \\ x_{n+1} = \frac{2}{3}x_n+\frac{1}{3}P_C[\frac{1}{3(n+1)}\cdot\frac{1}{2}x_n+(1-\frac{1}{3(n+1)})Sz_n], \quad\forall n\geq1, \end{cases} \end{eqnarray} | (4.2) |
where for each n\geq1 , C_n and \tau_n are chosen as above. Then, by Theorem 3.7, we know that \{x_n\} converges to 0\in\Omega = {\rm Fix}(S)\cap{\rm Fix}({\mathcal G})\cap{\rm VI}(C, A) .
In this paper, we introduce the modified Mann-like subgradient-like extragradient implicit rules with linear-search process for finding a common solution of the SGEP, VIP and FPP. The proposed algorithms are on the basis of the subgradient extragradient rule with linear-search process, Mann implicit iteration approach, and hybrid deepest-descent technique. Under mild restrictions, we demonstrate the strong convergence of the suggested algorithms to a common solution of the SGEP, VIP and FPP, which is a unique solution of a certain HVI defined on their common solution set. In addition, an illustrated example is provided to show the feasibility and implementability of our proposed rule.
This research was supported by the 2020 Shanghai Leading Talents Program of the Shanghai Municipal Human Resources and Social Security Bureau (20LJ2006100), the Innovation Program of Shanghai Municipal Education Commission (15ZZ068) and the Program for Outstanding Academic Leaders in Shanghai City (15XD1503100). Li-Jun Zhu was supported by the National Natural Science Foundation of China [grant number 11861003], the Natural Science Foundation of Ningxia province [grant number NZ17015], the Major Research Projects of NingXia [grant numbers 2021BEG03049]and Major Scientific and Technological Innovation Projects of YinChuan [grant numbers 2022RKX03 and NXYLXK2017B09].
The authors declare no conflicts of interest.
[1] | J. C. Ion, Laser processing of engineering materials: principles, procedure and industrial application, Elsevier, 2005. |
[2] |
Z. X. Guo, S. Kumar, Discrete-ordinates solution of short-pulsed laser transport in two-dimensional turbid media, Appl. Opt., 40 (2001), 3156–3163. https://doi.org/10.1364/AO.40.003156 doi: 10.1364/AO.40.003156
![]() |
[3] |
X. Y. Jiang, C. M. Soukoulis, Time dependent theory for random lasers, Phys. Rev. Lett., 85 (2000), 70. https://doi.org/10.1103/PhysRevLett.85.70 doi: 10.1103/PhysRevLett.85.70
![]() |
[4] | A. R. Jha, Infrared technology: applications to electro-optics, photonic devices and sensors, New York: Wiley, 2000. |
[5] | B. A. Lengyel, Introduction to laser physics, New York: Wiley, 1966. |
[6] | J. Ohtsubo, Semiconductor lasers: stability, instability and chaos, Berlin, Heidelberg: Springer, 2013. https://doi.org/10.1007/978-3-642-30147-6 |
[7] |
M. J. Weber, Science and technology of laser glass, J. Non-Cryst. Solids, 123 (1990), 208–222. https://doi.org/10.1016/0022-3093(90)90786-L doi: 10.1016/0022-3093(90)90786-L
![]() |
[8] | P. W. Milonni, J. H. Eberly, Lasers physics, New York: Wiley, 2010. https://doi.org/10.1002/9780470409718 |
[9] | H. Haken, Synergetics, Berlin, Heidelberg: Springer, 1983. https://doi.org/10.1007/978-3-642-88338-5 |
[10] | S. H. Strogatz, Nonlinear dynamics and chaos with student solutions manual: with applications to physics, biology, chemistry, and engineering, Boca Raton: CRC Press, 2018. https://doi.org/10.1201/9780429399640 |
[11] | J. Guckenheimer, P. Holmes, Nonlinear oscillations, dynamical systems and bifurcation of vector fields, New York: Springer, 1983. https://doi.org/10.1007/978-1-4612-1140-2 |
[12] | Y. A. Kuznetsov, Elements of applied bifurcation theorey, New York: Springer, 2004. https://doi.org/10.1007/978-1-4757-3978-7 |
[13] |
Z. Y. Hu, Z. D. Teng, L. Zhang, Stability and bifurcation analysis of a discrete predator-prey model with nonmonotonic functional response, Nonlinear Anal. Real World Appl., 12 (2011), 2356–2377. https://doi.org/10.1016/j.nonrwa.2011.02.009 doi: 10.1016/j.nonrwa.2011.02.009
![]() |
[14] |
A. Q. Khan, J. Y. Ma, D. M. Xiao, Bifurcations of a two-dimensional discrete time plant-herbivore system, Commun. Nonlinear Sci. Numer. Simul., 39 (2016), 185–198. https://doi.org/10.1016/j.cnsns.2016.02.037 doi: 10.1016/j.cnsns.2016.02.037
![]() |
[15] |
A. Q. Khan, J. Y. Ma, D. M. Xiao, Global dynamics and bifurcation analysis of a host-parasitoid model with strong Allee effect, J. Biol. Dyn., 11 (2017), 121–146. https://doi.org/10.1080/17513758.2016.1254287 doi: 10.1080/17513758.2016.1254287
![]() |
[16] |
Z. J. Jing, J. P. Yang, Bifurcation and chaos in discrete-time predator-prey system, Chaos Solitons Fract., 27 (2006), 259–277. https://doi.org/10.1016/j.chaos.2005.03.040 doi: 10.1016/j.chaos.2005.03.040
![]() |
[17] |
C. H. Zhang, X. P. Yan, G. H. Cui, Hopf bifurcations in a predator-prey system with a discrete delay and a distributed delay, Nonlinear Anal. Real World Appl., 11 (2010), 4141–4153. https://doi.org/10.1016/j.nonrwa.2010.05.001 doi: 10.1016/j.nonrwa.2010.05.001
![]() |
[18] |
M. Sen, M. Banerjee, A. Morozov, Bifurcation analysis of a ratio-dependent prey-predator model with the Allee effect, Ecol. Complex., 11 (2012), 12–27. https://doi.org/10.1016/j.ecocom.2012.01.002 doi: 10.1016/j.ecocom.2012.01.002
![]() |
[19] |
C. D. Huang, J. Wang, X. P. Chen, J. D. Cao, Bifurcations in a fractional-order BAM neural network with four different delays, Neural Netw., 141 (2021), 344–354. https://doi.org/10.1016/j.neunet.2021.04.005 doi: 10.1016/j.neunet.2021.04.005
![]() |
[20] |
E. Kaslik, I. R. Radulescu, Stability and bifurcations in fractional-order gene regulatory networks, Appl. Math. Comput., 421 (2022), 126916. https://doi.org/10.1016/j.amc.2022.126916 doi: 10.1016/j.amc.2022.126916
![]() |
[21] |
J. Alidousti, Stability and bifurcation analysis for a fractional prey-predator scavenger model, Appl. Math. Model., 81 (2020), 342–355. https://doi.org/10.1016/j.apm.2019.11.025 doi: 10.1016/j.apm.2019.11.025
![]() |
[22] | J. H. E. Cartwright, Nonlinear stiffness, Lyapunov exponents, and attractor dimension, Phys. Lett. A, 264 (1999), 298–302. https://doi.org/10.1016/S0375-9601(99)00793-8 |
[23] |
J. L. Kaplan, J. A. Yorke, Preturbulence: a regime observed in a fluid flow model of Lorenz, Commun. Math. Phys., 67 (1979), 93–108. https://doi.org/10.1007/BF01221359 doi: 10.1007/BF01221359
![]() |
[24] | S. N. Elaydi, An introduction to difference equations, New York: Springer, 1996. https://doi.org/10.1007/978-1-4757-9168-6 |
[25] | S. Lynch, Dynamical systems with applications using Mathematica, Boston: Birkhäuser, 2007. |
1. | Muhammad Bux, Saleem Ullah, Muhammad Bilal Khan, Najila Aloraini, A novel iterative approach for resolving generalized variational inequalities, 2023, 8, 2473-6988, 10788, 10.3934/math.2023547 | |
2. | Zai-Yun Peng, Zhi-Ying Peng, Gang Cai, Gao-Xi Li, Inertial subgradient extragradient method for solving pseudomonotone variational inequality problems in Banach spaces, 2024, 103, 0003-6811, 1769, 10.1080/00036811.2023.2267576 | |
3. | Habib ur Rehman, Debdas Ghosh, Jen-Chih Yao, Xiaopeng Zhao, Extragradient methods with dual inertial steps for solving equilibrium problems, 2024, 103, 0003-6811, 2941, 10.1080/00036811.2024.2330509 | |
4. | Cong-Shan Wang, Lu-Chuan Ceng, Bing Li, Sheng-Long Cao, Hui-Ying Hu, Yun-Shui Liang, Modified Inertial-Type Subgradient Extragradient Methods for Variational Inequalities and Fixed Points of Finite Bregman Relatively Nonexpansive and Demicontractive Mappings, 2023, 12, 2075-1680, 832, 10.3390/axioms12090832 | |
5. | Yali Zhao, Ziru Zhao, Inertial iterative algorithms for common solution of variational inequality and system of variational inclusion problems, 2024, 32, 0971-3611, 2889, 10.1007/s41478-024-00766-9 | |
6. | Mohammad Farid, Pronpat Peeyada, Rehan Ali, Watcharaporn Cholamjiak, Extragradient method with inertial iterative technique for pseudomonotone split equilibrium and fixed point problems of new mappings, 2024, 32, 0971-3611, 1463, 10.1007/s41478-023-00695-z | |
7. | Prasit Cholamjiak, Zhongbing Xie, Min Li, Papinwich Paimsang, Double inertial subgradient extragradient algorithm for solving equilibrium problems and common fixed point problems with application to image restoration, 2024, 03770427, 116396, 10.1016/j.cam.2024.116396 |