- Research
- Open access
- Published:
Modified inertial viscosity extrapolation method for solving quasi-monotone variational inequality and fixed point problems in real Hilbert spaces
Journal of Inequalities and Applications volume 2024, Article number: 38 (2024)
Abstract
In this paper, we introduce and study a viscous-type extrapolation algorithm for finding a solution of the variational inequality problem and a fixed point constraint of quasi-nonexpansive mappings under the scope of real Hilbert spaces when the underlying cost operator is quasi-monotone. The method involves inertial viscosity approximation and a constructed self-adjustable step size condition that depends solely on the information of the previous step. We establish a strong convergence result of the proposed method under certain mild conditions on the algorithm parameters. Finally, to demonstrate the gain of our method, some numerical examples are presented in comparison with some related methods in literature.
1 Introduction
Let \(\mathcal{K}\) be a nonempty, closed and convex subset of a real Hilbert space \(\mathscr{H}\) endowed with the inner product \(\langle \cdot ,\cdot \rangle \) and induced norm \(\| \cdot \| \). Let \(\mathcal{L}:\mathscr{H}\to \mathscr{H}\) be a real single-valued nonlinear mapping. Then the variational inequality problem (VIP) is formulated as follows:
VIP (1.1) is a very important tool in optimization theory and other fields of applied mathematical sciences (see [2, 5, 9, 10, 12, 13, 15] and references therein). The notion of variational inequality can be traced back to the Italian mathematician Stampacchia [28]. It was used as a tool for modeling problems in mechanics (the VIP was also independently formulated by Fichera [13]). The theory of variational inequalities is a crucial one in studying a wide range of problems in pure and applied sciences in a simple, natural and unified framework. There is a known relationship between the VIP and the fixed point problem (FPP). Finding the solution of VIP (1.1) is the same as finding the solution of the FPP:
where \(0<\mu \in \mathbb{R}\) and \(P _{\mathcal{K}}\) is a metric projection of \(\mathscr{H}\) onto \(\mathcal{K}\). There are several algorithms for finding the solution of VIP (1.1). Predominant among these are the projection algorithms. Projection algorithms leverage on projections onto the feasible set \(\mathcal{K}\) or onto some related sets in order to iteratively obtain a solution. Simplest among these methods is the gradient projection technique by Goldstein [16]:
where \(\mathcal{L}\) is μ-strongly monotone and L-Lipschitzian and \(P_{\mathcal{K}}\) is a metric projection of \(\mathbb{R}^{n}\) onto \(\mathcal{K}\). It has been established that the sequence \(\{q_{n}\}\) constructed by (1.3) converges uniquely to the solution of VIP (1.1). The stringent hypothesis associated with the cost operator in (1.3) limits application and efficiency of this gradient method. To mitigate this, many researchers have attempted to weaken some conditions on the cost operator \(\mathcal{L}\). For instance, Korpelevich [19] and Antipin [4] proposed and analyzed the extragradient method (EM) for approximating the solution of VIP (1.1) in finite-dimensional Euclidean spaces when the associated operator \(\mathcal{L}\) is monotone and Lipschitz continuous. They proposed the iterative method
They proved that the sequence \(\{q_{n}\}\) generated by (1.4) converges weakly to the solution of VIP (1.1) in a finite-dimensional space. The iterative EM has been extensively studied and extended by many researchers. Authors like He et al. [18], Nadezhkina and Takahashi [21], Noor [22], Popov [25], and many others extended EM to infinite-dimensional real Hilbert spaces with better conditions on the cost operator \(\mathcal{L}\), for instance when \(\mathcal{L}\) is pseudo-monotone or quasi-monotone. One copious defect of the EM is the calculation of orthogonal projections onto the feasible set \(\mathcal{K}\) twice per iteration. This can pose a serious deficiency if the feasible set \(\mathcal{K}\) is structurally complex. Thus, to overcome this drawback, in 2011, Censor et al. [6] proposed an improved iterative method termed the subgradient extragradient method (SEGM); among its main objectives are reducing the number of projections onto \(\mathcal{K}\) per iteration and achieving convergence under certain conditions in infinite-dimensional Hilbert spaces. The authors achieved the first objective by replacing the second projection in (1.4) with projection onto constructible half-space \(\chi _{n}\). The iterative algorithm is as follows:
The authors established that the sequence \(\{q_{n}\}\) generated by (1.5) converges weakly to the solution of VIP (1.1) under some conditions on the algorithm parameters. Due to the gain of SEGM (1.5), the iterative method witnessed several modifications (see [6–8, 33] and references therein) with some conditions imposed on the algorithm parameters which guaranteed strong convergence in infinite-dimensional Hilbert spaces.
Also, to mitigate the drawback associated with the EM, Tseng introduced another variant of the EM known as Tseng’s EM [37]. The iterative algorithm is as follows:
where \(\lambda \in (0,\frac{1}{L} )\) and the underlying operator \(\mathcal{L}\) is maximally monotone. Tseng’s EM has enjoyed several modifications and extensions by many authors (see [30–32, 36, 39, 41, 42, 44, 45] and references therein).
Recently, researchers have devoted their studies to improving the rate of convergence of iterative methods. One of the techniques of achieving faster convergence is by the introduction of an inertial term into the iterative scheme. This has been shown to be an efficient technique for accelerating the convergence of such iterative methods. The inertial technique emanated from a discrete analog of a second-order dissipative dynamical system which is known for its efficiency in improving the convergence rate of iterative methods. The well-known Polyak heavy ball algorithm [24] in convex optimization, which is an inertial extrapolation process for minimizing a smooth convex function, is the first such method. This popular technique has been used by many authors in different methods for approximating the solution of VIP (1.1) and other related optimization problems (see [1, 2, 8, 20, 27] for details).
Recently, Gang et al. [14] proposed the following modified Tseng extragradient iterative algorithm for solving VIPs in real Hilbert spaces when the underlying operator is pseudo-monotone and non-Lipschitzian:
where the step size \(\lambda _{n}\) is updated using the Armijo line search rule. The authors show that under some conditions, the sequence \(\{q_{n}\}\) generated by (1.7) converges strongly to the unique solution of (1.1).
We have noticed that many modified Tseng extragradient iterative algorithms mostly entail that the cost operator \(\mathcal{L}\) is either monotone and Lipschitzian or pseudo-monotone. Now a pertinent question arises: Can we propose a modified Tseng extragradient algorithm with a more robust cost operator \(\mathcal{L}\), say, a modified Tseng extragradient method when the cost operator is quasi-monotone? And can we extend the iterative method to solving the FPP when the underlying operator is quasi-nonexpansive?
Motivated by the work of Gang et al. [14] and many others in this direction, we give affirmative answers to these questions. We propose a modified Tseng extragradient algorithm for approximating the solution of quasi-monotone VIPs with a fixed point constraint of quasi-nonexpansive mappings in the framework of real Hilbert spaces. Our algorithm is a viscous iterative one that has an inertial extrapolation term incorporated and is embellished with relaxation. In this work, the cost operator \(\mathcal{L}\) is assumed to be quasi-monotone instead of the usual condition that \(\mathcal{L}\) is monotone, as in [37], or pseudo-monotone, as in [14] and most literature. We obtain strong convergence results for the sequence constructed by this method under some mild assumptions on the algorithm parameters. Finally, we give some numerical examples to demonstrate the applicability and efficiency of our proposed method.
2 Preliminaries
In this section, we will recall some definitions and present results that will help us in our convergence analysis in the subsequent section. Let \(\mathscr{H}\) be a real Hilbert space and let \(\mathcal{K}\) be a nonempty, closed and convex subset of \(\mathscr{H}\). For every \(q\in \mathscr{H}\) there exists a unique nearest point in \(\mathcal{K}\) denoted by \(P _{\mathcal{K}}q\) such that
\(P _{\mathcal{K}}\) is termed the metric projection of \(\mathscr{H}\) onto \(\mathcal{K}\) and is known to be nonexpansive. Moreover, \(P _{\mathcal{K}}\) is associated with the following characterization.
Lemma 2.1
([18])
Let \(P _{\mathcal{K}}:\mathscr{H}\to \mathcal{K}\) be a metric projection of \(\mathscr{H}\) onto \(\mathcal{K}\). Then the following hold:
-
(i)
\(\| P _{\mathcal{K}}q- P _{\mathcal{K}}s\| ^{2}\leq \langle P _{\mathcal{K}}q- P _{\mathcal{K}}s, q-s \rangle \), \(\forall q,s\in \mathscr{H}\);
-
(ii)
\(z = P _{\mathcal{K}}q\) if and only if \(\langle q-z,s-z \rangle \leq 0\), \(\forall s\in \mathcal{K}\);
-
(iii)
\(\| q- P _{\mathcal{K}}s\| ^{2}+ \| P _{\mathcal{K}}s-s\| ^{2}\leq \| q-s\| ^{2}\), \(\forall q\in \mathscr{H}\), \(s\in \mathcal{K}\).
Definition 2.1
Let \(\mathscr{H}\) be real Hilbert space and let \(\mathcal{K}\) be a nonempty, closed and convex subset of \(\mathscr{H}\). Let \(\mathcal{F}:\mathscr{H}\to \mathscr{H}\) be a real single-valued mapping. Then \(\mathcal{F}\) is said to be:
-
(a)
α-Lipschitz continuous if there exists \(\alpha >0\) such that
$$\begin{aligned} \Vert \mathcal{F} q - \mathcal{F} s \Vert \leq \alpha \Vert q-s \Vert , \quad \forall q,s\in \mathscr{H}; \end{aligned}$$ -
(b)
nonexpansive if
$$\begin{aligned} \Vert \mathcal{F} q - \mathcal{F} s \Vert \leq \Vert q-s \Vert ,\quad \forall q,s\in \mathscr{H}; \end{aligned}$$ -
(c)
quasi-nonexpansive if the fixed point set of \(\mathcal{F}\), \(F (\mathcal{F})\), is nonempty and
$$\begin{aligned} \Vert \mathcal{F} q - p \Vert \leq \Vert q-p \Vert , \quad\forall q\in \mathscr{H}, p\in F (\mathcal{F}); \end{aligned}$$ -
(d)
β-contraction if there exists \(\beta \in [0,1)\) such that
$$\begin{aligned} \Vert \mathcal{F} q - \mathcal{F}s \Vert \leq \beta \Vert q-s \Vert ,\quad \forall q,s\in \mathscr{H}. \end{aligned}$$
Definition 2.2
([34])
Let \(\mathcal{G}:\mathscr{H}\to \mathscr{H}\) be a real single-valued operator and let \(\mathcal{K}\) be a nonempty, closed and convex subset of \(\mathscr{H}\). Then \(\mathcal{G}\) is said to be:
-
(i)
γ-strongly monotone on \(\mathcal{K}\) if there exists \(\gamma >0\) such that
$$\begin{aligned} \langle \mathcal{G} q -\mathcal{G}s, q-s \rangle \geq \gamma \Vert q-s \Vert ^{2},\quad \forall q,s\in \mathcal{K}; \end{aligned}$$ -
(ii)
monotone on \(\mathcal{K}\) if
$$\begin{aligned} \langle \mathcal{G} q -\mathcal{G}s, q-s \rangle \geq 0,\quad \forall q,s \in \mathcal{K}; \end{aligned}$$ -
(iii)
pseudo-monotone on \(\mathcal{K}\) if
$$\begin{aligned} \langle \mathcal{G} q , s-q \rangle \geq 0\quad \implies\quad \langle \mathcal{G}s , s-q \rangle \geq 0, \quad \forall q,s \in \mathcal{K}; \end{aligned}$$ -
(iv)
quasi-monotone on \(\mathcal{K}\) if
$$\begin{aligned} \langle \mathcal{G} q , q-s \rangle > 0\quad \implies\quad \langle \mathcal{G} s , q-s \rangle \geq 0,\quad \forall q,s\in \mathcal{K}; \end{aligned}$$ -
(v)
sequentially weakly continuous if for each sequence \(\{q_{n}\}\) that converges weakly to a point \(q^{*}\in \mathscr{H}\) the sequence \(\{\mathcal{G}q_{n}\}\) converges weakly to a point \(\mathcal{G}q^{*}\).
Lemma 2.2
([38])
Let \(\mathcal{H}\) be a real Hilbert space with \(\mu ,\nu \in \mathscr{H}\) and \(\alpha \in \mathbb{R}\). Then the following hold:
-
(i)
\(2 \langle \mu ,\nu \rangle =\| \mu \| ^{2}+ \| \nu \| ^{2}-\| \mu -\nu \| ^{2}= \| \mu +\nu \| ^{2}-\| \mu \| ^{2}- \| \nu \| ^{2}\);
-
(ii)
\(\| \mu -\nu \| ^{2}\leq \| \mu \| ^{2}+2 \langle \nu , \nu -\mu \rangle \);
-
(iii)
\(\| \alpha \mu + (1-\alpha )\nu \| ^{2}=\alpha \| \mu \| ^{2}+(1-\alpha )\| \nu \| ^{2}-\alpha (1-\alpha )\| \mu -\nu \| ^{2}\).
Lemma 2.3
Let \(\mathcal{K}\) be a nonempty, closed and convex subset of a Hilbert space \(\mathscr{H}\) and let \(\mathcal{L}:\mathscr{H}\to \mathscr{H}\) be an L-Lipschitzian and quasi-monotone operator. Let \(z\in \mathcal{K}\). If for some \(q^{*}\in \mathcal{K}\) we have \(\langle \mathcal{L}(z), q^{*}-z \rangle \geq 0\), then at least one of the following must hold:
Lemma 2.4
[26] Let \(\{\phi _{n}\}\) be a sequence of positive real numbers. Let \(\{\delta _{n}\}\) and \(\{\rho _{n}\}\) be sequences in \((0,1 )\) with \(\underset{n=1}{\overset{\infty}{\sum}}\delta _{n}=\infty \). Suppose that \(\phi _{n}\) satisfies the inequality
If \(\underset{k\to \infty}{\limsup}\rho _{n_{k}}\leq 0\) for every subsequence \(\phi _{n_{k}}\) of \(\phi _{n}\) satisfying the condition \(\underset{k\to \infty}{\liminf} (\phi _{n_{k}+1}-\phi _{n_{k}} )\geq 0\), then \(\underset{n\to \infty}{\lim}\phi _{n}=0\).
3 Main results
Throughout this work, we shall use \(q_{n} \to q^{*}\) (respectively \(q_{n} \rightharpoonup q^{*}\)) to denote that the sequence \(\{q_{n}\}\) converges strongly (respectively weakly) to a point \(q^{*}\) as \(n \to \infty \). For the purpose of convergence analysis of our method, we shall make the following assumptions.
Assumption 3.1
Suppose that:
-
(B1)
the real Hilbert space \(\mathscr{H}\) has a nonempty, closed and convex subset \(\mathcal{K}\);
-
(B2)
the operator \(\mathcal{L}:\mathcal{H}\to \mathscr{H}\) is quasi-monotone, \(\ell _{0}\)-Lipschitzian and sequentially weakly continuous;
-
(B3)
\(\mathcal{T}:\mathscr{H}\to \mathscr{H}\) is a quasi-nonexpansive mapping which is semiclosed at the origin such that \(F (\mathcal{T})\neq \emptyset \), \(\mathcal{S}_{i}:\mathscr{H}\to \mathscr{H}\) is a sequence of nonexpansive mappings and \(h:\mathscr{H}\to \mathscr{H}\) is a λ-contraction;
-
(B4)
the solution set of VIP (1.1) \(VI (\mathcal{K},\mathcal{L} )\neq \emptyset \) and \(\Gamma := VI (\mathcal{K},\mathcal{L} )\cap F (\mathcal{T})\neq \emptyset \);
-
(B5)
the control sequences \(\{\sigma _{n}\}\) and \(\{\xi _{n}\}\) are positive real sequences in \((0,1 )\) with \(\{\sigma _{n}\} \) satisfying the property \(\underset{n\to \infty}{\lim}\sigma _{n}=0\), \(\underset{n=1}{\overset{\infty}{\sum}}\sigma _{n}=\infty \), with \(\xi _{n}=\circ (\sigma _{n})\).
Algorithm 3.2
Initialization: Choose \(\zeta _{0}>0\), \(\lambda \in [0,1)\), \(\theta \in [0,1]\), \(q_{0},q_{1}\in \mathscr{H}\).
Iterative process steps: Given the iterates \(q_{n-1}\) and \(q_{n}\) for each \(n\geq 1\), choose \(\theta _{n}\) such that \(0\leq \theta _{n}\leq \bar{\theta}_{n}\), where
Step 1: Set \(n=1\). We calculate the iterate \(q_{n+1} \) as follows:
and we compute
If \(z_{n}=w_{n}\), stop; \(z_{n}\) is the required solution. Else, execute step 2.
Step 2: Compute
where
and update \(\{\zeta _{n}\}\) as follows:
Set \(n:=n+1\) and return to Step 1.
Remark 3.3
We first highlight some novelties of Algorithm 3.2 with respect to others in the literature.
-
(i)
In [14, 44], the authors introduced modified Tseng extragradient-type algorithms for solving VIPs in the framework of real Hilbert spaces. We observed that their iterative methods used a class of pseudo-monotone operators, while in this work we propose a modified inertial Tseng extragradient algorithm for solving VIPs in real Hilbert spaces when the associated operator is quasi-monotone and there is a fixed point constraint of quasi-nonexpansive mappings. It can be observed that the class operators considered in this work are more general and include some important classes of operators; specifically, the class of quasi-monotone operator includes strongly monotone operators, monotone operators, pseudo-monotone operators, and others (see [23] for more details).
-
(ii)
In the proposed algorithm, the self-adjustable step size condition \(\zeta _{n+1}\in \min \{\frac{\mu \|w_{n}-z_{n}\| }{\|\mathcal{L}w_{n}-\mathcal{L}z_{n}\|},\zeta _{n} \}\) applied in Algorithm 3.2 is very simple and does not possess any inner loop, unlike the line search technique employed in [14, 33], which uses inner loops and might consume additional computation time for determining the step size. It also does not require prior knowledge of the operator norm \(\|\mathcal{L}\|\).
-
(iii)
The proposed method involves inertial and relaxation terms, which are vital in improving the rate of convergence for solving VIP (1.1).
-
(iv)
The proof of our convergence analysis (that is, strong convergence of Theorem 4.6) does not follow the usual “two cases approach” as seen in many papers handling optimization problems (see [3, 11, 27, 29, 44]).
4 Convergence analysis
Lemma 4.1
(See for instance Lemma 3.1 in [40])
The sequence \(\{\zeta _{n}\}\) generated by Algorithm 3.2is monotonically nonincreasing and bounded below by \(\min \{\frac{\mu}{\ell _{0}},\zeta _{0} \}\). Moreover, \(\|\mathcal{L}z_{n}-\mathcal{L}w_{n}\|\leq \frac{\mu}{\zeta _{n+1}}\|z_{n}-w_{n}\|\), \(\forall n\geq 1\).
Proof
From the construction of the sequence \(\{\zeta _{n}\}\), it is obvious to see that \(\{\zeta _{n}\}\) is monotone decreasing. Since the operator \(\mathcal{L}\) is \(\ell _{0}\)-Lipschitz continuous, we have for \(\mathcal{L}w_{n}\neq \mathcal{L}z_{n}\)
By induction it is clear that \(\{\zeta _{n}\}\) is bounded below with lower bound \(\min \{\frac{\mu}{\ell _{0}},\zeta _{0} \}\). Since every nonincreasing monotone sequence that is bounded is convergent, it follows that the limit of \(\{\zeta _{n}\}\) exists and we can denote this by
□
Lemma 4.2
Let \(\{p_{n}\}\), \(\{z_{n}\}\) and \(\{w_{n}\}\) be sequences generated by Algorithm 3.2under Assumption 3.1. Then \(\{p_{n}\}\) satisfies the inequality
Proof
Let \(q^{*}\in \Gamma \). Using the definition of \(p_{n}\) in Step 2 of Algorithm 3.2, Lemma 2.1, Lemma 2.3 and Lemma 4.1, we have
Since \(\mathcal{L}\) is quasi-monotone on \(\mathcal{K}\) and \(z_{n}\in \mathcal{K}\), by Lemma 2.3 we have
Using the characterization of the metric projection in Lemma 2.1, we obtain
With the help of (4.3) and (4.4), we get
Also, from our adaptive step size (3.1),
which yields the desired result. □
Lemma 4.3
Let \(\{q_{n}\}\) be a sequence generated by Algorithm 3.2under Assumption 3.1. Then \(\{q_{n}\}\) is bounded.
Proof
Let \(q^{*}\in \Gamma \). Then from the definition of \((w_{n})\) in Algorithm 3.2, we have
Since \(\frac{\theta _{n}}{\sigma _{n}}\|q_{n}-q_{n-1}\|\leq \frac{\xi _{n}}{\sigma _{n}}\), from which by assumption (B5) it follows that \(\underset{n\to \infty}{\lim} \frac{\theta _{n}}{\sigma _{n}}\|q_{n}-q_{n-1}\|=0\), the sequence \(\{ \frac{\theta _{n}}{\sigma _{n}}\|q_{n}-q_{n-1}\| \}\) is bounded. Thus, there exists a real constant \(K_{1}>0\) such that \(\frac{\theta _{n}}{\sigma _{n}}\|q_{n}-q_{n-1}\|\leq K_{1}\) for all \(n\geq 1\). Hence,
Now, using the definition of sequence \(\{q_{n+1}\}\) in Algorithm 3.2, Lemma 4.2 and (4.10), we deduce
Thus, by induction
Since the sequence \(\{\|q_{n}-q^{*}\|\}\) is bounded, it follows that \(\{q_{n}\}\) is bounded. Hence, \(\{w_{n}\}\), \(\{z_{n}\}\), \(\{p_{n}\}\) and \(\{\mathcal{T}p_{n}\}\) are all bounded. □
Lemma 4.4
Let \(\{z_{n}\}\) and \(\{w_{n}\}\) be sequences generated by Algorithm 3.2under Assumption 3.1. Suppose \(\{z_{n_{\ell}}\}\) and \(\{w_{n_{\ell}}\}\) are subsequences of \(\{z_{n}\}\) and \(\{w_{n}\}\), respectively, with \(\{z_{n_{\ell}}\}\) converging weakly to a point \(q^{*}\in \mathscr{H}\) and \(\underset{\ell \to \infty}{\lim}\|w_{n_{\ell}}-z_{n_{\ell}}\|=0\). Then \(q^{*}\in VI (\mathcal{K},\mathcal{L} )\).
Proof
We know that \(z_{n_{\ell}}= P _{\mathcal{\ell}} (w_{n_{\ell}}- \zeta _{n_{\ell}}\mathcal{L}w_{n_{\ell}} )\). Hence, from the characterization of the projection operator (see Lemma 2.1), we get
This implies
and hence
Since \(\{z_{n_{k}}\}\) converges weakly to a point \(q^{*}\in \mathscr{H}\) and \(\underset{\ell \to \infty}{\lim}\|w_{n_{\ell}}-z_{n_{\ell}}\|=0\), it follows that \(\{w_{n_{\ell}}\}\) and \(\{\mathcal{L}w_{n_{\ell}}\}\) are bounded sequences. We have established also from Lemma 4.1 that \(\underset{\ell \to \infty}{\lim}\zeta _{n_{\ell}}=\zeta >0\). So from (4.13), we obtain
We observe that
Recall that the operator \(\mathcal{L}\) is \(\ell _{0}\)-Lipschitzian, so
Combining (4.14), (4.15) and (4.16), we infer
Using (4.17), we shall look at the following two cases.
Case A: Suppose that \(\underset{k\to \infty}{\limsup} \langle \mathcal{L}z_{n_{\ell}}, q-z_{n_{\ell}} \rangle >0\), \(\forall q\in \mathcal{K}\). Then there exists a subsequence \((z_{n_{\ell _{m}}})\) of sequence \((z_{n_{\ell}})\) such that \(\underset{m\to \infty}{\lim} \langle \mathcal{L}z_{n_{\ell _{m}}}, q-z_{n_{\ell _{m}}} \rangle >0\). This implies that one can find \(m_{0}\geq 1\) such that
By the quasi-monotonicity of the mapping \(\mathcal{L}\), it follows that
If we pass to the limit as \(m\to \infty \) in (4.18), we get
Hence, \(q^{*}\in VI (\mathcal{K},\mathcal{L} )\).
Case B: Suppose in (4.17),
We construct a nonincreasing positive sequence \(\{\eta _{\ell}\}\) defined by
Clearly \(\eta _{\ell}\to 0\) as \(\ell \to \infty \), so combining (4.19) and (4.20), we get
Since \(\{z_{n_{\ell}}\}\subset \mathcal{K}\), this implies that \(\{\mathcal{L}z_{n_{\ell}}\}\) is strictly nonzero. We let \(\underset{\ell \to \infty}{\lim}\|\mathcal{L}z_{n_{\ell}}\|=K_{2}>0\).
We can infer that
Also, we let \(\{\psi _{n_{\ell}}\}\) be a sequence given by \(\psi _{n_{\ell}}= \frac{\mathcal{L}z_{n_{\ell}}}{\|\mathcal{L}z_{n_{\ell}}\|^{2}}\). It follows that
By combining (4.21) and (4.23), we obtain
so
Since \(\mathcal{L}\) is quasi-monotone on \(\mathscr{H}\), we get
But we note that
Combining (4.25) and (4.26) and applying the Cauchy–Schwartz inequality, we obtain
Since \(\mathcal{L}\) is \(\ell _{0}\)-Lipschitz continuous, we have
Combining (4.22) and (4.28) and taking into consideration the definition of the sequence \(\{\varepsilon _{n_{\ell}}\}\), we have
Using (4.29), since \(\eta _{\ell}\to 0\) and \(z_{n_{\ell}}\to q^{*}\) as \(\ell \to \infty \), we get
Thus, \(q^{*}\in VI (\mathcal{K},\mathcal{L} )\), completing the proof. □
Lemma 4.5
Let \(\{q_{n}\}\) be a sequence generated by Algorithm 3.2under Assumption 3.1. Then for all \(n\geq 1\), \(\{q_{n}\}\) satisfies the inequality
where \(\Lambda _{n+1}:=\| q_{n+1}-q^{*}\|^{2}\), \(\Lambda _{n}:=\|q_{n}-q^{*}\|^{2}\), \(\varrho _{n}:=(1-\lambda )\sigma _{n}\) and \(\partial _{n}:= \frac{K_{3}K_{4}+ \langle h(q^{*})-q^{*}, q_{n+1}-q^{*} \rangle}{1-\lambda}\) for some real constants \(K_{3}\), \(K_{4}\).
Proof
Indeed, using the definition of the sequence \(\{q_{n+1}\}\) in Algorithm 3.2, Lemma 2.2 and Lemma 4.2, we have
But we observe that
where \(K_{3}:=\sup (2\| q_{n}-q^{*}\|+\theta _{n}\|q_{n}-q_{n-1}\| )<\infty \).
By combining (4.32) and (4.33),
which gives the desired result. □
Theorem 4.6
Let \(\{q_{n}\}\) be a sequence generated by Algorithm 3.2under Assumption 3.1. Then \(\{q_{n}\}\) converges strongly to \(q^{*}\in \Gamma := VI (\mathcal{K},\mathcal{L} )\cap F (\mathcal{T}) \iff \underset{n\to \infty}{\lim}\|q_{n+1}-q_{n}\|=0\).
Proof
Let \(q^{*}\in \Gamma \). To show that \(q_{n}\to q^{*}\in \Gamma \), we shall apply Lemma 2.4 and (4.35) and show that
for every subsequence \(\{\|q_{n_{\ell}}-q^{*}\|\}\) of the sequence \(\{\|q_{n}-q^{*}\|\}\) satisfying the condition
Consider \(\{\|q_{n_{\ell}}-q^{*}\|\}\) to be the subsequence of \(\{\|q_{n}-q^{*}\|\}\) satisfying condition (4.36). Then
This implies
Thus,
so
We observe from the definition of the sequence \(\{w_{n_{\ell}}\}\) that
Since \(\mathcal{L}\) is uniformly continuous and \(\ell _{0}\)-Lipschitzian, using (4.39), we get
Also, using (4.39) and (4.42),
Furthermore, using the definition of \(\{q_{n_{\ell}+1}\}\) in Algorithm 3.2, we deduce
which immediately implies that
Since \(\sigma _{n_{\ell}}\to 0\), \(q_{n_{\ell}+1}-q_{n_{\ell}}\to 0\) as \(\ell \to \infty \), using (4.40) and (4.44), we get
Since the sequence \(\{q_{n_{\ell}}\}\) is bounded, there exists a subsequence \(\{q_{n_{\ell _{k}}}\}\) of \(\{q_{n_{\ell}}\}\) such that \(\{q_{n_{\ell _{k}}}\}\) converges weakly to \(z^{*}\in \mathscr{H}\) as \(k\to \infty \). Using (4.41), we have \(w_{n_{\ell _{k}}}\rightharpoonup z^{*}\in \mathscr{H}\) as \(k\to \infty \). Also, since \(\{p_{n_{\ell}}\}\) is bounded, there exists a subsequence \(\{p_{n_{\ell _{k}}}\}\) which converges weakly to \(z^{*}\in \mathscr{H}\). Since \(\mathcal{T}\) is semiclosed at the origin, by (4.45), we infer that \(z^{*}\in F (\mathcal{T})\). Thus, by (4.45) and Lemma 4.4, \(z^{*}\in \Gamma \).
Furthermore, since \(q_{n_{\ell _{k}}}\rightharpoonup z^{*}\), it follows that
However, since \(z^{*}\) is a unique solution in Γ, this implies that
Furthermore,
Using Assumption 3.1 (B5) and (4.48),
It follows from Lemma 2.4 that \(\underset{n\to \infty}{\lim}\|q_{n}-q^{*}\|=0\), completing the proof. □
5 Numerical experiments
In this section, we will give some special numerical examples which show the behavior of our proposed iterative method in comparison to Algorithm (3.4) of Gang et al. [14], Algorithm (3.6) of Thong et al. [35] and Algorithm (3.2) of Zhao et al. [44].
Example 5.1
Let \(\mathscr{H}=\ell _{2} (\mathbb{R} )\) be a real Hilbert space of square summable sequences of real numbers equipped with the property
Let \(\mathcal{L}:\mathcal{K}\to \mathcal{K}\) be a mapping defined by
where \(\mathcal{K}:=\{q\in \mathscr{H}:\|q\|\leq \gamma \}\), \(\sigma ,\gamma \in \mathbb{R}\), such that \(\gamma ,\sigma >0\). It is can be easily seen that \(\mathcal{L}\) is weakly sequentially continuous on \(\mathscr{H}\) and \(VI (\mathcal{K},\mathcal{L})=\{0\}\). Now for \(q, s\in \mathcal{K}\), we get
This implies \(\mathcal{L}\) is \(\ell _{0}\)-Lipschitzian with \(\ell _{0}=(\sigma +2\gamma )\). It is also easily verifiable that the operator \(\mathcal{L}\) is quasi-monotone for all \(0<\frac{\sigma}{2}<\gamma <\sigma \). To see this, suppose \(\langle \mathcal{L}q, s-q \rangle >0\) for \(q,s\in \mathcal{K}\) such that
Then from the definition of the feasible set \(\mathcal{K}\), it follows that
The operator \(\mathcal{L}\) is quasi-monotone on \(\mathcal{K}\). We compute the metric projection on \(\mathcal{K}\) as follows:
We define a map \(\mathcal{T}:\ell _{2} (\mathbb{R} )\to \ell _{2} ( \mathbb{R} )\) by \(\mathcal{T}q=\frac{q\sin q}{2}\). As seen in [44], \(\mathcal{T}\) is quasi-nonexpansive and semiclosed at 0 with \(F (\mathcal{T})=\{0\}\). Finally, let \(h:\mathscr{H}\to \mathscr{H}\) be a mapping defined by \(h(q)=\frac{1}{2}q\). It can be seen easily that h is \(\frac{1}{2}\)-contractive. We take \(F (q)=\frac{1}{2}q\) [44]. Suppose for this example \(\zeta _{0}=\frac{1}{2}\), \(\mu =0.5\), \(\gamma =2.0\), \(\theta =\frac{2}{5}\), \(\zeta _{n+1}=\frac{100}{(n+1)^{1.3}}\), \(\sigma _{n}=\frac{1}{100n^{2}+1}\) for Algorithm 3.2. Also, if we consider \(TOL=\|q_{n}-q_{n-1}\|\leq 10^{-5}\) as the stopping criterion and choose the following different initial points:
-
Case A: \(q_{0}= (1,\frac{1}{2}, \frac{1}{4},\ldots )\), \(q_{1}= (-2, 1,-\frac{1}{2},\ldots )\),
-
Case B: \(q_{0}= (1,\frac{1}{2}, \frac{1}{4},\ldots )\), \(q_{1}= (-3, 1,-\frac{1}{3},\ldots )\),
-
Case C: \(q_{0}= (5,1,\frac{1}{5},\ldots )\), \(q_{1}= (-5, 1,- \frac{1}{5},\ldots )\),
-
Case D: \(q_{0}= (2,1,\frac{1}{2}, \ldots )\), \(q_{1}= (-3, 1,- \frac{1}{3},\ldots )\),
then we obtain the results in Table 1.
Example 5.2
([33])
Let \(\mathscr{H}= L _{2} ([0,1] )\) be an infinite-dimensional linear space of square integrable functions associated with the inner product \(\langle q,s \rangle =\int _{0}^{1}q(t)s(t)\,dt \forall q\), \(s \in L _{2} ([0,1] )\), \(t\in [0,1]\), and induced norm \(\|q\|_{L_{2}}= (\int _{0}^{1}|q(t)|^{2}|\,dt )^{\frac{1}{2}}\), \(\forall q\in L _{2} ([0,1] )\). Consider a closed unit ball \(\mathcal{K}:= \{q\in L _{2} ([0,1] ):\|q\| \leq 1 \}\). Define a projection from \(L _{2} ([0,1] )\) onto \(\mathcal{K}\) by
Let \(\mathcal{L}: L _{2} ([0,1] )\to L _{2} ([0,1] )\) be an operator defined by \((\mathcal{L}q)(t)=\max \{0,\frac{q(t)}{2} \}\), \(\forall q \in L _{2} ([0,1] )\), t\(\in [0,1]\). It is easily verifiable that \(\mathcal{L}\) is pseudo-monotone (so, quasi-monotone), and \(VI(\mathcal{L},\mathcal{K})=\{0\}\). Also, let \(\mathcal{T}: L _{2} ([0,1] )\to L _{2} ([0,1] )\) be a mapping given by \((\mathcal{T}q)(t)=\int _{0}^{1}tq(s)\,ds\), \(t\in [0,1]\). The operator \(\mathcal{T}\) defined here is nonexpansive and hence quasi-nonexpansive (see [33]). It can also be easily seen that \(0\in F (\mathcal{T})\). Thus, \(\Gamma :=VI(\mathcal{L},\mathcal{K})\cap F (\mathcal{T}) \neq \emptyset \). Let \(h: L _{2} ([0,1] )\to L _{2} ([0,1] )\) be a mapping defined by \((h(q))(t)=\frac{1}{2}q(t)\), \(t\in [0,1]\). Then h is a contraction. If we take \(TOL=\|q_{n}-q_{n-1}\|\leq 10^{-5}\) as the stopping criterion, we obtain the following table and graphs using the following as starting points:
Case A: \(q_{0}(t)=2t^{3}+t\), \(q_{1}(t)=\sin (t^{2})\),
Case B: \(q_{0}(t)=2e^{t}+t^{2}\), \(q_{1}(t)=\sin (2t)\),
Case C: \(q_{0}(t)=2e^{t+1}+t\), \(q_{1}(t)=\sin (2t^{2})\),
Case D: \(q_{0}(t)=e^{t^{2}}+t^{3}\), \(q_{1}(t)=\cos (2t^{3})+e^{t}-2t^{3}\).
6 Conclusion
A modified Tseng inertial iterative algorithm for solving quasi-monotone variational inequality and fixed point problems when the underlying operator is quasi-nonexpansive in real Hilbert spaces was introduced and studied. We established strong convergence of the proposed Algorithm 3.2 without prior knowledge of the Lipschitz constant of the cost operator under the adaptive step size condition and other certain mild assumptions on the algorithm parameters. The algorithm, which is embellished with inertial extrapolation and viscosity terms as well as good relaxation of the cost operator (quasi-monotone operator), generally exhibited enhanced efficiency and applicability. Finally, we presented some numerical experiments to demonstrate the applicability and the advantages of our algorithm. Tables 1 and 2 and Figs. 1 and 2 reveal that our algorithm perform more favourable in comparison to some related methods in literature. Our results complement and extend some recent results in the literature.
Data availability
Not applicable.
References
Abuchu, J.A., Ugwunnadi, G.C., Narain, O.K.: Inertial Mann-type iterative method for solving split monotone variational inclusion problem with applications. J. Ind. Manag. Optim. 19(4), 3020–3043 (2023)
Abuchu, J.A., Ugwunnadi, G.C., Narain, O.K.: Iterative method for solving pseudomonotone variational inequality and fixed point problems. Adv. Nonlinear Var. Inequal. 26(1), 1–24 (2023)
Anh, P.K., Anh, T.V., Muu, L.D.: On bilevel split pseudomonotone variational inequality problems with applications. Acta Math. Vietnam. 42, 413–429 (2017)
Antipin, A.S.: On a method for convex programs using a symmetrical modification of the Lagrange function. Èkon. Mat. Metody 12, 1164–1173 (1976)
Byrne, C.: A unified treatment of some iterative algorithms in signal processing and image reconstruction. Inverse Probl. 20(103) (2003)
Censor, Y., Gibali, A., Reich, S.: The subgradient extragradient method for solving variational inequalities in Hilbert space. J. Optim. Theory Appl. 148, 318–335 (2011)
Censor, Y., Gibali, A., Reich, S.: Extensions of Korpelevich’s extragradient method for the variational inequality problem in Euclidean space. Optimization 61, 1119–1132 (2012)
Cholamjiak, P., Thong, D.V., Cho, Y.J.: A Novel Inertial Projection and Contraction Method for Solving Pseudomonotone Variational Inequality Problems. Acta Appl. Math. (2019)
Cottle, R.W., Yao, J.C.: Pseudo-monotone complementarity problems in Hilbert space. J. Optim. Theory Appl. 75, 281–295 (1992)
Dong, L.Q., Yang, J.F., Yuan, H.B.: The projection and contraction algorithm for solving variational inequality problems in Hilbert space. J. Nonlinear Convex Anal. 20, 111–122 (2019)
Dong, Q.L., Cho, J.Y., Rassias, T.M.H.: The projection and contraction methods for finding common solutions to variational inequality problems. Optim. Lett. 12, 1871–1896 (2018)
Ferris, M.C., Pang, J.S.: Engineering and economic applications of complementarity problems. SIAM Rev. 39(4), 669–713 (1997)
Fichera, G.: Sul pproblem elastostatico di Signorini con ambigue condizioni al contorno. Atti Accad. Naz. Lincei, Rend. Cl. Sci. Fis. Mat. Nat. 34, 138–142 (1963)
Gang, C., Shehu, Y., Iyiola, O.S.: Inertial Tseng’s extragradient method for solving variational inequality problems of pseudo-monotone and non-Lipschitz operators. J. Ind. Manag. Optim. 18(4), 2873–2902 (2022)
Godwin, E.C., Alakoya, T.O., Mewomo, O.T., Yao, J.C.: Relaxed inertial Tsengextragradient method for variational inequality and fixed point problems. Appl. Anal. https://doi.org/10.1080/00036811.2022.2107913
Goldstein, A.A.: Convex programming in Hilbert space. Bull. Am. Math. Soc. 70, 709–710 (1964)
Hadjisavvas, N., Schaible, S.: Quasimonotone variational inequalities in Banach spaces. J. Optim. Theory Appl. 90, 95–111 (1996)
He, B.S., Yang, Z.H., Yuan, X.M.: An approximate proximal-extragradient type method for monotone variational inequalities. J. Math. Anal. Appl. 300, 362–374 (2004)
Korpelevich, G.M.: An extragradient method for finding sadlle points and for other problems. Èkon. Mat. Metody 12, 747–756 (1976)
Maingé, P.E.: Convergence theorems for inertial KM-type algorithms. J. Comput. Appl. Math. 219(1), 223–236 (2008)
Nadezhkina, N., Takahashi, W.: Weak convergence theorem by an extragradient method for nonexpansive mappings and monotone mappings. J. Optim. Theory Appl. 128, 191–201 (2006)
Noor, M.: Extragradient methods for pseudomonotone variational inequalities. J. Optim. Theory Appl. 117, 475–488 (2003)
Ogwo, G.N., Izuchukwu, C., Shehu, Y., Mewomo, O.T.: Convergence of relaxed inertial subgradient extragradient methods for quasimonotone variational inequality problems. J. Sci. Comput. 90 (2022)
Polyak, B.T.: Some methods of speeding up the convergence of iteration methods. Comput. Math. Math. Phys. 4(5), 1–17 (1964)
Popov, L.D.: A modification of the Arrow–Hurwicz method for searching for saddle points. Mat. Zametki 28, 777–784 (1980)
Saejung, S., Yotkaew, P.: Approximation of zeros of inverse strongly monotone operators in Banach spaces. Nonlinear Anal. 75, 742–750 (2012)
Shehu, Y., Vuong, P.T., Zemkoho, A.: An inertial extrapolation method for convex simple bilevel optimization. Optim. Methods Softw. (2019)
Stampacchia, G.: Variational inequalities. In: Theory and Applications of Monotone Operators. Proceedings of the NATO Advanced Study Institute, pp. 102–192 (1968). Venice, Italy (Edizioni Odersi, Gubbio, Italy)
Tan, B., Liu, L., Qin, X.: Self adaptive inertial extragradient algorithms for solving bilevel pseudomonotone variational inequality problems. Jpn. J. Ind. Appl. Math. (2020)
Thong, D.V., Hieu, D.V.: Weak and strong convergence theorems for variational inequality problems. Numer. Algorithms 78, 1045–1060 (2018)
Thong, D.V., Hieu, D.V.: New extragradient methods for solving variational inequality problems and fixed point problems. J. Fixed Point Theory Appl. 20, 1–29 (2018)
Thong, D.V., Hieu, D.V.: Modified Tseng’s extragradient algorithms for variational inequality problems. J. Fixed Point Theory Appl. 20 (2018)
Thong, D.V., Hieu, D.V.: Inertial subgradient extragradient algorithms with line-search process for solving variational inequality problems and fixed point problems. Numer. Algorithms 80, 1283–1307 (2019)
Thong, D.V., Li, X.H., Dong, Q.L., Cho, Y.J., Rassias, T.M.: A projection and contraction method with adaptive step sizes for solving bilevel pseudomonotone variational inequality problems. Optimization (2020)
Thong, D.V., Shehu, Y., Iyiola, O.S.: Weak and strong convergence theorems for solving pseudo-monotone variational inequalities with non-Lipschitz mappings. Numer. Algorithms 84, 795–823 (2020)
Thong, D.V., Triet, N.A., Li, X.-H., Dong, Q.-L.: Strong convergence of extragradient methods for solving bilevel pseudo-monotone variational inequality problems. Numer. Algorithms 83, 1123–1143 (2020)
Tseng, P.: A modified forward-backward splitting method for maximal monotone mappings. SIAM J. Control Optim. 38, 431–446 (2000)
Ugwunnadi, G.C., Izuchukwu, C., Jolaoso, L.O., Okeke, C.C., Aremu, K.O.: A modified inertial projection and contraction method for solving bilevel split variational inequality problems. Appl. Set-Valued Anal. Optim. 4(1), 55–71 (2022)
Wang, F., Xu, H.-K.: Weak and strong convergence theorems for variational inequality and fixed point problems with Tseng’s extragradient method. Taiwan. J. Math. 16, 1125–1136 (2012)
Yang, J., Liu, H.: Strong convergence result for solving monotone variational inequalities in Hilbert space. Numer. Algorithms, 741–752 (2019)
Yao, Y., Iyiola, O.S., Shehu, Y.: Subgradient extragradient method with double inertial steps for variational inequalities. J. Sci. Comput. 90(2), 71 (2022)
Yao, Y., Shahzad, N., Yao, J.C.: Convergence of Tseng-type self-adaptive algorithms for variational inequalities and fixed point problems. Carpath. J. Math. 37, 541–550 (2021)
Zegeye, H., Shahzad, N., Yao, Y.: Minimum-norm solution of variational inequality and fixed point problem in Banach spaces. Optimization 64, 453–471 (2015)
Zhao, T., Wang, D., Ceng, L., He, L., Wang, C., Fan, H.: Quasi-inertial Tseng’s extragradient algorithms for pseudomonotone variational inequalities and fixed point problems of quasi-nonexpansive operators. Numer. Funct. Anal. Optim. 42, 69–90 (2021)
Zhao, X., Yao, Y.: Modified extragradient algorithms for solving monotonevariational inequalities and fixed point problems. Optimization 69, 1987–2002 (2020)
Zheng, L.: A double projection algorithm for quasimonotone variational inequalities in Banach spaces. J. Inequal. Appl. (2018)
Funding
This work did not receive any external funding.
Author information
Authors and Affiliations
Contributions
J.A.A. and A.E.O. made conceptualization, methodology and writing draft preparation. G.C.U. and H.I. performed the formal analysis, writing-review and editing. O.K.N. made investigation, review and validation. All authors read and approved the final version.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
This study does not involve any experiments with human participants or animals.
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Abuchu, J.A., Ofem, A.E., Işık, H. et al. Modified inertial viscosity extrapolation method for solving quasi-monotone variational inequality and fixed point problems in real Hilbert spaces. J Inequal Appl 2024, 38 (2024). https://doi.org/10.1186/s13660-024-03113-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-024-03113-5