- Research
- Open Access
- Published:
Convergence analysis of a general inertial projection-type method for solving pseudomonotone equilibrium problems with applications
Journal of Inequalities and Applications volume 2021, Article number: 63 (2021)
Abstract
In this paper, we introduce a new algorithm by incorporating an inertial term with a subgradient extragradient algorithm to solve the equilibrium problems involving a pseudomonotone and Lipschitz-type continuous bifunction in real Hilbert spaces. A weak convergence theorem is well established under certain mild conditions for the bifunction and the control parameters involved. Some of the applications to solve variational inequalities and fixed point problems are considered. Finally, several numerical experiments are performed to demonstrate the numerical efficacy and superiority of the proposed algorithm over other well-known existing algorithms.
1 Introduction
Let C be a closed and convex subset of a real Hilbert space \(\mathbb{H}\). The inner product and the induced norm on \(\mathbb{H}\) are denoted by \(\langle \cdot , \cdot \rangle \) and \(\|\cdot \|\), respectively. Assume that \(f : \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) is a bifunction with \(f (y, y) = 0\) for all \(y \in C\). The equilibrium problem (EP) for a bifunction f on C is defined in the following way [10, 17]:
Moreover, \(S_{\operatorname{EP}(f, C)}\) stands for the solution set of an equilibrium problem over the set C and \(\xi ^{*}\) is an arbitrary element of \(S_{\operatorname{EP}(f, C)}\). A bifunction \(f : \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) is said to be (see for more details [8, 10]):
-
(1)
strongly monotone on C if there exists \(\gamma > 0\) such that
$$ f(x_{1}, x_{2}) + f(x_{2}, x_{1}) \leq -\gamma \Vert x_{1} - x_{2} \Vert ^{2}, \quad \forall x_{1}, x_{2} \in C; $$ -
(2)
monotone on C if
$$ f(x_{1}, x_{2}) + f(x_{2}, x_{1}) \leq 0,\quad \forall x_{1}, x_{2} \in C; $$ -
(3)
strongly pseudomonotone on C if there exists \(\gamma > 0\) such that
$$ f (x_{1}, x_{2}) \geq 0 \quad \Longrightarrow\quad f(x_{2}, x_{1}) \leq -\gamma \Vert x_{1} - x_{2} \Vert ^{2}, \quad \forall x_{1}, x_{2} \in C; $$ -
(4)
pseudomonotone on C if
$$ f (x_{1}, x_{2}) \geq 0\quad \Longrightarrow\quad f(x_{2}, x_{1}) \leq 0, \quad \forall x_{1}, x_{2} \in C. $$
It is clear from the above definitions that the following implications hold:
In general, the reverse implications do not hold. A bifunction \(f : \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) is said to be Lipschitz-type continuous [28] on C if there exist two constants \(c_{1}, c_{2} > 0\) such that
The above-defined problem (EP) is a general mathematical problem in the sense that it unifies a number of mathematical problems, i.e., the fixed point problems, the vector and scalar minimization problems, the variational inequality problems (VIP), the complementarity problems, the saddle point problems, the Nash equilibrium problems in non-cooperative games, and the inverse optimization problems [9, 10, 30]. The problem (EP) is also known as the well-known Ky Fan inequality [17]. Many authors have established and generalized several results on the existence and nature of the solution of an equilibrium problem (see for more details [5, 9, 17]).
A number of effective algorithmic schemes have been well established along their convergence analysis to solve the equilibrium problems in finite and in-finite dimensional spaces. The regularization method is one of the most important approaches to solving various ill-posed problems in different fields of pure and applied mathematics. A significant feature of the regularization methodology is that it has been applied to solve monotone equilibrium problems, and the original problem is transformed into strongly monotone sub-problems. Thus, each sub-problem is strongly monotone and guarantees the existence of a unique solution. In particular, the formalized sub-problem can be resolved more effectively than the original monotone problem, and the sequence of regularization solutions converges to one solution of the initial problem once the regularization variables appear to have an appropriate limit. The proximal point method and Tikhonov’s regularized method are two famous regularization approaches. Recently, these approaches have been extended in the case of equilibrium problems (see [20, 25, 29, 31] for more details) and others types on the method in [1–3, 13, 21–24, 34, 35, 39–41].
The proximal point method is used to solve the problem (EP) and is also known as the two-step extragradient method in [37] due to the previous contribution of Korpelevich [26] to solve the saddle point problems. Tran et al. [37] established a weakly convergent iterative sequence \(\{x_{n}\}\) to solve the monotone equilibrium problem in a real Hilbert space. The method has the following form:
where \(0 < \lambda < \min \{\frac{1}{2 c_{1}}, \frac{1}{2 c_{2}} \}\). On the other hand, inertial-like methods are two-step iterative methods, and the next iteration is obtained from the previous two iterations (see [33] for more details). An inertial extrapolation term is used to enhance the iterative sequence performance in order to improve its rate of convergence. Numerical results suggest that inertial effects improve algorithmic efficiency during running time and the number of iterations. Recently, several inertial methods have been developed to solve different classes of equilibrium problems [14–16, 19, 38, 42].
In this paper, we first introduce a new inertial subgradient algorithm to solve a pseudomonotone equilibrium problem involving the Lipschitz-type condition in a real Hilbert space. This algorithm is designed around three methods: the extragradient method [37], the subgradient extragradient method [12], and the inertial method [33]. The weak convergence of the resulting algorithm is well established under mild conditions. Some of the applications to resolve variational inequalities and fixed point problems are considered. Numerical results are provided to show the computational effectiveness of our algorithm. Finally, the numerical evaluation demonstrates that the new method is more effective than the family of existing methods [19, 37, 38, 42].
The rest of the article has been arranged as follows: Sect. 2 includes some preliminary and basic results. Section 3 includes our proposed method and its convergence analysis. In Sect. 4 we present two mathematical applications of our proposed scheme, variational inequalities, and fixed points. Finally, in Sect. 5 we provide numerical examples with comparison with other related results in the literature in order to illustrate the validity and practical advantages of our method.
2 Preliminaries
The normal cone of C at \(x \in C\) is defined by
Let \(h : C \rightarrow \mathbb{R}\) be a convex function. The subdifferential of h at \(x \in C\) is defined by
The metric projection \(P_{C}(x)\) of \(x \in \mathbb{H}\) onto a closed and convex subset C of \(\mathbb{H}\) is defined by
Lemma 2.1
([7])
For any \(x, y \in \mathbb{H}\) and \(\kappa \in \mathbb{R}\), the following relationship is true:
Lemma 2.2
([36])
Let \(h: C \rightarrow \mathbb{R}\) be a subdifferentiable, convex, and lower semi-continuous function on C. An element \(x \in C\) is said to be a minimizer of a function h iff
Lemma 2.3
([6])
Let \(\{\vartheta _{n}\}\), \(\{\theta _{n}\}\), and \(\{\gamma _{n}\}\) be nonnegative sequences that satisfy the following condition:
where \(\sum_{n=1}^{+\infty } \gamma _{n} < +\infty \), and let \(\theta >0\) be such that \(0 \leq \theta _{n} \leq \theta < 1\) for each \(n \in \mathbb{N}\). Then we have
-
(i)
\(\sum_{n=1}^{+\infty } [\vartheta _{n} - \vartheta _{n-1}]_{+} < +\infty \) with \([a]_{+} := \max \{a, 0\}\) for any \(a \in \mathbb{R}\);
-
(ii)
\(\lim_{n \rightarrow +\infty } \vartheta _{n} = \vartheta ^{*} \in [0, \infty )\).
Lemma 2.4
([32])
Let \(\{\eta _{n}\}\) be a sequence in \(\mathbb{H}\) and \(C \subset \mathbb{H}\) such that
-
(i)
for every \(\eta \in C\), \(\lim_{n\rightarrow \infty } \|\eta _{n} - \eta \|\) exists;
-
(ii)
each weak sequentially cluster point of the sequence \(\{\eta _{n}\}\) belongs to C.
Then \(\{\eta _{n}\}\) converges weakly to an element of C.
In order to study convergence analysis, we assume that \(f: \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) satisfies the following conditions:
-
(A1)
\(f (y, y) = 0\) for all \(y \in C\) and f is pseudomonotone on a feasible set C.
-
(A2)
f satisfies the Lipschitz-type condition on \(\mathbb{H}\) with \(c_{1}>0\) and \(c_{2}>0\).
-
(A3)
\(\limsup_{n\rightarrow \infty } f(x_{n}, y) \leq f(p^{*}, y)\) for all \(y \in C\) and \(\{x_{n}\} \subset C\) satisfy \(x_{n} \rightharpoonup p^{*}\).
-
(A4)
\(f (x,\cdot )\) is subdifferentiable and convex on \(\mathbb{H}\) for every fixed \(x \in \mathbb{H}\).
3 Main result
In this section, we introduce the main method and prove a weak convergence result. The main algorithm is in the following form:
(An inertial subgradient extragradient method for problem (EP))
Lemma 3.1
Let a bifunction \(f : \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) satisfy conditions (A1)–(A4). For each \(\xi ^{*} \in S_{\operatorname{EP}(f, C)}\), we have
Proof
By using Lemma 2.2, we have
Thus, there exist \(\omega \in \partial f(y_{n}, z_{n})\) and \(\overline{\omega } \in N_{H_{n}}(z_{n})\) such that
Thus, the above implies that \(\langle \varrho _{n} - z_{n}, y - z_{n} \rangle = \lambda \langle \omega , y - z_{n} \rangle + \langle \overline{\omega }, y - z_{n} \rangle \) for all \(y \in H_{n}\). Since \(\overline{\omega } \in N_{H_{n}}(z_{n})\), it follows that \(\langle \overline{\omega }, y - z_{n} \rangle \leq 0\) for all \(y \in H_{n}\). Thus, we have
Since \(\omega \in \partial f(y_{n}, z_{n})\) and by using the subdifferential definition, we have
Combining expressions (4) and (5), we obtain
By substituting \(y=\xi ^{*}\) in expression (6), we obtain
It is given that \(\xi ^{*} \in S_{\operatorname{EP}(f, C)}\) implies that \(f(\xi ^{*}, y_{n}) \geq 0\) and \(f(y_{n}, \xi ^{*}) \leq 0\) due to the pseudomonotonicity of the bifunction f. From expression (7), we have
Due to the Lipschitz-type continuity of a bifunction f, we have
Combining expressions (8) and (9), we have
Since \(z_{n} \in H_{n}\) and by the definition of \(H_{n}\), we obtain
which implies that
Since \(t_{n} \in \partial _{2} f(\varrho _{n}, y_{n})\), by using the subdifferential definition, we have
By substituting \(y = z_{n}\) in the above expression, we obtain
It follows from inequalities (11) and (12) that
We have the following equalities:
and
The above facts and (14) complete the proof. □
Now, we are in a position to prove our weak convergence theorem.
Theorem 3.2
The sequences \(\{\varrho _{n}\}\), \(\{x_{n}\}\), \(\{y_{n}\}\), and \(\{z_{n}\}\) generated by Algorithm 1 are weakly convergent to an element \(\xi ^{*} \in S_{\operatorname{EP}(f, C)}\).
Proof
From the value of \(x_{n+1}\) and Lemma 2.1, we have
By using Lemma 3.1, we have
Combining expressions (15) and (16), we obtain
where \(b=\max \{2c_{1}, 2c_{2}\}\). From the value of \(x_{n+1}\), we have
Combining (17) and (18), we have
Due to the condition on λ, we infer that \(\frac{1-b\lambda }{2} \geq \frac{1}{6}\), and expression (19) is converted into
By taking the value of \(\varrho _{n}\), we have
By taking the value of \(\varrho _{n}\), we have
where \(\rho _{n}=\frac{1}{\delta \beta _{n}+\vartheta _{n}}\). It follows from (20), (21), and (23) that
where
By taking the value \(\{\rho _{n}\}\), we have
Set \(\Psi _{n} = \|x_{n} - \xi ^{*} \|^{2} - \vartheta _{n} \|x_{n-1} - \xi ^{*}\|^{2} + \gamma _{n} \|x_{n} - x_{n-1}\|^{2} \). By using (25), we have
Observe that
From expressions (2), (3), and (27), we have
It follows that
The above implies that the sequence \(\{\Psi _{n}\}\) is nonincreasing. From \(\Psi _{n+1}\) we have
From the value of \(\Psi _{n}\), we have
From expression (33), we obtain
Combining expressions (33) and (34), we obtain
Following inequalities (31) and (35), we can write
By letting \(k \rightarrow +\infty \) in (36), we get
which implies that
From the value of \(x_{n+1}\), we have
and from (39) and (40), we have
By using the triangular inequality with expressions (38) and (39), we obtain
and
From (24), (37), and Lemma 2.3, it follows that
From expressions (42) and (43), we have
Thus, Lemma 3.1 implies that
with expressions (44) and (45), we have
The above implies that the sequences \(\{x_{n}\}\), \(\{\varrho _{n}\}\), \(\{y_{n}\}\), and \(\{z_{n}\}\) are bounded, and for each \(\xi ^{*} \in S_{\operatorname{EP}(f, C)}\), \(\lim_{n\rightarrow \infty } \|x_{n} - \xi ^{*}\|^{2}\) exists. Next, our aim is to prove that the solution set \(S_{\operatorname{EP}(f, C)}\) contains all sequential weak cluster points of the sequence \(\{x_{n}\}\). Let z be an arbitrary weak cluster point of the sequence \(\{x_{n}\}\). Then there exists a subsequence \(\{x_{n_{k}}\}\) of \(\{x_{n}\}\) such that \(\{x_{n_{k}}\}\) weakly converges to z. It follows from (42) and (43) that the subsequences \(\{y_{n_{k}}\}\) and \(\{z_{n_{k}}\}\) are weakly convergent to \(z \in C\). Next, we show that \(z \in S_{\operatorname{EP}(f, C)}\). Due to inequality (6), the Lipschitz-type continuity of f and (13), we obtain
where y is an arbitrary element in \(H_{n}\). It follows from (41), (42), (43), and the boundedness of \(\{x_{n}\}\) that the right-hand side goes to zero. From \(\lambda > 0\), condition (A3), and \(y_{n_{k}} \rightharpoonup z\), we have
The above implies that \(f(z, y) \geq 0\) for all \(y \in C\), and hence \(z \in S_{\operatorname{EP}(f, C)}\). This completes the proof. □
Setting \(\vartheta _{n}=0\) in Algorithm 1, we have the following variant of Anh et al. [4].
Corollary 3.3
Let \(f : \mathbb{H} \times \mathbb{H} \rightarrow \mathbb{R}\) be a bifunction satisfying conditions (A1)–(A4). Then \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) are the sequences generated in the following manner:
-
Initialization: Let \(x_{0} \in \mathbb{H}\) arbitrarily and choose \(0 < \lambda < \min \{\frac{1}{2c_{1}}, \frac{1}{2c_{2}} \}\), and \(\{\beta _{n}\}\) is a control parameter.
-
Step 1: Compute \(y_{n} = \arg \min _{y \in C} \{ \lambda f(x_{n}, y) + \frac{1}{2} \|x_{n} - y\|^{2} \} \). If \(y_{n} = x_{n}\), then stop and \(x_{n}\) is the solution of the problem (EP). Otherwise go to the next step.
-
Step 2: Next, construct a half-space \(H_{n} = \{z \in \mathbb{H} : \langle x_{n} - \lambda t_{n} - y_{n}, z - y_{n} \rangle \leq 0 \} \), where \(t_{n} \in \partial _{2} f(x_{n}, y_{n})\), and then compute \(z_{n} = \arg \min _{y \in H_{n}} \{ \lambda f(y_{n}, y) + \frac{1}{2} \|x_{n} - y\|^{2} \} \).
-
Step 3: Evaluate \(x_{n+1} = (1- \beta _{n}) x_{n} + \beta _{n} z_{n} \), where \(\{\beta _{n}\}\) is a real sequence and there exist \(\beta , \delta , \sigma > 0\) such that \(0 < \beta \leq \beta _{n} \leq \frac{1}{6 \sigma }\). Then the sequences \(\{x_{n}\}\), \(\{y_{n}\}\), and \(\{z_{n}\}\) are weakly convergent to \(\xi ^{*} \in S_{\operatorname{EP}(f, C)}\).
4 Applications
4.1 Variational inequalities
In this subsection we apply our results for solving variational inequality problems involving a pseudomonotone and Lipschitz-type continuous operator. Let us recall that an operator \(G : \mathbb{H} \rightarrow \mathbb{H}\) is said to be
-
(1)
monotone on C if
$$ \bigl\langle G(x) - G(y), x - y \bigr\rangle \geq 0,\quad \forall x, y \in C; $$ -
(2)
L-Lipschitz continuous on C if
$$ \bigl\Vert G(x) - G(y) \bigr\Vert \leq L \Vert x - y \Vert ,\quad \forall x, y \in C; $$ -
(3)
pseudomonotone on C if
$$ \bigl\langle G(x), y - x \bigr\rangle \geq 0 \quad \Longrightarrow \quad \bigl\langle G(y), x - y \bigr\rangle \leq 0, \quad \forall x, y \in C. $$
The variational inequality problem is defined as follows:
If we define \(f(x, y) := \langle G(x), y - x \rangle \) for all \(x, y \in C\), then the equilibrium problem becomes the problem of variational inequality described above where \(L = 2c_{1} = 2 c_{2}\). From the above value of the bifunction f, we have
The value \(z_{n}\) in Algorithm 1 is converted into
Since \(t_{n} \in \partial _{2} f(\varrho _{n}, y_{n})\) and by the subdifferential definition, we obtain
which further implies that \(0 \leq \langle G(\varrho _{n}) - t_{n}, z - y_{n} \rangle \) for all \(z \in \mathbb{H}\). Thus
Suppose that G satisfies the following conditions:
-
(G1)
G is pseudomonotone on C and a solution set \(\operatorname{VI}(G, C) \neq \emptyset \);
-
(G2)
G is L-Lipschitz continuous on C through a positive constant \(L > 0\);
-
(G3)
\(\limsup_{n\rightarrow \infty } \langle G(x_{n}), y-x_{n} \rangle \leq \langle G(p^{*}), y-p^{*} \rangle \) for all \(y \in C\), where \(\{x_{n}\} \subset C\) satisfies \(x_{n} \rightharpoonup p^{*}\).
As a consequence of the results in Sect. 3, we have the following results.
Corollary 4.1
Let \(G : C \rightarrow \mathbb{H}\) be a mapping satisfying conditions (G1)–(G3). Let \(\{\varrho _{n}\}\), \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) be the sequences generated in the following way:
-
Initialization: Let \(x_{-1}, x_{0} \in \mathbb{H}\), \(0 < \lambda < \frac{1}{L}\), \(\{\vartheta _{n}\}\) and \(\{\beta _{n}\}\) are control parameters.
-
Step 1: Compute \(y_{n} = P_{C} (\varrho _{n} - \lambda G(\varrho _{n}))\), where \(\varrho _{n} = x_{n} + \vartheta _{n}(x_{n} - x_{n-1})\). If \(y_{n} = \varrho _{n}\), then \(\varrho _{n}\) is a solution of the problem (VIP).
-
Step 2: Construct a half-space \(H_{n} = \{z \in \mathbb{H} : \langle \varrho _{n} - \lambda G ( \varrho _{n}) - y_{n}, z - y_{n} \rangle \leq 0 \} \) and compute \(z_{n} = P_{H_{n}} (\varrho _{n} - \lambda G(y_{n}))\).
-
Step 3: Compute \(x_{n+1} = (1- \beta _{n}) \varrho _{n} + \beta _{n} z_{n}\), where \(\{\vartheta _{n}\}\) and \(\{\beta _{n}\}\) are real sequences. Here, we assume that the sequence \(\{\vartheta _{n}\}\) is nondecreasing with \(0 \leq \vartheta _{n} \leq \vartheta < 1\) for each \(n \geq 1\) and that there exist \(\beta , \delta , \sigma > 0\) such that
$$ \delta > \frac{6 \vartheta [ \vartheta (1+\vartheta ) + \sigma ]}{1 - \vartheta ^{2}} $$(53)and
$$ 0 < \beta \leq \beta _{n} \leq \frac{\delta - 6 \vartheta [ \vartheta (1+\vartheta ) + \sigma + \frac{1}{6} \vartheta \delta ]}{6 \delta [ \vartheta (1+\vartheta ) + \sigma + \frac{1}{6} \vartheta \delta ]}. $$(54)
Then the sequences \(\{\varrho _{n}\}\), \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) converge weakly to \(\xi ^{*} \in \operatorname{VI}(G, C)\).
Corollary 4.2
Let \(G : C \rightarrow \mathbb{H}\) be a mapping satisfying conditions (G1)–(G3). Then \(\{x_{n}\}\) is the sequence generated in the following manner:
-
Initialization: Choose \(x_{0} \in \mathbb{H}\), \(0 < \lambda < \frac{1}{L}\), and \(\{\beta _{n}\}\) is a control parameter.
-
Step 1: Compute \(y_{n} = P_{C} (x_{n} - \lambda G(x_{n}))\). If \(y_{n} = x_{n}\), then \(x_{n}\) is the solution of the problem.
-
Step 2: Next, construct a half-space \(H_{n} = \{z \in \mathbb{H} : \langle x_{n} - \lambda G (x_{n}) - y_{n}, z - y_{n} \rangle \leq 0 \} \) and compute \(z_{n} = P_{H_{n}} (x_{n} - \lambda G(y_{n})) \).
-
Step 3: Evaluate \(x_{n+1} = (1- \beta _{n}) x_{n} + \beta _{n} z_{n} \), where \(\{\beta _{n}\}\) is a real sequence and there are \(\beta , \delta , \sigma > 0\) such that \(0 < \beta \leq \beta _{n} \leq \frac{1}{6 \sigma }\). Then the sequences \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) converge weakly to \(\xi ^{*} \in \operatorname{VI}(G, C)\).
Note that condition (G3) could be deleted when G is monotone. Indeed, this condition, which is a particular case of condition (A3), is only used to prove (49). Without condition (G3), inequality (49) can be obtained by imposing the monotonicity of G. In that case, we can write
By the use of \(f(x, y) = \langle G(x), y-x \rangle \) in (48), we have
Combining (55) and (56), we obtain
Let \(y_{t} = (1 - t)z + ty\) for all \(t \in [0, 1]\). Due to the convexity of the set C, \(y_{t} \in C\) for every \(t \in (0, 1)\). Since \(y_{n_{k}} \rightharpoonup z \in C\) and \(\langle G(y), y - z \rangle \geq 0\) for every \(y \in C\), we have
Thus, \(\langle G(y_{t}), y - z \rangle \geq 0\) for all \(t\in (0, 1)\). Since \(y_{t} \rightarrow z\) as \(t \rightarrow 0\) and due to the continuity of G, we have \(\langle G(z), y - z \rangle \geq 0\) for all \(y \in C\), which gives that \(z \in \operatorname{VI}(G, C)\).
Remark 4.1
From the above discussion, it can be concluded that Corollary 4.1 and Corollary 4.2 still hold, even if we remove condition (G3) in the case of monotone bifunctions.
4.2 Fixed points
Following Sect. 3, we show how our results can be applied for solving fixed point problems involving a κ-strict pseudocontraction mapping. Let us recall that a mapping \(T : C \rightarrow C\) is said to be
-
(i)
κ-strict pseudocontraction [11] on C if
$$ \Vert Tx - Ty \Vert ^{2} \leq \Vert x - y \Vert ^{2} + \kappa \bigl\Vert (x - Tx) - (y - Ty) \bigr\Vert ^{2}, \quad \forall x, y \in C, $$that is equivalent to
$$ \langle Tx - Ty, x - y \rangle \leq \Vert x - y \Vert ^{2} - \frac{1- \kappa }{2} \bigl\Vert (x - Tx) - (y - Ty) \bigr\Vert ^{2}, \quad \forall x, y \in C; $$ -
(ii)
sequentially weakly continuous on C if
$$ T(x_{n}) \rightharpoonup T\bigl(x^{*}\bigr) \quad \text{for any sequence in } C \text{ satisfying } x_{n} \rightharpoonup x^{*}\ ( \text{weakly converges}). $$
If we consider that the mapping T is a κ-strict pseudocontraction and weakly continuous, then \(f(x, y) = \langle x - Tx, y - x \rangle \) satisfies conditions (A1)–(A4) (see [43] for details) and \(2 c_{1} = 2 c_{2} = \frac{3 - 2 \kappa }{1 - \kappa }\). The values of \(y_{n}\) and \(z_{n}\) turn into the following:
As a consequence of the results in Sect. 3, we have the following results.
Corollary 4.3
Let C be a nonempty, convex, and closed subset of a Hilbert space \(\mathbb{H}\), and \(T : C \rightarrow C\) be a κ-strict pseudocontraction and weakly continuous with a solution set \(\operatorname{Fix}(T) \neq \emptyset \). Let \(\{x_{n}\}\) be the sequence generated in the following way:
-
Initialization: Let \(x_{-1}, x_{0} \in \mathbb{H}\), \(0 < \lambda < \frac{1 - \kappa }{3 - 2 \kappa }\), \(\{\vartheta _{n}\}\) and \(\{\beta _{n}\}\) are control parameters.
-
Step 1: Compute \(y_{n} = P_{C} [ \varrho _{n} - \lambda (\varrho _{n} - T ( \varrho _{n})) ] \), where \(\varrho _{n} = x_{n} + \vartheta _{n}(x_{n} - x_{n-1})\). If \(y_{n} = \varrho _{n}\), then \(\varrho _{n}\) is a solution of the fixed point problem.
-
Step 2: Construct a half-space \(H_{n} = \{z \in \mathbb{H} : \langle (1 - \lambda ) \varrho _{n} + \lambda T(\varrho _{n}) - y_{n}, z - y_{n} \rangle \leq 0 \}\) and calculate \(z_{n} = P_{H_{n}} [ \varrho _{n} - \lambda (y_{n} - T (y_{n})) ] \).
-
Step 3: Compute \(x_{n+1} = (1- \beta _{n}) \varrho _{n} + \beta _{n} z_{n}\), where \(\{\vartheta _{n}\}\) and \(\{\beta _{n}\}\) are real sequences. Here, we assume that the sequence \(\{\vartheta _{n}\}\) is nondecreasing with \(0 \leq \vartheta _{n} \leq \vartheta < 1\) for each \(n \geq 1\) and that there exist \(\beta , \delta , \sigma > 0\) such that
$$ \delta > \frac{6 \vartheta [ \vartheta (1+\vartheta ) + \sigma ]}{1 - \vartheta ^{2}} $$(60)and
$$ 0 < \beta \leq \beta _{n} \leq \frac{\delta - 6 \vartheta [ \vartheta (1+\vartheta ) + \sigma + \frac{1}{6} \vartheta \delta ]}{6 \delta [ \vartheta (1+\vartheta ) + \sigma + \frac{1}{6} \vartheta \delta ]}. $$(61)
Then the sequences \(\{\varrho _{n}\}\), \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) converge weakly to \(\xi ^{*} \in \operatorname{Fix}(T)\).
Corollary 4.4
Let C be a nonempty, convex, and closed subset of a Hilbert space \(\mathbb{H}\) and \(T : C \rightarrow C\) be a κ-strict pseudocontraction and weakly continuous with a solution set \(\operatorname{Fix}(T) \neq \emptyset \). Let \(\{x_{n}\}\) be the sequences generated in the following way:
-
Initialization: Choose \(x_{0} \in \mathbb{H}\), \(0 < \lambda < \frac{1 - \kappa }{3 - 2 \kappa }\), and \(\{\beta _{n}\}\) is a control parameter.
-
Step 1: Compute \(y_{n} = P_{C} [ x_{n} - \lambda (x_{n} - T (x_{n})) ]\). If \(y_{n} = x_{n}\), then \(x_{n}\) is a solution of the fixed point problem.
-
Step 2: Construct a half-space \(H_{n} = \{z \in \mathbb{H} : \langle (1 - \lambda ) x_{n} + \lambda T(x_{n}) - y_{n}, z - y_{n} \rangle \leq 0 \} \) and calculate \(z_{n} = P_{H_{n}} [ x_{n} - \lambda (y_{n} - T (y_{n})) ] \).
-
Step 3: Evaluate \(x_{n+1} = (1- \beta _{n}) x_{n} + \beta _{n} z_{n}\), where \(\{\beta _{n}\}\) is a real sequence and there are \(\beta , \delta , \sigma > 0\) such that \(0 < \beta \leq \beta _{n} \leq \frac{1}{6 \sigma }\). Then the sequences \(\{y_{n}\}\), \(\{z_{n}\}\), and \(\{x_{n}\}\) converge weakly to \(\xi ^{*} \in \operatorname{Fix}(T)\).
5 Numerical examples
In this section we present four numerical examples with comparisons to related results in the literature. The examples are both in infinite and infinite dimensional spaces. The MATLAB implementations are done via MATLAB version 9.5 (R2018b) on the Intel(R) Core(TM)i5-6200 CPU PC @ 2.30 GHz 2.40 GHz, RAM 8.00 GB.
-
(1)
For Tran et al. [37] (egA), we use
$$ \lambda = \min \biggl\{ \frac{1}{3c_{1}}, \frac{1}{3c_{2}} \biggr\} , \qquad D_{n} = \Vert x_{n} - y_{n} \Vert ^{2}. $$ -
(2)
For Hieu et al. [19] (iHegA), we use
$$\begin{aligned}& \theta =0.45,\qquad \lambda = \frac{1}{2c_{2}+8c_{1}}, \\& D_{n} = \max \bigl\{ \Vert x_{n+1} - y_{n} \Vert ^{2}, \Vert x_{n+1} - w_{n} \Vert ^{2} \bigr\} . \end{aligned}$$ -
(3)
For Rehman et al. [38] (iRegA), we use
$$\begin{aligned}& \alpha _{n}=0.20,\qquad \beta _{n}=0.80, \\& \lambda = 0.8 \biggl( \frac{\frac{1}{2}-2\alpha -\frac{1}{2}\alpha ^{2}}{(c_{1}+c_{2})(1-\alpha )^{2}} \biggr),\qquad D_{n} = \Vert w_{n} - y_{n} \Vert ^{2}. \end{aligned}$$ -
(4)
For Vinh et al. [42] (iVegA), we use
$$\begin{aligned}& \epsilon _{n}=\frac{1}{n^{2}}, \qquad \theta =0.45, \\& \lambda = \min \biggl\{ \frac{1}{3c_{1}}, \frac{1}{3c_{2}} \biggr\} , \qquad D_{n} = \Vert w_{n} - y_{n} \Vert ^{2}. \end{aligned}$$ -
(5)
For Algorithm 1 (Algo1), we use
$$\begin{aligned}& \vartheta _{n}=0.50,\qquad \beta _{n}=0.80, \\& \lambda = \min \biggl\{ \frac{1}{3c_{1}}, \frac{1}{3c_{2}} \biggr\} , \qquad D_{n} = \Vert w_{n} - y_{n} \Vert ^{2}. \end{aligned}$$
Example 5.1
Assume that the bifunction \(f: C \times C \rightarrow \mathbb{R}\) is defined by
where \(c \in \mathbb{R}^{n}\) and P, Q are matrices of order n. The matrix P is symmetric positive semi-definite and the matrix \(Q - P\) is symmetric negative semi-definite with Lipschitz-type constants \(c_{1} = c_{2} = \frac{1}{2}\| P - Q\|\) (see [37] for details). The matrices P, Q are taken randomly.Footnote 1 The constraint set \(C \subset \mathbb{R}^{n}\) is defined by
Numerical results are shown in Figs. 1–5 by assuming \(x_{-1}=x_{0}=y_{0}=(1,\ldots , 1)\) and \(\mathit{TOL}=10^{-8}\).
Example 5.1 when \(n=5\): the number of iterations is 63, 56, 46, 38, 44, respectively
Example 5.1 when \(n=50\): the number of iterations is 137, 145, 120, 119, 126, respectively
Example 5.1 when \(n=200\): the number of iterations is 154, 150, 135, 129, 113, respectively
Example 5.1 when \(n=400\): the number of iterations is 165, 163, 146, 143, 129, respectively
Example 5.1 when \(n=600\): the number of iterations is 173, 171, 153, 146, 133, respectively
Example 5.2
Suppose that \(\mathbb{H} = L^{2}([0, 1])\) is a Hilbert space with the induced norm
and the inner product \(\langle x, y \rangle = \int _{0}^{1} x(t) y(t) \,dt\) for all \(x, y \in \mathbb{H}\). Assume that \(C := \{ x \in L^{2}([0, 1]): \|x\| \leq 1 \}\). Let \(F : C \rightarrow \mathbb{H}\) be defined by
where \(H(t, s) = \frac{2ts e^{(t+s)}}{e \sqrt{e^{2}-1}}\), \(f(x) = \cos x\), \(g(t) = \frac{2t e^{t}}{e \sqrt{e^{2}-1}}\). As stated in [18], the operator F is monotone and Lipschitz continuous with \(L = 2\). Figures 6 and 7 show the numerical results obtained with \(x_{-1}=x_{0}=y_{0}\).
Example 5.2 when \(x_{0}=1\): the number of iterations is 72, 65, 55, 44, 56, respectively
Example 5.2 when \(x_{0}=t\): the number of iterations is 83, 76, 69, 59, 65, respectively
Example 5.3
Let \(G : \mathbb{R}^{n} \rightarrow \mathbb{R}^{n}\) be an operator defined by
where A is a degree n symmetric semi-definite matrix. The proximal mapping \(B(x)\) derives from the function \(h(x) = \frac{1}{4} \|x\|^{4}\) such that
The above description implies that F is monotone on C [27]. The feasible set C is defined by
Figures 8–11 show the numerical results by assuming \(x_{-1}=x_{0}=y_{0}\).
Example 5.3 when \(x_{0}=(1,1,1,1,1)^{T}\): the number of iterations is 68, 63, 49, 45, 47, respectively
Example 5.3 when \(x_{0}=(1,2,3,4,5)^{T}\): the number of iterations is 88, 81, 61, 56, 50, respectively
Example 5.3 when \(x_{0}=(1,2,3,4,0)^{T}\): the number of iterations is 73, 67, 50, 46, 42, respectively
Example 5.3 when \(x_{0}=(-1,2,0,3,4)^{T}\): the number of iterations is 87, 80, 60, 56, 53, respectively
Example 5.4
Let \(\mathbb{H} = l_{2}\) be a real Hilbert space having sequences of real numbers that are square-summable with \(\|x\|= \sqrt{\sum_{i} |x_{i}|^{2}}\) and \(C= \{ x \in \mathbb{H} : \|x\| \leq 10 \}\). Let a bifunction f be defined by
It is easy to see that \(S_{\operatorname{EP}(f, C)} \neq \emptyset \) and meets condition (A3). Next, we need to prove that f is Lipschitz-type continuous. In fact, we have successively
where \(x, y, w \in C\) and \(c_{1}=c_{2}=\frac{33}{2}\). We show that the bifunction is pseudomonotone. Let \(x, y \in C\) be such that \(f(x, y) = (13 - \|x\|) \langle x, y-x \rangle \geq 0\), which means that \(\langle x, y-x \rangle \geq 0\). Thus, we have
Moreover, we prove that f is not monotone. In fact, let \(x = (\frac{13}{2}, 0, 0, \ldots , 0, \ldots )\) and \(y = (10, 0, 0, \ldots , 0, \ldots )\) such that
A metric projection \(P_{C}\) upon C is defined by
Numerical results regarding Example 5.4 are shown in Figs. 12–14 and Table 1.
Example 5.2 when \(x_{-1}=x_{0}=y_{0}=(2,\ldots ,2_{500},0,\ldots )\)
Example 5.2 when \(x_{-1}=x_{0}=y_{0}=(e^{1},e^{2},\ldots ,e^{500},0,\ldots )\)
Example 5.2 when \(x_{-1}=x_{0}=y_{0}=(1^{2},2^{2},\ldots ,500^{2},0,\ldots )\)
Availability of data and materials
Not applicable.
Notes
Two diagonal matrices randomly \(A_{1}\) and \(A_{2}\) with entries from \([0,2]\) and \([-2, 0]\), respectively. Two random orthogonal matrices \(O_{1}=\operatorname{RandOrthMat}(n)\) and \(O_{2}=\operatorname{RandOrthMat}(n)\) are generated. Then a positive semi-definite matrix \(B_{1}=O_{1}A_{1}O_{1}^{T}\) and a negative semi-definite matrix \(B_{2}=O_{2}A_{2}O_{2}^{T}\) are achieved. Finally, set \(Q=B_{1}+B_{1}^{T}\), \(S=B_{2}+B_{2}^{T}\) and \(P=Q-S\).
References
Alakoya, T.O., Jolaoso, L.O., Mewomo, O.T.: Modified inertial subgradient extragradient method with self adaptive stepsize for solving monotone variational inequality and fixed point problems. Optimization 70(3), 545–574 (2021)
Alansari, M., Ali, R., Farid, M.: Strong convergence of an inertial iterative algorithm for variational inequality problem, generalized equilibrium problem, and fixed point problem in a Banach space. J. Inequal. Appl. 2020, 42 (2020)
Alansari, M., Kazmi, K.R., Ali, R.: Hybrid iterative scheme for solving split equilibrium and hierarchical fixed point problems. Optim. Lett. 14(8), 2379–2394 (2020)
Anh, P.N., An, L.T.H.: The subgradient extragradient method extended to equilibrium problems. Optimization 64(2), 225–248 (2012)
Antipin, A.: Equilibrium programming: proximal methods. Comput. Math. Math. Phys. 37(11), 1285–1296 (1997)
Attouch, F.A.H.: An inertial proximal method for maximal monotone operators via discretization of a nonlinear oscillator with damping. Set-Valued Var. Anal. 9, 3–11 (2001)
Bauschke, H.H., Combettes, P.L.: Convex Analysis and Monotone Operator Theory in Hilbert Spaces, 2nd edn. CMS Books in Mathematics. Springer, Berlin (2017)
Bianchi, M., Schaible, S.: Generalized monotone bifunctions and equilibrium problems. J. Optim. Theory Appl. 90(1), 31–43 (1996)
Bigi, G., Castellani, M., Pappalardo, M., Passacantando, M.: Existence and solution methods for equilibria. Eur. J. Oper. Res. 227(1), 1–11 (2013)
Blum, E.: From optimization and variational inequalities to equilibrium problems. Math. Stud. 63, 123–145 (1994)
Browder, F., Petryshyn, W.: Construction of fixed points of nonlinear mappings in Hilbert space. J. Math. Anal. Appl. 20(2), 197–228 (1967)
Censor, Y., Gibali, A., Reich, S.: The subgradient extragradient method for solving variational inequalities in Hilbert space. J. Optim. Theory Appl. 148(2), 318–335 (2010)
Dong, Q.-L., Kazmi, K.R., Ali, R., Li, X.-H.: Inertial Krasnosel’skiı̌–Mann type hybrid algorithms for solving hierarchical fixed point problems. J. Fixed Point Theory Appl. 21(2), 57 (2019)
Dong, Q.L., Cho, Y.J., Zhong, L.L., Rassias, T.M.: Inertial projection and contraction algorithms for variational inequalities. J. Glob. Optim. 70(3), 687–704 (2017)
Dong, Q.L., Huang, J.Z., Li, X.H., Cho, Y.J., Rassias, T.M.: MiKM: multi-step inertial Krasnosel’skiı̌–Mann algorithm and its applications. J. Glob. Optim. 73(4), 801–824 (2018)
Fan, J., Liu, L., Qin, X.: A subgradient extragradient algorithm with inertial effects for solving strongly pseudomonotone variational inequalities. Optimization 69(9), 2199–2215 (2020)
Fan, K.: A minimax inequality and applications. In: Shisha, O. (ed.) Inequalities III. Academic Press, New York (1972)
Hieu, D.V., Anh, P.K., Muu, L.D.: Modified hybrid projection methods for finding common solutions to variational inequality problems. Comput. Optim. Appl. 66(1), 75–96 (2016)
Hieu, D.V., Cho, Y.J., Xiao, Y.-B.: Modified extragradient algorithms for solving equilibrium problems. Optimization 67(11), 2003–2029 (2018)
Hung, P.G., Muu, L.D.: The Tikhonov regularization extended to equilibrium problems involving pseudomonotone bifunctions. Nonlinear Anal., Theory Methods Appl. 74(17), 6121–6129 (2011)
Izuchukwu, C., Mebawondu, A.A., Mewomo, O.T.: A new method for solving split variational inequality problems without co-coerciveness. J. Fixed Point Theory Appl. 22(4), 98 (2020)
Izuchukwu, C., Ogwo, G.N., Mewomo, O.T.: An inertial method for solving generalized split feasibility problems over the solution set of monotone variational inclusions. Optimization (2020). https://doi.org/10.1080/02331934.2020.1808648
Jolaoso, L.O., Alakoya, T.O., Taiwo, A., Mewomo, O.T.: Inertial extragradient method via viscosity approximation approach for solving equilibrium problem in Hilbert space. Optimization 70(2), 387–412 (2021)
Jolaoso, L.O., Taiwo, A., Alakoya, T.O., Mewomo, O.T.: A strong convergence theorem for solving pseudo-monotone variational inequalities using projection methods. J. Optim. Theory Appl. 185(3), 744–766 (2020)
Konnov, I.: Application of the proximal point method to nonmonotone equilibrium problems. J. Optim. Theory Appl. 119(2), 317–333 (2003)
Korpelevich, G.: The extragradient method for finding saddle points and other problems. Matecon 12, 747–756 (1976)
Kreyszig, E.: Introductory Functional Analysis with Applications, 1st edn. Wiley, New York (1978)
Mastroeni, G.: On auxiliary principle for equilibrium problems. In: Nonconvex Optimization and Its Applications, pp. 289–298. Springer, New York (2003)
Moudafi, A.: Proximal point algorithm extended to equilibrium problems. J. Nat. Geom. 15(1–2), 91–100 (1999)
Muu, L., Oettli, W.: Convergence of an adaptive penalty scheme for finding constrained equilibria. Nonlinear Anal., Theory Methods Appl. 18(12), 1159–1166 (1992)
Oliveira, P., Santos, P., Silva, A.: A Tikhonov-type regularization for equilibrium problems in Hilbert spaces. J. Math. Anal. Appl. 401(1), 336–342 (2013)
Opial, Z.: Weak convergence of the sequence of successive approximations for nonexpansive mappings. Bull. Am. Math. Soc. 73, 591–598 (1967)
Polyak, B.: Some methods of speeding up the convergence of iteration methods. USSR Comput. Math. Math. Phys. 4(5), 1–17 (1964)
Taiwo, A., Alakoya, T.O., Mewomo, O.T.: Halpern-type iterative process for solving split common fixed point and monotone variational inclusion problem between Banach spaces. Numer. Algorithms 86, 1359–1389 (2021)
Taiwo, A., Jolaoso, L.O., Mewomo, O.T.: Inertial-type algorithm for solving split common fixed point problems in Banach spaces. J. Sci. Comput. 86(1), 12 (2021)
Tiel, J.V.: Convex Analysis: An Introductory Text, 1st edn. Wiley, New York (1984)
Tran, D.Q., Dung, M.L., Nguyen, V.H.: Extragradient algorithms extended to equilibrium problems. Optimization 57(6), 749–776 (2008)
ur Rehman, H., Kumam, P., Abubakar, A.B., Cho, Y.J.: The extragradient algorithm with inertial effects extended to equilibrium problems. Comput. Appl. Math. 39(2), 100 (2020)
ur Rehman, H., Kumam, P., Cho, Y.J., Suleiman, Y.I., Kumam, W.: Modified Popov’s explicit iterative algorithms for solving pseudomonotone equilibrium problems. Optim. Methods Softw. 36(1), 82–113 (2021)
ur Rehman, H., Kumam, P., Cho, Y.J., Yordsorn, P.: Weak convergence of explicit extragradient algorithms for solving equilibrium problems. J. Inequal. Appl. 2019(1), 282 (2019)
ur Rehman, H., Kumam, P., Kumam, W., Shutaywi, M., Jirakitpuwapat, W.: The inertial sub-gradient extra-gradient method for a class of pseudo-monotone equilibrium problems. Symmetry 12(3), 463 (2020)
Vinh, N.T., Muu, L.D.: Inertial extragradient algorithms for solving equilibrium problems. Acta Math. Vietnam. 44(3), 639–663 (2019)
Wang, S., Zhang, Y., Ping, P., Cho, Y., Guo, H.: New extragradient methods with non-convex combination for pseudomonotone equilibrium problems with applications in Hilbert spaces. Filomat 33(6), 1677–1693 (2019)
Acknowledgements
The authors are heartily grateful to the reviewers for their valuable remarks which greatly improved the results and presentation of the paper.
Funding
The authors acknowledge the financial support provided by the Center of Excellence in Theoretical and Computational Science (TaCS-CoE), KMUTT. Moreover, this research project is supported by Thailand Science Research and Innovation (TSRI) Basic Research Fund: Fiscal year 2021 under project number 64A306000005. In particular, Habib ur Rehman was financed by the Petchra Pra Jom Doctoral Scholarship Academic for Ph.D. Program at KMUTT [grant number 39/2560].
Author information
Authors and Affiliations
Contributions
All authors contributed equally and significantly in writing this article. All authors read and approved the final manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
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
Rehman, H.u., Kumam, P., Gibali, A. et al. Convergence analysis of a general inertial projection-type method for solving pseudomonotone equilibrium problems with applications. J Inequal Appl 2021, 63 (2021). https://doi.org/10.1186/s13660-021-02591-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-021-02591-1
Keywords
- Pseudomonotone bifunction
- Equilibrium problem
- Weak convergence
- Lipschitz-type conditions
- Variational inequality problem