A Kantorovich-type convergence analysis of the Newton-Josephy method for solving variational inequalities (Q607520)

This paper presents a Kantorovich-type semilocal convergence analysis of the Newton-Josephy method for solving a certain class of variational inequalities. It focuses on the problem of approximating a solution \(x^* \in D\) of variational inequalities \(VI (D,F)\), such that \[ (y-x^*)^T F(x^*) \geq 0,\text{ for all }y \in D,\tag{1} \] where \(F\) is a Fréchet-differentiable operator defined on a non-empty, convex, and closed subset \(D\) of \(\mathbb{R}^n\), with values in \(\mathbb{R}^n\). A large number of problems in applied mathematics, engineering and economics, such as analysis of elastoplastic structures, Walrasian equilibrium, Nash Equilibrium, financial equilibrium problems, are solved by finding the solution of problem (1). Moreover, the nonlinear complementarity problem is a special case of problem (1), which leads to find out the element \(x^* \in D = \mathbb{R}^n_+\) such that \[ x^{*^T} F(x^*)=0,\quad F(x^*) \in D. \] If \(F(x) = \nabla G(x) (x \in D)\) in (1), where \(G:D \rightarrow \mathbb{R}\) is a real differentiable function, the problem (1) corresponds to the first order necessary optimality condition of the optimization problem \(\min_{x \in D} G(x)\). The Newton-Josephy method (NJM) generates a sequence \(\{x_K\}\), so that \(x_{K+1}\) solves the \(F_K\) k-th linearized subproblem \(VI (D,F_K): F_K(x) = F(x)+ F'(x_K)(x-x_K)\), \((K \geq 0)\), and \((F'(x))_{ij} = (\partial F_i (x) /\partial x_j)\). Main result: The authors expand the applicability of (NJM). Firstly Lipschitz and center-Lipschitz conditions on \(F'\) are introduced. The authors provide a new semilocal convergence analysis of (NJM) (using a combination of the two and own new idea of recurrent functions) under less or the same computational cost with the advantages: \(1^o\) Weaker sufficient convergence conditions. \(2^o\) Larger convergence domain. \(3^o\) Finer error bounds on the distantes \(\|x_{K+1} - x_K\|, \|x_K - x^*\| (K \geq 0).\) \(4^o\) An at least as precise information on the location of the solution \(x^*\). Finally, the results of some numerical examples for the comparison of the authors error bounds with the corresponding ones in the works of \textit{Zh. Wang} and \textit{Z. Shen} [Appl. Math. Mech., Engl. Ed. 25, No. 11, 1291--1297 (2004; Zbl 1100.49012)] are also presented.