Block monotone iterative algorithms for variational inequalities with nonlinear operators (Q942876)

The authors study a question of monotone iterative algorithms for variational inequalities with nonlinear operators. The following finite dimensional variational inequality of finding a \(u \in K\) such that \[ (Au + f(u))^T (v-u) \geq 0,\text{ for any }v \in K,\tag{1} \] is considered. Here \(A \in \mathbb R^{n \times n}\) is an M-matrix, \(f(v)=(f_j(v_j))^n_{j=1}\) is continuously differentiable that satisfies \({f'}_j (v_j) \geq 0\) \((j = 1, \dots,n)\), and \(K\) is the convex set defined by: \(K=\{v \in \mathbb R^n :\Phi \leq v \leq \Psi\}\), with given vectors \(\Phi\) and \(\Psi \in\mathbb R^n\) satisfying \(\Phi < \Psi\). Some block iterative algorithms to solve problem (1) are presented, and some useful properties of the algorithms are obtained. Main result: The authors' algorithms can be regarded as discrete multiplicative and additive Schwarz algorithms. (They are special cases of domain decomposition methods). Especially, when the problem (1) only involves an affine operator, the algorithms are also special cases of matrix multisplitting methods. The monotone convergence of the algorithms and comparison theorems among the algorithms are obtained. The theoretical comparison results show that for the same initial value and the partition of \(N = \{ 1,2,\dots, n\}\), the iterative sequence generated by multiplicative algorithm converges faster than that generated by additive algorithm with initial in \(\widetilde{S}\) (or initial in \(\widehat{S})\).