Some predictor-corrector algorithms for multivalued variational inequalities (Q5942224)

This article deals with the problem of finding \(u\in H\), \(g(u)\in K\), \(v \in T(u)\) such that \(\langle v,g(v)- g(u)\rangle\geq 0\) \((g(v)\in K)\). Here \(T: H\to C(H)\) is a multivalued operator from a real Hilbert space \(H\) into the family \(C(H)\) of all nonempty compact subsets of \(H\), \(K\) is a nonempty closed and convex set in \(H\). The following algorithm for solving this variational inequality is studied: \[ \langle\rho\eta_n+ g(u_{n-1}- g(w_n), g(v)- g(u_{n+1}))\rangle\geq 0\quad (\forall g(v)\in K), \] \[ \eta_n\in T(w_n): \|\eta_{n+1}- \eta_n\|\leq M(T(w_{n+1}, T(w_n)), \] \[ \langle\beta \xi_n+ g(w_n- g(y_n), g(v)- g(w_n))\rangle\geq 0\quad (\forall g(v)\in K), \] \[ \xi_n\in T(y_n): \|\xi_{n+1}- \xi_n\|\leq M(T(y_{n+1}, T(y_n)), \] \[ \langle\mu v_n+ g(y_n- g(u_n), g(v)- g(y_n))\rangle\geq 0\quad (\forall g(v)\in K), \] \[ v_n\in T(u_n):\|v_{n+1}- v_n\|\leq M(T(u_{n+1}, T(u_n)), \] (\(\rho\), \(\mu\), \(\beta\) are positive constants, \(u_0\in H\)). The main result of the article is a theorem about the convergence of this algorithm in the case when \(H\) is finite-dimensional, \(g\) is injective, \(T\) an \(M\)-Lipschitz continuous operator, and \(\rho\) sufficiently small.