No label defined (Q4034274)

This paper is concerned with the problem of solving, on a closed convex subset of a Hilbert space, the variational inequality (1) \(\langle Tu,v- u\rangle+b(u,v)-b(u,u)\geq 0\), where \(T\) is nonlinear and continuous, and \(b(\cdot,\cdot)\) satisfies a number of conditions. In particular, \(b\) is nondifferentiable, linear in its first argument and convex in the second. The problem is approached by considering, for given \(u\), an auxiliary problem (2) \(\langle A(w),v-w\rangle+\rho b(u,v)-\rho b(u,w)\geq\langle A(u),v-w\rangle-\rho\langle Tu,v-w\rangle\), in which \(\rho>0\) is constant and \(A\) is a nonlinear operator. The idea is that when \(w=u\), then the solution \(w\) of this problem solves the original variational inequality. The following algorithm is proposed for the solution of (1): at step \(n\) the auxiliary problem (2) is solved for \(w_{n+1}\), with \(u=w_ n\). The algorithm terminates when the error \(| w_{n+1}-w_ n|\) is sufficiently small. It is shown that \(w_ n\to w\) provided that the operators \(A\) and \(T\) are both strongly monotone and Lipschitz continuous, and provided that the monotonicity and Lipschitz constants satisfy a rather complicated inequality.