A new double projection algorithm for variational inequalities

We present a modification of a double projection algorithm proposed by Solodov and Svaiter for solving variational inequalities. The main modification is to use a different Armijo-type linesearch to obtain a hyperplane strictly separating current iterate from the solutions of the variational inequalities. Our method is proven to be globally convergent under very mild assumptions. If in addition a certain error bound holds, we analyze the convergence rate of the iterative sequence. We use numerical experiments to compare our method with that proposed by Solodov and Svaiter.