A Predictor-Corrector Algorithm for Monotone Linear Complementarity Problems in a Wide Neighborhood

We propose a new primal-dual interior-point predictor-corrector algorithm in Ai and Zhang's wide neighborhood for solving monotone linear complementarity problems (LCP). Based on the understanding of this neighborhood, we use two new directions in the predictor step and in the corrector step, respectively. Especially, the use of new corrector direction also reduces the duality gap in the corrector step, which has good effects on the algorithm's convergence. We prove that the new algorithm has a polynomial complexity of O(root nL), which is the best complexity result so far. In the paper, we also prove a key result for searching for the best step size along some direction. Considering local convergence, we revise the algorithm to be a variant, which enjoys both complexity of O(root nL) and Q-quadratical convergence. Finally, numerical result shows the effectiveness and superiority of the two new algorithms for monotone LCPs.