A globalization strategy for Interior Point bound-constrained methods for mixed complementarity problems

Recent works have shown that a wide class of Interior Point methods employing linesearch along the Newton step may manifest a weakness of convergence. In order to alleviate such drawback, in this paper we introduce a new globalization strategy. It performs backtracking along a piecewise linear path which can be easily constructed. The proposed strategy is embedded into an Interior Point method. Computational results show that the resulting procedure is remarkably successful, shows fast local rate of convergence and low computational cost.