Partial projected Newton method for a class of stochastic linear complementarity problems (Q652329)

The class of linear complementarity problems with finitely many realizations is formulated as a minimization problem by introducing a slack variable and making use of the Fischer-Burmeister function. A partial projected Newton method is presented for solving the minimization problem. The advantage of the proposed method is that the algorithm will continue to generate conjugate directions even if a negative curvature direction is found. The global and quadratic convergence of the algorithm is given under mild assumptions. Some numerical results on some randomly generated stochastic linear complementarity problems are reported.