Smoothing Newton method for NCP with the identification of degenerate indices (Q989148): Difference between revisions

A new smoothing Gauss-Newton method for nonlinear complementarity prolems (NCP) by using a NPC-function for reformulation of the original problem. The method, based on active-set strategy, contains an estimating technique for the determination of the degenerate set of indices at the solution \(x^*\) under suitable regularity conditions. The active-set strategy reduces the scale of the problem. The method is proven to be globally convergent and under the above mentioned regularitgy locally superlinearly convergent. In the proposed and well defined algorithm the Gauss-Newton method is applied for the reduced smoothing NCP-function (fast step) and if this step is not successfull a line search along a modified direction is done (ensuring global convergence). The smoothing parameter is updated under some convergence conditions. If the norm of the gradient of the original NCP is small enough the algorithm stopps. The method is numerically tested for four examples with known degenerate solutions. It is demonstrated that the method detects the degenerate indices and that the fast steps are done in at least the half of the iterations. In the contrary, the known Newton method for the zero of the smoothing NCP function system with the same line search and update of the smoothing parameter needs more iterations and more function evaluations and fails some times.