A regularization semismooth Newton method based on the generalized Fischer-Burmeister function for \(P_0\)-NCPs (Q939548)

In this work, the authors present a regularization method for solving nonlinear complementary problems (NCPs). They analyze the algorithm and test it on a problem set. The theory is a direct extension of \textit{D. Sun}'s work [Appl. Math. Optimization 40, No.~3, 315--339 (1999; Zbl 0937.90110)], who used a Fischer-Burmeister function as complementary function (NCP function) together with a Tikhonov regularization to solve the problem with a generalized Newton method. Here the NCP function can be chosen out of a more general class. Instead of relying on the 2-norm it allows for a \(p\)-norm, where \(p \in (1, \infty)\). The convergence analysis from the case \(p=2\) [Zbl 0937.90110] is extended to arbitrary \(p\) with the same result of superlinear (for certain cases quadratic) convergence. A \textit{Matlab} implementation of the algorithm has been tested on the set of complementary problems MCLIB. For arbitrary \(p\) it is capable to solve nearly all problems from this package. In many runs the iteration-count for various \(p\) does not differ significantly. However, for certain problems small values of \(p\) are found to accelerate the convergence.