Gobally convergent variable metric method for nonconvex nondifferentiable unconstrained minimization (Q5956437)

Motivated by the convex variable metric method due to \textit{L. Luksan} and \textit{J. Vlcek} [J. Optim. Theory Appl. 102, 593-613 (1999; Zbl 0955.90102)], the authors propose a nonsmooth variable metric method for finding stationary points of locally Lipschitz functions. The convex variable metric method by Luksan-Vlcek mentioned above performs well in practice, but without assumption of convexity it may be not globally convergent, while the nonsmooth variable metric method here is also globally convergent in the nonconvex case. The algorithm presented here involves a special line search procedure which provides stepsizes satisfying conditions required for global convergence. Note that time-consuming quadratic programming subproblems do not need to be solved. An implementation of this algorithm and some numerical experiments are given as well.