An approximate subgradient algorithm for unconstrained nonsmooth, nonconvex optimization (Q2483018)

The authors propose a minimization algorithm based on subgradients which can be applied to minimize locally Lipschitz functions. The descent directions are computed by solving a system of linear inequalities. This part is based on previous papers of Bagirov on approximating subdifferentials. The new algorithm may use more than one approximate subgradient in each iteration - similar to bundle methods. However, unlike bundle methods, it does not use polyhedral underestimators of the objective function. An Armijo-type line search technique is applied to find the step lengths. The authors prove that the algorithm converges in case of quasidifferentiable, semismooth functions, including convex, concave, max-type and min-type functions. Moreover, some numerical experiments are presented and discussed.