Descent methods for quasidifferentiable minimization (Q1109687)

The problem of unconditional minimization of a continuous and uniformly directional differentiable function f(x), \(x\in R^ m\) is considered. In addition it is supposed that the derivative in the direction \(f_ x'(d)\) can be represented in the following way: \[ f_ x'(d)=\min_{w\in W(x)}f_ x(d,w),\quad and\quad \bar f_ x(d,w)=\max_{v\in V(x,w)}(v,d), \] where W(x) is a compact subset of \(R^ m\) and V(x,w) is compact for every w from W(x). The point \(\bar x\) is called inf- stationary for f, if \(f'_{\bar x}(d)\geq 0\) for every \(d\in R^ m.\) A descent method is now given for minizing the nondifferentiable function which can be locally approximated by pointwise minima of convex functions as shown above. At each iterate the algorithm finds several directions by solving several linear or quadratic programming subproblems. These directions are then used in an Armijo-like search for the next iterate. A feasible direction extension to inequality constrained minimization problems is also presented. The convergence of this algorithm to inf-stationary points is proved.