Implementace metody sečen pro nehladkou optimalizaci

Abstract

This work is devoted to methods for minimization of nonsmooth function. Nonsmooth functions are functions that are not differentiable. Problem of minimization can be solved by replacing original function with continuously differentiable aproximation or by using algorithms, that are appropriate for minimization of nonsmooth function. Subdifferential methods and bundle methods are some of them. Another type of algorithm is a r-secant method. The goal of this work is to describe the basics features of this method, to cover the whole algorithm, to implement the code of program and then to test it on suitable problems. This work is also covering the necessary theory of Clarke calculus. On top of that we are also comparing a r-secant method with subdifferential methods.