Group for Research in Decision Analysis

A Short Introduction to Derivative-Free Optimization

Mathilde Peyrega Polytechnique Montréal, Canada

Derivative-Free Optimization (DFO) is a field of Operational Research where the derivatives are not used. DFO algorithms have been developed to solve problems where the derivatives of the objective function and the constraints are unavailable or does not even exist. Classical algorithms can not be applied to this class of problems, and nowadays some industrial problems can only be solved with DFO algorithms since the expressions of the functions are sometimes unknown. This short presentation will show the characteristics of DFO problems, present one real DFO problem which has been studied in GERAD and outline the main DFO algorithms.