[Home]   [  News]   [  Events]   [  People]   [  Research]   [  Education]   [Visitor Info]   [UCSD Only]   [Admin]
Home > Events > CCoM > Abstract
Search this site:

A Piecewise Differentiable Line Search for Projected Search Optimization Methods

Minxin Zhang


Line search methods for unconstrained optimization based on satisfying the Wolfe conditions impose a restriction on the value of the directional derivative of the objective function at the new iterate. Projected search methods for bound-constrained optimization involve a line search along a continuous piecewise-linear path, which makes it impossible to apply the conventional Wolfe conditions. We propose a new quasi-Wolfe line search for piecewise differentiable functions. The behavior of the line search is similar to that of a conventional Wolfe line search, except that a step is accepted under a wider range of conditions. These conditions take into consideration steps at which the line search function is not differentiable. Some basic results associated with a conventional Wolfe line search are established for the quasi-Wolfe case. After identifying the practical considerations needed for converting a Wolfe line search into a quasi-Wolfe line search, details of the implementation along with some numerical results will be presented.

Tuesday, October 15, 2019
11:00AM AP&M 2402