Mathematica Eterna
Open Access

Abstract

An improved nonmonotone feasible direction method

Ke Su and Shibo Tang

In this paper, a new sequential quadratic programming (SQP) method of feasible directions is proposed and analyzed for nonlinear programming,where a feasible direction of descent can be derived from solving only one QP subproblem. The algorithm has no demond on initial point, moreover it avoids using a penalty function or a filter. So it is more flexible and easier to implement. To avoid Maratos effect, a revised direction is computed by solving a linear system. Under some reasonable conditions, the global convergence is shown.