Inexact restoration for nonlinear programs

Abstract

Inexact Restoration is a well-known technique for dealing with constrained nonlinear programs. Roughly speaking, each step of an Inexact Restoration algorithm can be divided into two basic parts. One is the restoration phase, where the feasibility of iterates is improved. This means that feasibility is treated here independently since it is often more important than optimality. Moreover, particular structures may be exploited better in this way. The second part of an Inexact Restoration step is an optimization phase in which a linearly constrained quadratic program has to be solved. An appropriate combination of both phases yields, under reasonable conditions, globally convergent algorithms.