O método de direções viáveis em duas estapas para programação não-linear e aplicações a programação quadrática

Abstract

Presents the computational implementation of the two-steps feasible directions algorithm for non-linear constrained optimization. It' s stated an assimptotic convergence method for quadratic programming, based on the sarne ideas. As a consequence of the computational implementation, a constrained line-search·procedure is developed. To avoid ili-conditionning and accelerate the convergence rate, new techniques are developed and tested by means of classical problems. For both methods, many versions that have different convergence rates are presented. Versions which use quasi-Newton techniques have superlinear convergence rate.