Regiões de confiança em programação matematica

AUTOR(ES)
DATA DE PUBLICAÇÃO

1994

RESUMO

Three trust region algorithms for constrained minimization are proposed in this work: RCARB, RCMRI and BOX, developed for dealing with arbitrary domains, equality constraints and simple bounds, respectively. Focusing on the algorithm RCARB, global convergence results (1st order) are proved and it is analysed with details the minimization in Euclidean balls, validated by a set of numerical experiments. As regards the algorithm RCMRI, local and global convergence results are proved (1st and 2nd order). It is applied for minimization in Euclidean spheres, particularly intending to solve the Initial Vector Problem in codification theory. Numerical experiments are included. When it comes to the algorithm BOX, both global convergence and identification of the active constraints are proved. It is made a thorough analysis of the algorithm in charge for the resolution of the subproblem (QUACAN), implemented to minimize general quadratics with bound constrained variables and especially developed for large scale problems. Finally, it is presented a strategy for minimizing convex functions with linear constraints, based on using the algorithm BOX

ASSUNTO(S)

programação (matematica)

ACESSO AO ARTIGO

Documentos Relacionados