Solução de problemas de autovalor generalizados através do método de iteração por subespaços em blocos e do algoritmo de Lanczos com ortogonalização seletiva

Abstract

The purpose of this work is to analyse the efficiency of some techniques to solve the generalized eigenproblem of large, sparse, symmetric matrix pencils. A block version of the Subspace Iteration Method, considering ar not convergence accelerators is studied. It is also examined a version of the Lanczos Algorithm with selective orthogonalization, in arder to avoid numerical instabilities. Those methods are applied to the free vibrations solution of a jacket platform simplified model, a self-elevative drilling unit anda jacket platform. Several questions concerning the studied methods are addressed, trying the stabilishement of the best suited method for each structure analysed.