A Parallel Condensation-based Method for the Structural Dynamic Reanalysis Problem

Abstract

In the dynamic analysis of a structure quite often a small number of the smallest eigenvalues and eigenmodes of a large and sparse general eigenvalue problem have to be determined assuming that good approximations to the demanded eigenmodes are available from previous computations. In this situation powerful approaches like Lanczos method or Jacobi-Davidson may be inferior to methods which in general are known to be slower. In this note we demonstrate this effect comparing P_ARPACK, the parallel version of ARPACK, with a parallel condensation method with general masters for a fnite element model of a container ship.