General masters in parallel condensation of eigenvalue problems

Abstract

In the dynamic analysis of structures using finite element methods very often prohibitively many degrees of freedom are required to model the structure sufficiently accurate. Condensation methods are often used to reduce the number of unknowns to manageable size. Substructuring and choosing the master variables as the degrees of freedom on the interfaces of the substructures yields data structures which are well suited to be implemented on parallel computers. In this paper we discuss the use of additional non nodal masters in substructuring. The data structure is preserved such that the condensed problem can be determined substructurewise.