METHODS OF SOLVING OPERATOR EIGENPROBLEMS IN PARALLEL DISTRIBUTED MEMORY SYSTEMS AS APPLIED IN ELECTROMAGNETICS

Abstract

This study presents numerical methods of solving operator eigenproblems, focusing on their applications in electromagnetics. The discussion concentrates on the analysis of new iterative algorithms or modifications of the existing ones, which are capable of finding a few eigenvalues from the point spectrum of a non-symmetric operator. The salient feature of the considered methods is a low computational cost and memory complexity as compared to alternative solutions. This paper also presents implementations of the investigated algorithms in parallel distributed memory systems, based on the message-passing parallel programming model and providing portable parallel eigensolvers. The discussion of the applied designs of the parallel algorithms is supported by the presentation of the results of performance tests in selected distributed memory environments, including scalable parallel supercomputer systems and networks of workstations. The results of these tests confirm high efficiency of the eigensolvers in the considered parallel environments. In this study attention is also drawn to the question of the applicability of the eigensolvers to problems of modelling of electromagnetic fields in dielectric waveguides. The results of numerical tests validating the methods in these applications determine the scope of problems which may be most effectively solved using the specific eigensolvers.