Accelerated implementation of the power method for analysis of electromagnetic fields

Abstract

This paper presents two kinds of optimizations to the well known power iteration method for finding dominant eigenvalue of matrices, with regard to special matrices often used in EM field measurement and processing. Based on the statistical analysis of actual measurement data, the first optimization suggests a method for improving the convergence of the power iteration algorithm. The second optimization is implementational: the acceleration of the power method by means of utilising modern vector processing features of the x86-64 architecture is evaluated. Various implementations using different parameters are evaluated by using the Intel MKL library, and a comparative analysis is presented for general and also specific matrices that are important in the EM field measurement and processing.