The inverse iteration with the WZ factorization used to the Markovian models

Abstract

In the article a new approach for solving Markovian linear systems is presented. It is the inverse iteration with the use of the WZ factorization. The paper contains the theoretical base of the method and numerical experiment results where the accuracy and the performance time were investigated. A modified (with the use of the SBLAS library and the Harwell-Boeing sparse matrix storage scheme) method was also presented. The experiments were carried out with the matrices generated for a Markovian model of a leaky bucket with tokens.