An Arnoldi-extrapolation algorithm for computing pagerank (Q989127)

The authors aim at improving the PageRank algorithm which is commonly used to determine the global importance of Web pages. The algorithm essentially needs to compute the principal eigenvector of the Google matrix and is obviously required to be very efficient. To this end, the paper proposes a new Arnoldi-Extrapolation algorithm, based on the Arnoldi-type algorithm already presented in literature. An extrapolation method based on Ritz values is developed and then conveniently introduced into the Arnoldi-type algorithm. A detailed convergence analysis is provided. Some enlightening numerical examples show the very good properties of the proposed method.