美国国家科学院院士、美国艺术与科学学院（AAAS）院士、Bolzano金奖获得者、著名矩阵计算权威、Stanford大学G.Golub教授http://www-sccm.stanford.edu/faculty/nf-golub.html将于10月26日下午2：30在数学中心201作学术报告。欢迎参加。

Abstract: We consider the problem of iteratively computing the stationary distribution vector of large finite Markov chains. It is assumed that the matrices involved are too large for a decompositional approach to be effective, and matrix-vector products must be used. The problem is motivated by Google's PageRank algorithm for large web databases. We consider an Arnoldi-type restarted algorithm based on a combination of Arnoldi and the SVD. Connection of the algorithm to other techniques such as the quadratic extrapolation method is discussed, and the sensitivity of the PageRank problem is also addressed. Numerical examples illustrate the performance and convergence behavior of the algorithm.