Efficient algorithms for eigenvalue problems

In computational science symmetric eigenvalue problems are central and the need for fast and accurate algorithms are high. When solving a symmetric eigenvalue problem the easiest way is to first transform the full matrix into a tridiagonal problem and then solve it. In this thesis we studie two algoritms for the symmetric tridiagonal eginvalue problem, Cuppen's Divide and Conquer and Dhillon's O(n²). These two algorithms show better performance than the classical Bisection followed by Inverse Iteration. Issues about implementation both serial and parallell are discussed.Validerat; 20101217 (root