The eigenvalue problem, Ax = /\x, is ubiquitous in Science and Engineering. Many numerical tools have been developed over the years to compute eigenvalues and eigenvectors. The FEAST eigenvalue solver belongs to a new generation of techniques that are ideally suited for large-scale parallel computing. FEAST requires the user to specify a search interval in the eigenvalue spectrum which may be difficult to know a priori for computing the extreme eigenvalues. This project proposes a tool for computing the search interval of M extreme lowest or largest eigenvalues. We employ procedures such as Gershgorin’s circle theorem and FEAST stochastic techniques in our approach. The resulting search interval which is obtained at a relatively low cost can be used as an input for FEAST.
More info: http://www.feast-solver.org/