Eigensolvers for Multi-core Processors and Massively Parallel Supercomputers
Eigensolvers for Multi-core Processors and Massively Parallel Supercomputers The problem of computing the eigenvalues and eigenvectors of large matrices comes up in many applications of computational engineering and science. We aim at developing theories and algorithms for efficiently and accurately solve large eigenproblems on emerging and parallel computer architectures. One of our objectives is to link the results to real applications. Much of the knowledge available to scientists and engineers is lost when the input matrix is formed: no information on the process that generates the matrix and on how the computed quantities are used is available to traditional solvers. Our goal is to devise high-quality eigensolvers by identifying and explicitly exploiting such knowledge.