Department of Computer Science and Engineering University of Minnesota
Room: DCH 1049
Sep 29, 2016 2:00 PM CST
Two filtering techniques are presented for solving large Hermitian eigenvalue problems by the method of spectrum slicing that consists of subdividing the spectrum in a number of subintervals and computing eigenvalues in each subinterval independently. In the first approach, the filter is a polynomial constructed as the least-squares approximation to an appropriately centered Dirac distribution. The second approach targets matrices whose spectral distribution is very irregular, as well as generalized eigenvalue problems. It is based on using a rational filter in a least-squares sense. The efficiency and robustness of the proposed methods are demonstrated through some Hamiltonian matrices from electronic structure calculations.