A parallel eigenvalue problem solver for computational nanoelectronics
Maxim Naumov, Ahmed Sameh · PAMM · 2007
Abstract A new parallel eigenvalue solver for finding the interior eigenvalues of a standard Hermitian eigenvalue problem arising in atomistic simulations in nanoelectronics is presented. It is based on the Tracemin algorithm which finds the p smallest eigenpairs of a generalized Hermitian eigenvalue problem. The original problem is modified using spectrum folding or a quadratic mapping so that the interior eigenvalues are mapped onto the smallest or the largest, respectively. In the latter case the solution of systems in every iteration of Tracemin is avoided and Chebyshev polynomials are used to speedup convergence. (© 2008 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)