Optimized Solver for Sequences of Sparse Eigenvalue Problems Arising in Ab Initio Computations


Recent trends in high performance computing have put an old algorithm for calculating the spectral decomposition back on the map: Subspace iteration accelerated via polynomials or rational functions. Modern formulations of accelerated subspace iteration can achieve better parallel speedup than Krylov-based methods. Our goal is to provide a solver based on rationally filtered subspace iteration for large and sparse matrices. We target high performance Density Function Theory codes as a potential area of application.