Parallel implementation of the CI-vector evaluation in full CI/CAS-SCF

Abstract

An implicitly parallel, direct reduced list method for integral-driven full configuration interaction (FCI) algorithms to be used in CAS-SCF is presented. The algorithm makes use of modern supercomputer hardware supporting both shared memory and distributed memory architectures. The applicability is demonstrated with a CAS-SCF(14,14) calculation on stilbene and a CAS-SCF(8,16) calculation on pentalene. The algorithm is memory efficient, and there is compressed storage of precomputed excitation lists in main memory. The usage of single excitation and double excitation lists is made.