A tiered approach to Monte Carlo sampling with self-consistent field potentials

Abstract

A "tiered" approach to Monte Carlo sampling of nuclear configurations is presented for ab initio, self-consistent field (SCF)-based potentials, including Hartree-Fock and density functional theory. Rather than Metropolis testing only the final SCF energy, individual cycle energies are tested in a tiered fashion, without approximation. Accordingly, rejected configurations are terminated early in the SCF procedure. The method is shown to properly obey detailed balance, and effective modifications are presented for cases in which the initial SCF guess is particularly poor. Demonstrations on simple systems are provided, including an assessment of the thermal properties of the neutral water dimer with B3LYP/6-31++G**. Cost analysis indicates a factor-of-two reduction in SCF cycles, which makes the method competitive with accelerated molecular dynamics sampling techniques, without the need for forces.