Multiple-Timestep ab Initio Molecular Dynamics Using an Atomic Basis Set Partitioning.

Abstract

This work describes an approach to accelerate ab initio Born-Oppenheimer molecular dynamics (MD) simulations by exploiting the inherent timescale separation between contributions from different atom-centered Gaussian basis sets. Several MD steps are propagated with a cost-efficient, low-level basis set, after which a dynamical correction accounts for large basis set relaxation effects in a time-reversible fashion. This multiple-timestep scheme is shown to generate valid MD trajectories, on the basis of rigorous testing for water clusters, the methanol dimer, an alanine polypeptide, protonated hydrazine, and the oxidized water dimer. This new approach generates observables that are consistent with those of target basis set trajectories, including MD-based vibrational spectra. This protocol is shown to be valid for Hartree-Fock, density functional theory, and second-order Møller-Plesset perturbation theory approaches. Recommended pairings include 6-31G as a low-level basis set for 6-31G** or 6-311G**, as well as cc-pVDZ as the subset for accurate dynamics with aug-cc-pVTZ. Demonstrated cost savings include factors of 2.6-7.3 on the systems tested and are expected to remain valid across system sizes.