Efficient quantum Monte Carlo energies for molecular dynamics simulations

Abstract

Abstract This paper opened the area of molecular dynamics to “on the fly” trajectory calculations with Q:tvlC accuracies. Using the continuous evolution of fixed-node diffusion QMC electronic structure calculations simultaneously with the motion of nuclei the authors were able to gain more than an order of magnitude in computational speed and make such calculations feasible for a variety of interesting systems. “On the fly” molecular dynamics simulations using QMC energies were thus advanced to compete with those using density functional theorya. In the simulations described the nuclear Hamiltonian and the electronic wave function, represented by QMC walkers, were updated at each molecular dynamics step. This replaced discrete QMC calculations and took advantage of correlated sampling of electron configurations. Relaxation of the electronic wave function to new nuclear positions was required, but as few as three QMC steps per molecular dynamics step were found adequate. Calculations were carried out using nonlocal pseudopotentials with trial wave functions for valence electrons consisting of Slater determinants and Jastrow functions. Variational QMC was used for optimizations. The efficiency and accuracy of the dynamical QMC approach was demonstrated in calculations for silicon hydride species SiH4 to Si14H20 at 1000 K, for the dissociation of a water molecule as H20 HO + H, and for a 32-molecule liquid-water system along with a single molecule for comparisons. The calculations were clearly successful in producing accurate molecular dynamics simulations at a modest cost in computational effort.