Petr Plechac, University of Delaware
We introduce approximations of ab-initio molecular dynamics derived from quantum mechanics.
Molecular dynamics simulations are often used to approximate canonical quantum correlation
observables in complex nuclei-electron systems. We present shallow random feature neural
networks and provide an analysis of their approximation properties. Furthermore, we describe
an adaptive sampling strategy that ensures a near-optimal distribution of features, thus
enabling controlled approximation of inter-atomic potentials for molecular dynamics simulations.
Finally, we demonstrate that the resulting molecular dynamics accurately approximate correlation
observables with quantifiable error estimates.