Predicting The One-Particle Density Matrix With Machine Learning

Abstract

Two of the most widely used electronic structure theory methods, namely Hartree-Fock and Kohn-Sham density functional theory, both requires the iterative solution of a set of Schr\"odinger-like equations. The speed of convergence of such self-consistent field process depends on the complexity of the system under investigation and on the initial guess for the density matrix. An initial density matrix close to the ground-state one will effectively allow one to cut out many of the self-consistent steps necessary to achieve convergence. Here, we predict the density matrix of Kohn-Sham density functional theory by constructing a neural network, which uses the atomic positions as only information. Such neural network provides an initial guess for the density matrix far superior to any other recipes available, in particular for molecules with metallic bonds. Furthermore, the quality of such neural-network density matrix is good enough for the evaluation of interatomic forces. This allows us to run accelerated {\it ab-initio} molecular dynamics with little to no self-consistent steps.