Accelerating Kohn-Sham DFT calculations at extreme conditions using machine-learned force fields with ab initio accuracy

We present a framework for computing equation of state and transport properties for materials at extreme conditions of temperature and pressure using on-the-fly machine learned force field (MLFF) molecular dynamics simulations. We employ an MLFF model based on the kernel method and Bayesian linear regression to compute the free energy, atomic forces, and pressure, with training data provided on-the-fly from Kohn-Sham DFT calculations during the course of the simulation. We compare to recent Kohn-Sham and path Integral Monte Carlo results in the literature and find that, using the on-the-fly MLFF approach, Kohn-Sham calculations of equations of state can be accelerated by up to two orders of magnitude, and calculations of transport properties can be accelerated by up to three orders of magnitude while retaining ab initio accuracy. This work was performed in part under the auspices of the U.S. DOE by LLNL under Contract DE-AC52-07NA27344.