Roberto Car
Department of Chemistry and Department of Physics, Princeton University, Princeton, New Jersey - USA
Using a technique for incremental learning recently introduced by our group (L. Zhang, et al., Phys Rev Mat 3, 023804 (2019)) we constructed a conservative force-field that mimics accurately classical ab-initio molecular dynamics trajectories based on the SCAN functional approximation of density functional theory over a wide range of pressures and temperatures, including molecular and ionic phases, in both the liquid and crystalline state. The machine learned potential can model water dissociation and describes proton transfer processes in presence of hydronium and hydroxyl ions in solution. I will use the phase diagram predicted by the new potential, which extends from room temperature and pressure conditions up to temperatures as high as 2500 K and to pressures in excess of 100 GPa, to illustrate the capability of the approach. Finally, I will discuss a few simulation studies made possible by the new approach.