Fig. 1: Schematic of the data generation and simulation workflow.
From: Low-index mesoscopic surface reconstructions of Au surfaces using Bayesian force fields
a DFT-sized input geometries for active learning in FLARE. b Active learning in FLARE, as described in more detail in refs. 7 and36; adapted with permission (CC BY 4.0). c ML-MD simulations performed in LAMMPS for various facets with defects, mechanical strains, and temperatures applied. d Nucleation kinetics and mechanisms underlying the transition towards the reconstructed phases are made available to determine the stoichiometric and mechanical regimes wherein each surface is reconstructed, and what is the resulting surface geometry.
