Self-learning kinetic Monte Carlo method: Application to Cu(111)

We present a method of performing kinetic Monte Carlo simulations that does not require an a priori list of diffusion processes and their associated energetics and reaction rates. Rather, at any time during the simulation, energetics for all possible (single- or multiatom) processes, within a specific interaction range, are either computed accurately using a saddle-point search procedure, or retrieved from a database in which previously encountered processes are stored. This self-learning procedure enhances the speed of the simulations along with a substantial gain in reliability because of the inclusion of many-particle processes. Accompanying results from the application of the method to the case of two-dimensional Cu adatom-cluster diffusion and coalescence on Cu(111) with detailed statistics of involved atomistic processes and contributing diffusion coefficients attest to the suitability of the method for the purpose.