Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 25: Interface Controlled Properties, Nanomaterials and Microstructure Design IV

MM 25.1: Vortrag

Dienstag, 19. März 2024, 11:45–12:00, C 230

Automatic exploration of elementary processes with adaptive kinetic Monte Carlo — •King Chun Lai, Sebastian Matera, Christoph Scheurer, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin

Lattice kinetic Monte Carlo (kMC) is very efficient for the simulation of rare event controlled dynamics but requires prior knowledge about the network of elementary processes. Currently, constructing such a network heavily relies on human intuition, which has its limitations in complex scenarios. We address this with a newly developed adaptive kMC (akMC) framework to automatically explore the reaction space. Elementary processes are identified solely on the basis of their energetics and without human bias. A key feature of our akMC framework is the on-the-fly labeling of local atomic environments, which is used to speed up the process searching algorithm. Demonstrating on island migration on Pd surfaces, we employ this labeling to structure the obtained elementary processes. Intriguingly, the process exploration readily identifies a number of non-trivial elementary steps, which involve collective motion of multiple atoms in a single event. Long time scale kMC simulations including these collective processes highlight the shortcomings of prevalent human-generated fixed process lists which typically center on intuitive single-atom events.

Keywords: Kinetic Monte Carlo; Nanomaterial; Classification; Transition states; Machine Learning