Regensburg 2022 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 10: Poster Session 1

MM 10.22: Poster

Montag, 5. September 2022, 18:00–20:00, P2

Benchmarking a Machine-Learning Enhanced Dimer Method for Transition State Search — •Nils Gönnheimer, King Chun Lai, Karsten Reuter, and Johannes T. Margraf — Fritz-Haber-Institut der Max-Planck-Gesellschaft, Berlin, Germany

The implementation of Machine-Learning (ML) surrogate models into established atomistic simulation types (e.g. molecular dynamics or geometry optimizations) offers the opportunity of significantly lowering their computational cost. The Dimer method for finding saddle points on high-dimensional potential surfaces is a prime example of this. This method is an important tool for locating transition states and exploring reaction mechanisms in heterogeneous catalysis, but applications are limited by its large computational cost. Jacobsen et al. recently showed that this can be overcome by combining Dimer search with a Gaussian Process Regression surrogate model in the Artificial Intelligence-Driven dimer (AID-TS) algorithm. To better understand the role of the ML surrogate in this method, we compare accuracy, efficiency and diversity of the found states, for AID-TS and conventional dimer search, using surface self-diffusion of Pd(100) as an example.