Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 57: New Methods: Theory
O 57.6: Vortrag
Mittwoch, 20. März 2024, 17:30–17:45, MA 043
Enhanced Sampling of Rare Events Using Boltzmann Generators — •David Greten1, Karsten Reuter1, and Johannes T. Margraf1,2 — 1Fritz-Haber-Institut der MPG, Berlin — 2University of Bayreuth
The kinetics of catalytic processes in surface chemistry are determined by rare events such as bond breaking or bond forming reactions. These events are highly challenging to describe in atomistic simulations because brute-force molecular dynamics (MD) simulations would require an unfeasible number of time steps to observe them. Biased MD methods like Metadynamics or Umbrella Sampling represent the most rigorous approach to overcome this limitation. Such enhanced sampling methods are thus the gold standard for the computation of free energy barriers. Nonetheless, they inherit some of the drawbacks of conventional MD, namely long equilibration periods and correlation between samples along the trajectory. In this context, generative machine learning models (such as Boltzmann Generators, BGs) may offer significant advantages, as they can directly generate statistically independent samples from the Boltzmann distribution. In this contribution, we explore how BGs can be used in enhanced sampling simulations of rare events.
Keywords: Machine Learning; ML; Enhanced Sampling; Generative Model; Generator