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

MM: Fachverband Metall- und Materialphysik

MM 8: Development of Computational Methods: Diverse Topics and Machine Learning

MM 8.4: Vortrag

Montag, 27. März 2023, 16:30–16:45, SCH A 251

Exploring Enhanced Sampling Concepts based on Boltzmann Generators — •David Greten, Karsten Reuter, and Johannes T. Margraf — FHI Theory Department, Berlin, DE

Computational surface science and catalysis research is still mainly conducted with static density functional theory (DFT) calculations. This approach is computationally convenient, but misses important aspects of surface chemistry, such as anharmonic free energy contributions. In principle, DFT-based molecular dynamics (MD) simulations (ideally combined with enhanced sampling algorithms) would allow a much more accurate description of these processes. Unfortunately, these are far too expensive to be routinely applied to complex surface/adsorbate systems. This is due to the fact that configurations in MD are generated sequentially. As a consequence, MD configurations are not statistically independent so that a very large number of samples is required to obtain converged ensemble properties. To overcome this limitation, Noé and co-workers recently proposed a generative machine learning model called the Boltzmann Generator, which was used to generate independent configurations of biomolecules. In this contribution, we explore how we can expand ML based sampling concepts utilizing Boltzmann Generators. In particular, training protocols and validation metrics will be discussed.