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

O: Fachverband Oberflächenphysik

O 69: Surface Reactions and Heterogeneous Catalysis 2

O 69.8: Vortrag

Donnerstag, 8. September 2022, 17:00–17:15, H4

Machine-learning Gaussian Approximation Potentials to solve a longstanding puzzle about RuO₂ surfaces — •Yonghyuk Lee, Jakob Timmermann, Christoph Scheurer, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin, Germany

Machine-learning Gaussian Approximation Potentials (GAPs) have recently evolved as a powerful class of surrogate models to computationally demanding first-principles calculations. Along with structure exploration techniques, they enable us to examine the potential energy surface of interest with a hitherto unforeseen combination of physical accuracy and computational efficiency and to achieve global surface structure determination (SSD) for increasingly complex systems. This can be leveraged e.g. to discover novel surface motifs which are critical in understanding the “living" state of heterogeneous catalysts and their degradation under dynamic operating conditions. In our preceding study, this versatility could be leveraged by a general and data-efficient iterative training protocol that allows for the on-the-fly generation of GAPs via the actual surface exploration process. The iterative refinement of GAPs identifies plenty of unknown low energy terminations of RuO₂ even within the restricted sub-space of (1×1) surface unit-cells. Moreover, by extending the protocol to larger surface unit-cells, we discovered new surface structures, which provide solutions to longstanding questions in heterogeneous catalysis.

[1] J. Timmermann et al., Phys. Rev. Lett. 125, 206101 (2020)

[2] J. Timmermann et al., J. Chem. Phys., 155, 244107 (2021)