Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 62: Surface Reactions and Heterogeneous Catalysis 1
O 62.4: Vortrag
Donnerstag, 8. September 2022, 11:15–11:30, H4
Predicting Binding Motifs of Complex Adsorbates Using Machine Learning with a Physics-inspired Graph Representation — •Wenbin Xu1, Karsten Reuter1, and Mie Andersen2 — 1Fritz-Haber-Institut der MPG, Berlin, Germany — 2Aarhus Institute of Advanced Studies and Department of Physics and Astronomy, Aarhus University, Denmark
Complex adsorbates are involved in many surface catalytic reactions such as Fischer-Tropsch, methanol, or higher oxygenate synthesis. The modeling of these species at transition metal catalysts must account for their ability to exhibit a wide range of adsorption motifs, including mono- and multi-dentate adsorption modes. Given the combinatorial explosion of possible adsorption motifs and the computational cost of density functional theory, it is desirable to develop machine learning (ML) models for predicting the binding motifs and their associated adsorption enthalpies. Most ML models to date are only applicable to simple adsorbates. In this work, we overcome this limitation and propose a kernelized ML model with a physics-inspired graph representation for the prediction of complex species. The model is data-efficient and its good extrapolation ability makes it promising for comprehensively exploring complex reaction networks on novel catalysts. Furthermore, we show that the outliers with large prediction errors can be reliably captured from an ensemble uncertainty prediction approach.