Parts | Days | Selection | Search | Updates | Downloads | Help

O: Fachverband Oberflächenphysik

O 52: Heterogeneous Catalysis and Surface Dynamics I

O 52.6: Talk

Wednesday, March 29, 2023, 12:00–12:15, TRE Phy

Machine learning interatomic potential for metallic and oxidized copper surfaces — •Felix Riccius, Nicolas Bergmann, Nicolas G. Hörmann, Hendrik H. Heenen, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin, Germany

Copper (Cu) is a promising catalyst for the electrochemical reduction of CO₂ to chemicals and fuels since it yields many possible reduction products. The selectivity between these products is hypothesized to be altered by partial oxidation of the catalysts* surface. The extent of Cu surface oxidation, as well as accompanying morphological transformations are, however, unclear. Predictive atomistic simulations could potentially uncover this surface chemistry but time and length scales necessary for the required sampling are intractable by first-principles methods. Machine learning interatomic potentials (MLIP) trained to first principle data can overcome this limitation by retaining predictive accuracy at a fraction of the computational cost. In this work, we use Gaussian Approximation Potentials to train a MLIP for metallic Cu and Cu oxides. We design a workflow tailored to the problem at hand. Via iterative parallel tempering simulations, we sample the relevant phase space for Cu oxidation at various oxidation states during training. We demonstrate the viability of our approach, which results in a reliable potential capable of describing the chemically important regions of the configuration space. The potential gives access to (surface)phase diagrams and surface reconstruction phenomena following surface reduction and oxidation. The atomic insight may shine light on the role of (surface)oxidation in Cu electrocatalysts.