Berlin 2024 – scientific programme

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

O: Fachverband Oberflächenphysik

O 16: Surface Dynamics & Electron-Driven Processes

O 16.5: Talk

Monday, March 18, 2024, 16:00–16:15, MA 141

Realistic Representations of IrO2 Catalyst Surfaces through Extensive Sampling — •Hao Wan, Hendrik H. Heenen, Christoph Scheurer, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin

Iridium oxide catalyzes the oxygen evolution reaction with unparalleled activity and stability, even under harsh acidic conditions. However, this performance is sensitively correlated to strong structural, compositional and morphological changes of the working catalyst. At the atomic level little is presently known about the true active state, if only that it is unlikely ideal rutile IrO2.

This situation spans a vast configurational space, the extensive sampling of which e.g. by means of parallel tempering would be intractable with direct predictive-quality first-principles calculations. Training of a machine-learning interatomic potential (MLIP) as an efficient surrogate is in turn challenged by an unprecedented required diversity of training structures if not even the bulk structure and composition can be assumed known. To this end, we create a comprehensive training set by first assembling prototype bulk structures for various IrO* stoichiometries from existing databases. In an active learning loop, this set is then augmented through extensive sampling of diverse surface structures created from the prototypes. The thus trained MLIP successfully reproduces the known stability reversal of the rutile (110) and (111) facets with increasing potential, while its computational efficiency allows to rapidly probe the activity and stability of the plethora of sampled surface sites using established descriptors.

Keywords: Machine learned interatomic potentials; IrO2 catalysis surface; active learning