ML-Accelerated Automatic Process Exploration Reveals Facile O-Induced Pd Step-Edge Restructuring on Catalytic Time Scales

Abstract

We combine automatic process exploration with an iteratively trained machine-learning interatomic potential to systematically identify elementary processes occurring during the initial oxidation of a Pd step edge. Corresponding process lists are a prerequisite to overcome prevalent predictive-quality microkinetic modeling approaches which consider only a minimum number of hand-selected and thus typically intuitive processes. The exploration readily generates close to 3000 inequivalent elementary processes and thus unveils a complexity far beyond current microkinetic modeling capabilities. Among these processes are numerous low-barrier processes involving the collective motion of several atoms that enable a facile O-mediated restructuring of the Pd step edge through the motion of larger Pd_xO_y units. The concomitant interconversion happens on time scales comparable to those of molecular processes of heterogeneous oxidation catalysis. This suggests a dynamic aspect of the operando evolution of the working interface reminiscent of the fluxionality discussed in nanocluster catalysis.