Computing Surface Reaction Rates by Adaptive Multilevel Splitting Combined with Machine Learning and Ab Initio Molecular Dynamics

Computing accurate rate constants for catalytic eventsoccurringat the surface of a given material represents a challenging task withmultiple potential applications in chemistry. To address this question,we propose an approach based on a combination of the rare event samplingmethod called adaptive multilevel splitting (AMS) and ab initio molecular dynamics. The AMS method requires a one-dimensional reactioncoordinate to index the progress of the transition. Identifying agood reaction coordinate is difficult, especially for high dimensionalproblems such as those encountered in catalysis. We probe variousapproaches to build reaction coordinates such as support vector machineand path collective variables. The AMS is implemented so as to communicatewith a density functional theory-plane wave code. A relevant casestudy in catalysis, the change of conformation and the dissociationof a water molecule chemisorbed on the (100) gamma-alumina surface,is used to evaluate our approach. The calculated rate constants andtransition mechanisms are discussed and compared to those obtainedby a conventional static approach based on the Eyring-Polanyiequation with harmonic approximation. It is revealed that the AMSmethod may provide rate constants that are smaller than those providedby the static approach by up to 2 orders of magnitude due to entropiceffects involved in the chemisorbed water molecule.