Reaction Pathway Detection using Machine-Learned Energy Potentials -- Decomposition of Energized CF$_3$CHOO

Abstract

Characterization of the decomposition products of energized Criegee intermediates is essential for assessing their impact on the chemical evolution of the atmosphere. Here, a generic and microscopically resolved approach is used to determine the molecular fragmentation pathways and products for CF$_3$CHOO. They include, among others, direct formation of CO$_2$ + CHF$_3$ (HFC-23), HF + CO$_2$ + CF$_2$, and fragmentation routes that are not evident from static reaction path calculations alone. The computed probability for formation of HFC-23 of 14 \% qualitatively agrees with a value of $(7.9^{+0.4}_{-0.2})$ \% from recent measurements, given the differences in the two approaches. Non-statistical dynamics is found for almost all decomposition pathways and the simulations show that excess energy can redirect reaction outcomes away from minimum-energy pathways. The results highlight the power of machine-learned PESs to elucidate multi-step reaction mechanisms of atmospherically relevant intermediates beyond traditional Master equation/electronic structure approaches to provide molecular-level understanding of the role of dynamics.