Theoretical Investigations of Rate Coefficients for H+O3 and HO2+O Reactions on a Full-Dimensional Potential Energy Surface.

In this work, a machine learning method is used to construct a high-fidelity multi-channel global reactive potential energy surface (PES) for the HO3 system from 21452 high-level ab initio calculations at the explicitly correlated multi-reference configuration interaction (MRCI-F12) level of theory. The permutation invariance of the PES with respect to the three identical oxygen atoms is enforced using permutation invariant polynomials (PIPs) in the input layer of a neural network (NN). This PIP-NN representation is highly faithful to the ab initio points, with a root mean square error of 0.20 kcal/mol. Using this PES, the kinetics of H + O3 → OH + O2 (R1) and HO2 + O → OH + O2 (R2) reactions were investigated using a quasi-classical trajectory method over a wide temperature range (200 - 2000 K). It was found that the calculated thermal rate coefficients of R1 and R2, exhibiting positive and negative temperature dependences, respectively, are in reasonably good agreement with most experimental measured values. These temperature dependences can be attributed to the presence and absence of an entrance channel potential barrier.