Reactive rate coefficients and machine learning predictions for O(3P) + C2(X1Σg+) collisions on an accurate PIP-NN potential energy surface

Abstract

A full-dimensional potential energy surface (PES) for the 3A″ state of the [CCO] system has been constructed using neural networks (NNs) with permutationally invariant polynomials. This global analytical PES was accurately fitted from 9293 ab initio energies at the MRCI + Q/aug-cc-pVTZ level of theory. Based on the newly developed surfaces, the microscopic chemical reaction mechanisms of the O(3P) + C2(X1Σg+) → CO(X1Σ+) + C(3P) reactive collision were investigated using the quasi-classical trajectory (QCT) method. The reaction cross sections and rate coefficients obtained from QCT calculations are in good agreement with available theoretical and experimental data reported in the literature. Rate coefficient calculations indicate that for O + C2 collisions, the results for the reactive channel are significantly higher than those for the inelastic channel across a wide temperature range of 1000–20 000 K. Finally, to reduce computational demands, we also established an NN-based model to predict cross section by combining QCT with NNs. The developed model accurately reproduces the original QCT results.