Improvement of Fourteen Coupled Global Potential Energy Surfaces of 3A' States of O + O2.

Abstract

We improved the potential energy surfaces for 14 coupled ³A' states of O₃ by using parametrically managed diabatization by deep neural network (PM-DDNN) with three improvements: (1) We used a new functional form for the parametrically managed activation function, which ensures the continuity of the coordinates used in the parametric management. (2) We used higher weighting for low-lying states to achieve smoother potential energy surfaces. (3) The asymptotic behavior of the coupled potential energy surfaces was further refined by utilizing a better low-dimensional potential. As a result of these improvements, we obtained significantly smoother potentials that are better suited for dynamics calculations. For the new version of 14 coupled ³A' surfaces, the entire set of 532,560 adiabatic energies are fit with a mean unsigned error (MUE) of 45 meV, which is only 0.7% of the mean energy in the data set, which is 6.24 eV.