Machine Learning-Assisted Mixed Quantum-Classical\nDynamics without Explicit Nonadiabatic Coupling: Application to the\nPhotodissociation of Peroxynitric Acid

We\nhave devised a hybrid quantum–classical scheme utilizing\nmachine-learned potential energy surfaces (PES), which circumvents\nthe need for explicit computation of nonadiabatic coupling elements.\nThe quantities necessary to account for the nonadiabatic effects are\ndirectly obtained from the PESs. The simulation of dynamics is based\non the fewest-switches surface-hopping method. We applied this scheme\nto model the photodissociation of both N–O and O–O bonds\nin a conformer of peroxynitric acid (HO<sub>2</sub>NO<sub>2</sub>).\nAdiabatic PES data for the six lowest states of this molecule were\ncomputed at the CASSCF level for various nuclear configurations. These\nserved as the training data for the machine-learning models for the\nPESs. The dynamics simulation was initiated on the lowest optically\nbright singlet excited state (S<sub>4</sub>) and propagated along\nthe two Jacobi coordinates J⃗1 and J⃗2 while accounting for the nonadiabatic effects\nthrough transitions between PESs. Our analysis revealed that there\nis a very high chance of dissociation of the N–O bond leading\nto the HO<sub>2</sub> and NO<sub>2</sub> fragments.