On application of deep learning to simplified quantum-classical dynamics in electronically excited states

Abstract Deep learning (DL) is applied to simulate non-adiabatic molecular dynamics of phenanthrene, using the time-dependent density functional based tight binding (TD-DFTB) approach for excited states combined with mixed quantum–classical propagation. Reference calculations rely on Tully’s fewest-switches surface hopping (FSSH) algorithm coupled TD-DFTB, which provides electronic relaxation in fair agreement various available experimental results. Aiming at describing electron-nuclei large systems, we then examine combination DL excited-state potential energy surfaces (PESs) a simplified trajectory propagation Belyaev–Lebedev (BL) scheme. We start assess accuracy TD-DFTB upon comparison optical spectrum and higher-level theoretical Using recently developed SchNetPack (Schütt et al 2019 J. Chem. Theory Comput. 15 448–55) applications, train several models evaluate their performance predicting energies forces. Then, main focus given analysis population low-lying computed aforementioned methods. determine timescales compare them data. Our results show that demonstrates its ability describe PESs. When BL scheme considered this study, it reliable description phenanthrene as compared either data or FSSH/TD-DFTB Furthermore, allows high-throughput negligible cost.