Nonadiabatic Coupling in Trajectory Surface Hopping: How Excited-State Reaction

Photochemical reactions are widely modeled using the popular trajectory surface hopping (TSH) method, an affordable mixed quantum-classical approximation to the full quantum dynamics of the system. TSH is able to account for nonadiabatic effects using an ensemble of trajectories, which are propagated on a single potential energy surface at a time and which can hop from one electronic state to another. The occurrences and locations of these hops are typically determined using the nonadiabatic coupling between electronic states, which can be assessed in a number of ways. In this work, we benchmark the impact of some approximations to the coupling term on the TSH dynamics for several typical isomerization and ring-opening reactions. We have identified that two of the schemes tested, the popular local diabatization scheme and a scheme based on biorthonormal wave function overlap implemented in the OpenMOLCAS code as part of this work, reproduce at a much reduced cost the dynamics obtained using the explicitly calculated nonadiabatic coupling vectors. The other two schemes tested can give different results, and in some cases, even entirely incorrect dynamics. Of these two, the scheme based on configuration interaction vectors gives unpredictable failures, while the other scheme based on the Baeck-An approximation systematically overestimates hopping to the ground state as compared to the reference approaches.

Automated summary

This study benchmarks the impact of different approximations to the nonadiabatic coupling term on the trajectory surface hopping (TSH) dynamics of isomerization and ring-opening reactions. It is found that the popular local diabatization scheme and a scheme based on biorthonormal wave function overlap in the OpenMOLCAS code can reproduce the dynamics obtained using explicitly calculated nonadiabatic coupling vectors at a reduced cost. However, the other two schemes tested can give different and even incorrect results.