Frozen Gaussian Wavepacket to Perform and Correct Nonadiabatic Mixed Quantum- Classical Dynamics

Nonadiabatic dynamics methods play a crucial role in investigating the reaction of molecular systems to being excited by photons and submitted to a new electronic potential energy surface. The topography of the excited state can be explored to propose chemical reactions facilitated by photoexcitation, study the molecular stability, and energy harvesting processes, among other purposes. To this day, many new methods for nonadiabatic dynamics and the electronic structure calculations associated with them are still being developed to improve accuracy, reduce computational costs, and expand the application domain. Despite their success, those methods often use approximations that may lead to wrong results, which can be hard to detect. In those cases, one can usually refer to experimental results to validate the model used to represent the system. Alternatively, one can perform multiple sets of dynamics with different methods to detect a trend and outliers, but this would have extended computational costs. To address those challenges of performing affordable, high-quality simulations while still keeping track of their occasional inaccuracies, we developed a research program branched into two main topics. The first one resulted in a computer program named Legion, a platform to facilitate the development of new methods for nonadiabatic dynamics and modify existing ones. It is built in Python so that the code can be easily expanded, and it is already interfaced with multiple commonly used electronic structure methods and software. We test the program with real molecules and use it to complement the discussion of fulvene dynamics and its dependence on initial conditions. The second research branch resulted in QDCT, a novel strategy for post-processing surface hopping results to obtain wavepacket dynamics results, as in multiple spawning. It uses classical trajectories to propagate the nuclear wavepacket. The method is indifferent to ad hoc approximations present in surface hopping and can be used to assess problematic systems. The method is tested against analytical models, and approximations necessary to work in multi-dimensional systems are presented. Both Legion and QDCT are integrated into the Newton-X platform and are freely available.