Databased accelerated on-the-fly hybrid quantum/classical dynamics

Abstract

Light-triggered processes, which are ubiquitous in nature and technology,
are inherently quantum. Phenomena such as photovoltaic effect, charge
migration, and proton-coupled electron transfer require quantum
mechanical description. Understanding and optimizing these processes is the
key to novel technologies such as optogenetics, photopharmacology, and
photoresponsive materials. However, an accurate description of these
processes in (supra)molecular systems is still a demanding task due to: i) the
need of high-level electronic structure calculations; (ii) coupled electronnuclear
dynamics; (iii) the importance of the environment.
Recent years have seen a continuous growth of the direct dynamics
approaches, e.g. semiclassical trajectory surface hopping or on-the-fly
quantum dynamics within variational multi-configuration Gaussian, to study
coupled electron-nuclear dynamics of photo-active molecules of moderate
size. Particularly troublesome for trajectory-based methods is the huge
amount of electronic structure calculations that need to be performed
during the simulation. In this work, we present an innovative solution, i.e.
both semiclassical and quantum direct dynamics formulations are combined
with a database, within which the amount of electronic structure
calculations is drastically reduced, by employing machine-learning
algorithms and methods borrowed from the realm of artificial intelligence.
On-the-fly direct dynamics can be further embedded into a quantum
mechanics/molecular mechanics, for the explicit inclusion of the
environment. The degree of sophistication, which can be achieved by such
an implementation will be illustrated with a test system.