Dr. Philipp Marquetand, University of Vienna, Institute of Theoretical Chemistry, Vienna, Austria

Excited-state Dynamics Simulations with Machine Learning

Light can induce a wealth of processes in electronically excited states but corresponding simulations are limited by the costly computations of potential energy surfaces. A solution to this problem will be presented, where machine learning potentials are used to carry out excited-state molecular dynamics. The dynamics is simulated with our surface hopping approach called SHARC (surface hopping including arbitrary couplings), which is able to treat not only kinetic dynamical couplings but also any other arbitrary coupling on an equal footing. Consequently, machine learning is employed not only for potentials but also for nonadiabatic couplings. These developments open up the possibility to simulate time scales in the nanosecond regime compared to a few picoseconds in conventional approaches.

Excited-state Dynamics Simulations with Machine Learning

Light can induce a wealth of processes in electronically excited states but corresponding simulations are limited by the costly computations of potential energy surfaces. A solution to this problem will be presented, where machine learning potentials are used to carry out excited-state molecular dynamics. The dynamics is simulated with our surface hopping approach called SHARC (surface hopping including arbitrary couplings), which is able to treat not only kinetic dynamical couplings but also any other arbitrary coupling on an equal footing. Consequently, machine learning is employed not only for potentials but also for nonadiabatic couplings. These developments open up the possibility to simulate time scales in the nanosecond regime compared to a few picoseconds in conventional approaches.