Prof. Hans Lischka: Machine Learning Allows First High-level Photodynamical Simulations of Polyenes

The nonadiabatic photodynamics of polyenes is a paradigmatic but extremely challenging process in Photochemistry because of the strongly differing ionic and covalent characters of the involved excited states. In recent work, we have shown that the multireference configuration interaction (MRCI) method of COLUMBUS is very well suited for a balanced description of these states. To make photodynamic simulations in the framework of non-adiabatic surface hopping feasible, Machine Learning (ML) approaches have been applied in a first example to all-trans hexatriene. An initial surface-hopping dynamics using the semiempirical AIQM1 method has been performed to describe the scope of the structural space of the photodynamics. Employing active learning techniques, this entire structural space has been reduced to ~3000 representative structures, for which MRCI calculations have been performed. These data served as input for a new Machine Learning Interatomic Potential denoted MLIP MRCI. New surface-hopping dynamics simulations gave good agreement with experimentally determined lifetimes. The possibilities to extend these simulations to larger polyenes will be discussed.

Hans Lischka is a distinguished theoretical and computational chemist and Research Professor at Texas Tech University. He earned his Ph.D. from the University of Vienna in 1969, where he pursued a distinguished academic career. His research focuses on the development of multireference methods, photodynamics, and machine learning, with major contributions to computational chemistry. He has held research and visiting professorships in Europe, the USA, and China. In 2023, he was elected a Corresponding Member of the Brazilian Academy of Sciences and received the EuChemS European Chemical Society Lifetime Achievement Award.