Machine Learning for Electronically Excited States of Molecules

Abstract

Electronically excited states of molecules are at the heart of photochemistry, photophysics, as well as photobiology and also play a role in material science. Their theoretical description requires highly accurate quantum chemical calculations, which are computationally expensive. In this review, we focus on not only how machine learning is employed to speed up such excited-state simulations but also how this branch of artificial intelligence can be used to advance this exciting research field in all its aspects. Discussed applications of machine learning for excited states include excited-state dynamics simulations, static calculations of absorption spectra, as well as many others. In order to put these studies into context, we discuss the promises and pitfalls of the involved machine learning techniques. Since the latter are mostly based on quantum chemistry calculations, we also provide a short introduction into excited-state electronic structure methods and approaches for nonadiabatic dynamics simulations and describe tricks and problems when using them in machine learning for excited states of molecules.

Automatic discovery of photoisomerization mechanisms with nanosecond machine learning photodynamics simulations

“…We picked two sets of the most accurate and efficient NNs (each set has one NN for energies and forces together, and another one for NACs) with distinct architectures. We implemented all NNs using the TensorFlow/Keras (v2.3) packages 33 34,35 None of these individual techniques can adequately sample the relevant potential energy surfaces of a photochemical reaction as the molecular structure become more complex and the degrees of freedom increase. For example, Wigner sampling is limited to the congurations immediately related to equilibrium geometries.…”

A Force Field for a Manganese-Vanadium Water Oxidation Catalyst: Redox Potentials in Solution as Showcase

“…The FF parameters that we obtained in this way performed significantly worse than the manually refined parameters, which is why we focus only on the latter in this work. Other possible alternatives to obtaining FFs for challenging systems include QM-derived FFs [62,63] and machine-learning-based potentials [64,65].…”

Predicting phosphorescence energies and inferring wavefunction localization with machine learning

“…24,31,32 ML for ground state chemistry has been explored at an appreciable depth, achieving models of high-accuracy theories [33][34][35] and extremely large systems. 36 However, its use for excited state processes [37][38][39] such as phosphorescence is a new area of research.…”