Photodynamics With Neural Networks and Kernel Ridge Regression

Photodynamics simulations can unravel the secrets and intricacies of photochemistry in an intuitive way. These simulations can be carried out efficiently, if machine learning is used to obtain the excited-state properties that steer the dynamics, like energies, forces, nonadiabatic couplings, spin-orbit couplings or dipole moments. In the present chapter, an introduction to such machine-learning based photodynamics simulations is given. The basics of the most commonly used machine learning techniques, neural networks and kernel ridge regression, are provided. The tricks and pitfalls when predicting excited-state properties with these machine learning techniques are discussed.