Dependence between Structural and Electronic Properties of CsPbI3: Unsupervised Machine Learning of Nonadiabatic Molecular Dynamics.

Using unsupervised machine learning on the trajectories from a nonadiabatic molecular dynamics simulation with time-dependent Kohn-Sham density functional theory, we elucidated the structural parameters with the largest influence on nonradiative recombination of charge carriers in CsPbI3, which forms the basis for solar energy and optoelectronic applications. The I-I-I angles between PbI6 octahedra, followed by the Cs-I distance, have the strongest impact on the bandgap and the nonadiabatic coupling. The importance of the Cs-I distance is unexpected, because Cs does not contribute to electron and hole wave functions. The nonadiabatic coupling is most influenced by static properties, which is also surprising, given its explicit dependence on atomic velocities.