Deep Learning for Molecular Orbitals

Abstract

The advancement of deep learning in chemistry has resulted in state-of-the-art models that incorporate an increasing number of concepts from standard quantum chemistry, such as orbitals and Hamiltonians. With an eye towards the future development of these deep learning approaches, we present here what we believe to be the first work focused on assigning labels to orbitals, namely energies and characterizations, given the real-space descriptions of these orbitals from standard electronic structure theories such as Hartree-Fock. In addition to providing a foundation for future development, we expect these models to have immediate impact in automatizing and interpreting the results of advanced electronic structure approaches for chemical reactivity and spectroscopy.