The electron density is obviously a key component in using Kohn-Sham density functional theory to obtain molecular energies. We use machine learning to create maps from the nuclear potential to the electron density, and subsequently train models for the electron density to energy maps. The latter models can also leverage a delta-learning approach to obtain density functionals that return ab initio (coupled-cluster) energies. I will show that with sufficient accuracy in predicted energies, we can use these models to generate molecular dynamics trajectories that sample strained geometries and conformer changes.
Back to Workshop IV: Monte Carlo and Machine Learning Approaches in Quantum Mechanics