Fig. 1

A schematic illustration of the ROSA feature. The top figure displays an ordinary DFT calculation, in which two self-consistent optimization loops (the electron density optimization and geometry optimization) determine the electronic structure of the ground state system. This ground state is identified by energy eigenvalues and total energy (the system “Energies”). The lower figure shows how ROSA feature approximate the properties of the ground state system: it extracts the system “Energies” before optimization and uses that to train a machine learning model to predict the desired properties