Data-driven density functional model for atomic nuclei

Through ensemble learning with multitasking and complex connection neural networks, we aggregated nuclear properties, including ground-state density distributions, charge radii, and binding energies obtained from the Kohn-Sham auxiliary single-particle systems. The root-mean-squared (RMS) error in describing binding energies is reduced to about 500 keV, while the RMS error in describing charge radii is decreased to about 0.017 fm. In addition, by using the correlation between densities and binding energies, we discuss the impact of various densities and nuclear radii on the binding energy and find the neutron-skin thickness of ²⁰⁸Pb to be around 0.2200.16em0exfm. The present paper provides a new way to accelerate the integration of machine learning into nuclear density functional theory.