Global prediction of nuclear charge density distributions using a deep neural network

A deep neural network (DNN) has been developed to generate the distributions\nof nuclear charge density, utilizing the training data from the relativistic\ndensity functional theory and incorporating available experimental charge radii\nof 1014 nuclei into the loss function. The DNN achieved a root-mean-square\n(rms) deviation of 0.0193 fm for charge radii on its validation set.\nFurthermore, the DNN can improve the description in both the tail and central\nregions of the charge density, enhancing agreement with experimental findings.\nThe model's predictive capability has been further validated by its agreement\nwith recent experimental data on charge radii. Finally, this refined model is\nemployed to predict the charge density distributions in a wider range of\nnuclide chart, and the parameterized charge densities, charge radii, and\nhigher-order moments of charge density distributions are given, providing a\nrobust reference for future experimental investigations.\n