Mirror symmetry is incorporated in the training process of the radial basis function (RBF) approach to improve the prediction accuracy of nuclear mass models. The RBF approach with mirror symmetry (RBFms) demonstrates significantly enhanced predictive performance, improving the accuracy of DZ31, FRDM12, KTUY, and WS4 mass models by 69-87% when evaluated against learning set. The evaluation of the test set demonstrates that the RBFms approach maintains exceptional performance in enhancing the predictive accuracy of the mass model. The effective extrapolation distance of the RBFms approach is farther than that of the RBF approach, based on the description of the decay energy of the ground-state two-proton decay nuclei. The KTUY model combined with the RBFms approach achieves a root-mean-square deviation of 0.076 MeV when describing the known masses of 217 atomic nuclei. The data training approach leveraging mirror symmetry may establish a novel learning paradigm for machine learning applications in nuclear physics.
Li et al. (Wed,) studied this question.