Study of nuclear low-lying excitation spectra with the Bayesian neural network approach

Abstract

Nuclear low-lying excitation spectra are studied with the Bayesian neural network (BNN) approach by taking 02+, 21+, and 41+ states as examples. The BNN approach can well describe the low-lying excitation energies in a large energy scale from about 0.1 MeV to about several MeV, by including an input related to nuclear collectivity besides proton and neutron numbers and employing the logarithm of excitation energy as the output. Comparing with the sophisticated microscopic collective Hamiltonian model, the BNN approach significantly improves the description of nuclear low-lying excitation energies, which can generally reproduce the experimental data within about 1.12 times including those of transitional nuclei and magic nuclei. Taking Mg, Ca, Kr, Sm, and Pb isotopes as examples, it is found that the BNN approach well describes the isotopic trend of low-lying excitation energies, including those abrupt increases at magic numbers due to the shell effect, very low excitation energies of 02+ states due to the shape coexistence, and complex nuclear shape evolution due to the shape phase transition.