Refining mass formulas for astrophysical applications: A Bayesian neural network approach

The masses of nuclei between the experimentally known region and the neutron drip line are key inputs for a variety of astrophysical applications. Particularly near the drip lines, unstable nuclei are also at the core of fundamental questions about the limits of nuclear binding. The authors use two existing mass models that capture the essential underlying physics and then refine their predictions by training an artificial neural network. The results significantly reduce the root-mean-square deviation relative to experiment. These newly refined mass tables are used to map the neutron drip lines and to study a few critical r-process nuclei.