Bayesian neural network prediction methods for fragment cross sections in proton-induced spallation reactions
Abstract
Nuclear spallation reactions can be induced by light particles with the incident energy greater than 10 MeV resulting in the production of various radioactive nuclides. The cross sections of fragment creation and fragment production are the fundamental and basic data, respectively for nuclear physics and nuclear applications in a wide variety of fields. Due to the complexity in spallation systems varying from light to heavy, incident energy ranging from MeV to GeV, fragment charge and mass numbers, the current models should be upgraded for increased precision. Using the Bayesian neural network (BNN), machine learning approaches have been established to predict the fragment cross sections in the proton-induced nuclear spallation reactions. The direct BNN method, together with the physical model guided BNN + SPACS and BNN + sEPAX methods, has been shown to improve the precision of fragment cross sections.
