Resonance systematics for capture reactions using Machine Learning

Nuclear data is used for a variety of purposes in our daily life, from basic science to advanced usages like nuclear power. Nuclear reaction data must be evaluated for all these purposes that require a detailed knowledge of the neutron-nucleus cross section. Below 1MeV incident energy, such cross sections show fluctuations, or resonances, that are not predictable and are characterized by several parameters like the resonance spacing, the distance in energy between two resonance peaks, and the resonance width, the width of the resonance peaks. These parameters can be measured experimentally, but for specific energies of the incident neutron this is not possible, because they cannot be resolved experimentally. Thus, all we can do is to calculate the average values of such parameters. The goal of my project is to study the average resonance widths, and I have developed a code to extract the average widths from the Atlas of Neutron Resonances, for all elements given their atomic species and mass number. Using machine learning, I have developed code that learns the average capture width and the capture degrees of freedom from the capture width survival function and the results show survival functions for each combination of allowed quantum numbers of orbital (L) and total (J) angular momentum. Problems related to small widths and deficient data in the Atlas have been also corrected using machine learning techniques to compare the resonance systematics for capture reactions and help to fill in the information to correct the misassignment. We identified and corrected several errors in the Atlas. The current goal is to use the extracted resonance widths and tabulated L and J assignments from the Atlas and so that we can produce average resonance widths for specific materials.