Resonance Capture Widths for the Bayesian Resonance Reclassifier

An accurate description of the neutron interaction cross sections for the most abundant isotopes of lead are essential for application in practical nuclear systems. These nuclear reaction cross sections demonstrate resonance phenomena in the presence of low-energy or thermal neutrons. The cross section shape is obtained by classifying the associated angular-momentum quantum numbers for each resonance. These classifications, however, are often subjective and not fully reproducible. This leads to incorrect assignments. In this project, we attempt to rectify assignments for ²⁰⁶Pb by using a machine learning (ML) approach. ML is a process that attempts to learn patterns and make predictions based on the given training data and set of features. Quality ML training requires abundant and diverse training data. The real data is often incomplete and contains errors, so instead, we build synthetic data that mimics the statistical properties of real resonances. For neutron capture reactions, we have found that a realistic distribution of resonance decay widths contributes substantially to the success of the ML algorithm. To accomplish this, we provide our synthetic data with capture widths sampled from a Porter-Thomas distribution, where the degrees of freedom ν is fit from the real data.