Estimation of Maxwellian averaged cross-sections with machine learning methods

Neutron capture cross-sections are crucial for many applications in nuclear physics, stellar astrophysics, and nuclear engineering are but acquiring the cross-section data can be expensive and difficult to perform. In particular, data for unstable (or even low natural abundance) nuclei are costly to obtain, limiting the amount of available experimental data. Thus, capture cross-section data for these nuclei are dependent on numeric estimates with varying degrees of success. In this work, machine learning methods are employed to develop a low order regression model for the temperature dependence of Maxwellian averaged cross-sections (MACS). We then use a neural network to learn the isotopic dependence of the regression model features from nuclei in a curated training set. The resulting model can be used to predict the temperature dependence of the MACS for all nuclei. Since the training set necessarily consists of data from stable nuclei, it is expected that the model works best near the valley of stability.