Applying machine learning to a Maxwellian-averaged n-capture cross section for improved estimates

Neutron-capture cross sections are relevant to many nuclear physics fields. These fields depend on the accuracy of both the existing measurements, as well as estimated values for unmeasured nuclei. For many applications the most relevant nuclear quantities are unmeasured, and will remain so for some time. The result is almost total reliance on estimates, with varying levels of success due to assumptions that, while necessary, might not be supported by the physics. As such, new methods are always of interest. Current methods include diverse theoretical formulations of mass, level density, and neutron separation energy. These often focus on a single nuclear quantity or bulk property (such as mass model), which heavily weights that quantity's importance by default. This work attempts to improve estimates of neutron capture cross sections, especially where few or no measurements exist by applying the predictive capability of machine learning (ML). As the neutron fluence in many applications can be treated as a weighted sum of Maxwellian spectra, we use machine learning to develop a regression model for the temperature dependence of the Maxwellian averaged-cross section, coupled to additional physical quantities where known. The methods and early results will be discussed.