Bayesian model mixing for nuclear ground state properties

Improvements in the predictions for nuclear properties has a direct impact on our understanding of the universe and the performance of nuclear technologies. Due to experimental limitations, these properties have been measured only for some nuclei in the nuclear chart and we must rely on theoretical predictions from different models to fill in the gaps. But such predictions can diverge substantially far from the valley of stability. Therefore, instead of relying on a single theoretical model, it can be a better choice to leverage the knowledge from many existing models and experimental values to get predictions with reliable uncertainty bands. There have been few efforts in this direction using Bayesian model averaging (BMA), but such early attempts mostly relied on combining different energy functionals treated at the single-reference energy density functional (SR-EDF) level. Increasing the predictive power of these methods requires going ‘beyond mean field’ by incorporating correlations effects that cannot be captured at the SR-EDF level. In this study, we will be presenting a Bayesian model mixing approach based on a few complementary variants of multi-reference energy density functional theory (MR-EDF) such as particle number, angular momentum projection or the generator coordinate method. It remains computationally prohibited to perform all these operations simultaneously at the scale of the mass table. However, since they provide corrections to the binding energy in different region of the nuclear chart, they are good candidates for Bayesian model mixing. The objective of this study is to leverage these three models to provide more accurate and precise overall predictions across the nuclear chart.