Bayesian Model Averaging of Nuclear Mass Models

Abstract

In this thesis we investigate the performance of Bayesian inference and Bayesian model averaging (BMA) applied to two nuclear mass models, the Duflo-Zuker 10 parameter model (DZ10) and the semi-empirical mass formula (SEMF). The DZ10 and SEMF models both have theoretically and experimentally motivated terms but the relative importance of them is less clear. Using Bayesian inference and BMA we have attempted to quantify model uncertainties and improve inference about nuclear masses. To explore the robustness of our BMA analysis we compare the results using different choices of parameter priors, and vary the assumed model discrepancy. The main focus is on the DZ10 model. We employ the Atomic Mass Evaluation from 1983 (AME83) for parameter estimation, then we evaluate the predictive power of the model using the Atomic Mass Evaluation from 2016 (AME16). In an attempt to determine the limits of stability of visible matter we also make a prediction for the neutron drip lines, in the tin isotopic chain (Z=50) using the DZ10 model trained on AME16. The 1-neutron drip line is predicted to neutron number N=123 [95, 125], and the 2-neutron drip line at N=115 [103, 125]. Where the error bar corresponds to a 68% degree of belief.