Novel Bayesian probability method in predictions of nuclear masses
Jiaze Xie, Keping Wang, Chuan Wang, Wanqing Gao, Min Ju, Jian Liu
Abstract
Machine learning methods have recently gained interest in the complexity of nuclear mass prediction. According to previous studies, we propose a continuous Bayesian probability (CBP) classifier combined with Bayesian model averaging (BMA) to refine the descriptions of sophisticated nuclear mass models. In the CBP method, the nuclear masses are considered continuous variables to generate prior and conditional probability density functions (PDFs), and the posterior PDFs are determined by the Bayesian formula. The global optimizations and the extrapolating analyses exhibit impressive improvements. Additionally, we employ the BMA method to consider the predictions of different models by assigning weights based on their predictive effectiveness for seven benchmark nuclei. By presenting predictions of the neutron drip line, we assess the reliability of the refinements of the BMA method. The methods proposed in this paper provide an effective way of predicting the nuclear mass in unknown regions and can be applied to other model-based extrapolations of nuclear properties.