Current browse context:
stat.ML
Change to browse by:
References & Citations
Nuclear Theory
Title: Statistical aspects of nuclear mass models
(Submitted on 11 Feb 2020 (v1), last revised 7 May 2020 (this version, v3))
Abstract: We study the information content of nuclear masses from the perspective of global models of nuclear binding energies. To this end, we employ a number of statistical methods and diagnostic tools, including Bayesian calibration, Bayesian model averaging, chi-square correlation analysis, principal component analysis, and empirical coverage probability. Using a Bayesian framework, we investigate the structure of the 4-parameter Liquid Drop Model by considering discrepant mass domains for calibration. We then use the chi-square correlation framework to analyze the 14-parameter Skyrme energy density functional calibrated using homogeneous and heterogeneous datasets. We show that a quite dramatic parameter reduction can be achieved in both cases. The advantage of Bayesian model averaging for improving uncertainty quantification is demonstrated. The statistical approaches used are pedagogically described; in this context this work can serve as a guide for future applications.
Submission history
From: Vojtech Kejzlar [view email][v1] Tue, 11 Feb 2020 00:47:22 GMT (2288kb,D)
[v2] Wed, 6 May 2020 14:45:36 GMT (2247kb,D)
[v3] Thu, 7 May 2020 00:39:02 GMT (2247kb,D)
Link back to: arXiv, form interface, contact.