A modified Generalized Least Squares method for large scale nuclear data evaluation

详细信息    查看全文

• 作者：Georg Schnabel^a ; ^b ; ; Helmut Leeb^b
• 关键词：Nuclear data evaluation ; Generalized Least Squares ; Bayesian statistics
• 刊名：Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
• 出版年：2017
• 出版时间：1 January 2017
• 年：2017
• 卷：841
• 期：Complete
• 页码：87-96
• 全文大小：794 K

文摘

Nuclear data evaluation aims to provide estimates and uncertainties in the form of covariance matrices of cross sections and related quantities. Many practitioners use the Generalized Least Squares (GLS) formulas to combine experimental data and results of model calculations in order to determine reliable estimates and covariance matrices. A prerequisite to apply the GLS formulas is the construction of a prior covariance matrix for the observables from a set of model calculations. Modern nuclear model codes are able to provide predictions for a large number of observables. However, the inclusion of all observables may lead to a prior covariance matrix of intractable size. Therefore, we introduce mathematically equivalent versions of the GLS formulas to avoid the construction of the prior covariance matrix. Experimental data can be incrementally incorporated into the evaluation process, hence there is no upper limit on their amount. We demonstrate the modified GLS method in a tentative evaluation involving about three million observables using the code TALYS. The revised scheme is well suited as building block of a database application providing evaluated nuclear data. Updating with new experimental data is feasible and users can query estimates and correlations of arbitrary subsets of the observables stored in the database.