Abstract
In the context of neutron science, Bayesian inference methods have been recently implemented within the MANTID framework [Monserrat D et al. 2015 J. Phys. Conf. Ser. 663 012009 (2015)]. In this contribution, we highlight the advantages of this software package for robust data analysis and subsequent model selection. To this end, we use the celebrated Rosenbrock function to illustrate its merits and strengths relative to classical fitting algorithms. We also introduce the latest additions implemented in MANTID, with a view to increasing its user friendliness as well as stimulating wider use. These include simulated-annealing schemes to reduce the need for initial guesses, as well as new options for multidimensional fitting.
Highlights
In the context of neutron science, the analysis of Quasielastic Neutron Scattering data is a good case in point, as the measured response almost invariably consists of the cumulative contribution of partially overlapping signals convolved with the instrument resolution in the presence of unwanted features such as backgrounds and/or so-called ‘spurions.’ In this situation, a robust analytical description of a given experimental data set requires the use of several and often complex spectral functions, whose precise number and associated line shapes carry vital information on the nature of the underlying properties of the material under investigation
Using the conceptual framework originally introduced by Sivia and Skilling [1], previous work has focused on the development of a Fitting Algorithm for Bayesian Analysis of DAta (FABADA) [2, 3]
This algorithm uses an adaptive Markov Chain Monte Carlo (MCMC) method to effect Bayesian model selection, and it is applicable to any type of data
Summary
- Neutron imaging data processing using the Mantid framework Federico M. - An approximate empirical Bayesian method for large-scale linear-Gaussian inverse problems Qingping Zhou, Wenqing Liu, Jinglai Li et al. This content was downloaded from IP address 147.83.170.62 on 20/11/2018 at 13:05
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have