Multimodal reconstruction of TbCo thin-film structure with Bayesian analysis of polarized neutron reflectivity

Abstract

Bayesian analysis has been applied to polarized neutron reflectivity data. Reflectivity data from a magnetic TbCo thin-film structure were studied using a combination of a Monte Carlo Markov-chain algorithm, likelihood estimation and error modeling. By utilizing Bayesian analysis, it was possible to investigate the uniqueness of the solution beyond reconstructing the magnetic and structure parameters. The expedience of this approach has been demonstrated, as several probable reconstructions were found (the multimodality case) concerning the isotopic composition of the surface cover layer. Such multimodal reconstruction emphasizes the importance of rigorous data analysis instead of the direct data fitting approach, especially in the case of poor statistically conditioned data typical for neutron reflectivity experiments. This article presents details of the analysis and a discussion of multimodality.