SCAMs@bristol

statistical chemical
analysis methods
@ the university of bristol

Jekyll — Theme from BDHU

Bayesian model selection

Frequently, in the analysis of chemical data, there is more than one model that could describe the phenomena being observed or we want to understand how many parameters in a given model can be allowed to vary without “fitting an elephant”.1 We use Bayesian model selection, to perform a rationale comparison between analytical models with different numbers of parameters to obtain a greater understanding of the analysis processes at hand.

We have used Bayesian model selection to develop a framework for the interpretation of reflectometry analysis, ensuring that the maximum information density is obtained from a given analysis.2 This approach is now being expanded to give an analytical understanding of the minimum resolvable lengthscales from reflectometry measurements3 and has been applied to rationally compare different analysis models to describe space-change layers in solid-electrolyte grain boundaries.4

References

  1. J. Mayer, K. Khairy, & J. Howard. Am. J. Phys., 78(6), 648-649, 2010. DOI: 10.1119/1.3254017. ↩

  2. A. R. McCluskey, T. Arnold, J. F. K. Cooper, & T. Snow. Mach. Learn.: Sci. Technol., 1(3), 035002, 2020. DOI: 10.1088/2632-2153/ab94c4. ↩

  3. N. Shiaelis, L. A. Clifton, & A. R. McCluskey. In Preparation. ↩

  4. J. M. Dean, S. W. Coles, W. R. Saunders, A. R. McCluskey, M. J. Wolf, A. B. Walker, & B. J. Morgan. Phys. Rev. Lett., 127(13), 135502, 2021. DOI: 10.1103/PhysRevLett.127.135502. ↩