CheKiPEUQ Intro 1: Bayesian Parameter Estimation Considering Uncertainty or Error from both Experiments and Theory

• Published in: ChemCatChem 2020-11, Vol.12 (21), p.5385-5400
• Main Authors: Savara, Aditya, Walker, Eric A.
• Format: Article
• Language: English
• Subjects: Bayesian analysis; Bayesian Parameter Estimation; CheKiPEUQ; Data Science; Fitting; INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CHEMISTRY; Parameter estimation; Parameter uncertainty; Software; Uncertainty
• ISSN: 1867-3880; 1867-3899
• DOI: 10.1002/cctc.202000953

A common goal is extraction of physico‐chemical parameter values such as pre‐exponentials and activation energies from experiment. Ever increasing knowledge from experiments and computations is enabling semi‐quantitative prior predictions of such values. When prior knowledge of physically realistic ranges is available, a method named Bayesian parameter estimation (BPE) enables more physically realistic parameter estimation relative to unsophisticated fitting by seeking the most probable value when considering together the uncertainties from prior knowledge, experimental data, and approximations in the model. An impediment to widespread use of BPE is a lack of understanding, training, and user‐friendly software. Along with this invited publication, a general software package for BPE is being released that is user‐friendly and that does not require understanding of the math behind the methodology. Two previously unpublished catalysis science examples are provided along with considerations and guidelines for successful application of BPE. Following this work, BPE can become more widespread to enable extraction of physically meaningful parameter values. Experiment and theory sitting in a tree… Extraction of physically realistic parameters from experiments, such as pre‐exponentials and activation energies, can be challenging. In many cases, fitting models results in physically unrealistic values. Bayesian parameter estimation considers the uncertainties from both experiment and theory; ‘getting them together’ to output the estimate with maximum probability. A software package is released with examples such that others may apply the same methodology to other systems.