Hierarchical Bayesian Change-Point Models for Chemical Properties Inference

• Activity Number: 448 - Statistics Impacting Challenges Within Academia, Industry, and Government
• Type: Invited
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Amanda Koepke* and Felix Jimenez and Kenneth Kroenlein and Chris Muzny
• Companies: National Institute of Standards and Technology and University of Colorado, NIST and National Institute of Standards and Technology and National Institute of Standards and Technology
• Keywords: Thermodynamics; Bayesian; change-point; random effects model; Hamiltonian Monte Carlo; measurement uncertainty

Abstract

While extensive historical data is often available to inform estimates of chemical properties, current methods for combining these disparate data are unclear. We seek to provide rigorous, defensible statistical methods that utilize all available data to inform both the estimate of the quantity of interest and its uncertainty. We estimate the triple point temperature for a well-studied chemical, naphthalene. Some of the literature provides information on temperature and pressure measurements, which can be understood using a change-point model with one model for vapor pressure (liquid phase) and a different model for sublimation pressure (crystal phase). The change-point occurs at the triple point temperature. Other works provide direct estimates and uncertainties for the triple point temperature. We develop a hierarchical Bayesian approach to estimate triple point temperature using both types of data. We implement model selection to choose between several different vapor pressure and sublimation pressure models from the literature and arrive at a comprehensive estimate of the uncertainty for the triple point temperature of naphthalene that incorporates all available information.