(675c) Quantification of Variations in Experimental and Molecular Simulation Data for CO2 Adsorption Isotherm By Hierarchical Bayesian Estimation.

Adsorptive separation is considered one of the promising methods to recover CO₂ in flue gas from various sources. For designing adsorptive separation processes, both equilibrium and kinetic data is required, and among them obtaining reliable adsorption equilibrium data is critical. To date, many researchers have conducted experimental and computational studies on CO₂adsorption using various porous materials including zeolites, porous carbons, mesoporous silicas, and metal-organic frameworks (MOFs) to identify efficient adsorbents. Such studies have generated a large amount of CO₂ adsorption equilibrium data on various adsorbents. For using such data effectively, open databases such as the one by the National Institute of Standards and Technology (NIST)^[1] can be useful resources for material screening and process design.

However, there still exist many challenges in utilizing adsorption databases for material screening and process design. Adsorption isotherm data obtained by different researchers often show substantial variations even for the same adsorption system that employ the same adsorbent and adsorbate^[2]. Such variations can be explained by many different reasons including the measurement methods, as well as imperfect adsorbent materials caused by defects in the crystals and insufficient activation. Moreover, experimental and computational isotherm data often show inconsistency, which can be due to assumptions and approximations in molecular simulation. Because of these variations and inconsistency, it is difficult to determine the most reliable adsorption isotherm model with one set of equilibrium parameters to develop a robust develop model.

In this study, the parameters in isotherm models were estimated using hierarchal Bayesian estimation, and their uncertainty was quantified as probability distributions. Two case studies of CO₂ adsorption on MIL-101(Cr) and zeolite 13X were considered. Hierarchical Bayesian estimation was implemented using Markov Chain Monte Carlo (MCMC) method to analyze multiple datasets obtained by different resources. Among these data sets, Grand Canonical Monte Carlo (GCMC) simulation data was used as a reference which assumes ideal conditions for adsorption. The proposed approach quantifies the deviation of each researcher i from the reference GCMC data using the coefficient R_i, as shown in Figure 1. We demonstrate that the proposed approach can be a powerful tool in resolving differences among data sets from different researchers, as well as providing insights into the deviations of experimental measurements from computationally generated data.

Reference

[1] NIST adsorption https://adsorption.nist.gov/isodb/index.php#home

[2] Park, J., Howe, J. D., & Sholl, D. S. (2017). Chemistry of Materials, 29(24), 10487–10495.