Discrepancy quantification between experimental and simulated data of CO2 adsorption isotherm using hierarchical Bayesian estimation

To quantitatively analyze the inconsistencies commonly observed between experimental and simulated adsorption isotherms, parameter estimation of adsorption isotherm models was conducted by hierarchical Bayesian estimation with parameter uncertainties being quantified as probability distributions. The estimation method was implemented using Markov Chain Monte Carlo (MCMC) to analyze multiple data sets obtained from different sources, including a publicly available database. To describe the discrepancies of experimental and simulated adsorption data, the simulation data was set as the reference to which experimental measurements were compared. We applied the proposed approach to analyze CO2 adsorption isotherms that are measured and simulated on zeolite 13X and MIL-101(Cr). In these case studies, the discrepancy of CO2 adsorption isotherm was successfully quantified between experimental measurements and predictions given by molecular simulations using Grand Canonical Monte Carlo (GCMC), where uncertainties were quantified as probability distributions. Furthermore, experimental data sets that agree well with the GCMC simulation have been identified, providing insights into experimental and measurement methods as well as choosing the right assumptions in the molecular simulation.

Automated summary

Quantitative analysis of inconsistencies between experimental and simulated adsorption isotherms was conducted using Bayesian estimation with parameter uncertainties as probability distributions. The method utilized Markov Chain Monte Carlo to analyze multiple datasets, including a publicly available database. By setting simulation data as the reference, the discrepancies between experimental measurements and molecular simulation predictions were quantified. Applying this approach to CO2 adsorption isotherms on zeolite 13X and MIL-101(Cr), the differences were successfully quantified and experimental data sets that aligned with simulation were identified.