(555i) Predicting Dielectric Constant of Carbon Dioxide-Water Medium Under Clay Nanoconfinement Using Machine Learning Models

Natural clay such as montmorillonite are potential materials for capturing carbon dioxide (CO₂) due to their large surface area, high porosity, abundant basic sites, excellent thermal and chemical stability and low cost.^[1] It is well accepted that clay nanoconfinements can affect properties of carbon dioxide-water medium such as dielectric constant^[2] and diffusion coefficient.^[3] Prior work reported strong evidence that nanoconfinement enhances water self-dissociation and a qualitative relation exists between strongly varying components of dielectric tensor with enhanced self-dissociation.^[4] Also, the parallel component of the dielectric tensor of water at interfaces increases significantly when approaching surface from the bulk.^[5] An estimation of the parallel and perpendicular components of the dielectric tensor under nanoconfinement is therefore essential for studying reaction involving CO₂. Machine learning algorithms have been tremendously successful at predicting chemical properties of materials or processes of interest.^[6] Supervised learning techniques such as artificial neural network, kernel regression and random forest algorithms have been applied across physical sciences with great successes in many areas. We use these models to predict the parallel and perpendicular component of dielectric tensor of CO₂-H₂O medium inside clay nanoconfinements. In this context, we generate training datasets by performing large numbers of molecular dynamics simulations at a wide range of initial configurations and conditions such as temperature, pressure, density, mole fraction, nanoconfinement width, surface charge and types of interlayer cations. We then calculate the dielectric constant of the medium for each simulation, train a machine learning model using these data, and test on the testing set until optimized parameters of the machine learning model are obtained. The final results are compared with semi-empirical dielectric constant correlations to obtain a parity plot.

References

[1] Krumlins et al., Minerals, 12 (3), 349, 2022.

[2] Bonthuis et al., Phys. Rev. Lett, 107 (16), 166102, 2011.

[3] Chiavazzo et al., Nat. Commun, 5 (1), 3565, 2014.

[4] Munoz-Santiburcio and Marx, Phys. Rev. Lett, 119 (5), 56002, 2017.

[5] Diebenbeck et al., Phys. Rev. Lett, 126 (13), 136803, 2021.

[6] Allen and Tkatchenko, Sci. Adv, 8, 7185, 2022.

Acknowledgement

Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.