(362l) Molecular Simulations and Machine Learning of Materials for Carbon Dioxide Capture, Conversion, and Storage

Climate change resulting from the emission of greenhouse gases especially carbon dioxide has become a major concern in recent years.^[1] Carbon capture technologies can be considered as a viable strategy for reducing CO₂ emissions. Different materials have been used for absorption, adsorption, and reaction in carbon capture technologies. Molecular simulation methods such as molecular dynamics (MD) and density functional theory (DFT) along with machine learning (ML) have shown tremendous potentials to advance various frontiers of carbon capture materials. In this perspective, we reveal how MD, DFT and ML can be implemented to study materials applied for CO₂ capture and conversion. In this work, we have explored the effect of different class of materials for CO₂ capture and conversion which can be divided into two broad categories.

Geological sequestration of CO₂: Clay and limestone-based minerals are present primarily in abundance within the subsurface of the earth. These minerals are reasonable to use in direct air capture (DAC) of CO₂ because they are inexpensive, have huge potential for industrial scale up, and can offer lower development and maintenance cost.

To study CO₂ intercalation and conversion in nanoconfinements we consider special class of minerals such as pyrophyllite, gibbsite, montmorillonite, and beidellite to sorb CO₂ and react with other species in the interlayer spaces or nanopores. We perform MD simulations using non-reactive (CLAYFF)^[2] and reactive (ReaxFF)^[3] force fields to study the intercalations of CO₂ in the interlayers/nanopores of these materials. We also use enhanced sampling metadynamics technique to delineate the free energies of reaction and intercalations^[4-9]. Two important phenomena have been observed under nanoconfinement – drastic change in dielectric of the medium and CO₂ mobility leading to formation of charge stabilized species. In this regard, we predict the dielectric constant of the carbon dioxide – water medium and diffusion coefficient of CO₂ under clay nanoconfinement as a function of temperature, pressure, density, mole fraction, nanoconfinement width, surface charge and types of interlayer cations using ML algorithms such as artificial neural network, kernel regression and random forest.

The second class of minerals which we study herein are impure limestones like wollastonite and forsterite which undergo natural weathering and over time can sequester carbon in the form of carbonates and bicarbonates. We perform reactive (ReaxFF) and ab-initio MD to study the surface leaching of these materials along with the enhanced sampling metadynamics to calculate the free energy of reaction. The subsequent bulk substitution reactions taking place after surface leaching is found out to be the driving factor for carbonate and bicarbonate nucleation and precipitation.^[10]

CO₂ adsorption on metal oxide surface: The adsorption-based capturing of CO₂ process on metal oxide surfaces have gained significant interest over the years.^[11] CO₂ can be used as feedstock for production of valuable chemicals, which in general will also help in curbing excess CO₂ levels in the atmosphere. Metal oxide surfaces play a pivotal role in CO₂ hydrogenation, as they provide additional advantages, such as selectivity and energy efficiency. We consider Gallium (Ga) and Indium (In) based oxide materials to reduce CO₂ to carbon monoxide and methanol using DFT and ab-initio MD. The surface energy of adsorption shows that the Ga-In based oxide alloys are even better substitutes for Ga₂O₃ and In₂O₃ metal oxides.

Another set of materials we consider are the alkaline earth metal oxides – calcium oxide (CaO) and magnesium oxide (MgO) which are used as sorbents at high temperature for CO₂ looping. We perform ReaxFF MD simulations and free energy calculations to obtain the adsorption profile of CO₂ on metal oxide clusters which are in good agreement with literature.^[12]

References

[1] Solomon et al., PNAS, 106 (6), 1704 – 1709, 2009.

[2] Cygan et al., J. Phys. Chem. B, 108, 1255 – 1266, 2004.

[3] Sentfle et al., npj Comput. Mater, 2, 15011, 2016.

[4] Muraleedharan et al., ACS Earth Space Chem, 5 (5), 1006 – 1019, 2021.

[5] Dasgupta et al., Phys. Chem. Chem. Phys, 24 (5), 3322 – 3337, 2022.

[6] Bhamra et al., J. Phys. Chem. C, 125 (33), 18395 – 18408, 2021.

[7] Ho et al., J. Phys. Chem. Lett, 14, 2901 – 2909, 2023.

[8] Dasgupta et al., J. Phys. Chem. Lett, 14, 1693 – 1701, 2023.

[9] Dasgupta et al., Nanoscale, under review, 2023.

[10] Asgar et al., Nanoscale, under review, 2023.

[11] Khdary et al., Catalysts, 12 (3), 300, 2022.

[12] Dou et al., J. Hazard Mater, 183, 759 – 765, 2010.

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.