(704e) Computational Design of Physical Solvents for Gas Separation

Recently, we have developed a computational approach ^[1,2] through the integration of the NIST experimental database, in-house computational database, in-house Monte Carlo and molecular dynamics simulations to screen physical solvents for CO₂ pre-combustion capture from the fuel gas, which mainly contains CO₂, H₂, and H₂O. About 23,000 organic compounds contained in the NIST database were screened and one especially promising compound was identified. Both the lab scale (solvent volume ~ 1 liter) and the recent preliminary bench scale (solvent volume ~ 60 liters) experimental data show that this solvent performs better than the commercial Selexol solvent for CO₂ pre-combustion capture application.

In this talk, we will discuss our recent computational driven efforts to design physical solvents with large CO₂ solubility along with other favorable properties, such as high CO₂/H₂ solubility selectivity, high hydrophobicity, low viscosity, low vapor pressure, and low tendency to foam. From the simple group contribution method, several functional groups with a large CO₂ solubility contribution were identified. Both experiments and simulations show that when the number of -CH₂- groups increase, CO₂ solubility decreases. This is due to the following two reasons. When the number of -CH₂- groups increase, the functional group concentration decreases. This leads to a smaller number of functional groups in the same volume of solvent, which in turn decreases CO₂ solubility. In addition, our simulations show that when the number of -CH₂- groups increase the solvent free volume fraction will decrease, which also decreases CO₂ loading. Furthermore, both our simulations and experimental data show that simply adding more functional groups to a solvent molecule could significantly decrease instead of increasing CO₂ solubility. When more functional groups were added to a solvent molecule, the solvent free volume fraction significantly decreases partly due to stronger solvent-solvent interactions, which in turn decrease CO₂ solubility.

Finally, our most recent results from computational screening of the NIST solvent database will be presented, such as H₂O, H₂S, H₂, and CO₂ interactions with all functional groups. It was found that H₂O interactions with all functional groups are stronger than CO₂ interactions with the same functional groups, and the two sets of interactions show a strong linear positive correlation coefficient of 0.88. New simulated and experimental CO₂ solubility data will also be compared with the results obtained from our previous machine leaning model ^[2].

References:

1. Wei Shi, Robert L. Thompson, Megan K. Macala, Kevin Resnik, Janice A. Steckel, Nicholas S. Siefert, and David P. Hopkinson, “Molecular Simulations of CO₂ and H₂ Solubility, CO₂ Diffusivity, and Solvent Viscosity at 298 K for 27 Commercially Available Physical Solvents”, Invited paper to the J. Chem. Eng. Data, DOI: 10.1021/acs.jced.8b01228, 2019
2. Wei Shi, Robert L. Thompson, Megan K. Macala, S. Tiwari, Kevin Resnik, Janice A. Steckel, Nicholas Siefert, David P. Hopkinson, “Integration of Data Mining, Molecular Modeling, and Machine Learning to Screen Physical Solvents for Gas Separation”, 2018 AIChE Annual meeting, David L. Lawrence Convention Center, Pittsburgh, PA, Oct. 28-Nov.2, 2018