Computational Screening of Physical Solvents for CO₂ Pre-combustion Capture

Abstract

A computational scheme was used to screen physical solvents for CO 2 pre-combustion capture by integrating the commercial NIST database, an in-house computational database, cheminformatics, and molecular modeling. A commercially available screened hydrophobic solvent, diethyl sebacate, was identified from the screening with favorable physical properties and promising absorption performance. The promising performance to use diethyl sebacate in CO 2 pre-combustion capture has also been confirmed from experiments. Water loading in diethyl sebacate is very low, and therefore, water is kept with H 2 in the gas stream. The favorable CO 2 interaction with diethyl sebacate and the intermediate solvent free volume fraction leads to both high CO 2 solubility and high CO 2 /H 2 solubility selectivity in diethyl sebacate. An in-house NETL computational database was built to characterize CO 2 , H 2 , N 2 , and H 2 O interactions with 202 different chemical functional groups. It was found that 13% of the functional groups belong to the strong interaction category with the CO 2 interaction energy between −15 and −21 kJ/mol; 62% of the functional groups interact intermediately with CO 2 (−8 to −15 kJ/mol). The remaining 25% of functional groups interact weakly with CO 2 (below −8 kJ/mol). In addition, calculations show that CO 2 interactions with the functional groups are stronger than N 2 and H 2 interactions but are weaker than H 2 O interactions. The CO 2 and H 2 O interactions with the same functional groups exhibit a very strong linear positive correlation coefficient of 0.92. The relationship between CO 2 and H 2 gas solubilities and solvent fractional free volume (FFV) has been systematically studied for seven solvents at 298.2 K. A skewed bell-shaped relation was obtained between CO 2 solubility and solvent FFV. When an organic compound has a density approximately 10% lower than its density at 298.2 K and 1 bar, it exhibits the highest CO 2 loading at that specific solvent density and FFV. Note that the solvent densities were varied using simulations, which are difficult to be obtained from the experiment. In contrast, H 2 solubility results exhibit an almost perfect positive linear correlation with the solvent FFV. The theoretical maximum and minimum physical CO 2 solubilities in any organic compound at 298.2 K were estimated to be 11 and 0.4 mol/MPa L, respectively. An examination of 182 experimental CO 2 physical solubility data and 29 simulated CO 2 physical solubilities shows that all the CO 2 physical solubility data are within the maximum and minimum with only a few exceptions. Finally, simulations suggest that in order to develop physical solvents with both high CO 2 solubility and high CO 2 / H 2 solubility selectivity, the solvents should contain functional groups which are available to interact strongly with CO 2 while minimizing FFV.

Interpretable Machine Learning Model for Predicting Interaction Energies between Dimethyl Sulfide and Potential Absorbing Solvents

“…Driven by both the expanded chemical database and the advanced algorithms, machine learning (ML) has been finding powerful functions and wide applications in designing molecules and infrastructure for broad engineering areas, including chemistry, material, biology, medicine, , environment, and electronics. , ML algorithms have been used to aid solvent discovery by predicting the solubilities of various species, diffusion coefficients, and reaction paths . Quantitative structure–activity relationship (QSAR) models were explored using extensive training data sets and descriptors. , Using sufficient solubility data, Orlov et al and Shi et al have successfully used ML methods to achieve solubility prediction and solvent identification for the absorption of H 2 S and CO 2 , respectively. However, the valid solubility data are still lacking for most of the environmentally unfriendly sulfides including volatile thioether compounds.…”

“…24 Quantitative structure−activity relationship (QSAR) models were explored using extensive training data sets and descriptors. 25,26 Using sufficient solubility data, Orlov et al 27 and Shi et al 28 have successfully used ML methods to achieve solubility prediction and solvent identification for the absorption of H 2 S and CO 2 , respectively. However, the valid solubility data are still lacking for most of the environmentally unfriendly sulfides including volatile thioether compounds.…”

Simulation and Experiment of CO2 Philicity and Separation in Carbonate-Rich Polymers

“…The interactions between CO 2 and polar groups have been investigated using computational chemistry methods. ,− Binding energy between the polar groups and CO 2 can be calculated to elucidate their interactions, and it can be used to screen functional groups to better design sorbents and membranes and to correlate the molecular structure of the polar groups with the membrane performance. For example, the functionalization of polydimethylsiloxane (PDMS) with amidoxime groups enhanced CO 2 permeability and CO 2 /N 2 selectivity, consistent with the enhanced binding between CO 2 and amidoxime groups …”

Virtual Issue on Carbon Dioxide: Physical Chemistry That Impacts Its Capture, Sequestration, and Conversion

“…Shi and co-workers used high throughput computational screening methods to search for physical solvents with good CO 2 capture properties. 6 Their work identified diethyl sebacate as having favorable physical properties and promising absorption performance. These results were experimentally confirmed by the researchers.…”

“…An alternative is to use solvents that have a high physical solubility for CO 2 , and thus a lower regeneration energy penalty. Shi and co-workers used high throughput computational screening methods to search for physical solvents with good CO 2 capture properties . Their work identified diethyl sebacate as having favorable physical properties and promising absorption performance.…”