Data-Driven Ionic Liquid Design for CO₂ Capture: Molecular Structure Optimization and DFT Verification

Abstract

To identify optimal ionic liquids (ILs) for CO2 capture, an efficient computer-aided IL design (CAILD) approach is desired. The traditional CAILD methods usually combine an equation of state with the UNIFAC-IL model to calculate gas solubility, which is computationally expensive and sometimes cannot give quantitatively satisfying results. In this contribution, a new CAILD approach is presented for the optimal design of ILs for CO2 capture, where mathematically simple and reliable data-driven models are applied to predict CO2 solubility. The IL design problem is formulated as a mixed-integer nonlinear programming (MINLP) problem with the objective of maximizing the CO2 solubility of ILs under prespecified conditions. Global optimal solutions are successfully obtained due to model simplicity. Moreover, to prevent misleading results led by poor extrapolability of data-driven models, multiple data-driven models are trained from the same experimental solubility database. These models are then adopted in different batches of the MINLP formulation. For each batch, the optimization problem is solved to generate top IL candidates. Only the ILs that repeatedly appear in different batches are considered as reliable solutions falling into the validity domain of the data-driven models. Such a new strategy can effectively enhance design reliability. The CO2 capture performance of the designed ILs is finally confirmed using density functional theory calculations. The applicability of the proposed method is illustrated in a case study of post-combustion carbon capture.

Reviewing and screening ionic liquids and deep eutectic solvents for effective CO2 capture

“…To screen ILs/DESs, properties are needed, which can be determined experimentally or predicted theoretically. Theoretical models have been developed to predict properties of ILs/DESs qualitatively or quantitively, and the models include Quantitative Structure Property Relationship (QSPR) method ( Eike et al, 2004 ), regular solution theory ( Kilaru et al, 2008 ), Molecular dynamics ( Kerlé et al, 2009 ), Monte Carlo ( Shah and Maginn, 2005 ), group contribution method ( Kim et al, 2005 ; Zhang et al, 2021b ; Zhou et al, 2021 ), Conductor-like Screening Model for Real Solvents (COSMO-RS) method ( Liu et al, 2021 ), statistical associating fluid theory (SAFT)-based equation of state ( Ji et al, 2012 ; Ji and Zhu, 2013 ; Ji et al, 2014 ; Shen et al, 2015 ; Sun et al, 2019 ; Alkhatib et al, 2020 ; Sun, 2020 ), etc. Based on the properties, ILs and DESs are screened with different criteria, for example, thermodynamic CO 2 absorption capacity (CO 2 solubility and Henry’s constant) ( Liu et al, 2021 ) and (Henry’s constant, selectivity, relative polarity, and molar volumes) ( Sumon and Henni, 2011 ), thermodynamic and kinetic performances for CO 2 absorption (CO 2 absorption capacity, viscosity, melting point, and Henry’s constant) ( Farahipour et al, 2016 ; Zhang et al, 2021a ) as well as CO 2 chemical absorption properties (chemical equilibrium constants, Henry’s constants, and reaction enthalpies) ( Moya et al, 2020 ), etc.…”

Rational Design and Screening of Ionic Liquid Absorbents for Simultaneous and Stepwise Separations of SO2 and CO2 from Flue Gas

“…Although true component in the target system has been considered, prediction range and accuracy of UNIFAC are still unsatisfactory. Therefore, a machine learning model that is proven by many researchers − to be suitable for multiple gases can improve prediction performance and obtain better solvents in our future work.…”

Understanding the Electronic Interactions, Vertical Excitation Analysis, and the Photovoltaic Properties of 5-(2-ethylhexyl)-1,3-di(furan-2-yl)-4H-thieno[3,4-c]pyrrole-4,6-dione

“…From the lowest unoccupied molecular orbital (LUMO) isosurface, one can infer that LUMO spreads all over the donor and acceptor backbone including the S atom in the acceptor unit, but N. Visuals from the plots of LUMO+1, LUMO+2 and HOMO-1 show that they are also distributed around the donor and acceptor groups, but HOMO-2 is virtually distributed all over the FTPF monomer. Negative energy terms was obtained for both HOMO and LUMO (-5.7509 eV and -2.3274 eV, respectively) using DFT method at CAM-B3LYP/6-31+G(d,p) levels employed; an indication that all calculation converged appropriately at this level [21,52,[60][61][62]. The calculated HOMO-LUMO energy gap is 3.4256 eV, a relatively small energy gap which shows that FTPF polymer is a good candidate for application as opto-electrical component [22,23] in organic electronic devices like OPV cells.…”