CO₂ Capture from Flue Gas: A High-Fidelity Force Field and Machine Learning Framework for Adsorbent Discovery.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Carbon capture (CO2) is crucial for mitigating fossil fuel emissions.
• Adsorbent performance directly impacts carbon capture efficiency and cost.
• Existing CO2 adsorption databases often use inaccurate force fields, limiting machine learning (ML) model reliability.

Purpose of the Study:

• To develop a physically accurate force field and high-fidelity database for CO2 adsorption.
• To introduce novel descriptors for improved ML predictive accuracy in adsorbent discovery.
• To identify high-performing COF/MOF adsorbents for efficient CO2 capture.

Main Methods:

• Developed a van der Waals force field using an Exp-PE potential.
• Constructed a high-fidelity CO2 adsorption database.
• Introduced quadrupole-responsive descriptors for ML models.

Main Results:

• The new force field and database address limitations of traditional Lennard-Jones potentials.
• Quadrupole-responsive descriptors significantly improve ML predictive accuracy.
• Identified COF-50 and COF-364 as high-performing adsorbents with superior working capacities.

Conclusions:

• A physically accurate computational framework enhances ML-driven adsorbent discovery for CO2 capture.
• The developed method overcomes the 'garbage in, garbage out' problem in ML training data.
• This approach accelerates the identification of advanced materials for efficient carbon capture.