Y.F. Meng; Y.B. Zhu; B. Yang; L.C. Zhou; S.C. Zhang; Y.R. Wang; X.S. Meng; B. Ge; L.B. Mao; S.H. Yu
Matter 8 (2025) 102203
Direct air capture (DAC) of CO2 is necessary for climate change mitigation, but it faces challenges from low
CO2 concentrations and competition from water vapor. Metal-organic frameworks (MOFs) hold significant
promise for DAC owing to their high surface area and adsorption-based capture processes. However, iden
tifying optimal MOFs is hindered by structural complexity and vast chemical diversity. Here, we introduced a
machine learning force field (MLFF) tailored for CO2 and H2O interactions in MOFs by fine-tuning a foundation
model. To address smoothing issues and catastrophic forgetting, we curated the GoldDAC dataset and intro
duced a continual learning scheme. We further developed DAC-SIM, a molecular simulation package inte
grated with MLFF, including a Widom insertion. Then, we screened an extensive MOF database, uncovering
high-performing MOFs and identifying chemical features for DAC applications. This approach overcomes
prior limitations in describing MOF-CO2 and MOF-H2O interactions, providing a scalable and accurate frame
work for DAC research of porous materials.