Chemical Accuracy in Computational Prediction of PFAS-Cyclodextrin Binding: A Multilevel SQM/QM Study With...

Achieving chemical accuracy in the prediction of host-guest binding free energies remains a central challenge in computational chemistry, particularly for flexible supramolecular systems in aqueous environments. Here, we demonstrate that a multilevel CREST/CENSO SQM/QM workflow based on Grimme's r²SCAN-3c methodology can predict β-cyclodextrin binding free energies for perfluoroalkyl carboxylic acids (PF4-PF8) with near-quantitative agreement with experiment. Calculated binding free energies were validated against isothermal titration calorimetry (ITC) measurements, yielding a mean absolute error of 0.86 kcal mol^-1 and thereby achieving chemical accuracy. Both computational and experimental results reveal a pronounced chain-length dependence, with binding affinity decreasing toward shorter PFAS. Energy decomposition analysis shows that selectivity is governed primarily by electronic interactions, dominated by dispersion along the fluorocarbon chain. Noncovalent interaction analysis further identifies carboxylate hydrogen bonding at the primary rim of β-cyclodextrin as a key anchoring interaction. These results establish an efficient and robust quantum-mechanical framework for predicting PFAS-cyclodextrin complexation and provide a rational basis for the computational screening and design of cyclodextrin-based sorbents for PFAS remediation.