How Accurate Is DFT for Alkanes and Water Adsorbed in Acidic Zeolites?
We assess the accuracy of electronic structure methods for modeling the dispersion-dominated adsorption of methane, ethane, and propane, and the adsorption of hydrogen-bonded water in the Brønsted acidic zeolite chabazite (H–CHA) using density functional theory (DFT) with periodic boundary conditions and a hybrid QM:QM (QM = quantum mechanics) approach. In addition to coupled-cluster (CCSD(T)) reference energies, which agree with available experimental values within chemical accuracy limits of ±4 kJ mol–1, we also provide reference adsorption structures obtained with Møller–Plesset perturbation theory (MP2). DFT mostly overestimates binding compared to CCSD(T), with root-mean-square deviations (RMSDs) ranging from 6–29 kJ mol–1 for alkanes and 5–33 kJ mol–1 for water. QM:QM calculations with hybrid functionals as high level can reduce RMSDs from the CCSD(T) reference. With B3LYP+D4 as a high-level method, the QM:QM energies deviate by 7–12 and 11–21 kJ mol–1 for alkanes and water, respectively, depending on the low-level method. With ωB97M-V as high-level method, the RMSDs are 4–10 and 4–15 kJ mol–1, and with MP2, they are 1–3 and 2–8 kJ mol–1, respectively. When ωB97M-V is used as a high-level method or r2SCAN+D4 as a low-level method, predictions for the protonation state of water can be qualitatively wrong. For DFT modeling of periodic systems, when hybrid functionals are not affordable, we recommend revPBE+D3. For QM:QM with hybrid DFT as high-level method, ωB97M-V:revPBE+D3 is the best combination. Only CCSD(T) single points on hybrid MP2 structures reliably deliver chemically accurate adsorption energies and qualitatively correct adsorption structures.