Understanding Water-Zeolite Interactions: On the Accuracy of Density Functionals

Water is ubiquitous in zeolite catalysis, and electronic structure calculations play a crucial role in arriving at an atomistic understanding of water-zeolite interactions. However, a critical evaluation of the performance of different electronic structure methods in describing the interactions between water and zeolites is still missing. Here, we model the adsorption of one water molecule in all-silica chabazite (CHA) and of one and two water molecules in the acidic zeolite SSZ-13 using different electronic structure methods, which include 11 density functional theory (DFT)- based methods and two post Hartree-Fock (HF) methods, namely, the random phase approximation (RPA) and second-order Moller-Plesset (MP2) perturbation theory. We find that all DFT functionals lead to similar structures as long as water is strongly coordinated to the adsorption site, but adsorption energies vary in a range of 50 kJ/mol between the used methods. Subsequently, we use ab initio molecular dynamics calculations to show that all methods reproduce the experimentally observed hydrophobicity of purely siliceous zeolites. Comparing DFT energetics with RPA and MP2 calculations shows that PBE and revPBED3 adsorption energies show the best agreement with RPA, while BEEF-vdW agrees the best with MP2 results. At the same time, the performance of PBE functional without any dispersion correction is less consistent with respect to different adsorption sites (BAS, LAS, or the zeolite wall of all-silica CHA) and the BEEF-vdW functional fails to reproduce relative stabilities of the protonation sites. For the adsorption of two water molecules, most methods agree on the formation of a protonated water dimer, and only vdW-DF, vdW-DF2, and BEEF-vdW prefer the formation of a neutral complex. Based on these results, we suggest using the revPBE-D3 functional model water adsorption in purely siliceous or protonated zeolites since it can correctly capture covalent and dispersion interactions, is computationally efficient, correctly predicts the formation of a positively charged water dimer, and is able to closely reproduce adsorption energies calculated at the RPA or MP2 level of theory.