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
Møller–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 revPBE-D3 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.

水在沸石催化中无处不在，而电子结构计算对于获取水-沸石相互作用的原子级认知具有至关重要的作用。然而，当前仍缺乏针对不同电子结构方法描述水与沸石相互作用性能的系统性批判性评估。在此，我们借助多种电子结构方法，对单水分子在全硅菱沸石（CHA）中的吸附，以及单、双水分子在酸性沸石SSZ-13中的吸附过程开展建模；所选用的方法涵盖11种基于密度泛函理论（DFT）的泛函，以及2种后哈特里-福克（HF）方法，即随机相近似（RPA）与二阶莫勒-普莱塞特（MP2）微扰理论。研究发现，当水分子与吸附位点保持强配位作用时，所有DFT泛函预测的结构均较为相近，但不同方法给出的吸附能差值可达50 kJ/mol。随后，我们通过从头算分子动力学计算证实，所有方法均能复现实验观测到的纯硅沸石疏水性。将DFT的能量计算结果与RPA、MP2方法进行对比后可知，PBE与revPBE-D3泛函的吸附能与RPA结果契合度最高，而BEEF–vdW泛函则与MP2结果匹配最佳。与此同时，未添加任何色散校正的PBE泛函，其性能在不同吸附位点（布仑斯惕酸性位点（BAS）、路易斯酸性位点（LAS）以及全硅CHA的沸石孔壁）间的一致性较差；且BEEF–vdW泛函无法复现质子化位点的相对稳定性。对于双水分子吸附过程，绝大多数方法均预测会形成质子化水二聚体，仅vdW-DF、vdW-DF2与BEEF–vdW三种泛函更倾向于生成中性络合物。基于上述研究结果，我们建议在模拟纯硅或质子化沸石中的水吸附过程时采用revPBE-D3泛函：该泛函既能准确捕捉共价作用与色散相互作用，又具备较高的计算效率，可正确预测带正电的水二聚体的形成，且能够高度复现RPA或MP2理论层级下计算得到的吸附能。