Restructuring of MFI Framework Zeolite Models and Their Associated Artifacts in Density Functional Theory Calculations

This
study compares and evaluates multiple orthorhombic silicalite
MFI framework structures using periodic density functional theory
(DFT) calculations implemented with a wide range of exchange–correlation
functionals and dispersion-correction schemes. Optimization of the
structure available from the International Zeolite Association (IZA)
yields only metastable forms, which restructure to arrangements 18–156
kJ mol–1 lower in energy (55 kJ mol–1 on average) through annealing and adsorption/desorption processes
without altering their connectivity. These restructuring events can
occur unintentionally during DFT studies of adsorptive and catalytic
properties, leading to very large artifacts in DFT-predicted adsorption,
reaction, and activation energies. Pre-annealing the IZA structure
prevents restructuring and these artifacts but forms MFI structures
which do not conform to the Pnma spacegroup symmetry
and have significantly perturbed sinusoidal and straight channel geometries.
These issues persist across a wide range of exchange–correlation
functionals, including common choices such as the Perdew–Burke–Ernzerhof
and Bayesian error estimation functionals, and dispersion-correction
schemes such as the D3 method. Direct optimization of structures generated
from the work of van Koningsveld et al. and Olson et al., in contrast,
yields structures that are extremely similar across all functionals,
restructure less often during annealing, and have smaller energy shifts
when they do restructure (5 kJ mol–1, on average).
Optimizing the unit cell parameters of these structures without constraining
atoms or the unit cell shape also yields more stable structures, though
often with unit cell parameters that do not closely match structures
found experimentally. Annealing of other commonly studied zeolites
(BEA, CHA, and LTA) does not yield structures with energy decreases
or structural changes as significant as those for MFI. This study
thus illuminates a potential source of significant error for DFT studies
of MFI and provides evidence-based solutions for a variety of DFT
methods.