Creation and Characterization of Periodic Slab Models of Amorphous Silica
M.P. Klein (TU Delft - Applied Sciences)
A.A. Kolganov – Mentor (TU Delft - Applied Sciences)
Evgeny A. Pidko – Mentor (TU Delft - Applied Sciences)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Amorphous silica is a widely used material with many applications. Industrially, it has found common use as a catalyst support or adsorbent. As it is an amorphous material, the lack of long-range periodicity makes it difficult to reason about what its surface looks like. As a consequence, when we construct atomic models, it is difficult to determine whether they are representative. Furthermore, this difficulty extends to the active sites, as there are many different possibilities with different local topologies and varying amounts of strain. This makes the computational modeling of the material a challenge to modern chemistry.
This work aims to generate periodic models of amorphous silica of varying roughness and strain and use the topological features of the created models as descriptors for strain. To generate these models, classical molecular dynamics is used to generate bulks and equilibrate cleaved surfaces using a randomly generated stochastic Fourier expansion. DFT is then used to optimize the geometry of the resulting surfaces and their saturated counterparts. The calculated energies are compared to those of the most relaxed states of the substituents the surfaces are composed of.
It was found that the method of cleaving surfaces resulted in varying roughness after re-equilibration and that roughness has a correlation with strain. Varying the roughness had the greatest effect on the amount of strained topological features in the model. Algorithmically saturating models showed that strain is generally decreased through the addition of water and strain is most effectively decreased through the removal of two-membered rings on the surface.
The main result of this study is that, using purely topological features, the strain of a model can be predicted using a multivariate linear regression. Using the coordination of O atoms, average bond lengths, and angles as descriptors, multivariate linear regression was found to result in an R² of 0.925.