Modeling hydroxylated nanosilica: Testing the performance of ReaxFF and FFSiOH force fields

Abstract

We analyze the performance of the FFSiOH force field and two parameterisations of the ReaxFF force field for modeling hydroxylated nanoscale silica (SiO). Such nanosystems are fundamental in numerous aspects of geochemistry and astrochemistry and also play a key role during the hydrothermal synthesis of technologically important nanoporous silicas (e.g., catalysts, absorbents, and coatings). We consider four aspects: structure, relative energies, vibrational spectra, and hydroxylation energies, and compare the results with those from density functional calculations employing a newly defined dataset (HND: Hydroxylated Nanosilica Dataset). The HND consists of three sets of (SiO)(HO) nanoparticles (NPs), each with a different degree of hydroxylation and each containing between 23 and 26 distinct isomers and conformers. We also make all HND reference data openly available. We further consider hydroxylated silica NPs of composition (SiO)(HO) with M = 4, 8, 16, and 24 and infinite surface slabs of amorphous silica, both with variable hydroxylation. For energetics, both ReaxFF and FFSiOH perform well for NPs with an intermediate degree of hydroxylation. For increased hydroxylation, the performance of FFSiOH begins to significantly decline. Conversely, for the lower degree of hydroxylation both parameterisations of ReaxFF do not perform well. For vibrational frequencies, FFSiOH performs particularly well and significantly better than ReaxFF. This feature also opens the door to inexpensively calculating Gibbs free energies of the hydroxylated nanosilica systems in order to efficiently correct density functional theory calculated electronic energies. We also show how some small changes to FFSiOH could improve its performance for higher degrees of hydroxylation.

Computing Free Energies of Hydroxylated Silica Nanoclusters: Forcefield versus Density Functional Calculations

“…However, we find that nanoclusters with lower (25%) hydroxylation are those for which FFSiOH provides the best match with the DFT-calculated G corrections. This result is in line with the capability of FFSiOH to better estimate relative Uelec values (with respect to DFT values) for this degree of hydroxylation [23].…”

“…However, we find that nanoclusters with lower (25%) hydroxylation are those for which FFSiOH provides the best match with the DFT-calculated G corrections. This result is in line with the capability of FFSiOH to better estimate relative U elec values (with respect to DFT values) for this degree of hydroxylation [23]. and Uvib is very close to DFT values, the total correction to U follows the same trends as the ZPE, and the differences among DFT and FFSiOH values are almost the same as for ZPE.…”

“…The average absolute differences between FFSiOH results and DFT for the rotational and translational entropies are 0.05 kJ/mol per SiO2 unit and 0.03 kJ/mol prer SiO2 unit respectively. These results are not surprising, since Srot and Strans depend only on the optimized geometric structure of the respective nanocluster, which is very similar when calculated with DFT and FFSiOH (see Reference [23]).…”

“…Specifically, we examine the performance of FFSiOH [22], which is a polarisable ionic FF, that was parametrized with respect to the results of DFT calculations of silica bulk phases and hydroxylated silica surfaces. We have recently shown that FFSiOH can provide results in good agreement with DFT calculations (with respect to structure, relative energies, and vibrational frequencies) for a range of hydroxylated silica nanoclusters of different sizes and degrees of hydroxylation [23]. Following this work, we have selected hydroxylated silica nanoclusters as a suitable test case for the present study.…”

“…The second set (B) of structures consist of five isomers with (SiO 2 ) 16 ·(H 2 O) 4 stoichiometry (see Figure 2). These structures are selected from the dataset in Reference [23] as representative low energy minima, as found by classical IP-based global optimization searches, followed by DFT refinement (see details in References [27][28][29][30]. The relative energies of the structures calculated at a DFT level using the B3LYP functional are noted in Figure 2.…”