Performance dei metodi computazionali GFN su nanoclusters di silice amorfa idrossilata

A new series of low-cost methods developed by Grimme et al. has been recently published with the acronym “GFN”, for the reasonable prediction of Geometries, vibrational Frequencies and Noncovalent interactions. xTB-GFN2, xTB-GFN1 and xTB-GFN0 are semiempirical quantum mechanical methods, while GFN-FF follow a force-field approach. The main feature of this methodologies is the universal applicability they want to reach, since the parametrization covers all elements up to Z=86 (radon). The aim of my Master Thesis is to evaluate the performance of these parametrized methods over amorphous silica nanoparticles. The dataset employed includes over hundred structures, with different degrees of hydroxylation. Five aspects are considered for the evaluation: structural features, relative energies, gyration radii, hydroxylation energies and vibrational frequencies. I compare GFN-results with those from PBEsol0-3c DFT functional, designed to achieve high accuracy in the description of solids. xTB-GFN1 misses the description of the fundamental Si-OH group, which are involved in all intra- and inter-molecular interactions, including hydrogen bonds, whereas the other GFN methods provide at least a reasonable description for the main properties. xTB-GFN2 shows the closest behaviour to the reference PBEsol0-3c, but it leads to an excessive compacting of the structures. GFN-FF structural and energetic behaviour is unexpectedly accurate and its general agreement with the PBEsol0-3c-optimized structures is very promising.