Modelování NMR vlastností zeolitů pomocí strojového učení a ab initio výpočtů

Willimetz, Daniel

This study introduces a double machine learning approach, which combines neural net- work potentials (NNP) based molecular dynamics simulations with kernel ridge regres- sion (KRR) model to predict NMR parameters in aluminosilicate zeolites under experi- mentally relevant conditions. By capturing local atomic environments and interactions, this method accurately predicts 29 Si and 27 Al chemical shifts across a range of zeo- lite frameworks (MFI, RTH, CHA, etc.). This double machine learning approach is validated against both DFT calculations and experimental data, demonstrating high reliability and precision. While the application part of this work focuses primarily on interpreting experimental 29 Si-27 Al D-HMBC NMR spectra in RTH zeolite, the method- ology is expected to be broadly applicable to other spectroscopic analyses and materials. Furthermore, it is shown that most aluminium atoms in RTH occupy the T2 position, a finding not accessible through standard experimental techniques. These advancements enable efficient, large-scale simulations of complex zeolite systems, with significant im- plications for catalyst design and broader materials research. Keywords Zeolites, NMR, machine learning, aluminium distribution, DFT

guest :: login Digital Repository
		Search		Submit		Help		About