Improving the Silicon Interactions of GFN-xTB

Area of Science:

• Computational chemistry
• Materials science
• Quantum chemistry

Background:

• The general-purpose density functional tight binding (GFN-xTB) method is increasingly used for accurate simulations beyond conventional ab initio methods.
• The original GFN1-xTB parametrization exhibits poor accuracy when describing organosilicon compounds.

Purpose of the Study:

• To address the limitations of the GFN1-xTB model for organosilicon compounds.
• To develop an improved GFN-xTB parametrization specifically for silicon-containing systems.

Main Methods:

• Re-fitting the silicon parameters of the GFN1-xTB model.
• Utilizing a dataset of 10,000 reference compounds.
• Geometry optimization using the revPBE functional for reference data generation.

Main Results:

• The new GFN1(Si)-xTB parametrization demonstrates enhanced accuracy.
• Improved prediction of system energies, nuclear forces, and molecular geometries for organosilicon compounds.
• Validation against a large dataset of reference calculations.

Conclusions:

• The re-parametrized GFN1(Si)-xTB model offers superior performance for silicon-containing systems.
• This improved model should be considered for all GFN-xTB Hamiltonian applications involving silicon.
• The findings facilitate more accurate computational studies in organosilicon chemistry.