Efficient Modeling for Large-Scale Chemical Problems: Force Field Development and Screening Workflows

Abstract

This thesis focuses on the implementation and evaluation of periodic boundary conditions (PBCs) in the force field model GFN-FF and its improvement for lanthanides and parameterization for actinides. The original model provides a robust and accurate description of geometries, frequencies, and non-covalent interactions (GFN) for molecular systems with up to 10.000 atoms, including elements up to radon. PBCs are implemented for solid-state systems to account for interactions with periodic images of the unit cell, up to specified cutoffs. The lattice parameters can be optimized with a periodic L-BFGS optimizer, and the Ewald summation ensures the convergence of the electrostatic energy with an optimal convergence factor. Existing molecular crystal benchmarks are expanded concerning atom types by compiling a benchmark including elements P, S, and Cl in addition to commonly included H, C, N, and O. A second molecular crystal benchmark is introduced, which includes peptides with an average molecular size of 96 atoms. A special run mode for molecular crystals (mcGFN-FF) is developed, significantly reducing overbinding for molecular crystals and showing performance comparable to the semiempirical tight-binding method GFN1-xTB on various molecular crystal benchmarks. Excluding the results for ten ice polymorphs, mcGFN-FF achieves a mean absolute relative deviation of 14.1 % for lattice energies and 6.4 % for unit cell volumes. The force field's applicability is further extended through reparameterization for lanthanides and the inclusion of actinides. In particular, the assignment of covalent bonds and ligand bond lengths are improved by optimized covalent radii. The force field, combined with tight-binding and DFT methods, was effectively used in a study of C60 in metal-organic frameworks for photocatalysis, revealing preferred C60 positions and the energetic improbability of solvent shell formation. The improved model enables robust molecular dynamics simulations for scientifically relevant molecules, can optimize lanthanide or actinide biocomplexes, and provides reasonable geometries and isomerization energies. <br/> In addition to this main part, a quantum chemical workflow for the accurate calculation of reaction free energies is developed, with a special focus on microsolvation and the inclusion of conformational entropy contributions. This workflow lays the foundation for large-scale screening of tweezer-like ligands, where mircrosolvation is crucial for accurate reaction free energy predictions. Averaged reaction free energies, calculated with varying numbers of explicit solvent molecules, show an average deviation of 2.3 kcal/mol from experimental data. This level of accuracy allows screening for candidate ligands with a reasonable threshold. This work significantly facilitates computational investigations in solid-state chemistry and presents a feasible and robust approach to handle chemical problems that require mircrosolvation.