Development and Application of Electronic Structure Methods for Noncovalent Interactions in Organic Solids

Abstract

This thesis reports on multilevel electronic structure approaches for the description of noncovalent interactions. They play an important role in various areas of chemistry and physics, ranging from bio-molecular applications to organic semi-conductors. The main focus lies on the cost-effective yet reasonably accurate description of noncovalently bound solids in the framework of organic crystal structure prediction (CSP). In principle, high-level quantum chemical wavefunction theory methods can seamlessly describe all of the local and nonlocal interactions but are computationally too demanding for large organic complexes, specifically for molecular crystals of larger molecules.<br /> London dispersion inclusive density functional theory (DFT-D) is state-of-the-art in molecular gas phase applications. However, its absolute accuracy for lattice energies and crystal geometries was still uncertain.<br /> The good performance of DFT-D on standard and newly compiled benchmark sets is shown to be close (or within) the chemical accuracy of 1 kcal/mol. Exact exchange and three-body dispersion overall improve the performance, e.g., the mean absolute relative deviation of the hybrid functional PBE0-D3^atm from the reference lattice energies of the X23 and ICE10 sets is 6.6% and 6.1%, respectively. Because the references are typically experimental sublimation energies and X-ray geometries at finite temperature, a correct treatment of zero-point and thermodynamic effects is mandatory.When compared to the experimental unit-cells, which are corrected for zero-point and thermal effects, the DFT-D unit cell volumes are accurate within 1--3%. Thus, DFT-D in principle is applicable to CSP, but the computational demands to sample a huge number of polymorphs, are too high.<br /> In the second part of this thesis, alternative low-cost methods are developed, extended to periodic boundary conditions, and evaluated on standard benchmarks. Two approaches, namely the London dispersion corrected density functional tight-binding (DFTB3-D3) and the corrected small basis set Hartree-Fock (HF-3c) are especially promising. The empiricism of HF-3c is comparable to modern density functionals (nine global parameters) while the tight binding Hamiltonian relies on element-specific parametrized pair potentials. Both schemes are shown to accurately model both solid- and gas phase inter- and intramolecular noncovalent interactions. The mean absolute deviation for interaction (lattice) energies are typically 1-3 kcal/mol (5--20%), that is, only about two times larger than those for DFT-D. At the same time, a speed-up of two to three orders of magnitude can be achieved. HF-3c yields very reasonable unit cell volumes (mass densities) within 3-5% error, while DFTB3-D3 yields larger errors up to 15%. However, the deviations of thermodynamic corrections to sublimation energies between the DFTB3-D3 and DFT-D level is below 0.5 kcal/mol and the tight binding model can be ideally used in a multilevel approach. One can, for instance, combine the thermal corrections of DFTB3-D3 with the electronic energy from DFT-D or use the computationally cheaper method to screen a huge number of possible conformations.<br /> The presented methods can be routinely applied to molecular crystals as demonstrated in the last part of the thesis. The correct description of a variety of crystal packing effects is presented. Specifically, the change of the molecular conformer of ethyl acetate, the stacking of pi-systems, the spin state of iron spin-crossover compounds, and the bond isomerization of certain zirconium complexes are computed in agreement with corresponding experiments.