Most ab initio simulation packages based on Density Functional Theory (DFT) use the Superposition of Atomic Densities (SAD) as a starting point of the self-consistent fi eld (SCF) iteration. However, this trial charge density without modeling atomic iterations nonlinearly may lead to a relatively slow or even failed convergence. This thesis proposes a machine learning-based scheme to improve the initial guess. We train a 3-Dimensional Convolutional Neural Network (3D CNN) to map the SAD initial guess to the corresponding converged charge density with simple structures. We show that the 3D CNN-processed charge density reduces the number of required SCF iterations at different unit cell complexity levels.

We propose a new method of calculating electronically excited states that combines a density functional theory based ground state calculation with a linear response treatment that employs approximations used in the time-dependent density functional based tight binding (TD-DFTB) approach. The new method termed time-dependent density functional theory TD-DFT+TB does not rely on the DFTB parametrization and is therefore applicable to systems involving all combinations of elements. We show that the new method yields UV/Vis absorption spectra that are in excellent agreement with computationally much more expensive TD-DFT calculations. Errors in vertical excitation energies are reduced by a factor of two compared to TD-DFTB.

Ab initio Hartree-Fock and density functional theory calculations were carried out to investigate the structures, energies, and vibrational spectra of indole and two of its hydrogen migration tautomers. The calculated results of indole are in good agreement with experiments. Rotational constants and infrared spectral features of 2H-indole and 3H-indole are predicted to assist future experimental identification of the two tautomers. Transition states of unimolecular isomerization among the three tautomers are also optimized and activation barriers of the isomerization reactions evaluated. The results indicate that unimolecular indole → 3H-indole proceeds via 2H-indole with an activation barrier of 51 kcal/mol.

The molecular structures and vibrational spectra of Dewar benzene, benzvalene, trimethylenecyclopropane, prismane and 3,4-dimethylenecyclobutene have been investigated by density functional theory using Becke's exchange with Lee-Yang-Parr's correlation functional and the 6-31G* basis set. Both the calculated structural parameters and vibrational frequencies are in good agreement with available experimental data. On the basis of the agreement between the calculated and experimental results, assignments of fundamental vibrational frequencies of Dewar benzene, benzvalene, and trimethylenecyclopropane were examined and some reassignments are proposed. The calculations also predict prominent IR and Raman spectral features of prismane and 3,4-dimethylenecyclobutene, which can assist experimental identification of these compounds and the assignment of observed spectral features when they are available.

Density functional theory (using Becke's exchange and Lee-Yang-Parr's correlation functionals (BLYP)) and ab initio Hartree-Fock calculations were carried out in order to investigate the molecular structure and vibrational spectra of 9,10-anthraquinone and its perdeuterated analog. The calculated structural and spectral features are in good agreement with the available experimental results. Most of the BLYP/6-31G* non-CH(D) stretching frequencies are slightly lower than reliable experimental assignments; the mean absolute deviation is about 14 cm-1. On the basis of agreement between calculated and experimental results, assignments of the fundamental vibrational modes were examined and some reassignments were proposed. The calculated results can serve as a guide for a future experimental search for the missing fundamentals of the target molecules.

Comparison of the observed and calculated vibrational frequencies of all-trans-octatetraene indicates that the density functional theory (DFT) using Becke's exchange and Lee-Yang-Parr's correlation functionals is as accurate as the Hartree-Fock (HF)-based scaled quantum mechanical force field approach in predicting fundamental vibrational frequencies. As the DFT calculation does not use any empirical parameters pertaining to the subject molecule and its computational cost scales more favorably than that of the HF theory, it is a more promising approach to molecular vibrational problems and should replace the empirical normal coordinate analysis for assisting vibrational assignments.

Small clusters of [6,6] phenyl-C61-butyric acid methyl ester (PCBM) molecules are analyzed with respect to their equilibrium geometries and associated electronic as well as energetic properties. Plane wave density functional theory (PWDFT) computations, assisted by molecular dynamics (MD) simulations, are performed on systems of the form PCBMn (n = 1-5). The bonding operative in these units is described as a cooperation between HO bonding, involving the C5H9O2 groups of the PCBM molecule, and fullerene-fullerene attraction. The maximally stable structures identified tend to include a dimer motif that combines both interaction modes. The great importance of van-der-Waals effects in stabilizing the studied clusters is demonstrated by comparing the PCBM3 series with and without inclusion of a van-der-Waals term in the PWDFT procedure. The two approaches yield reverse orders of stability. A decreasing tendency in the Kohn-Sham HOMO-LUMO gaps of PCBMn with the cluster size may be used to monitor PCBM aggregation in the active layer of organic photovoltaic devices by optical spectroscopy.

A combination of quantum chemical calculation and molecular dynamics simulation is applied to compute refractive indices of liquid-forming silicon oils. The densities of these species are obtained from molecular dynamics simulations based on the NPT ensemble while the molecular polarizabilities are evaluated by density functional theory. This procedure is shown to yield results well compatible with available experimental data, suggesting that it represents a robust and economic route for determining the refractive indices of liquid-forming organic complexes containing silicon.

The structures of endohedral complexes of the polyhedral oligomeric silsesquioxane (POSS) cage molecule (HSiO 3/2) 12, with both D 2d and D 6h starting cage symmetries, containing the atomic or ionic species: Li 0, Li +, Li -, Na 0, Na +, Na -, K 0, K +, K -, F -, Cl -, Br -, He, Ne, Ar were optimized by density functional theory using B3LYP and the 6-311G(d,p) and 6-311 ++G(2d,2p) basis sets. The exohedral Li +, Na +, K +, K -, F -, Cl -, Br -, He, Ne, Ar complexes, were also optimized. The properties of these complexes depend on the nature of the species encapsulated in, or bound to, the (HSiO 3/2) 12 cage. Noble gas (He, Ne and Ar) encapsulation in (HSiO 3/2) 12 has almost no effect on the cage geometry. Alkali metal cation encapsulation, in contrast, exhibits attractive interactions with cage oxygen atoms, leading to cage shrinkage. Halide ion encapsulation expands the cage. The endohedral X@(HSiO 3/2) 12 (X = Li +, Na +, K +, F -, Cl -, Br -, He and Ne) complexes form exothermically from the isolated species. The very low ionization potentials of endohedral Li 0, Na 0, K 0 complexes suggest that they behave like "superalkalis". Several endohedral complexes with small guests appear to be viable synthetic targets. The D 2d symmetry of the empty cage was the minimum energy structure in accord with experiment. An exohedral fluoride penetrates the D 6h cage to form the endohedral complex without a barrier.

Computational investigations on semiconductor (silicon or germanium) clusters (Sinor Gen) in combination with transition metal (M) impurities are reviewed in this contribution. Emphasis is placed on investigations that focus on the size evolution features of MmSi n(or MmGen) such as the critical ligand number for the transition from exohedral to endohedral equilibrium geometry. Geometric, energetic, electronic, and magnetic characteristics of MmSi n or MmGen systems are discussed. It is pointed out that selected MmSin systems with n = 12 and n = 16 and MmGen with n = 10 or 12 and n = 16 emerge from present computational research in the size region of n ≤ 20 as the most promising candidates for building blocks of novel nanomaterials. In addition, comparison is made between MmSin and MmGen clusters.