Approaches to Increase the Accuracy of Molecular Dynamics Simulations : From Classical to Tight Binding and First Principles Methods

The predictive power of molecular dynamics simulations is determined by the accessible system sizes, simulation times and, above all, the accuracy of the underlying potential energy surface. Tremendous progress has been achieved in recent years to extend the system sizes via multi-scale approaches and the accessible simulation times by enhanced sampling techniques. Such improvements on the sampling issues shift the focus of development on the accuracy of the potential energy surface. A first goal of this thesis was to assess the accuracy limits of a variety of computational methods ranging from classical force fields over the self-consistent-charge density functional tightbinding method (SCC-DFTB), various density functional theory (DFT) methods, to wave function based ab initio methods in identifying the correct lowest energy structures. As experimental benchmark data to assess the performance of different computational methods we used high-resolution conformer-selective vibrational spectra, measured by cold-ion spectroscopy, of protonated tryptophan and gramicidin S in the group of T. Rizzo at EPFL. These studies showed that most empirical force fields performed rather poorly in describing the correct relative energetics of these molecules, possibly due to the limited transferability of their underlying parameters. With the goal to increase the accuracy of classical molecular mechanics force fields we implemented a recently developed force-matching protocol for an automated parametrisation of biomolecular force fields from mixed quantum mechanics/molecular mechanics (QM/MM) reference calculations in the CPMD software package. Such a force field has an accuracy that is comparable to the QM/MM reference, but at the greatly reduced computational cost of the MM approach. We have applied this protocol to derive in situ FF parameters for the retinal chromophore in rhodopsin embedded in a lipid bilayer. In a similar effort, we employed iterative Boltzmann inversion to derive repulsive potentials for SCC-DFTB. We used reference data at the DFT/PBE level to derive highly accurate parameters for liquid water at ambient conditions, a particularly challenging case for conventional SCCDFTB. The newly determined parameters significantly improved the structural and dynamical properties of liquid water at the SCC-DFTB level. In a third project of the thesis we explored possible accuracy improvements in the context of DFT methods. Dispersion Corrected Atom Centered Potentials (DACPs) are a recently developed method to cure the failure of DFT methods within the generalised gradient approximation to describe dispersion interactions. Here, we complemented the existing library of DCACP parameters by the halogens and compared the performance of various dispersion