Andrey Laktionov
Density Functional Theory (DFT) and its time-dependent extension (TDDFT) have become two of the most popular approaches for computer simulations of the electronic structure and response properties of quantum systems. A reasonable compromise between accuracy and computational cost allows to apply DFT to a wide range of systems from small molecules to biological complexes. Despite the in principle exact nature of DFT and TDDFT, practical calculations require the use of approximate DFT exchange-correlation (XC) functionals and TDDFT kernels and the accuracy of the obtained results is determined by the accuracy of the chosen XC description. These approximations inevitably introduce some limitations in the use of DFT-based methods.The inability of the local spin density approximation, generalized gradient approximations (GGA) and even some hybrid functionals to properly describe charge transfer (CT) excitations and predict intermolecular interaction energies of weakly bound complexes are two major drawbacks of current XC descriptions. This thesis is therefore devoted to the improvement of XC functionals for the special cases of weak interactions and charge-transfer excitations. Many original strategies have been suggested to cure DFT calculations from these failures. In particular, the Dispersion-Corrected Atom-Centered Potentials (DCACP) approach provides an accurate description of dispersion forces within generalized gradient approximations for the exchange-correlation functional. The DCACP method has been extensively used for the last seven years and has shown an excellent performance for a large class of applications. However, one of the drawbacks of the current implementation of DCACP is that the correct R-6 asymptotics of dispersion interaction is not reproduced. A first goal of this thesis was to enable DCACP to recover the R-6 asymptotic limit. To this end, we have designed a new 2-channel version of DCACP and carried out test calculations on both small molecules and large macromolecular complexes. The obtained results demonstrate the excellent performance and transferability of the DCACP approach. Moreover, for large macromolecular complexes, in which the binding energy is dominated by dispersion, [pi]-stacking, or hydrogen bonding, 2-channels DCACP were found to be the best method overall for correcting the popular BLYP functional. Our study clearly shows that account of R-6 asymptotics is crucial for the description of large complexes and that 2-channels DCACP are fully able to capture these effects. On the contrary, an account of R-6 asymptotics is of little importance for small molecules since the remaining errors of the underlying GGA functional are dominating. With the aim of deepening our understanding of the DCACP concept and the reasons for its excellent performance, we explored the properties of the two DCACP parameters and were able to establish some systematic trends. It turned out that variational tuning of the DCACP can be done in an analytical manner that enables the easy generation of DCACP potentials for the full periodic table. Furthermore, since DCACP have little but crucial impact on the electronic density, dispersion energies can be obtained from non-self-consistent electron densities. These empirical findings suggest that the high transferability of DCACP is due to their atom-centered form and the intrinsically weak nature of dispersion interactions. [...]
EPFL2014
