Recent DFTB Extension for Improving Accuracy and Boosting the Efficiency for Computational Applications to Nanomaterials
Speaker
Prof. Thomas Frauenheim
University of Bremen, Bremen Center for Computational Materials Science
Abstract

The new release of DFTB+ as a density-functional (DFT)-based approach, combining DFT-accuracy and Tight-Binding (TB) efficiency, is reported; http//:www.dftb.org. Methodological details and recent extensions including 3rd order DFTB and going beyond the monopole approximation to improve reliability and accuracy will be described briefly. Advanced functions include spin degrees of freedom, time dependent methods for excited state dynamics, and multi-scale QM/MM/Continuum-techniques to treat reactive processes in nanostructures under environmental conditions.
Very recently the time-dependent density functional based tight-binding (TD-DFTB) approach has been generalized going beyond the Mulliken approach. An additional on-site correction leads to marked qualitative and quantitative improvements over the original method. Especially, the known failure of TD-DFTB for the description of σ → π* and n → π* excitations is overcome. Benchmark calculations on a large set of organic molecules confirm a much better description of triplet states. As application we report on excited-state theoretical simulations and experimental studies to investigate the degradation of nitric oxide and acetaldehyde on TiO2 under VIS and UV irradiation revealing charge transfer complexes on TiO2 as new source for visible light activity. Surprisingly, the on-site correction as well improves the accuracy in the performance of hydrogen bonds.
For boosting the efficiency I report on a radical extension of density functional based molecular dynamics (MD) by at least three orders of magnitude. Our targets are 106-atom and 106-time step simulations with a wall-clock time of about 0.1 second per time step for a few thousands of atoms. Our goals will be achieved by combining and further developing, a number of new ground breaking ideas and techniques, which include 1) a new extended Lagrangian Born-Oppenheimer MD that reduces self-consistent field costs to the absolute minimum, 2) new accelerated, low pre-factor, linear scaling solvers and 3) novel graph based parallel approaches to sparse matrix algebra tailored for hybrid architectures.
In the last months there has been ongoing effort to adapt range-separated hybrid exchange correlation functionals and the particle-particle random phase approximation (pp-RPA) applying an approximate pairing matrix fluctuation formalism to determine different types of vertical excitation energies and the corresponding oscillator strengths, including charge-transfer (CT) and double excitations [Yang Yang, Helen van Aggelen and Weitao Yang, J. Chem. Phys. 139, 224125 (2013)]. The full diagonalization of the pp-RPA matrix, built within the Mulliken approach, has been implemented in a development version of the DFTB+ code, and the parameters for the elements O, N, H, C and S have been obtained. Preliminary tests show that pp-DFTB can describe double and single excitations of small organic molecules with fair accuracy comparable to ab-initio pp-RPA using B3LYP as reference. Furthermore, CT excitation energies of a set of donor-acceptor complexes have been benchmarked against pp-RPA (cam-B3LYP) and TD-DFT (cam-B3LYP). Our method outperforms TD-DFT for the description of these CT complexes whereas returns a mean unsigned error similar to pp-RPA. In the future, a Davidson algorithm for the resolution of the eigenvalue problem within pp-DFTB will be implemented. This will allow for the study of larger systems, for which the full diagonalization of the pp-RPA matrix is unfeasible.

 

About the Speaker
Date&Time
2015-11-19 3:00 PM
Location
Room: A303 Meeting Room
CSRC 新闻 CSRC News CSRC Events CSRC Seminars CSRC Divisions