Berlin 2018 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 57: Methods in Computational Materials Modelling (methodological aspects, numerics)
MM 57.3: Vortrag
Donnerstag, 15. März 2018, 12:15–12:30, TC 006
Basis Set Selection for Advanced Density Functionals and Quantum Chemistry via Compressed Sensing — •Niklas Menzel1, Luca M. Ghiringhelli1, Gitta Kutyniok2, and Matthias Scheffler1,3 — 1Fritz-Haber-Institut der MPG, Berlin, DE — 2Technische Universität, Berlin, DE — 3UC Santa Barbara, USA
The selection of basis functions is an important issue in density functional theory and quantum chemistry. The main task is to minimize the computational costs while maintaining the accuracy. Commonly used basis sets are not sufficiently accurate to represent the eigenfunctions for advanced exchange-correlation treatments. This leads to basis set extensions, such as the correlation-consistent basis sets by Dunning [JCP 90, 1007 (1989)]. We propose a new method based on Compressed Sensing (CS), a recently developed signal processing technique. In CS, sparse signals are recovered using ℓ0-norm or ℓ1-norm regularization. Similarly, the key component of our approach is to find sparse real-space representations of self-consistently converged eigenfunctions (reference orbitals). We have developed a method for the selection of continuously parametrized basis functions (like Gaussian- or Slater-type basis functions). For the reference orbitals we used numeric atom-centered orbital basis functions. The reference orbitals are generated for the free atom and homonuclear dimers.
With our new approach we can robustly determine accurate basis sets for all atoms.