Dresden 2006 – scientific programme
Parts | Days | Selection | Search | Downloads | Help
O: Oberflächenphysik
O 12: Methods
O 12.6: Talk
Monday, March 27, 2006, 16:15–16:30, WIL B321
Efficiency of numerical atom-centered basis functions in all-electron electronic-structure theory: the absolute-convergence limit — •Volker Blum, Jörg Behler, Ralf Gehrke, Karsten Reuter, and Matthias Scheffler — Fritz-Haber-Institut, Faradayweg 4-6, D-14195 Berlin, Germany
The most fundamental choice in practical electronic structure calculations is the underlying basis set. For structures with large empty regions (surfaces, clusters, molecules), numeric atom-centered orbitals (NAO’s) are attractive [1,2]. However, the question of basis set convergence is often skirted by relying on fixed basis set libraries and error cancellation; a straightforward route to establish the basis set limit does not exist.
We here investigate the routine feasibility of meV-level absolute convergence in all-electron total-energy calculations with NAO basis sets (confined atomic, ionic, or hydrogen-like), using as examples small molecules and clusters (N2, Cu2, Cu4, Cu10). (i) We show how to establish a systematic (over-)converged NAO basis set limit. (ii) To construct minimal-sized converged basis sets, we pick individual functions one by one from large candidate function pools, using non-selfconsistent superposition potentials for maximum efficiency. ∼50 basis functions per atom are needed for meV-level convergence in each case. (iii) We find that converged basis sets are transferable to self-consistent potentials, and from one structure to another. The total energy errors from underconverged basis sets depend systematically on the interatomic distance.
[1] B. Delley, J. Chem. Phys. 92, 508 (1990).
[2] J.M. Soler et al., J. Phys.: Condens. Matter 14, 2745 (2002)