Dresden 2006 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

CPP: Chemische Physik und Polymerphysik

CPP 29: Computational Techniques

CPP 29.3: Vortrag

Freitag, 31. März 2006, 11:00–11:15, ZEU 160

Some computational techniques for first-principles simulations on condensed matter systems — •Alberto Castro¹, M. A. L. Marques², Heiko Appel¹, Angel Rubio^3,1, and E. K. U. Gross¹ — ¹Institüt fur Theoretische Physik, Arnimallee 14, Fachbereich Physik, 14195 Berlin (Germany). — ²Laboratoire de Minéralogie-Cristallographie de Paris (France). — ³Donostia International Physics Center and Departamento de Física de Materiales, Facultad de Ciencias Químicas, 20018 San Sebastián (Spain).

The real-space techniques for computational simulations in the condensed matter realm are usually praised for, at least, two strong advantages: On the first hand, the intrinsically local character of the "basis set" should permit, in principle, to allow for large scale parallelization by dividing the space in domains. This locality is also the basis for the use of techniques aiming at the linear-scaling of the computational effort. On the second hand, the real space mesh on which the magnitudes are represented may be locally adapted to the needs of each region – one feature difficult to translate to the more traditional plane wave representation. We will present our approach to these two time-saving computational techniques: (i) One possible route to implement curvilinear coordinates, able to adapt the local resolution to the needs of each region in space; (ii) A multiple-way parallelization scheme that may divide the work among a given number of processors, splitting the tasks either in k-points, in Kohn-Sham states (assuming a density functional formulation of the condensed matter problem), in regions in real-space, or in a combination of all of them.