Berlin 2008 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

O: Fachverband Oberflächenphysik

O 49: Particles and Clusters

O 49.1: Vortrag

Mittwoch, 27. Februar 2008, 15:15–15:30, MA 005

Comparing the sampling efficiency of genetic and basin-hopping algorithms in the structural optimization of Lennard-Jones clusters — •Vladimir Froltsov and Karsten Reuter — Fritz-Haber-Institut, Faradayweg 4-6, D-14195 Berlin (Germany)

Detailed knowledge about structural properties is an indispensable prerequisite en route towards a future exploitation of the intriguing materials properties of small atomic clusters. Crucial for the corresponding global geometry optimization problem is a reliable sampling of the high-dimensional potential-energy surface (PES) to identify the (meta)stable minima. In view of the high computational cost when the PES is evaluated with predictive quality by first-principles methods, utmost efficiency with a minimum dependence on the initial starting configuration is an additional demand on the sampling scheme. With this focus we compare the efficiency of the two prevalent algorithms for PES sampling, namely genetic algorithms (GA) and the basin-hopping (BH) approach, using the determination of the ground-state structure of model Lennard-Jones clusters as criterion. We find an intriguing cross-over between the two approaches, with BH more efficient than GA for cluster sizes up to 65, and GA more efficient for larger clusters. We rationalize this finding with the ability of the population-based GA's to explore in parallel disjunct parts of the PES, which becomes increasingly important for the increasingly high-dimensional PESs of larger clusters.