Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 17: Development of Calculation Methods
MM 17.1: Talk
Wednesday, March 19, 2025, 10:15–10:30, H22
Premature Convergence, It’s Nothing to be Embarrassed About: Solving Performance Issues with Swarm-Based Global Optimization to Generate Pt Nanoparticle Ensembles — •Julian Holland1, Malgorzata Makos3, Difan Zhang4, Mal-Soon Lee3, Roger Rousseau3, Chris-Kriton Skylaris2, and Vanda Glezakou3 — 1Fritz-Haber-Institut der MPG, Berlin — 2University of Southampton, Southampton, UK — 3Oak Ridge National Laboratory, Oak Ridge, USA — 4Pacific Northwest National Laboratory, Three Cities, USA
Swarm-based global optimisation (GO) algorithms have proven successful in exploring potential energy surfaces (PESs) of chemical systems. However, they are often limited by their serial implementation. Our GO software, pyGlobOpt, uses an asynchronously parallel artificial bee colony (ABC) methodology, mitigating this limitation. We enhance pyGlobOpt further by tuning parameters against a new, general, ensemble generation assessment criterion. Using this criterion, we were also able to demonstrate how to overcome premature convergence, an issue pervading the use of the ABC algorithm for some systems, using a clustering-based methodology. We demonstrate that using the clustering algorithm alongside tuned pyGlobOpt parameters can lead to a 5-fold increase in the number of unique low-energy structures found as well as more than halving the average energetic distance from the global minimum. We produce ensembles of thermodynamically relevant Pt nanoparticles with varying hydrogenation using our enhanced software and compare to experimental results.
Keywords: Global Optimization; Nanoparticle; DFT