GLOBAL OPTIMIZATION STRATEGIES FOR EFFICIENT ENERGY LANDSCAPE EXPLORATION IN ATOMISTIC SIMULATIONS

Abstract

Genetic algorithms (GAs) are widely used in materials science to explore complex potential energysurfaces and identify optimal atomic configurations. This study develops a comprehensive GAbased workflow to optimize nanoparticle structures, incorporating both traditional and machine learning-accelerated approaches. The process begins with the generation of candidate structures, followed by structural relaxation using the Vienna Ab initio Simulation Package (VASP). We evaluate and compare five selection strategies: tournament, elitism, ranking, elitism with ranking, and elitism with tournament. These strategies are compared using hypothesis testing to assess statistically significant differences in optimization performance. This analysis is conducted on two systems: • An FCC(111) 3×3×4 Ag slab with a single atom of (Pd, Pt, Ru, or Rh), • A Ni₆Pd₄ cluster, where the goal is to discover the atomic arrangement that minimizes total energy. Various crossovers are benchmarked following the selection strategy comparison to further refine the evolutionary process. The best-performing GA configuration is then coupled with a Gaussian Process (GP) surrogate model, enabling accelerated prediction of structural stability and significantly reducing the number of expensive DFT evaluations. Key findings from this study include: vii • Ranking and tournament selection strategies effectively balance exploration and exploitation, yielding consistently low-energy configurations. • Integrating a Gaussian Process model reduces the number of VASP evaluations needed while achieving similar or better minima with minimal loss in accuracy for the cluster system. • Simple Cut Splice and Half Uniform crossovers produced the most favorable distributions among the crossover operators tested. These results highlight the critical importance of selecting appropriate evolutionary strategies and operators and demonstrate the potential of surrogate modeling to improve computational efficiency significantly in nanoparticle structure optimization. The combined GA-GP framework developed here offers a promising path forward for accelerating the discovery and design of advanced materials. The custom genetic algorithm operators (selection, crossover, mutation) and Gaussian Process surrogate modeling scripts developed as part of this work are available at: https://github.com/gunasooriya-lab/cluster