Structural Determination of (Al2O3)(n) (n = 1-15) Clusters Based on Graphic Processing Unit.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Global optimization is crucial for determining molecular and atomic cluster structures.
• Cluster structure prediction is a computationally intensive, non-deterministic polynomial problem.
• Graphic Processing Units (GPUs) offer superior computational power over Central Processing Units (CPUs).

Purpose of the Study:

• To develop and evaluate a GPU-based genetic algorithm for efficient atomic cluster structure prediction.
• To assess the acceleration ratio of the GPU algorithm compared to traditional CPU implementations.
• To predict low-lying structures of aluminum oxide clusters.

Main Methods:

• Implementation of a genetic algorithm optimized for GPU parallel processing.
• Utilizing analytical interatomic potentials for cluster energy calculations.
• Testing the algorithm on (Al2O3)n clusters across various GPU operation dimensions (1D, block, 2D).

Main Results:

• Achieved significant acceleration ratios: up to 220x (single precision) and 103x (double precision) on 1D operations.
• Peak acceleration ratios reached 240x (single precision) and 107x (double precision) on block operations.
• The overall GPU genetic algorithm program showed a ~35x speedup over CPU in double precision.
• Successfully located known structures of (Al2O3)n clusters (n=1-10) and predicted new low-lying structures (n=11-15).

Conclusions:

• GPU-based genetic algorithms provide a highly efficient approach for atomic cluster structure prediction.
• The developed method significantly accelerates the search for global minimum structures.
• This approach enables the exploration of larger and more complex cluster systems.