Memetic figure selection for cluster expansion in binary alloy systems

Cluster expansion provides a powerful tool in materials modeling. It has enabled an efficient prediction of the atomic properties of materials with the combination of the modern quantum calculation theory. To construct an accurate cluster expansion model, a few important cluster figures should be identified. This paper proposes a novel figure selection method based on memetic algorithm (MA), which is a synergy of genetic algorithm (GA) and orthogonal matching pursuit (OMP) based memetic operation. The memetic operation is designed to fine-tunes the solutions of GA and accelerate the convergence of the search. The performance of the proposed method is evaluated on two binary alloy datasets. Comparative study to other state-of-the-art figure selection methods demonstrates that the proposed method is capable of obtaining better or competitive prediction accuracy and searching the figure space efficiently.