2022年第69回応用物理学会春季学術講演会

講演情報

一般セッション(口頭講演)

23 合同セッションN「インフォマティクス応用」 » 23.1 合同セッションN「インフォマティクス応用」

[24a-E203-1~10] 23.1 合同セッションN「インフォマティクス応用」

2022年3月24日(木) 09:00 〜 11:45 E203 (E203)

沓掛 健太朗(理研)、志賀 元紀(岐阜大)

09:15 〜 09:30

[24a-E203-2] Application of combinatorial group theory to the atomic substitution problem

〇(DC)Genki Imam Prayogo1、Andrea Tirelli2、Keishu Utimula1、Kenta Hongo3、Ryo Maezono1、Kosuke Nakano1,2 (1.Sch. Info. Sci. JAIST、2.SISSA、3.RCACI, JAIST)

キーワード:materials informatics, high-throughput

Supercell model is often utilized to study a disordered system within periodic ab-initio framework. In this model, a supercell is created in which a certain number of target elements are substituted. The key problem in the generation of supercells is how to identify and eliminate symmetry-equivalent structures from a vast number of substitution patterns. These structures are physically identical, and thus redundant for ab-initio simulations. By reducing the number of considered structures to only those which are symmetrically unique, the computational cost can be reduced by various order of magnitude.
In this work, we mapped the atomic substitution problem into a graph coloring problem, implementing canonical augmentation as the isomorph rejection technique. In this approach, symmetry-equivalent structures are pruned in a search tree, covering the space of all possible substitutions, without directly comparing the structures. To further improve rejection efficiency, we also incorporated quantities derived from the structural information into the algorithm.
Built upon these concepts, we developed a python package, called SHRY (Suite for High-throughput generation of models with atomic substitutions implemented by python), whose main function is to identify and select, from any given structures, a single representative structure for each symmetry-equivalent class of structures. We benchmarked its performance against existing approaches, confirming its significant efficiency over existing approaches.