Presentation Information
[11a-PA3-3]Crystal Structure Prediction from Substituted Compositions Based on a Structural Similarity Metric for Aliovalent Substitution
〇(D)Tomiya Yamamoto1, Yosuke Harashima1,2, Shogo Takasuka1, Tomoaki Takayama1,3, Fujii Mikiya1,2,3 (1.NAIST, 2.NAIST DSC, 3.NAIST CMP)
Keywords:
Crystal Structure Prediction,Aliovalent substitution,Structural similarity
In this work, we construct a metric that quantifies structural similarity under aliovalent substitution and propose a method that performs post-substitution structure prediction by predicting this similarity from the substituted composition alone. Targeting 106,318 oxide substitution pairs from a Materials Project, we defined the similarity as a weighted average of six sub-metrics, including polyhedral type distribution and coordination number distribution, and built a machine learning model to predict it. The output of this model is then provided as a condition to a crystal structure generation model to generate the substituted structures.