WyCryst: Wyckoff inorganic crystal generator framework

• Main Authors: Zhu, Ruiming, Nong, Wei, Yamazaki, Shuya, Hippalgaonkar, Kedar
• Other Authors: School of Materials Science and Engineering
• Format: Article
• Language: English
• Published: 2024
• Subjects: Engineering; Design; Models
• Online Access: https://hdl.handle.net/10356/180690
• Institution: Nanyang Technological University

Recent advancements in property-directed generative design of inorganic materials account for periodicity and global Euclidian symmetry through translations, rotations, and reflections; however, they do not account for symmetry constraints within allowed space groups. To address this, we introduce a generative design framework (WyCryst) composed of three components: (1) a Wyckoff position-based inorganic crystal representation, (2) a property-directed variational autoencoder (VAE) model, and (3) an automated density functional theory (DFT) workflow for structure refinement. Our framework selectively generates materials by encoding the Wyckoff representation for each space group. As validation, we reproduce a variety of existing materials, CaTiO3, CsPbI3, BaTiO3, and CuInS2, for both ground-state and polymorphic crystal structure predictions. We also generate several ternary materials not found in the training database, which are proven to retain their symmetry and are phononically stable using our automated DFT workflow. We believe our symmetry-aware WyCryst takes a vital step toward AI-driven inorganic materials discovery.