AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-21
CrystalAR: Autoregressive Modeling for Symmetric Crystal Generation
Incorporating space-group symmetry into crystal generative models increases the combinatorial complexity of stoichiometric relationships between atomic species and crystallographic sites, and creates tension between exploring discrete crystallographic templates and realizing them as continuous 3D structures. CrystalAR models Wyckoff symmetry templates and complete crystal structures in a single autoregressive sequence, reformulating continuous coordinate generation as a symmetry-conditioned hierarchy of discrete spatial localization decisions across resolution levels and enabling constraint-aware search. It retains strong symmetry-template generation while achieving competitive de novo generation performance and template-free crystal structure prediction accuracy.