Structural Constraint Integration in Generative Model for Discovery of Quantum Material Candidates

Compared to the billions of discovered organic molecules, inorganic materials have long faced the challenge of data scarcity, with only a tiny fraction of possible candidates having been discovered. Recent advancements in machine-learning-based generative models, particularly diffusion models, show great promise for generating new, stable materials. However, challenges remain in integrating lattice types into material generation. Here, we introduce Structural Constraint Integration in the GENerative model (SCIGEN), which sets up constraint and unconstraint channels and combines them during the diffusion process, generating new structures while maintaining preset lattice constraints. We generate eight million compounds using Archimedean lattices (AL) as prototypes, with over 10% surviving a multi-stage stability pre-screening. High-throughput density functional theory (DFT) on 26,000 compounds shows over 50% were stable at the DFT level. Since the properties of quantum materials are closely related to the lattice types, SCIGEN provides a general framework for generating quantum materials candidates.