Enhancing materials discovery with valence-constrained design in generative modeling.

Valence-constrained generative model CrysVCD achieves 85% metastability in crystal design, significantly improving chemical validity in materials discovery.

• Why it matters: Ensuring chemical validity is crucial for reliable materials design, yet many models overlook essential constraints like oxidation-state balance, leading to invalid structures.
• What they did: CrysVCD integrates chemical rules into a diffusion-based generative framework, using a transformer-based elemental model for valence-balanced compositions and a diffusion model for crystal structures, fine-tuned on stability metrics.
• The result: This approach enables efficient validation, supports conditional functional material generation, and can be integrated into various pipelines, advancing scientifically grounded materials discovery.