CrysVCD enforces chemistry constraints in crystal generation
bravo_abad · x · 2026-08-27
Mouyang Cheng et al. introduce CrysVCD, a generative framework that builds oxidation-state balance directly into crystal generation rather than filtering invalid structures afterward.
The architecture separates the problem typically treated jointly by crystal diffusion models:
- Composition Generation: A Transformer-based elemental language model generates the composition. Tokens encode element and oxidation state, with embeddings initialized from electronic configurations. Charge neutrality is enforced here.
- Structure Generation: An E(3)-equivariant diffusion model generates the 3D crystal structure conditioned on that composition.
This separation is computationally efficient, as composition generation requires only a few autoregressive steps versus 1,000 denoising steps.
Related event: MIT's CrysVCD Boosts Chemically Stable AI-Designed Materials to 68%(2 posts)→
More from Research
- IJCAI 2026 Tutorial: How LLMs are reshaping mathematical optimization — StanfordAILab · 2026-08-27
- Research predicts optimal model size and data allocation for pre-training — yoavgo · 2026-08-27
- ARC-AGI-3 learnings: Key designs for self-learning AI agents — GregKamradt · 2026-08-27
- Recommended Reads: Poggio on Minds & Machines, Welling on GenAI Thermodynamics — neurovium · 2026-08-27
- Category Theory Master: AI Expands Mathematical Truth but Doesn't Change Core — begusgasper · 2026-08-27
- Tempus Cancer Model Boosts Survival Prediction AUC with 1.67M Patient Dataset — neuroecology · 2026-08-27