Physics-informed diffusion models for extrapolating crystal structures beyond known motifs

2025-10-27

AI Summary

Crystal structure prediction (CSP) often relies on known structural prototypes, limiting the discovery of genuinely novel topologies. Method: We propose a physics-informed diffusion generative model that integrates compactness and local environment diversity descriptors, augmented with chemical validity checking and physical constraint enforcement; this model is further co-optimized with conventional CSP. Contribution/Results: Our approach generates structures exhibiting unprecedented topological novelty: 67% fall outside the top 100 most common prototypes, and 97% are reconstructible via CSP—of which 66% correspond to previously unreported, low-energy, stable frameworks. The method significantly enhances both the efficiency and quality of discovering unknown stable crystal prototypes, establishing a scalable, generative paradigm for de novo functional material design.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Discovering materials with previously unreported crystal frameworks is key to achieving transformative functionality. Generative artificial intelligence offers a scalable means to propose candidate crystal structures, however existing approaches mainly reproduce decorated variants of established motifs rather than uncover new configurations. Here we develop a physics-informed diffusion method, supported by chemically grounded validation protocol, which embeds descriptors of compactness and local environment diversity to balance physical plausibility with structural novelty. Conditioning on these metrics improves generative performance across architectures, increasing the fraction of structures outside 100 most common prototypes up to 67%. When crystal structure prediction (CSP) is seeded with generative structures, most candidates (97%) are reconstructed by CSP, yielding 145 (66%) low-energy frameworks not matching any known prototypes. These results show that while generative models are not substitutes for CSP, their chemically informed, diversity-guided outputs can enhance CSP efficiency, establishing a practical generative-CSP synergy for discovery-oriented exploration of chemical space.

Problem

• Generating novel crystal structures beyond known motifs
• Balancing physical plausibility with structural novelty
• Enhancing crystal structure prediction efficiency through generative models

Innovation

• Physics-informed diffusion embeds compactness and diversity descriptors
• Conditioning on metrics boosts novel structure generation
• Generative models enhance crystal structure prediction efficiency