Reproduce “DiffSyn: A Generative Diffusion Approach to Materials Synthesis Planning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Elton Pan, Soonhyoung Kwon, Sulin Liu, Mingrou Xie, Alexander J. Hoffman, Yifei Duan, Thorben Prein, Killian Sheriff and 4 more
A generative diffusion model named DiffSyn predicts viable synthesis routes for crystalline materials, particularly zeolites, by accurately modeling complex, multi-modal parameter distributions. It establishes a new state-of-the-art in synthesis prediction, notably enabling the successful experimental synthesis of an unknown UFI zeolite with an Si/Al ratio of 19.0.
Key insights
- Diffusion models are superior at capturing complex, multi-modal distributions in synthesis parameter space compared to previous generative models.
- The model implicitly learns and adheres to fundamental chemical and physical principles governing materials formation, such as Arrhenius kinetics and Villaescusa's rule.
- Combining robust generative models with chemical guidance allows for the prediction of phase-selective synthesis conditions and optimal routes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.