Generative model designs crystals to order; its flagship synthesis turned out to be a known compound
MatterGen generates crystal structures conditioned on target properties, and its headline experimental validation was later shown to match a compound reported in 1972 that sits in the model's own training data.
- Model
- MatterGen
- Field
- Materials science
- Date
- 2025-01-16
Sources
Original work
Announcement
Challenge
What was found
MatterGen is a diffusion model over crystal structures (atom types, coordinates and lattice jointly), trained on about 608,000 stable materials from the Materials Project and Alexandria, and fine-tunable to condition on a target property. Conditioned on a bulk modulus of 200 GPa it proposed a structure that collaborators at the Shenzhen Institutes of Advanced Technology synthesised, measuring 169 GPa, within 20% of the specification. Published in Nature.
Novelty check
Generative models for crystals existed (CDVAE, and screening pipelines like GNoME); the contribution claimed is property-conditioned generation validated experimentally. The novelty question is precisely what is disputed below: whether the synthesised compound was new.
Caveats and known objections
A published critique in Materials Horizons argues the synthesised disordered phase Ta₁⁄₃Cr₂⁄₃O₂ is the same material as Ta₁⁄₂Cr₁⁄₂O₂ reported in 1972, that this compound is present in MatterGen's training set, and that its composition differs from the reported TaCr₂O₆, i.e. the flagship validation recovered a known material rather than a new one. The generative method itself is not refuted; the disputed part is the experimental novelty claim, and the entry is graded on the strength of that published objection. This registry keeps entries like this one rather than removing them.
Independent checks
Materials Horizons: Continued challenges in high-throughput materials predictions: disputed the novelty of the synthesised compound; identified it as present in the training data · link ↗
Disagree with these grades?
Bring a citation: a grade moves on evidence, not on argument.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Disputed and Search scaffold.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded disputed for verification and search scaffold for autonomy. What these mean.
Cite this entry
whataifound.org. (2025). Generative model designs crystals to order; its flagship synthesis turned out to be a known compound. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2025-01-16-mattergenBibTeX
@misc{whataifound-microsoftresearch-2025-mattergen,
title = {Generative model designs crystals to order; its flagship synthesis turned out to be a known compound},
author = {{whataifound.org}},
year = {2025},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Microsoft Research. Verification: Disputed. Autonomy: Search scaffold.},
url = {https://whataifound.org/finding/2025-01-16-mattergen}
}