DMFlow: Disordered Materials Generation by Flow Matching
Liming Wu, Rui Jiao, Qi Li, Mingze Li, Songyou Li, Shifeng Jin, Wenbing Huang · 2026
Read DMFlow: Disordered Materials Generation by Flow Matching by Liming Wu, Rui Jiao, Qi Li, Mingze Li, Songyou Li, Shifeng Jin, Wenbing Huang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04734.txt.
Description
Liming Wu and 6 co-authors generate disordered materials with a model called DMFlow in a 2026 paper. Most generative models design perfectly ordered crystals only. Disordered crystals, where atoms swap places or shift positions, matter greatly but lacked their own tools.
DMFlow uses one scheme covering ordered, substitutionally disordered, and positionally disordered crystals. A flow matching model generates all parts of the structure jointly. A Riemannian version with a spherical reparameterization keeps disorder weights valid as probabilities. A graph network with symmetry-aware message passing learns the flow. A two-stage step turns smooth weights into concrete atom choices. On structure prediction and de novo generation tasks, DMFlow beat adapted baselines by clear margins.
Catalog record
| CREATOR | Liming Wu, Rui Jiao, Qi Li, Mingze Li, Songyou Li, Shifeng Jin, Wenbing Huang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04734