Structural Constraint Integration in a Generative Model for the Discovery of Quantum Materials

Published in Nature Materials, 2025

Paper / MIT News / GitHub / Data page /

SCIGEN enforces geometric constraints such as honeycomb and kagome lattices within diffusion-based generative models, generating ten million candidate compounds with Archimedean and Lieb lattices; over 10% pass multistage stability screening, and a GNN classifier detects magnetic ordering in 41% of DFT-relaxed structures.

Recommended citation: Okabe, R., Cheng, M., Chotrattanapituk, A., Mandal, M., Mak, K., Cordova Carrizales, D., Hung, N.T., Fu, X., Han, B., Wang, Y., Xie, W., Cava, R.J., Jaakkola, T.S., Cheng, Y., Li, M. (2025). Structural Constraint Integration in a Generative Model for the Discovery of Quantum Materials. Nature Materials 25, 223-230.
Download Paper