AI Workflow Adds 1.3M Compounds to Materials Database

A groundbreaking multi-stage AI workflow for discovering computational materials has achieved a remarkable 99% success rate in pinpointing compounds within 100 meV/atom of thermodynamic stability, marking a threefold enhancement over earlier techniques. This innovative approach integrates the Matra-Genoa generative model, the Orb-v2 universal machine learning interatomic potential, and the ALIGNN graph neural network for energy forecasting. It produced 119 million candidate structures and contributed 1.3 million DFT-validated compounds to the ALEXANDRIA database, which now boasts 5.8 million structures, including 74,000 newly identified stable materials. The database features 175,000 compounds on the convex hull, with predicted structural disorder rates (37-43%) aligning with experimental databases, unlike other recent AI-generated collections. Analysis uncovers essential patterns in space group distributions and phase stability networks.