AI and the World's Hard Problems · Chemistry & materials · Nature Machine Intelligence · 2026-10-23
Efficient equivariant framework for universal charge density prediction
Density functional theory is computationally expensive, which limits electronic structure calculations used in materials discovery. The authors present EdenGNN, a machine learning model that predicts charge density and augmentation occupancies, and EdenGNN-Uni, a universal version trained on the Materials Project database for non-magnetic structures across the periodic table. The model is more than an order of magnitude faster and accurately predicts band structures for most unseen materials, showing greater robustness than universal ML Hamiltonian methods.