CHGNet

A universal graph neural network force field trained on 1.5 million structures to predict multi-element material stability with DFT-level precision.

CHGNet (Crystal Hamiltonian Graph network) represents a breakthrough in materials informatics by incorporating formal oxidation states into a universal machine learning force field. Trained on the massive Materials Project database (encompassing 10 years of density functional theory calculations), it handles 94 elements and predicts energy, forces, stresses, and magnetic moments simultaneously. Researchers use it to bypass expensive quantum mechanical simulations, enabling the rapid screening of lithium-ion battery cathodes and high-entropy alloys at speeds 1,000 times faster than traditional methods.

https://github.com/CederGroupHub/CHGNet