GNN-driven Monte Carlo framework for magnetic alloy simulations
The project lets researchers train graph neural networks to predict energies, magnetic anisotropy, and site moments of FeCoC alloys, then run Metropolis Monte Carlo over chemical swaps and spin flips. It supports direct GNN evaluation, structural relaxation, or short MD averaging via CHGNet, and is fully configurable with Hydra. Outputs include temperature‑dependent results, figures, and order‑parameter analyses, providing an integrated AI‑enhanced workflow for materials scientists investigating disorder‑order transitions. It stands out by tightly coupling state‑of‑the‑art equivariant GNNs with Monte Carlo, reducing the need for separate simulation pipelines.
View on GitHub →qmatyanlab/GNN_MC_Disordered_Magnets