AI-accelerated quantum exploration of high-anisotropy magnetic materials MnBiX

We present a hybrid machine-learning and quantum-mechanical framework for discovering and optimizing MnBiX compounds (X = O, F, ...) with high magnetic anisotropy energy (MAE). A crystal graph convolutional neural network (CGCNN) predicts the MAE directly from crystal structures and compositional descriptors, enabling rapid exploration of a large ternary MnBiX chemical space. We validate promising low-symmetry candidate compounds selected from the model predictions using density functional theory (DFT), computing their magnetocrystalline anisotropy constants, saturation magnetizations, and energies above the convex hull. We then feed these first-principles results back into the ML model to further improve its predictive performance. This iterative workflow accelerates the design of rare-earth-free magnetic materials exhibiting strong anisotropy and high magnetization, thereby supporting the development of sustainable hard magnets for energy applications.