High-throughput discovery of high Curie point two-dimensional ferromagnetic materials

Databases for two-dimensional materials host numerous ferromagnetic materials without Curie temperatures because computing Tc requires manually intensive workflows. The authors develop a fully automated, hardware-accelerated, dynamic-translation-based code that performs first-principles calculations followed by Heisenberg-model Monte Carlo simulations to estimate Tc directly from crystal structures. Applying this to a database of 786 materials, they discover 26 with Tc above 400 K. They further build an end-to-end machine learning model using generalized chemical features and an exhaustive model/hyperparameter search to rapidly predict Tc, discovering additional candidates from other sources. The material informatics results agree with recent experiments and are expected to foster practical applications of 2D magnetism.