[SY-D5] Understanding pairwise magnetic interactions in Fe-based materials with machine learning techniques
Magnetic interactions are crucial to the stability of structural phases as well as for various thermophysical effects such as magnetocalorics. Despite their importance, there is no experimental procedure which allows for the understanding of magnetic interactions at the atomic level, and there is no exact theoretical model capable of describing them precisely except for expensive ab initio methods. It has been recently suggested that the Heisenberg Landau model has sufficient versatility to map the contribution of complex magnetic interactions of Fe-based materials to the free energy. Its original form, however, contains a high number of parameters which make it prone to overfitting. In this study, we mapped the magnetic interactions created from spin-polarised DFT calculations to extended Heisenberg-Landau models via various machine learning regression techniques. The free energy contribution of the magnetic interactions is then determined through Monte Carlo simulations for millions of atoms which would otherwise not be achievable with ab initio methods. The results enable us to understand the overall effects of impurities contained within microstructures on the magnetism in iron based materials.