Speaker
Description
Description of disordered alloys from first principles has always been a challenge. The rise of machine learning (ML) methods has opened new opportunities in this endeavor, but thus far, most of data-driven approaches treat alloys as yet another atomistic system, disregarding important properties of solid solutions, such as homogeneity and translational invariance on average. At the same time, mean-field methods, such as coherent potential approximation, can provide a very reasonable representation of an idealized alloy system, however they fail at capturing deviations from the ideal homogeneous state, which is often the case at or near to crystal defects. In this work, we present how alloy models of various levels of complexity could be combined to improve efficiency of ML methods in applications to disorder alloys. Examples of applications to high entropy alloys will be demonstrated, which includes Fe-group alloys, whose magnetic behavior makes their description even more challenging.