学术文献库

The Heusler compounds have provided a playground of material candidates for various technological applications based on their highly diverse and tunable properties, controlled by chemical composition and crystal structure. However, physical exploration of the Heusler chemical space en masse is impossible in practice, hindering the exploration of the chemical composition vs. proprieties relationship. Many of these applications are related to the Heuslers electron transport characteristics, which are embedded in their electronic band structure (EBS). Here we we created a Heuslers dataset using the Materials Project (MP) database -- retrieving both chemical composition and their EBSs. We then used machine learning to develop a model correlating the composition vs. number of Dirac points in the EBS for Heuslers and also other Cubic compounds by identifying said Dirac points using an automated algorithm as well as generating chemical composition and global crystal structure features. Our ML model captures the overall trend, as well as identifies significant electronic and global crystal structure features, however, the ML model suffered from significant variance due to the lack of site specific features. Future work on a methodology for handling atomic site specific features will allow ML models to better match the underlying quantum mechanics governing the properties (also based on site specific properties) and capture the electronic properties in a more generalized approach.