Discovering electronic signatures for phase stability of intermetallics via machine learning

In this paper, we identify the signatures of the density of states (DOS) spectra which control the bulk modulus via a hybrid informatics driven analysis. The signatures of the DOS spectra then constitute the electronic structure fingerprint of the material. This provides an important step in the “inverse design” process because if we are able to compute bulk modulus from the DOS, then we can also compute the DOS from the bulk modulus, and in this way create a “virtual” DOS based on optimized properties. In this paper, we identify the signatures for bulk modulus, and associate the signatures with specific chemistry and crystal structure. Further, we identify the details in the electronic structure that result in Ni₃Al and Co₃Al having such different stabilities in L1₂ structure although they are seemingly isoelectronic. This paper lays out the methodology for extracting these features and has significant implications, such as in the identification of critical element substitutions, by developing a framework for accelerated and targeted materials design.