Learning structure-property relationship in crystalline materials: A study of lanthanide–transition metal alloys

We have developed a descriptor named Orbital Field Matrix (OFM) for representing materialstructures in datasets of multi-element materials. The descriptor is based on the informationregarding atomic valence shell electrons and their coordination. In this work, we develop anextension of OFM called OFM1. We have shown that these descriptors are highly applicable inpredicting the physical properties of materials and in providing insights on the materials spaceby mapping into a low embedded dimensional space. Our experiments with transitionmetal/lanthanide metal alloys show that the local magnetic moments and formation energiescan be accurately reproduced using simple nearest-neighbor regression, thus confirming therelevance of our descriptors. Using kernel ridge regressions, we could accurately reproduceformation energies and local magnetic moments calculated based on first-principles, with meanabsolute errors of 0.03 B and 0.10 eV/atom, respectively. We show that meaningful low-dimensional representations can be extracted from the original descriptor using descriptivelearning algorithms. Intuitive prehension on the materials space, qualitative evaluation on thesimilarities in local structures or crystalline materials, and inference in the designing of newmaterials by element substitution can be performed effectively based on these low-dimensional representations.