Machine-learning-assisted prediction of magnetic double perovskites

Abstract

Magnetism is an important property of materials that plays a key role in many different applications. In the present paper, we use a combination of computational tools: a machine-learning technique for screening of stable candidates, an evolutionary algorithm for crystal structure determination, and first-principles calculations for characterization of electronic and magnetic properties to make predictions on magnetic double perovskites, which are yet to be synthesized. Out of 412 scanned candidates of ${A}_{2}{BB}^{\ensuremath{'}}{O}_{6}$ composition with $3d$ and $4d$ or $5d$ transition metals at B and ${B}^{\ensuremath{'}}$ sites, we found 33 compounds to form stable double-perovskite structures, 25 of which were further considered for characterization of their structure and properties. Our exercise predicted 21 double perovskites of varying magnetic and electronic properties, ranging from ferromagnetic half metals to ferri- and antiferromagnetic insulators to ferromagnetic metals and a rare example of antiferromagnetic metals. Our computational study is expected to help in discovering new magnetic double perovskites.