Predicting the stability of ternary intermetallics with density functional theory and machine learning

We use a combination of machine learning techniques and high-throughput density-functional theory calculations to explore ternary compounds with the AB₂C₂ composition. We chose the two most common intermetallic prototypes for this composition, namely, the tI10-CeAl₂Ga₂ and the tP10-FeMo₂B₂ structures. Our results suggest that there may be ∼10 times more stable compounds in these phases than previously known. These are mostly metallic and non-magnetic. While the use of machine learning reduces the overall calculation cost by around 75%, some limitations of its predictive power still exist, in particular, for compounds involving the second-row of the periodic table or magnetic elements.