ML-guided discovery of La-Co-Pb ternary compounds involving immiscible pairs of Co and Pb elements

Abstract

Ternary compounds with an immiscible pair of elements are relatively unexplored but promising for novel quantum materials discovery. Exploring what third element and its ratio that can be added to make stable ternary compounds out of an immiscible pair of elements remains a great challenge because the number of possible combinatorial combinations of chemical compositions and crystal structures is enormous. In this work, we combine a machine learning (ML) method with ab initio calculations to efficiently search for the energetically favorable ternary La-Co-Pb compounds containing immiscible elements Co and Pb. Three previously reported structures are correctly captured by our approach even though these structures are excluded from the database for ML. Moreover, we predict a new stable La3CoPb compound and 57 low-energy La-Co-Pb ternary compounds whose formation energies are less than 100 meV/atom above the convex hull. Attempts to synthesize La3CoPb via multiple techniques ranging from arc melting and annealing, solid state reactions, and high pressure synthesis produce mixed or multi-phases samples with, at best, ambiguous signals of the predicted lowest-energy La3CoPb and the second lowest-energy La18Co28Pb3 phases.