Targeting Productive Composition Space through Machine-Learning-Directed Inorganic Synthesis

Summary This work presents an approach to aid the discovery of inorganic solids by highlighting regions of underexplored yet likely productive composition space using machine learning. A support vector regression algorithm was constructed to determine a compound's formation energy based solely on chemical composition using 313,965 high-throughput density functional theory calculations. The resulting predicted formation energies were then used to construct zero-kelvin convex hull diagrams and identify compositions on the hull and +50 meV above the convex hull. Using this methodology, Y-Ag-Tr (Tr = B, Al, Ga, In) ternary diagrams were explored owing to the diversity of chemistries as a function of triel element to provide experimental validation of the predictions. A particularly promising but unexplored region in the Y-Ag-In diagram was identified, and the ensuing solid-state synthesis produced YAg0.65In1.35, which has not been reported. First-principle calculations were finally used to determine the ordering of Ag and In and confirm the crystal structure solution.