Rapid estimation of synthesizability windows of inorganic materials from first principles

Abstract

Fast prediction of the synthesizability conditions of materials remains challenging, even assuming synthesis under thermodynamic equilibrium. We combine density functional theory (DFT) with machine-learned interatomic potentials to enable high-throughput generation of phase predominance diagrams as a function of temperature and partial pressures of the gaseous reactants. These diagrams can immediately be used by experimentalists to translate computational predictions into real synthesis parameters in the lab. Predominance diagrams are generated for a diverse set of binary compounds and for 48 more complex ternary metal phosphosulfide systems, but the method is in principle scalable to any inorganic material class. The calculated predominance diagrams generally show good agreement with the experimental synthesis literature, with a drastic reduction in computational cost compared to a full DFT approach. We find several examples of compounds that appear as metastable in a zero-temperature stability hull picture, but that become thermodynamically stable under well-defined synthesis windows.