Phase stability and thermodynamics of the Co-Se system from DFT-informed machine-learning potentials

The first CALPHAD-type thermodynamic assessment of the Co-Se system was developed in this work based on knowledge of the published experimental phase diagram, including invariant phase reactions. Due to the scarcity of liquid-phase thermochemical data, the mixing enthalpy of the liquid phase from a machine learning potential trained on ab initio molecular dynamics data was evaluated at 1000, 1300, and 1500 K and incorporated into a CALPHAD assessment along with the DFT enthalpies of formation of the CoSe, CoSe 2 , and Co 9 Se 8 phases. The resulting model reproduces the main features of the experimental phase diagram and provides reasonable agreement with the ab initio and experimental 𝛥 ⁢ 𝐻 𝑚 𝑖 𝑥 values. At 298 K, the calculated composition dependence of the formation enthalpy of the Co-Se phases is consistent with available literature data, supporting the reliability of the optimized Gibbs-energy functions across the whole composition range.