Machine-learned interatomic potential for Fe–Co–X (X = Re, W) alloys: Ductility–brittleness behavior at finite temperatures

Fe–Co alloys combine high strength with poor ductility, limiting their structural applications at elevated temperatures. To overcome this limitation, we developed a machine-learned interatomic potential within the Neuroevolution Potential (NEP) framework for the Fe–Co–Re–W system, trained via active learning on ab initio and molecular dynamics datasets. The model reproduces DFT-level energies, forces, and elastic properties with high fidelity, enabling large-scale finite-temperature simulations. Across 300–1500 K, the elastic response follows a hierarchy of for stiffness and the inverse for ductility, with all phases satisfying the Born–Huang stability criteria. At 300 K, Re and W doping enhances ductility by increasing and , while FeCoW shows a pronounced increase in the elastic ductility indicators and approaches the empirical ductile–brittle thresholds around 1000 K. W induces stronger ductility in B2 and A1, whereas Re improves toughness in A2. This work provides an accurate and efficient framework for exploring and designing ductile Fe–Co–X (X = Re, W) high-temperature structural alloys within the tested conditions. • A machine-learned NEP interatomic potential is developed for the Fe-Co-Re-W alloy system. • The model achieves DFT-level accuracy for energies, forces, virials, and elastic constants. • Temperature-composition-structure coupling is mapped across B2, A2, and A1 phases up to 1500 K. • Re and W additives systematically enhance Fe-Co ductility, with Re improving A2 toughness most. • W induces stronger softening in B2 and A1 phases, enabling early ductile-brittle transition at high temperature.