Accelerated discovery of Cr-based A2+B2 superalloys across 11 elements with a deep-learning CALPHAD surrogate

Chromium is an attractive platform for next-generation refractory alloys, but its practical use is limited by room-temperature brittleness. Here, we present a multi-criteria discovery workflow for Cr-based A2+B2 superalloys. A deep-neural-network surrogate reproduces CALPHAD phase fractions and A2-matrix chemistries with a per-field-averaged root-mean-square error below 2 percentage points and provides a speed-up exceeding 10,000×, enabling thermodynamic screening of >108 compositions in an 11-element design space. Surrogate-feasible compositions are then revalidated by direct CALPHAD. Targeted experiments provide a feasibility check: of eight synthesized compositions, six homogenize to single-phase A2 and three form the intended A2+B2 microstructure upon aging, underscoring both the potential of accelerated screening and the sensitivity of predictions near the A2–B2 ordering boundary. We further assess two screening-friendly indicators for Cr-based A2 matrices: an edge-dislocation strength model and valence electron concentration (VEC). The strength indicator captures the overall experimental strength scale but not the composition-dependent ranking, implicating screw-dislocation strengthening as a missing contribution. Density functional theory calculations show that, in Cr-rich alloys, increasing VEC raises the Rice intrinsic ductility parameter (favoring dislocation emission over cleavage) by increasing the Fermi-level d-state density, either through a rigid-band shift or through partial pseudogap filling. These results provide an electronic-structure rationale consistent with the Re-like ductilizing effect reported in Cr alloys and support the use of VEC as an intrinsic-ductility indicator in Cr-rich A2 phases.