E248–E255 — the review's experiments: one per open question, parameters fixed (registered 2026-09-27 22:50)

EXPERIMENTS.md · lines 17327–17333

E248–E255 — the review's experiments: one per open question, parameters fixed (registered 2026-09-27 22:50)

The external review (Jan, 2026-09-26) asked questions that could show the ladder is wrong. The ones not already answered become these eight experiments, run one at a time in this order. Each gets its own entry, with bars, before it runs; the parameters and predictions below are fixed now. Already settled: R0 is a ranking (J1–J2); the Mo–Ta reference is corrected (J5); the requirement is one disordered phase (J14, E247).

EXPERIMENTS.md · lines 17365–17376

E251 — Ordering energies against DFT on matched cells, unrelaxed and relaxed (J4, J7, J12.)

• Compositions, first round: Mo₅₀Ta₅₀, Mo₂₅Nb₂₅Ta₂₅W₂₅, Mo₃₅Ta₃₃W₂₂Nb₅Ti₅ (the E245c case that ordered at 1044 K), Mo₃₅Ti₃₁Nb₂₄Ta₅W₅ (Ti-rich), Mo₃₅Ta₂₂W₂₂Nb₇V₆Hf₆ (V and Hf) and Cr–Ta–Ti–W (Cr).
• Per composition, 54-atom cells: the annealed ground state (E249), the next two distinct low-energy states from the anneals, and three random cells with identical counts (random_cell.py --counts).
• DFT: Quantum ESPRESSO, PBE, 60/720 Ry, k 4×4×4, Marzari–Vanderbilt 0.02 Ry, ideal sites at the Vegard constant, one SCF each. Then the ground state and one random cell relaxed (vc-relax; forces < 10⁻³ Ry/bohr).
• Measured: per-cell errors, paired ΔE_ord errors, how much relaxation changes ΔE_ord, and whether the relaxed cells stay bcc.
• Predictions: the model overstates ΔE_ord on at least four of six (as on the 16-atom sets); relaxation lowers ΔE_ord by 5–30 %; every relaxed ground state stays bcc. ~130 A100-hours on Colab.