Retargeted from ranking Fe-W prediction errors (none exist beyond the alpha-Fe anchor) to the structure that actually failed: the validated WFe2 C14 prototype, CIF 0db9981c-a143-4523-9fda-88b0de667edb.
One matched pair, frozen v1 settings otherwise unchanged:
Arm A: FM seed, all Fe moments +2.0 uB
Arm B: antiparallel seed, alternating +/-2.0 uB on the 6h Fe sublattice
ecutwfc 50 Ry, DZP, PBE, kspacing 0.3 1/A, scf_thr 1e-6 Ha, scf_nmax 200, mixing_beta 0.2, mixing_beta_mag 0.05, smearing mp 0.05 eV
Quantity | Arm A (FM seed) | Arm B (AFM seed) |
|---|---|---|
SCF convergence | FAIL at step 200 (charge sloshing on Fe majority band) | FAIL at step 200 (moment oscillation, amplitude 0.4 uB) |
Total energy at last step (Ha/f.u.) | -1284.31 | -1284.29 |
Per-site moments at last step (uB) | Fe: 1.8-2.4 spread | Fe: +/-1.5-2.6 oscillating |
Ms (mu_0 T) | not reportable | not reportable |
Does magnetic-state initialization rescue the terminal SCF failure? No. Both seeds terminate at scf_nmax 200 without meeting 1e-6 Ha. The failure mode is electronic (small C14 interband gaps near E_F with mp smearing 0.05 eV), not seed-dependent.
Does state choice change the result? Not meaningfully. Last-step energies differ by 0.02 Ha/f.u. (~0.5 eV/f.u.), far above the 1e-6 Ha convergence threshold, so the difference is unconverged noise, not a physical FM-vs-AFM splitting. No valid ordering conclusion is possible from this pair.
The WFe2 C14 prototype remains a pipeline-invalid case at frozen v1 settings. The two-temperature bound contemplated in the calibration plan is NOT computable from this pair. A future arm with reduced mixing_beta (0.05/0.02) or increased smearing (0.2 eV) with a convergence-first protocol would be required, which is outside the frozen v1 contract and therefore not attempted here.
CIF: 0db9981c-a143-4523-9fda-88b0de667edb (validated WFe2 C14 prototype)
Both arms logged as distinct actions with terminal-step diagnostics attached