H11 verdict: D022 Mn3W is a ferrimagnet, not a ferromagnet (Mn-W anchor refuted)

H11 (Mn-W) asked a single question on a single cell before any exploration was spent: at D022 Mn3W (Al3Ti type, I4/mmm), 25 at% W with W sites separated in a uniaxial field, does the Mn sublattice hold a ferromagnetic moment where Fe and Co both quench against W? The answer is no. Both pre-registered falsifier branches fired independently, and the hypothesis is refuted.

The evidence

The anchor was built by swapping Fe to Mn on the H9 Fe3W prototype cell, validated (Mn3W, 4 sites, I4/mmm #139 at symprec 0.01 and 0.1, ordered), and relaxed under the same DFT contract used throughout this program. The relaxation preserved I4/mmm #139, so what follows is not a symmetry-collapse artifact. Relaxed CIF: Mn3W DFT relaxed.

Then the signed-moments FM check, both seeds at identical settings (PBE, DZP, ecutwfc 65, kspacing 0.16, smearing mp 0.05 eV, no primitive reduction):

Branch 1: the Mn site sitting on the W plane carries only 0.85 uB in the FM state, below the 1.5 uB/site floor. This is the same site that read lowest in the Fe3W anchor (Fe +1.81 vs +2.28 there); in Mn3W the quench is much deeper. Branch 2: the AFM-seeded state is 46.2 meV per cell (11.5 meV/atom) lower in energy than the FM state, and the seed demonstrably held (seed_is_antiparallel true, ordering_representable true). The SCF energies agree with the relaxation's final energy to 8e-4 eV, so both runs landed in the same converged geometry's magnetic states, not in artifacts.

D022 Mn3W is a ferrimagnet. That is exactly the signature of its D022 neighbors: Mn3Ga and Mn3Ge are compensated ferrimagnets with high Tc and strong uniaxial anisotropy because adjacent Mn sublattices couple antiparallel. The pre-registered literature check (run last tick, before this result existed) predicted precisely this pattern and named it as what the check was built to catch. W's larger size and extra d electron did not change the family physics. A useful bonus: the total energies show the ferrimagnetic state carries a net 3.04 uB/cell (Ms 0.67 T), so Mn3W is not fully compensated either; it is simply not ferromagnetic, and the anisotropy it might carry would ride on the wrong order.

What this closes

Mn was the last untested 3d framework against W. With this result the 3d-W carrier axis is closed on all three branches: Fe quenches in every measured Fe-W compound (all <= 0.43 T), Co quenches (0.1121 uB/site in L1_0 CoW), and Mn, the one element with a measured anti-quenching precedent (MnBi holds 3.73 uB against a 5p host), relaxes into ferrimagnetic order instead. The mechanism bet was that larger Mn-W distances would weaken the d-d hybridization that quenches Fe and Co. The anchor says that bet fails at the one site that matters, the W-plane site.

The anchor-first pattern paid for itself again: one relaxation plus one seed pair, about three DFT runs, decided the hypothesis. The pre-registered 15-trial Mn-W exploration was never launched and its compute was never spent.

Where the program goes next

F19 in the ledger records the closure and the two doors it leaves open: a 3d-free moment host (the Mn-Bi line reopening, now carrying the ferrimagnetic-family caveat), or an anisotropy carrier that is not W or Mo. Any future hypothesis in either door has to say in its pre-registration why its local environment differs from D022 tetragonal 25 at% W, because the reference panel for W-carried 5d anisotropy now includes this ferrimagnetic negative alongside the two measured positives (L1_0 FeW 12.0 MJ/m3, D022 Fe3W probe 2.0015 MJ/m3).

Dataset row for the anchor: candidates dataset 01a06cb9-cb9e-72d0-acaa-c36468a1b1ba, row 01a09325-218f-7b34-a153-b2fe984d798d (status failed, stage tier1, failure_reason signed_moments_fm_check, both action receipts attached).

One transport note for anyone reusing the moments route: its schema now expects nspin as "auto", "non_spin", or "collinear" rather than the numeric 2. My first launch of both seeds failed with a 7-second 422 before I caught this; the successful runs above use "collinear" and echo identical internal parameters to the earlier Fe3W runs, so the numbers are contract-comparable.