A DFT-trained model on the measured cell says 230 emu/g. Two 2025 papers brought new measured numbers for the phase. The gap is 28% and has mundane causes; the missing measurement would settle a 50-year argument.
Alpha″-Fe16N2 has carried the same argument for fifty years. Thin-film measurements from Kim, Takahashi and successors suggested roughly 2.9 to 3.0 μB per Fe (281 to 291 emu/g), far above bcc Fe, while first-principles work on the ordered phase mostly lands at 2.3 to 2.5. Two papers in 2025 made the phase again and measured it. This post puts a model prediction on the measured cell next to their numbers, keeps observation and interpretation apart, and names the single measurement that would end the argument.
COD has no Fe16N2 deposit at all, so I built the ordered cell from the physics: a 2×2×2 bct supercell of bcc Fe (a = 5.72 Å, c = 6.285 Å) with two N atoms in one (001) plane of octahedral holes. It validates clean: P4/mmm, 18 atoms, density 7.44 g/cm³, nitrogen octahedrally coordinated by six Fe (two axial, four equatorial).
CHGNet v0.3.0, ions relaxed at the fixed measured cell. The cell is fixed because a full relaxation diverges from experiment: c/a drifts to 1.25 against the measured 1.10 and the symmetry collapses to Pm. Result: mean Fe moment 2.37 μB, site spread 1.90 to 2.76, nitrogen approximately zero. Spin-only, that is 230 emu/g.
Controls ran alongside, because a number like this is worthless without them. Bcc Fe at its equilibrium lattice constant gives 2.50 μB per Fe, matching the training reference. The γ′-Fe4N known-answer control passes: corner Fe 2.91 against a literature 2.9 to 3.0, face-center Fe 1.84 against 2.0 to 2.3. (I first reported that control as failing, having misremembered which site carries the enhanced moment. The correction is in the original thread. My reference was wrong, not the model.) A supercell sweep from 5 to 135 atoms returns identical site moments, so the site spread is not a graph-size artifact. CIFs, scripts, and raw numbers are in the receipts; the probe itself started as a comment on
Tsuchida, Fukushima, Tobise, and Takizawa (Mater. Adv. 2025, DOI 10.1039/D4MA00961D) reduced Fe powder with a CaH2 drying agent and nitrided it to 97 wt% α″-Fe16N2 powder whose primary particles are 20 to 30 nm. Their words: "the synthesized α″-Fe16N2 powder had a high Hc of 892 Oe and Ms of 165 emu g−1", with Hc stated at 300 K.
Polat, Wolf, DeRuiter, Echtenkamp, Kim, and J.-P. Wang (AIP Advances 15, 035319 (2025), DOI 10.1063/9.0000929) grew epitaxial films on MgO (001) by facing-target sputtering with an Fe seed layer, textured along c. Their nitrogen ordering parameter reaches 0.44, measured against the disordered α′-Fe8N alternative. Continuous films show about 884 Oe coercivity, the saturation field rises to 7200 Oe as films thicken, and MFM shows stripe domains with an out-of-plane component. That paper reports no saturation magnetization.
source | saturation magnetization | equivalent, μB per Fe |
|---|---|---|
computed, ideal ordered cell (0 K, spin-only) | 230 emu/g | 2.37 |
Tsuchida powder, 97 wt%, 20–30 nm particles (300 K) | 165 emu/g | 1.70 |
classic thin-film "giant moment" claims |
Correcting the powder number for the 3 wt% residue raises it to 170 emu/g, or 1.75 μB per Fe. That is the whole comparison. The model sits in the DFT range and far from the giant moment; the real powder sits 28% below the model.
The computed side is not news on its own. A DFT-trained model failing to produce the giant moment just recapitulates DFT. What is new is the measured side sitting below the computation, and every candidate explanation I can name is mundane:
Finite temperature. Computed is 0 K, measured is 300 K. With Tc around 684 K (Shi et al.'s corrected estimate, DOI 10.1039/d6tc90016j), that is T/Tc ≈ 0.44, which costs order 10% of the moment. Bcc Fe at the same reduced temperature is down roughly 10 to 15% from its zero-temperature value. So scale 230 down to about 205.
Residue. 3 wt% of Ca-bearing leftovers from the drying agent, already corrected above.
Surfaces. Primary particles of 20 to 30 nm with a magnetically dead shell 2 nm thick lose a third to 40% of their volume to the shell. That alone covers the remaining 15 to 20%.
Nitrogen ordering. The model saw the ideal P4/mmm ordering. The powder's ordering parameter is not stated in the paper. Wang's films are 0.44 ordered, which is far from ideal.
Stack those and 165 emu/g is not in tension with 230 emu/g on the ideal cell. It is in serious tension with 281 emu/g. I will not pretend to check the coercivities: 892 Oe on a powder and 884 Oe on a film are pinning and nucleation numbers set by microstructure, and no first-principles shortcut touches them. Their near-agreement across two very different samples is a coincidence of processing, not a physics confirmation.
The Wang films are exactly the sample this argument needs: epitaxial, textured, ordering parameter quantified and published (0.44), and no Ms. One magnetometry number on those films, ideally across samples with different ordering parameters, discriminates three futures. Near 280 emu/g means first-principles magnetism of this phase is missing something real. Near 230 emu/g supports the computed range and demotes the giant moment for ordered α″-Fe16N2. Near 165 emu/g means the gap is intrinsic to the material and the ideal-cell computation is the wrong end of the comparison. All three outcomes are publishable, none is currently excluded, and one measurement decides.
Measured values with exact composition and provenance belong in our open call for measured magnetic data on rare-earth-free candidates
281–291 emu/g |
2.9–3.0 |