This loop asked whether iron and zirconium with a third element hold a rare-earth-free permanent magnet that the campaign has not already measured.
No new candidate. One cell clears both magnetic targets, Zr₂Fe₇CoB₂ with cobalt on the in-plane iron site at 1.93 MJ/m³ and 1.09 T, but cobalt is 51 meV lower in energy on the other iron site, and there the same composition has no easy axis.
The parent, ZrFe₄B, was computed this year by another group as a 1.25 MJ/m³ magnet. On our routes its anisotropy is zero at the magnetisation our calculation gives it and passes 1 MJ/m³ only if the true magnetisation is about 5 % lower, which nobody has measured; the published value and ours are points on one curve. Of 48 substitutions none brings an iron-rich cell nearer to stability than the parent, which sits about 42–50 meV/atom above iron, ZrFe₃ and ZrB₂. What the loop adds is a lever: the in-plane iron site decides the anisotropy, and a cobalt there supplies an easy axis that silicon buys only by giving up magnetisation and nickel does not supply at all. Yttrium on the zirconium sites brings the cell toward stability and takes the easy axis away at the same rate. Whether a real alloy, with cobalt spread over both iron sites, keeps enough anisotropy is the open question. The campaign's standing results are unchanged: Mn₁₆Al₁₅Ti and ZrFe₉Co₂Ti.
PBE, DZP basis, k-spacing 0.2 Å⁻¹, each cell on its own DFT-relaxed geometry at its own moment unless a held moment is given; anisotropy is the energy with the moment in the plane minus along c, at a band-sum width of 0.1 eV, with 0.05 eV in brackets.
Cell | Geometry | Novelty | Moment (μB per 12 atoms) | Js (T) | K₁ (MJ/m³) | Orb hull (meV/atom) |
|---|---|---|---|---|---|---|
Targets: one easy axis with K₁ ≥ 1.0 MJ/m³, Js ≥ 1.0 T, within 75 meV/atom of the hull. The in-plane cobalt cell is phonon-stable, costs 13.5 $/kg in raw materials and is 68 meV/atom up against published PBE references. Its two cobalt arrangements differ by 51 meV per cobalt in the same DFT setup, the between-layers one lower. The anisotropy route has no calibration in this structure type: the made member, LuFe₄B, is reported easy-axis, and our DFT does not yet support Lu.
ZrFe₄B with its moment held, same relaxed cell:
Moment held (μB per 12 atoms) | Js (T) | K₁ (MJ/m³) |
|---|---|---|
12.5 | 1.00 | +1.41 (+1.51) |
Stability of ZrFe₄B, seven PBE single points in one setup: +42 meV/atom above Fe + ZrFe₃ + ZrB₂, where published PBE gives +47 for the same reaction and +50 against its own hull. The setup reproduces a published reaction among these phases to 6 meV/atom where ZrFe₂ and Fe₂B stay out of it and misses by 24–36 where one enters. ZrFe₄B is phonon-stable and costs 10.6 $/kg. Swapping one boron with one in-plane iron costs 1.7 eV per cell on the Orb potential, so the ordered cell is the right model.
Zr₂Fe₇CoB₂ with cobalt in the plane of one Zr layer, the cell that reads 1.93 MJ/m³. Iron sits on two kinds of site: two in the plane of a Zr (here one is cobalt) and six between the layers; boron is in the plane of the other Zr.
The same composition with cobalt between the layers, 51 meV lower and with no easy axis.
Cell + ionic relaxation with ABACUS DFT; 0.04 eV/Å threshold; final energy = -30943.7946 eV; energy change = -0.0220 eV
The substitution screen, 48 cells relaxed and placed on the hull with the Orb potential (free): hafnium for zirconium raises the cell by 11–15 meV/atom, titanium, niobium and magnesium by 50 or more; cobalt, nickel, manganese, aluminium, copper and chromium on the in-plane iron site raise it by 15–60; carbon, silicon, aluminium, gallium, germanium and phosphorus in place of boron by 45–140. Only silicon or gallium on both in-plane iron sites comes down, with iron at half the atoms.
The structure type is stable with a larger atom than zirconium, and yttrium is the cheap one (31 $/kg on the cost route's table, against 57.5 for neodymium). These cells are on Orb-relaxed geometries, so their anisotropy is a first reading, not a converged value. Hull distances are Orb, with published-PBE references in brackets.
Cell | Js (T) | K₁ (MJ/m³) | Hull (meV/atom) |
|---|---|---|---|
Yttrium takes the site in the iron-only layer by 358 meV. With cobalt in the plane, going from no yttrium to half to all moves the hull distance from 57 to 40 to 22 meV/atom and the anisotropy from +1.93 to +1.07 to −0.15 MJ/m³: no composition on that line is both near the hull and above 1 MJ/m³. Yttrium also brings in Y₂Fe₁₄B as the competing phase, a known easy-axis magnet. One thing improves: cobalt's preference for the between-layers site falls from 38 meV in the zirconium cell to 12 in the yttrium one (Orb).
ZrFe₄B reads easy-axis at 1 MJ/m³ and 1 T on our route: refuted at the free moment. Predicted +0.5 to +1.6; measured 0.00.
Its anisotropy is set by its moment, and at the published moment it is easy-axis: supported. All five held-moment points fell inside their predicted ranges. The paper (Cheng et al., J. Phys. D 59, 2026, doi:10.1088/1361-6463/ae71de) reports +1.25 at 1.08 T, and in its own scan +0.17 at 1.15 T and −0.51 at 1.19 T.
A substitution brings the family within 25 meV/atom at 1 T: refuted. Predicted the best iron-majority cell at 28–45 on published-PBE references; it is 59, level with the parent.
Silicon on half the in-plane iron gives an easy axis at the cell's own moment: half held. +0.99 at 0.90 T: easy-axis, short on magnetisation. With silicon on both sites, +2.25 at 0.46 T.
Cobalt on the in-plane site gives an easy axis at 1 T at the cell's own moment: supported for that cell. Predicted +0.8 to +2.0 at 1.02–1.10 T; measured +1.93 at 1.085 T. Predicted cobalt would prefer the other site by 20–80 meV; it does, by 51, and reads nothing there. Nickel, the control, was predicted to help and does not.
Yttrium brings the cell toward the hull and leaves it easy-axis: refuted on the anisotropy. The half-yttrium cell was predicted at 15–35 meV/atom and −0.4 to +0.6 MJ/m³; it is 34 and −0.52. Every yttrium reading came in below its predicted range.
In Fe–Zr–X the databases hold nothing iron-rich and uniaxial near the hull beyond what earlier loops measured: across 24 third elements every such cell is a Laves-type ordering, soft or easy-plane.
ZrFe₄B is the second iron-only family here, after the 1:12 cells, whose easy axis turns on a few per cent of moment. A single calculated anisotropy for such a cell is one point on a curve.
The in-plane site is the lever. Cobalt there adds an easy axis of its own; removing the iron for silicon helps less and costs magnetisation; nickel does nothing.
Lowering the cell's moment by alloying is not the same as holding it down in the calculation: cobalt between the layers lowers the moment by one μB and leaves the anisotropy at zero.
Stability and anisotropy pull against each other on the large-atom site: yttrium buys the first with the second.
Zr₂Fe₇CoB₂ is not put forward as a candidate. The cell that meets the targets is not the lowest arrangement of its own composition; at an annealing temperature roughly one cobalt in six would sit on the in-plane site. Its Curie temperature was not run, its stability rests on the Orb hull and published references, and the anisotropy route is uncalibrated in this structure type. What would change this: a larger cell with cobalt on both sites in thermal proportion that still reads above 1 MJ/m³.
Earlier loops: loop 37
computed-only (Cheng 2026) |
14.26 |
1.14 |
0.00 (−0.07) |
13.50 |
1.09 |
+1.93 (+2.27) |
57 |
same composition |
13.25 |
1.07 |
+0.15 and −0.03, no easy axis |
not checked |
12.45 |
1.00 |
+0.01 (−0.05) |
54 |
11.16 |
0.90 |
+0.99 (+1.01) |
not checked |
5.60 |
0.46 |
+2.25 (+2.38) |
1.04 |
+1.29 (+1.58) |
13.5 | 1.08 | +0.96 (+1.18) |
13.75 | 1.10 | +0.73 (+0.86) |
14.0 | 1.12 | +0.44 (+0.51) |
Cell + ionic relaxation with ABACUS DFT; 0.04 eV/Å threshold; final energy = -30943.7437 eV; energy change = -0.0124 eV; not fully force/stress converged
1.16 |
−0.52, easy-plane |
34 (41) |
1.18 | −0.91, easy-plane | 8 (16) |
1.11 | 40 (46) |
1.12 | −0.15 (−0.35) | 22 (29) |
Fe–Zr with a third element: the layered boride ZrFe₄B read on our routes, substituted on every site, and the one substitution that makes it an easy-axis magnet.