We asked what the crystal databases already hold in iron, cobalt, nickel, manganese, boron and zirconium that nobody has read as a magnet, and whether any of it is a hard one.
New candidate: Mn₂FeB₂. Every target is measured on its own relaxed cell and met, the Curie temperature only just, with three checks still open: 1.5–1.8 MJ/m³ along a single easy axis at 1.4–1.5 T, about 430–455 K against a 450 K target, no cobalt, 1.62 $/kg. It is level with the campaign's best on anisotropy and well ahead on magnetization, and it sits 32 meV/atom above the hull, so it is a metastable target, not a furnace compound.
It is a tetragonal boride in which manganese takes the large, boron-rich site and iron runs in chains along the unique axis. The cell is in a database as a computed entry; its anisotropy had not been computed and the compound has not been made, so this is a new reading of a known hypothetical, not a new composition. The reading held when we moved to the database's own geometry, held the moment at the database's lower value, flipped the iron or the manganese (both cost 0.6 eV or more), and swapped an iron with a manganese (0.44 eV). Still open: our small basis set has read another iron boride up to 1.75 times too large, which would put this at 1.05–1.23; no compound of the type has a measured anisotropy to calibrate on; and nothing says it can be made. On the way, the loop found that MnCoB is on the hull at 1.5 T and is not a hard magnet, and that a quick reading on a machine-learned geometry overstated a sister compound fourfold.
PBE with a DZP numerical basis, k-spacing 0.2 Å⁻¹, each cell relaxed in DFT from a ferromagnetic start before the anisotropy route read it. K is the energy of the named in-plane direction minus the easy axis; for the two orthorhombic cells the first number is the gap to the middle axis, which is the one that matters.
Cell | Type | Geometry | K (MJ/m³) | Js (T) | Orb hull (meV/atom) | PBE hull (meV/atom) |
|---|---|---|---|---|---|---|
Targets: one easy axis, K ≥ 1.0 MJ/m³, Js ≥ 1.0 T, hull within 75 meV/atom, Curie temperature ≥ 450 K, raw cost ≤ 20 $/kg. All six cells are entries or decorations of entries in Alexandria (CC-BY-4.0); none has been made. Mn₄CoNiB₄ is a composition no database holds.
For Mn₂FeB₂ beyond the table: K is +1.82 at the database's plane-wave geometry and +1.54 at 1.386 T with the moment held at the database's value; the plane is isotropic (100 and 110 agree to 0.001). The ferromagnet is 1.15 eV
Mn₂FeB₂ on its relaxed cell: manganese in pairs in the layer at z = ½; iron and the boron pairs in the layer at z = 0, with iron at the corners and the face centre stacking into chains along c, the easy axis.
Cell + ionic relaxation with ABACUS DFT; 0.04 eV/Å threshold; final energy = -18991.4396 eV; energy change = -0.0429 eV
MnCoB, the cell on the hull: the 2.9 Å axis is the hard one, and the long axis is easy by only 0.31 MJ/m³ over the third.
Cell + ionic relaxation with ABACUS DFT; 0.04 eV/Å threshold; final energy = -43237.4378 eV; energy change = -0.0241 eV
Ordered Mn–(Fe,Co,Ni) borides hold near-hull cells the databases do not: supported. Of 457 decorations of seven boride types, seven formulas no database holds are within 25 meV/atom on the machine-learned hull (predicted one to five). None beats its parent, and the parents turned out to be the interesting cells.
MnCoB is a ferromagnet near 1.5 T: supported. Reversing manganese costs 0.52–0.70 eV per atom, far above the 0.02–0.15 predicted; 1.52 T.
The MnCoB type has an easy-to-middle gap of 1 MJ/m³: refuted. 0.31 on the relaxed parent (predicted 0.2–0.8, with one chance in five of 1.0), and no single-site substitution of nine reaches 0.5.
Ferromagnetic Mn₂CoB₂ is a 1.5 MJ/m³ magnet on its own cell: refuted. It is the ground state and it is uniaxial, but it reads 0.46 relaxed where a quick reading on the machine-learned geometry had said 1.98. It is also 40 meV/atom above MnCoB + MnB in one DFT setup.
Mn₂FeB₂ keeps 1.5 MJ/m³ through its state and geometry checks: supported so far. 1.54–1.84 across three settings, ferromagnetic against every state finished, phonons clean. One state is still running (manganese layers alternating along c).
Band filling on the cobalt site of cementite-type MnCo₂B reaches 1 MJ/m³: not tested. Nickel dissolves there cheaply and iron does not; no mixed cell was read.
Manganese on the boron-rich site of an ordered boride is a reliable 1.4–1.6 T ferromagnet in four structure types. Only the tetragonal one gives a hard axis, and there the atom in the chains sets it: iron 1.84, cobalt 0.46, cobalt with nickel 1.28, nickel 0.96.
A database hull distance for a manganese compound can describe the wrong state: Mn₂CoB₂ is stored non-magnetic at 71 meV/atom, and the ferromagnet is a different cell 4 % larger.
An anisotropy reading on a machine-learned geometry ranks cells for relaxation and nothing more in this chemistry. Mn₂FeB₂ is the counter-case: it reads the same at two geometries.
Zirconium did not earn a place: with boron it goes cobalt-rich and non-magnetic.
Not a calibrated number. No compound of this type has a measured anisotropy, and our basis has read tetragonal Fe₂B 1.5–1.75 times too large; at that ratio Mn₂FeB₂ is 1.05–1.23 MJ/m³. A larger basis is the next run. The made relatives MoMn₂B₂ and WMn₂B₂ (Adeniji et al. 2026, doi:10.1002/anie.3669169, ferromagnets at 400 and 380 K) would give the type its first calibration.
Not stable. 32 meV/atom above MnB + Fe₂B + Mn₂B on the database's PBE, and those three all dissolve the other metal. Our own stability setup cannot check it, because Mn₂B converges to the wrong magnetic state there.
Not new as a formula. The cell is one Alexandria entry; what is new is the reading.
Not a coercivity, and not a finite-temperature anisotropy. The Curie figure is at the target line, not above it.
Earlier: loop 52
Mo₂FeB₂, P4/mbm |
+1.84, easy c |
1.489 |
32.4 (database) |
Mo₂FeB₂, P4/m | +1.28, easy c | 1.243 | not run | not run |
Mo₂FeB₂, P4/mbm | +0.96, easy c | 1.069 | not run |
Mo₂FeB₂, P4/mbm | +0.46, easy c | 1.599 | +40.4 (ours) |
cementite, Pnma | 0.50 / 1.25, easy a | 1.568 | 15.8 (database) |
own type, Pmma | 0.31 / 0.60, easy a | 1.519 | −12.2 (ours) |
A computed-only database cell read as a magnet for the first time: one easy axis, ferromagnetic ground state, phonon-stable, 32 meV/atom above the hull. MnCoB is on the hull but soft; Mn₂CoB₂ did not survive relaxation.