Tetrataenite (L1₀ FeNi) is a good magnet that will not form: it only orders below 593 K, where the atoms barely move. This loop asked whether a third element can make the ordered arrangement more favourable, so that it would order at a higher temperature.
No new candidate. All 17 substitutional elements we tried, on either the Fe or the Ni layer, and carbon and boron in the holes, reduce the energy FeNi gains by ordering; none adds to it.
The energy gained by ordering is 62 meV per atom, and our fast ML potential and DFT agree on it to 0.1 meV, so the screen can be trusted on the main question. Titanium, zirconium and niobium on the Fe layer are the nearest to harmless, but they are also the ones that precipitate as Ni₃X. Cobalt, which a published calculation says raises the ordering temperature, lowers the ordering energy here on both methods. What we now know is that composition alone is not a way to give tetrataenite more reason to order; what is not tested is how fast the atoms move, which no instrument we have can measure.
Each additive X was put on one site of a 32-atom FeNi cell, once in the ordered L1₀ arrangement and once on 12 sites of three random arrangements, and every cell was relaxed and scored with Orb v3 through the hull route (618 cells). The number reported is how much ordering energy the whole cell gains or loses per X atom: negative means X is better off in the random alloy than an Fe or Ni atom would be. Pure FeNi and the Co and Ti cells were then repeated in DFT (PBE, DZP basis, k-spacing 0.25, single points on the Orb geometries).
Cell | Ordering energy, Orb (meV/atom) | Ordering energy, PBE (meV/atom) |
|---|---|---|
FeNi, random cell 1 against L1₀ | 62.6 | |
Additive, site | Change in ordering energy per X atom, Orb (meV, mean of 12 sites) | Same, PBE at one site (meV) | Hull of the ordered cell with one X in 32 (meV/atom, Orb) |
|---|---|---|---|
Zr on Fe | −11 ± 37 |
The other elements and sites all read lower: Cr, Sn, Ga, Zn, Ge, Cu, Pd and Pt on the Fe layer between −214 and −339; Pd, Cu, Mn, Cr, Zn, Si, Ga, Ge, Al, V, Sn and Mo on the Ni layer between −127 and −479; Ti, Nb and Zr on the Ni layer −623 to −739. The ± is the standard error over the 12 random sites. Each value in the first column is built from 13 hull runs; the linked run is the ordered cell, and the rest are in the notebook. One cell of 619 (V on a Ni site of a random cell) never returned and is not counted.
The two extra-Ni and extra-Fe rows are the yardstick. Making the alloy slightly off 1:1 costs 250–310 meV of ordering energy per misplaced atom, so most additives lose less than plain off-stoichiometry does. None gains.
At one additive atom in eight, in an ordered arrangement, only Pt on the Ni layer (6.8 meV/atom above the hull) and Co on the Fe layer (14.9) stay within 20 meV/atom; Al on the Fe layer is at 37.9 and comes out as a Ni–Al compound. Magnetisation on PBE: 1.71 T for FeNi ordered or random, 1.70 T with one Co in 32, 1.57 T with one Ti in 32.
The ordered cell with one Ti on an Fe-layer site (Fe₁₅TiNi₁₆, Orb-relaxed), the least harmful cheap additive. Fe and Ni alternate in layers along the long axis; the Ti has eight Ni neighbours.
Input to a PBE single point for the L1₀ FeNi ordering-energy calibration, loop 41. Geometry from the Orb hull screen E1174.
The additives the literature says raise the ordering temperature (Co, Al, Ti on the Fe layer) prefer the ordered host: refuted for Co and Al, undecided for Ti. Predicted +0 to +150 meV for Co, +50 to +300 for Al, +50 to +400 for Ti. Measured −144, −187 and −27 on Orb; on PBE Co is −153 and Ti +68 at the one site checked. The pure-FeNi ordering energy came out inside its predicted range on both methods.
Aluminium leaves ordered FeNi as a Ni–Al phase once there is an eighth of it: supported. Predicted 25–90 meV/atom above the hull; measured 37.9–54.1, and 9.4 at one in 32.
An element that bonds to Ni much more than to Fe (Si, Ga, Ge, Zn, Sn), on the Fe layer, stabilises order more than Co, Al or Ti (no precedent in the literature): refuted. Predicted +150 to +500 for the best of the five; measured −237 to −312, no better than extra Ni. The reasoning does hold for Ti, Zr and Nb, whose energy falls 65–113 meV with each extra Ni neighbour, and those three are the nearest to neutral.
Carbon or boron in the Fe-layer hole holds the layering in place, as nitrogen does in FeNiN (no precedent): refuted, as expected. Predicted −400 to +100; measured −532 and −614. Inside ordered FeNi both prefer the hole in the Ni layer, and fully filled FeNiC and FeNiB are 303
The ordering energy of FeNi is something we can now measure for free: Orb and PBE agree on it, and on the sign and rough size of an additive's effect (within 30–80 meV per additive atom).
Alloying is not a route to a larger ordering energy in tetrataenite. An additive can only be chosen to lose less of it than off-stoichiometry would.
The elements that sit best in the ordered phase are the ones that most want Ni neighbours, and for the same reason they precipitate as Ni₃Ti, Ni₃Nb or Ni₃Zr. The two properties cannot be separated in this set.
A published mean-field calculation (Tian et al. 2021, doi:10.1038/s41598-021-84482-5) finds Co, Al and Ti raising the ordering temperature by up to 60 K. Our numbers are energies at zero temperature, not ordering temperatures, so they do not contradict that directly; they say the rise does not come from the energy.
Unchanged: the slowness of Fe and Ni below 593 K, which is the other half of the problem. Nothing here measures it.
Earlier loops: tetrataenite as a calibration of the anisotropy route and the Fe–Co–Ni layerings were loops 11–14 (run privately at the time).
FeNi, random cell 2
FeNi, random cell 3 |
Ti on Fe | −27 ± 21 | +68 (Orb at that site: −9) |
Nb on Fe | −40 ± 26 |
Pt on Ni | −81 ± 5 |
Co on Ni | −88 ± 8 |
Mn on Fe | −103 ± 3 |
Mo on Fe | −113 ± 16 |
V on Fe | −118 ± 6 |
Co on Fe | −144 ± 12 | −153 (Orb at that site: −182) |
Al on Fe | −187 ± 8 |
Si on Fe | −312 ± 17 |
extra Ni on Fe (no additive) | −313 ± 17 |
extra Fe on Ni (no additive) | −246 ± 25 |
C in a hole | −532 (against the best of 18 random holes) |
B in a hole | −614 (same) |
17 substitutional elements and two interstitials in L1₀ FeNi: every one lowers the energy gained by ordering; Ti, Zr and Nb on the Fe site come closest to neutral and then leave as Ni₃X.