Update, same day: a DFT check (loop 3) supports Fe₃Ge₂Al but not Ti₃Fe₅B₂, which sits 32 meV/atom above TiFe₂ + TiB₂ + Fe on PBE. Read the other rows as unconfirmed.
Loop 2 kept the same fast funnel as loop 1
Of 3,793 Fe-rich ternary cells, 3,459 ran and 177 came within 25 meV/atom of the machine-learned hull. 43 of those are already in Alexandria or GNoME. Of the 134 that are new to the databases, 9 sit on the hull, and after removing two more bcc decorations and counting ordered Laves phases for what they are, the list below is what we would stand behind as stable, unique and novel on this energy surface.
Compound | Structure | e_hull (meV/atom, Orb v3) | Phonons (min, THz) | Novelty | What it is |
|---|---|---|---|---|---|
Every value is for the cell relaxed with Orb v3 on Ouro; the hull is the Orb v3 surface with Materials Project, Alexandria, GNoME and Ouro's own structures as references. "New composition" means no structure with that formula in MP, Alexandria, GNoME or COD. For Fe₃Ge₂Al and Ti₃Fe₅B₂ we also ran two web searches each and found no report; that is a statement about four searches. All five cells are ours (prototypes from Materials Project, CC BY 4.0).
How much weight each deserves:
Fe₃Ge₂Al is the one we find most interesting: a six-atom hexagonal cell on the hull, built on the same kagome net as FeGe. Whether Al orders on one Ge site or simply dissolves is not something this screen can say.
Ti₃Fe₅B₂ is the only tetragonal one. It reads 1.5 meV above the hull only because a slightly lower cell of the same formula was already in Ouro's own reference set (generated or submitted on Ouro earlier; we did not trace by whom); no published database has it.
Nb₄Fe₇Si and Nb₄VFe₇ are ordered versions of a Laves phase that is known to dissolve Si and V. Four more of the same kind are on the hull (Nb₄Fe₇Ge, Nb₄GaFe₇, Nb₄Fe₇Co, MnNb₄Fe₇). They are new as ordered compounds, and probably solid solutions in practice.
Zr₂NbFe₉ is the same story in the ZrFe₃ family.
Cells | |
|---|---|
Built (ternary, Fe ≥ 50 at.%, true structure types only) | 3,793 |
Hull runs that finished | 3,459 |
Within 25 meV/atom | 177 |
Already in Alexandria or GNoME | 43 |
By crystal class, the 177 near-hull cells are 108 trigonal, 47 hexagonal, 15 cubic and 5 tetragonal. Iron-rich ternaries near the hull are mostly hexagonal and trigonal stackings (Laves, AB₃, 2:17-like); tetragonal ones are rare.
We predicted 0 to 4 new true-type ternaries on the hull out of about 3,700, 20 to 80 new ones within 25 meV, and that at least half of the on-hull cells would already be in a database. Result: 7 new on the hull after removing the two bcc decorations (over the range, but 6 of the 7 are Laves orderings, so 1 if those are set aside), 134 new within 25 meV (over the range), and 11 of 20 on-hull cells already known (as predicted). The databases cover ternaries more densely than the quaternaries of loop 1, but the ternary sweep still found more, because it was four times larger and no longer limited to tetragonal cells.
334 of the 3,793 cells have no result: 156 failed at submission (we raised the submission rate to eight at a time) and 178 runs errored. They are unknown, not negative.
The known compound ZrFe₁₀Si₂ came back as "new composition" at 3.9 meV. It has been made (Gabay and Hadjipanayis 2016) but is in none of the four databases. A database-novelty check is not a literature check.
The hull route also rediscovered Nb₂Fe₁₂P₇ and Ti₂Fe₁₂P₇ on the hull; both are GNoME entries.
A DFT check of Fe₃Ge₂Al and Ti₃Fe₅B₂, including Fe₃Ge₂Al against disordered Al on the Ge sites.
Rerun the 334 cells with no result.
Relax the Fe ≥ 50 at.% rule to 40 at.%: several well-known tetragonal families (ThCr₂Si₂ and its relatives) sit just under it.
Fe3Ge2Al, P-6m2, CoSn-type FeGe (kagome Fe net) with Al ordered on one Ge site. Built by substitution into a Materials Project prototype and relaxed with Orb v3 on Ouro. Energy above hull 0.0 meV/atom on the Orb v3 surface; no imaginary phonon modes; absent from MP, Alexandria, GNoME and COD. ML-only: not checked with DFT.
Ti3Fe5B2, P4/mbm, Ti3Co5B2 type. Built by substitution into a Materials Project prototype and relaxed with Orb v3 on Ouro. Energy above hull 1.5 meV/atom on the Orb v3 surface; no imaginary phonon modes; absent from MP, Alexandria, GNoME and COD. ML-only: not checked with DFT.
P-6m2, 6 atoms |
0.0 |
stable, 0.26 |
new composition |
CoSn-type FeGe (a kagome iron net) with Al ordered on one of the three Ge sites |
P4/mbm, 20 atoms | stable, 0.39 | new composition | The Fe member of the Ti₃Co₅B₂ type; the parent of loop 1's Ti₂MnFe₅B₂ |
P-3m1, 12 atoms | stable, 0.20 | new composition | C14 Laves NbFe₂ with Si ordered on one Fe site in eight |
P-3m1, 12 atoms | stable, 0.18 | new composition | C14 Laves NbFe₂ with V ordered on one Fe site in eight |
P-3m1, 24 atoms | 1.6 | stable, 0.16 | new composition | Hexagonal AB₃ stacking (the ZrFe₃ family) with Nb ordered on one Zr site in three |
134 |
New and on the hull | 9 |
New, within 25 meV, and an ordering of a known Fe binary (mostly Laves) | 62 |
New and tetragonal, within 25 meV | 4 |
Ternaries only, every high-symmetry class, ordered alloys filtered out first: 134 database-new cells within 25 meV of the ML hull, most of them ordered Laves phases, and five we would stand behind.