The question: if we put one light atom (H, B, C, N or O) into the empty holes of binary compounds that someone has already made, do we get ternary compounds that are stable, and that nobody has reported?
New candidate: In₄BPt₆ (Pt₆In₄B), a boride of the double-kagome metal Pt₃In₂. It sits 11.9 meV/atom below every competing phase in our PBE calculation and has no unstable phonon modes. It is in no database, and we found no report of it in the literature. It meets every target on measurement. Two checks are still open: a tighter basis or a full DFT relaxation, and whether it can be made. A second cell, Nb₃BRh₉, is 14.9 meV/atom below its hull, but a related cubic NbRh₃Bₓ has been made, so its novelty is still open. This is the campaign's first loop, so there is no earlier best to compare with.
Why: we filled every suitable hole in 927 made binary intermetallics, 5,106 cells in all, and screened them with a machine-learned potential. 395 came within 25 meV/atom of the hull. Most were already in a database, or belong to families chemists have filled for decades: stuffed Mn₅Si₃- and Cr₅B₃-type phases, antiperovskites, η-carbides. The few that came from hosts nobody had filled were nearly all borides. Three of them went through a DFT check. That check was first tested on published reactions in each chemical system, and two of the three cleared its error by a wide margin. What we learned is where this search pays: boron in the octahedral holes of late-metal-rich hosts. It does not pay in the families the screen ranks highest.
Screen: Orb v3 relaxation and hull distance on the hull route, every reference on the same surface. DFT check: PBE single points in ABACUS (DZP basis, k-spacing 0.2 Å⁻¹, non-spin-polarised) on the screened cell and on every competing phase both hulls name. Each balance uses compounds only. Phonons: NequIP-OAM-XL, 2×2×1 supercell.
Cell | Geometry | Novelty | Orb hull (meV/atom) | PBE vs compound hull (meV/atom) | Decomposes into | Imaginary modes |
|---|---|---|---|---|---|---|
Targets: three elements, at or below the PBE hull, new to the databases and the literature, and no imaginary phonon modes.
How far to trust the DFT column: before reading each margin, we reran published reactions among the competing phases in the same setup. In Nb–Rh–B it reproduces them to within 1 meV/atom. In In–Pt–B it is within 0–6 meV/atom, apart from the order of two BPt₂ polymorphs. In Tl–Pt it is 14 meV/atom off. So the In and Nb margins hold, and Tl₄BPt₆'s −2.7 is not settled either way.
In₄BPt₆: Pt₃In₂ is a double-layer kagome metal of the Fe₃Sn₂ family. Look at the boron atom between the two platinum kagome sheets, in a Pt₆ octahedron. There is one boron for every two Pt₃In₂.
target in a PBE stability check; lab/notebook/campaigns/sun-ternary/work/e002/relaxed/In4BPt6mp-510439v1_B.cif. Primitive cell, sites in the route's read order.
Nb₃BRh₉: boron in one Rh₆ octahedron in three, in the hexagonal stacking of made NbRh₃.
target in a PBE stability check; lab/notebook/campaigns/sun-ternary/work/e002/relaxed/Nb3BRh9mp-1191559v2_B.cif. Primitive cell, sites in the route's read order.
Filling holes puts cells near the hull at a rate of a few percent, mostly in families already known to take interstitials: mixed. The rate was higher than predicted: 7.9 % within 25 meV/atom (we said 1–5 %) and 3.0 % on the hull. The family part held, with 234 of the 395 on Mn₅Si₃/Cr₅B₃-type hosts. The filler part failed: boron gave the fewest near-hull cells (12 of 933), not more than oxygen.
At least one cell from a host nobody has filled is new and below the PBE hull: supported. Two of three are, by more than the setup's error. All three are phonon-stable.
Other boron arrangements in the same hosts are higher: supported, in part. Boron in any other hole of these hosts costs 30–127 meV/atom more on the screen, and half-filled cells sit 13–54 above the hull. Mixed arrangements across hole types were not enumerated.
Filling holes finds new stable compounds where substitution stopped finding them, but only away from the families the screen ranks highest. Those are mostly known.
Boron in octahedral holes of platinum- and rhodium-rich hosts is the lever worth following. The next test is the other Pt₃In₂-type double-kagome hosts.
The DFT check's error depends on the chemistry, from 1 to 20 meV/atom. It has to be measured in each system before a margin is quoted.
In₄BPt₆ is a calculation, not a synthesis.
Its margin comes from single points at the machine-learned geometry, in a mid-sized basis. A full DFT relaxation would most likely lower the cell further, but has not been run.
We have not checked whether Pt₃In₂ itself is stable only in a narrow temperature window, or what boron does on cooling.
Nb₃BRh₉ could turn out to be one stacking of the reported cubic NbRh₃Bₓ phase. That paper has not been read at source.
Tl₄BPt₆ needs a calibration that holds in Tl–Pt.
Earlier SUN work: the Fe-rich SUN campaign's carbides
Orb-relaxed |
4 InPt + BPt₂ |
Orb-relaxed | Nb₂BRh₆ + NbRh₃ |
Orb-relaxed | 4 TlPt + BPt₂ |
5,106 filled binary cells screened; In₄BPt₆ and Nb₃BRh₉ sit below the PBE hull and are phonon-stable.