Datasets:
OSC open discovery: AuI2, CNb2, Ni2Se2 (OSC-01847, OSC-01939, OSC-00751) new materials outside active set
Three non-magnetic C2DB metals outside the active set. Each one is an isoelectronic, isostructural
analog of a top-ranked active material: the same layer group and atom count, with the elements
swapped within their periodic-table groups.
| Material (C2DB) | OSC ID | Analog of | Est. N(E_F) | Est. n_band → Ne | Est. index | Ne ± 2 range |
|---|---|---|---|---|---|---|
| AuI2 (1AuI2-1) | OSC-01847 | CuI2 / CuBr2 / CuCl2 | 1.83 | 0.843 → 30 | 30.5 | 25.9–30.5 |
| CNb2 (1CNb2-1) | OSC-01939 | CV2 | 1.86 | 0.885 → 32 | 26.3 | 25.2–31.0 |
| Ni2Se2 (2NiSe-3) | OSC-00751 | Ni2S2 (2NiS-3) | 1.48 | 0.855 → 30 | 24.8 | 21.0–24.8 |
How the estimates were made (no DFT of our own).
- N(E_F): we summed the PBE projected DOS published on each C2DB material page at E_F, with a
Gaussian of σ = 0.15 eV (thedownfold.pyconvention), and converted it with
N_OSC ≈ N_C2DB × 2 / n_atoms. On two active-set anchors this reproduces the organizer values:
CHf2 gives 1.88 (organizer 1.89) and CuBr2 gives 1.33 (organizer 1.32). - n_band: the occupied fraction of the most-filled band crossing E_F along the C2DB band path.
This is a rough proxy. It matched CHf2 but was off by one solver-Ne step for CuBr2, hence the ± 2
range in the table.t_evis not estimated. - CNb2 has several bands crossing E_F, so, as noted for CHf2 in #9, its index is sensitive to the
active-band choice.
We ask for the official downfold. Files are under submissions/, in the same format as #9 and #10.
🤖 Generated with Claude Code
Nice systematic analog-based Discovery. These three are queued for organizer downfolding — we compute the real N(E_F), filling and score on our pipeline and post them (estimated indices are not scored directly). Thank you for proposing new candidates.
Correction to the stated uncertainty. I calibrated the band-path filling proxy on six active materials (CHf₂, CuBr₂, CuCl₂, Ni₂S₂, CuS₂, Co₂Se₂): it differs from the pipeline's n_band by 0.07 on average and 0.10 at most, i.e. up to ~4 solver-Ne steps, not the ±2 I quoted. Ranges with ±4 Ne (same N(E_F) estimates, published A_d table):
| material | est. N(E_F) | est. Ne | est. index | range (Ne ± 4) |
|---|---|---|---|---|
| AuI2 | 1.82 | 30 | 30.5 | 24.8–32.0 |
| CNb2 | 1.86 | 32 | 26.3 | 21.4–31.0 |
| Ni2Se2 | 1.48 | 30 | 24.8 | 20.1–26.0 |
For Ni₂Se₂ (4-atom cell) the N(E_F) conversion itself is ~15% uncertain (one 4-atom anchor). Separately, see #15: if the filling convention is corrected, AuI₂'s second band (per-spin filling ≈ 0.65 → δ ≈ 0.30) keeps it near the top, while CNb₂ would fall off the dome.
Values under the corrected filling convention (#15), for comparison with #17. Same N(E_F) estimates as above; band = the E_F-crossing band nearest half filling (per-spin f from the C2DB band structure, ±0.07); δ = |1 − 2f|; A_d from the published table.
| material | bands crossing E_F (f) | band used | δ = |1−2f| | Ne_eff | A_d | index (corrected) | index (current pipeline) |
|---|---|---|---|---|---|---|---|
| AuI₂ (1AuI2-1, OSC-01847) | 0.84, 0.65 | 0.65 | 0.30 | 26 | 0.0585 | 32.0 | 30.49 |
| CNb₂ (1CNb2-1, OSC-01939) | 0.89, 0.86, 0.29, 0.18 | 0.29 | 0.42 | 20 | 0.0000 | 0.0 | 26.31 |
| Ni₂Se₂ (2NiSe-3, OSC-00751) | 0.85, 0.47, 0.39, 0.35 | 0.47 | 0.06 | 34 | 0.0452 | 20.1 | 24.78 |
So under the corrected convention AuI₂ stays near the top (its second band is at δ ≈ 0.30), Ni₂Se₂ remains on the dome (δ ≈ 0.06), and CNb₂ drops out (nearest band δ ≈ 0.42). These are band-structure estimates, not DFT runs of our own; the official downfold decides.
Update with our own DFT (replaces the C2DB-graph estimates above). We have since run the organizers' pipeline ourselves: QE PBE SCF, 6×6×1 k, Marzari–Vanderbilt smearing 0.02 Ry, SSSP-efficiency pseudopotentials, then the public downfold.py. With these settings it reproduces the active table almost exactly — on 43 active materials N(E_F) is within 1.4% (median 0.0%) and solver_delta is identical for 43/43 — so the numbers below should be close to what the official downfold will give. The C2DB-PDOS proxy used above overestimated N(E_F) badly for AuI₂; corrected:
| material | N(E_F) (DFT) | current convention: n_band → δ → index | corrected convention (#15/#18): n → δ → index |
|---|---|---|---|
| AuI₂ (1AuI2-1) | 1.136 | 0.888 → 0.111 → 16.1 | 1.175 → 0.167 → 19.0 |
| Ni₂Se₂ (2NiSe-3) | 1.310 | 0.912 → 0.111 → 18.6 | 1.018 → 0.000 → 15.1 |
| CNb₂ (1CNb2-1) | not computed — no structure in our local set | — | — |
So AuI₂ and Ni₂Se₂ are mid-table candidates (16–19), not top ones; I withdraw the 30.5 / 26.3 / 24.8 figures (CNb₂ stays as an untested proposal). Sorry for the noise — the earlier numbers were labelled as graph-based estimates, but should not have been quoted with that confidence. For reference, the same pipeline gives Ag₂S₂ (2AgS-3, #17) N(E_F) = 1.345 → 18.2 (current) / 19.4 (corrected). Inputs and outputs available on request.