427 Ir(III) complexes · two data surfaces (see below) · Uniconf torsional seeds (MK≤3000) → GFN2-xTB geometry optimization at the correct per-complex charge/spin → RMSD (0.5 Å) + energy (0.1 kcal/mol) dedup · CREST low-energy ensemble coverage →
8,777
Canonical source-faithful geometries
Every conformer geometry in the source ensembles across all 427 complexes (conformers.xyz) — the full source geometry set the GFN2-xTB single-point workflow evaluates, at the correct per-complex charge/spin. This is the source total, not a completeness claim.
6,285
Browser / QM-refined entries
Distinct conformers served in this browser: GFN2-xTB-optimized coordination-valid minima (RMSD 0.5 Å + energy 0.1 kcal/mol deduped), plus single-point fallback ensembles for the coordination-caveat and Architector-rescued complexes.
Coordination: 426/427 browser records are coordination-valid.
Complex 5750 is a documented source-topology caveat — its source structure has only 4–5 Ir donors within bonding distance (one Ir–donor elongated to ~2.56–2.63 Å) and no valid 6-coordinate octahedral form exists, so it is flagged coordination_ok=False and served with its actual-topology geometry plus GFN2-xTB single-point energies. This is not a 427/427 octahedral claim.
CREST cross-validation reports empirical coverage of the sampled low-energy ensemble — it does not prove that every mathematically possible conformer has been enumerated.
The per-conformer energies below are GFN2-xTB electronic energies (lowest first) — a collection-wide electronic ordering, not a free-energy ordering. For the 282 decision-critical complexes (near-degenerate electronic top-1/top-2, gap < 1 kcal/mol) a subset-only CENSO xtb-backend ΔG re-ranking (implicit-water Gsolv + GFN2 bhess mRRHO) is shown on the card and flagged with a ΔG badge (118 change their top conformer under ΔG). Non-subset complexes remain GFN2-xTB electronic only.
427
Complexes
6285
Total Conformers
14.7
Avg / Complex
327
With ≥3 Conformers
87
Rigid (1 conformer)
Conformer generation for Ir(III) C,N-chelate complexes.
Method: Uniconf v1.0.1 (UFF/NLOPT) seeds.
Then GFN2-xTB geometry optimization at the correct per-complex charge/spin, with RMSD+energy deduplication to distinct minima.
SDF sorted by GFN2-xTB energy (lowest first).
Regenerate with scripts/build_zenith_conformers.py.
Independent CREST / GFN2-xTB conformer sampling on the GAP-004 cross-validation subset — run on Hive T07 with per-complex charge/spin from charge_uhf_map.json.
25 complexes attempted: 21 produced CREST ensembles (1974 total frames), 4 timed out with no crest_conformers.xyz.
Every complex here is also in the distinct-conformer set above. Full validation report →
25
Complexes attempted
21
With CREST ensembles
4
No output (timeout)
1974
Total CREST frames
7/18
squick / mquick
CREST metadynamics conformer search (--squick / -mquick) at GFN2-xTB. Each card renders the full CREST trajectory — drag the slider to step through conformers; the iridium centre is the enlarged teal sphere. Energies are GFN2-xTB total energies (Hartree). Source: results/crest_crossval_gap004_20/.