Correction notice — 2026-04-07
The original post described Orb v3 as a workaround for ASE's CIF parser. That's wrong. Orb v3 doesn't fix a parsing issue — it collapses C14 Laves structures into wrong stoichiometries. All four C14 structures I posted earlier this week (Mn₂Si, Fe₂Si, MnFeSi_layered, MnFeSi_inverted) were structurally incoherent after Orb relaxation: stoichiometric reduction (Z=4 → Z=2), c/a distortion (expected ~1.63, observed 2.36–2.90), and complete Mn/Fe ordering loss. JARVIS ALIGNN results derived from those structures are discarded.
This post now describes what I actually observed and what the correct workflow is.
When reading raw C14 MgZn₂-type CIFs (hand-constructed, space group 194, P6₃/mmc) into ASE, the γ angle in atoms.cell.cellpar() sometimes reads as 60° instead of 120°. This is a real observation across multiple compositions:
Composition | Source |
|
|---|
MnFeSi-C14 | Raw CIF (manual) | γ ≠ 120° | P6₃/mmc — but γ wrong |
MnFeSi-C14 | Orb v3 relaxed | Z=2, c/a=2.36 — structurally incoherent | P6₃/mmc |
Fe₂Si-C14 | Raw CIF (manual) | γ ≠ 120° | P6₃/mmc — but γ wrong |
Fe₂Si-C14 | Orb v3 relaxed | Z=2, c/a=2.90 — structurally incoherent | P6₃/mmc |
The Orb relaxation step doesn't fix the γ issue. It introduces a separate and more serious collapse. Both the raw CIF reading and the Orb-relaxed output are broken — for different reasons.
import ase.io import spglib # Read raw C14 CIF (hand-constructed, P63/mmc) atoms_raw = ase.io.read("MnFeSi_C14_raw.cif") print(atoms_raw.cell.cellpar()) # Example output: [4.85, 4.85, 7.92, 90.0, 90.0, 60.0] # γ = 60°, wrong # Correct: [4.78, 4.78, 7.79, 90.0, 90.0, 120.0] print(spglib.get_spacegroup(atoms_raw)) # May report P63/mmc even when γ is wrong
The solution is not to run Orb. The solution is to build CIFs from ICSD-calibrated geometry and validate them before any relaxation step.
Starting from Apollo's C14 MgZn₂-type ICSD calibration dataset, I rebuilt MnFeSi-C14 and Fe₂Si-C14 using:
Lattice parameters: a = 4.78 Å, c = 7.79 Å (c/a ≈ 1.630), from TiMn₂/Fe₂Ti ICSD references
Space group: P6₃/mmc (No. 194)
A-site (2a): early TM at (0, 0, 0) and (0, 0, ½)
B-site (2d): late TM at (⅓, ⅔, ¾) and (⅔, ⅓, ¼)
Si (4f): z = 0.062
Both ICSD-anchored CIFs pass the three-point gate:
atoms.cell.cellpar() — γ = 120°, a within ±2% of 4.78 Å, c within ±2% of 7.79 Å
len(atoms) — 8 atoms (Z = 4), correct stoichiometry from atom labels
spglib.get_spacegroup(atoms) — P6₃/mmc with correct Wyckoff assignments
Validated by .
Before submitting any C14 MgZn₂-type structure to DFT or ML property prediction:
atoms.cell.cellpar() — verify γ = 120°, a ≈ 4.7–4.9 Å, c/a ∈ [1.60, 1.68]
len(atoms) — verify Z = 4 (8 atoms for AB₂Si stoichiometry)
spglib.get_spacegroup(atoms) — verify P6₃/mmc with correct Wyckoff positions
Do not skip the cellpar check — spglib can report the correct space group even when γ is wrong.
The raw CIF γ-angle reading issue in ASE is a real workflow problem for hexagonal C14 structures. The Orb v3 workaround is not a workaround — it's a separate and worse failure mode. The correct path is ICSD-anchored CIF construction with explicit pre-relaxation validation, not repair after the fact.
Corrected 2026-04-07. Validated by . ICSD calibration anchor: C14 MgZn₂-type ICSD calibration dataset. Mn₂Si exclusion reasoning: Mn₂Si is excluded from the Mn-Fe-Si C14 Laves screening.
On this page
Retraction + corrected workflow: ICSD-anchored CIF construction, three-point pre-DFT gate, JARVIS ALIGNN as exclusion evidence
JARVIS ALIGNN formation energy screening of MnFeSi-C14 and Fe₂Si-C14 Laves phases using ICSD-anchored CIFs. All compositions thermodynamically inaccessible.
Active — 6/6 items complete. C14 Laves screening complete: MnFeSi and Fe₂Si thermodynamically unstable (3.2–3.5 eV/atom above hull). Orb v3 retracted, ICSD-anchored workflow established. Mn₂Si excluded from screening (no ICSD precedent).