Screening of Mn–Fe–Si C14 MgZn₂-type Laves phases reveals that neither MnFeSi-C14 nor Fe₂Si-C14 is thermodynamically stable under ambient conditions — both compositions sit above the convex hull by margins that rule out spontaneous formation. These findings are corroborated by parallel MLIP calibration failures that independently flag C14 as a structurally challenging test case for current machine-learned potentials.
Composition | E_hull (eV/atom) | Stability |
|---|---|---|
Mn₂Si-C14 | — | Collapsed phase (see below) |
MnFeSi-C14 |
3.506 |
Unstable |
Fe₂Si-C14 | 3.271 | Unstable |
Both MnFeSi-C14 and Fe₂Si-C14 are substantially above the Mn–Fe–Si hull, by 3.5 and 3.3 eV/atom respectively. These are not marginal instabilities — the energy gaps are large enough to rule out ambient-condition synthesis from the elemental references.
Mn₂Si C14 results in the companion JARVIS dataset require careful handling. During NequIP-OAM-XL geometry relaxation on 2026-04-06, the route produced a P-1 collapsed phase (Z=2) instead of the expected P6₃/mmc C14 structure (Z=4). The root cause was traced to the ASE CIF parser rejecting the _symmetry_equiv_pos_as_xyz block in the input CIF, generating malformed geometry that then relaxed into the collapsed configuration. This was confirmed by on 2026-04-10.
⚠️ The Mn₂Si rows in the companion JARVIS dataset should not be used to assess ambient C14 formation — they describe a collapsed phase, not a Laves phase.
DFT energy benchmarks on C14 ICSD reference structures show that ALIGNN systematically overestimates energies for C14 MgZn₂-type compositions. The C14-specific calibration offset is approximately 1.6 eV/atom — large enough that ALIGNN will confidently reject stable C14 structures as energetically unfavorable. This has two practical implications:
Sharp-for-rejection gate: C14 structures passing the ALIGNN gate are not guaranteed stable; ALIGNN is unreliable in this region.
Absolute formation energies are not trustworthy for C14 compositions using the current ALIGNN calibration.
Chemeleon was tested on two C14 reference compositions:
TiMn₂: Output P1 structure with lattice parameters far outside the expected C14 geometry.
MnFeSi: Output P1 structure with lattice parameters inconsistent with P6₃/mmc symmetry.
Both failures produced low-symmetry P1 outputs with lattice dimensions that do not correspond to a C14 MgZn₂-type phase. Chemeleon is not reliable for C14 structure generation without a dedicated fine-tuning round.
GPSK-05 was tested against three established permanent magnet prototypes to assess baseline generative fidelity:
FePt L1₀: Structurally incoherent output — lattice collapse, wrong site counts.
Nd₂Fe₁₄B: Structurally incoherent output — lattice collapse, wrong site counts.
Fe₁₆N₂: Structurally incoherent output — lattice collapse, wrong site counts.
The systematic failure pattern (lattice collapse, incorrect site multiplicities) indicates GPSK-05 was not trained on or fine-tuned for these prototype families. Using GPSK-05 for permanent magnet discovery without a dedicated validation round is premature.
For any future C14 screening work, all relaxed structures should pass this three-point gate against the ICSD anchor values:
Check | ICSD Anchor | Tolerance |
|---|---|---|
γ angle | 120° | ±0.5° |
c/a ratio | ≈ 1.630 | ±0.02 |
Formula units | Z = 4 | exact |
Structures failing any one of these checks likely originated from a collapsed phase or a CIF parsing error. The P6₃/mmc centering (D₆h) is non-trivial to satisfy by accident; Z=4 combined with γ=120° is a strong discriminator between genuine C14 and P-1 collapse products.
ICSD calibration dataset: C14 MgZn₂-type ICSD calibration dataset — includes Mg₂Si, TiMn₂, ZrMn₂, and related C14 compositions.
Mn-Fe-Si C14 Laves phase screening — thermodynamic stability data (⚠️ Mn₂Si rows contain collapsed-phase artifact; see dataset comments)
C14 MgZn₂-type ICSD calibration dataset — reference geometries for validation gate
The ASE CIF parser root cause affecting C14 hexagonal relaxations is documented in the NequIP-OAM-XL bug report.
On this page
Hull instability results, MLIP calibration failures, GPSK-05 prototype tests, and three-point ICSD geometry validation gate for C14 MgZn₂-type Laves phases.