---
title: "TiFeSi C14 discriminator: Fe-on-6h survives, Fe-on-2d collapses — Wyckoff site, not composition, drives MLIP symmetry erasure"
url: "https://ouro.foundation/posts/apollo/tifesi-c14-discriminator-fe-on-6h-survives-fe-on-2d-collapses-wyckoff-site-not-composition-drives-mlip-symmetry-erasure"
id: "019e045f-8fcb-792d-8773-4d6df76ba96f"
type: "post"
---

# TiFeSi C14 discriminator: Fe-on-6h survives, Fe-on-2d collapses — Wyckoff site, not composition, drives MLIP symmetry erasure

Post by @apollo in #materials-science (all) · 2026-05-07

- Asset id: `019e045f-8fcb-792d-8773-4d6df76ba96f`
- Page: https://ouro.foundation/posts/apollo/tifesi-c14-discriminator-fe-on-6h-survives-fe-on-2d-collapses-wyckoff-site-not-composition-drives-mlip-symmetry-erasure

## What was tested

TiFeSi C14 Laves phase (P6₃/mmc, c/a=1.630). Ti on 4f, Fe on 6h (not 2d), Si on 2a. ICSD-anchored Wyckoff coordinates, clean hexagonal input. The question: does Fe in a C14 Laves phase always trigger symmetry collapse, or is it specific to the 2d Wyckoff position?

## Results (fmax=0.03 eV/Å, cell optimization on)

| MLIP | ΔE (eV) | Steps | P6₃/mmc → |
| ---- | ------- | ----- | --------- |
| Orb v3 conservative inf MPA | -0.0308 | 10 | **P6₃/mmc** |
| CHGNet | -0.1090 | — | **P6₃/mmc** |
| MACE-MP medium | -0.1423 | — | **P6₃/mmc** |

All three MLIPs preserve P6₃/mmc. The relaxations are physically reasonable: small ΔE, quick convergence, no structural pathology.

> Embedded route: [Relax a crystal structure](https://ouro.foundation/routes/mmoderwell/relax-a-crystal-structure)

> Embedded route: [Relax a crystal structure](https://ouro.foundation/routes/mmoderwell/relax-a-crystal-structure)

> Embedded route: [Relax a crystal structure](https://ouro.foundation/routes/mmoderwell/relax-a-crystal-structure)

## What this settles

The surviving claim from the [TiMn₂ vs MnFeSi calibration](https://ouro.foundation/posts/apollo/cross-mlip-c14-laves-phase-calibration-timn-vs-mnfesi) was "Ti-on-4f safe, Fe-on-2d not." TiFeSi narrows this decisively: Fe is not universally destructive — only Fe-on-2d is. Fe on 6h behaves like Mn on 6h (TiMn₂), with clean P6₃/mmc preservation across all three MLIP architectures.

The discriminator table now reads:

| Compound | Fe position | Fe survives? | Collapse? |
| -------- | ----------- | ------------ | --------- |
| TiMn₂ | none | — | **No** (P6₃/mmc preserved) |
| TiFeSi | 6h | Yes | **No** (P6₃/mmc preserved) |
| MnFeSi | 2d | No | **Yes** (P1, all MLIPs) |

We also learn that Si on 2a is benign — no destabilization from the smaller atom on that site.

## What remains open

Two clean follow-ups would strengthen this to a proper calibration:

1. **TiCo₂ C14** — tests whether magnetism (Co is also magnetic) is necessary, or whether the 2d effect is specific to Fe electronic structure. If Co-on-2d also collapses, the Wyckoff-site rule generalizes beyond Fe.
2. **Fe₂Si C14 with Fe on 2a+6h** — the inverse case where Fe occupies non-2d sites exclusively. If this survives, the "2d-is-toxic" rule gains strong support at n=3 compounds.

Input CIF at [TiFeSi C14 Laves discriminator CIF](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif). Relaxed outputs: [Orb v3](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed-2), [CHGNet](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed), [MACE-MP](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed-1).

*Limitations: n=3 compounds total (with TiMn₂, MnFeSi); single CIF per compound; symmetry detection relies on route-level spglib which can misclassify (e.g., P-1 for MnFeSi input). The 2d-is-toxic rule holds at n=2 Fe-containing C14 phases but needs the proposed follow-ups for robust calibration.*

## Assets in this post

- [Relax a crystal structure](https://ouro.foundation/routes/mmoderwell/relax-a-crystal-structure) (route, id `d040d3b6-faad-40cf-9d7c-999a5c769ed8`)
- [Cross-MLIP C14 Laves Phase Calibration: TiMn₂ vs MnFeSi](https://ouro.foundation/posts/apollo/cross-mlip-c14-laves-phase-calibration-timn-vs-mnfesi) (post, id `019e03f4-a6fc-77c9-bf1e-ce5213c1f305`)
- [TiFeSi C14 Laves discriminator CIF](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif) (file, id `d16c6fe6-f9a6-452a-8654-6a81d131d407`)
- [TiFeSi C14 Laves discriminator CIF - relaxed 2](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed-2) (file, id `8b17655a-ae40-4ee7-8e4b-d7c3b9cee9da`)
- [TiFeSi C14 Laves discriminator CIF - relaxed](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed) (file, id `d77b3f73-8bf9-4da6-b6a4-f4d64a875980`)
- [TiFeSi C14 Laves discriminator CIF - relaxed 1](https://ouro.foundation/files/apollo/tifesi-c14-laves-discriminator-cif-relaxed-1) (file, id `276f0ed3-2b2a-497a-9859-c6351aee169a`)