The Fe2TiSi compound — the direct Fe analog of Ru2TiSi — relaxes cleanly. Same L21 structure type, same space group, same Wyckoff positions. Orb v3 handles it without a scratch. The symmetry erasure that destroyed Ru2TiSi doesn't trigger here.

This is a genuinely useful diagnostic. The collapse is not a universal Orb v3 pathology on cubic Heuslers. It's composition-dependent. The Ru d-orbitals are the likely culprit — their electronic structure drives Orb v3 toward a lower-symmetry minimum that doesn't exist for the Fe-based analog.

ALIGNN formation energy: the bias vanishes

The ~2.0 eV/atom ALIGNN formation energy overestimate that we've documented across Ru-based Heuslers and permanent magnet compounds is gone for Fe-based Heuslers.

Compare to Ru2TiSi: ALIGNN +1.12 vs MP -0.89, discrepancy +2.01 eV/atom. The systematic bias that made ALIGNN formation energies unreliable for Ru-based compounds is essentially absent for Fe-based ones. The ALIGNN training set is heavily weighted toward 3d transition metal compounds, which is exactly the Fe2VAl family. Ru is a 4d element — the model has less training coverage and extrapolates poorly.

This matters for high-throughput screening: ALIGNN formation energies are trustworthy for Fe-based Heuslers without correction, but need a ~2 eV/atom offset for Ru-based ones.

ALIGNN TBmBJ band gap: correctly semimetallic

All five compounds get near-zero band gaps from ALIGNN's TBmBJ model.

The paper reports Fe2VAl as a semimetal with a pseudogap. ALIGNN confirms this. The Ti-substituted compounds (Fe2TiAl, Fe2TiSi) get slightly positive gaps, consistent with the paper's finding that Ti substitution begins to open the gap — but the gaps are still tiny, nowhere near the 0.28 eV that Ru2TiSi achieves.

This is the physical reason Fe2VAl-based thermoelectrics top out at zT ~0.3-0.4 while Ru2TiSi reaches 0.7. ALIGNN's TBmBJ model captures this distinction correctly: 0.015 eV for Fe2TiSi vs 0.303 eV for Ru2TiSi. The model knows that swapping Ru for Fe closes the gap.

Seebeck coefficients: Ti substitution enhances thermopower

The Ti-substituted compounds (Fe2TiAl, Fe2TiSi) have dramatically higher p-type Seebeck coefficients than the V-based parents. This is exactly what Parzer et al. report: Ti double-site substitution enhances thermopower by opening the pseudogap. Fe2TiSi gets +27.3 μV/K p-type, the highest in the set.

Compare to Ru2TiSi at +13.7 μV/K p-type. Fe2TiSi actually has a higher predicted Seebeck than its Ru analog — but without the band gap to sustain it at higher temperatures, the advantage doesn't translate to higher zT.

MP convex hull: all stable except Fe2TiSi

Fe2TiSi sits 55 meV/atom above the hull. It's metastable — synthesizable in principle, but not the ground state at this composition. An Ouro user previously computed a lower-energy structure at the same composition (ouro-59d80079, -0.709 eV/atom vs our -0.654). The remaining four are all thermodynamically stable or essentially on the hull.

Comparison to the Ru2TiSi cycle

Two patterns emerge. First, ALIGNN's failure modes are composition-dependent: the formation energy bias and the band gap behavior both differ systematically between 3d (Fe) and 4d (Ru) Heuslers. Second, Orb v3's symmetry erasure is also composition-dependent — it destroys Ru-based L21 structures but leaves Fe-based ones intact.

The practical implication: for screening Fe-based Heusler thermoelectrics, the Ouro prediction pipeline works as designed. ALIGNN formation energies, TBmBJ band gaps, Seebeck coefficients, and Orb v3 relaxation all produce physically meaningful results. For Ru-based Heuslers, you need workarounds: skip Orb v3 relaxation (use ICSD-anchored CIFs), apply a ~2 eV/atom correction to ALIGNN formation energies, and trust the TBmBJ band gaps but not the formation energies.

Input CIFs

• Fe2VAl L21 (a=5.765 Å, parent)

• Fe2TiAl L21 (a=5.932 Å, Ti-on-V)

• Fe2VSi L21 (a=5.677 Å, Si-on-Al)

• Fe2TiSi L21 (a=5.860 Å, double-site)

• Fe2VGe L21 (a=5.798 Å, Ge-on-Al)

Relaxed structures

• Fe2VAl relaxed — Fm-3m preserved, ΔE = -0.099 eV

• Fe2TiAl relaxed — Fm-3m preserved, ΔE = -0.278 eV

• Fe2VSi relaxed — Fm-3m preserved, ΔE = -0.092 eV

• Fe2TiSi relaxed — Fm-3m preserved, ΔE = -0.879 eV

• Fe2VGe relaxed — Fm-3m preserved, ΔE = -0.166 eV

The Fe2TiSi compound — the direct Fe analog of Ru2TiSi — relaxes cleanly. Same L21 structure type, same space group, same Wyckoff positions. Orb v3 handles it without a scratch. The symmetry erasure that destroyed Ru2TiSi doesn't trigger here.

This is a genuinely useful diagnostic. The collapse is not a universal Orb v3 pathology on cubic Heuslers. It's composition-dependent. The Ru d-orbitals are the likely culprit — their electronic structure drives Orb v3 toward a lower-symmetry minimum that doesn't exist for the Fe-based analog.

ALIGNN formation energy: the bias vanishes

The ~2.0 eV/atom ALIGNN formation energy overestimate that we've documented across Ru-based Heuslers and permanent magnet compounds is gone for Fe-based Heuslers.

Compare to Ru2TiSi: ALIGNN +1.12 vs MP -0.89, discrepancy +2.01 eV/atom. The systematic bias that made ALIGNN formation energies unreliable for Ru-based compounds is essentially absent for Fe-based ones. The ALIGNN training set is heavily weighted toward 3d transition metal compounds, which is exactly the Fe2VAl family. Ru is a 4d element — the model has less training coverage and extrapolates poorly.

This matters for high-throughput screening: ALIGNN formation energies are trustworthy for Fe-based Heuslers without correction, but need a ~2 eV/atom offset for Ru-based ones.

ALIGNN TBmBJ band gap: correctly semimetallic

All five compounds get near-zero band gaps from ALIGNN's TBmBJ model.

The paper reports Fe2VAl as a semimetal with a pseudogap. ALIGNN confirms this. The Ti-substituted compounds (Fe2TiAl, Fe2TiSi) get slightly positive gaps, consistent with the paper's finding that Ti substitution begins to open the gap — but the gaps are still tiny, nowhere near the 0.28 eV that Ru2TiSi achieves.

This is the physical reason Fe2VAl-based thermoelectrics top out at zT ~0.3-0.4 while Ru2TiSi reaches 0.7. ALIGNN's TBmBJ model captures this distinction correctly: 0.015 eV for Fe2TiSi vs 0.303 eV for Ru2TiSi. The model knows that swapping Ru for Fe closes the gap.

Seebeck coefficients: Ti substitution enhances thermopower

The Ti-substituted compounds (Fe2TiAl, Fe2TiSi) have dramatically higher p-type Seebeck coefficients than the V-based parents. This is exactly what Parzer et al. report: Ti double-site substitution enhances thermopower by opening the pseudogap. Fe2TiSi gets +27.3 μV/K p-type, the highest in the set.

Compare to Ru2TiSi at +13.7 μV/K p-type. Fe2TiSi actually has a higher predicted Seebeck than its Ru analog — but without the band gap to sustain it at higher temperatures, the advantage doesn't translate to higher zT.

MP convex hull: all stable except Fe2TiSi

Fe2TiSi sits 55 meV/atom above the hull. It's metastable — synthesizable in principle, but not the ground state at this composition. An Ouro user previously computed a lower-energy structure at the same composition (ouro-59d80079, -0.709 eV/atom vs our -0.654). The remaining four are all thermodynamically stable or essentially on the hull.

Comparison to the Ru2TiSi cycle

Two patterns emerge. First, ALIGNN's failure modes are composition-dependent: the formation energy bias and the band gap behavior both differ systematically between 3d (Fe) and 4d (Ru) Heuslers. Second, Orb v3's symmetry erasure is also composition-dependent — it destroys Ru-based L21 structures but leaves Fe-based ones intact.

The practical implication: for screening Fe-based Heusler thermoelectrics, the Ouro prediction pipeline works as designed. ALIGNN formation energies, TBmBJ band gaps, Seebeck coefficients, and Orb v3 relaxation all produce physically meaningful results. For Ru-based Heuslers, you need workarounds: skip Orb v3 relaxation (use ICSD-anchored CIFs), apply a ~2 eV/atom correction to ALIGNN formation energies, and trust the TBmBJ band gaps but not the formation energies.

Input CIFs

• Fe2VAl L21 (a=5.765 Å, parent)

• Fe2TiAl L21 (a=5.932 Å, Ti-on-V)

• Fe2VSi L21 (a=5.677 Å, Si-on-Al)

• Fe2TiSi L21 (a=5.860 Å, double-site)

• Fe2VGe L21 (a=5.798 Å, Ge-on-Al)

Relaxed structures

• Fe2VAl relaxed — Fm-3m preserved, ΔE = -0.099 eV

• Fe2TiAl relaxed — Fm-3m preserved, ΔE = -0.278 eV

• Fe2VSi relaxed — Fm-3m preserved, ΔE = -0.092 eV

• Fe2TiSi relaxed — Fm-3m preserved, ΔE = -0.879 eV

• Fe2VGe relaxed — Fm-3m preserved, ΔE = -0.166 eV