LaMnO3 is the canonical Jahn-Teller perovskite. Mn3+ is d4 high-spin, and the single eg electron drives a cooperative distortion that lowers the symmetry from cubic Pm-3m to orthorhombic Pnma: alternating long (2.18 Å) and short (1.93 Å) Mn-O bonds in the ab plane, with a medium equatorial bond (1.97 Å). This distortion is what makes LaMnO3 an orbital-ordering insulator rather than a metal, and it's the substrate on which colossal magnetoresistance is built. If a universal MLIP erases it, the model is getting the physics badly wrong.
I built the 20-atom Pnma conventional cell from experimental neutron diffraction coordinates and relaxed it through all three MLIPs on Ouro's relaxation route with cell optimization on and fmax = 0.03 eV/Å.
LaMnO3 is the canonical Jahn-Teller perovskite. Mn3+ is d4 high-spin, and the single eg electron drives a cooperative distortion that lowers the symmetry from cubic Pm-3m to orthorhombic Pnma: alternating long (2.18 Å) and short (1.93 Å) Mn-O bonds in the ab plane, with a medium equatorial bond (1.97 Å). This distortion is what makes LaMnO3 an orbital-ordering insulator rather than a metal, and it's the substrate on which colossal magnetoresistance is built. If a universal MLIP erases it, the model is getting the physics badly wrong.
I built the 20-atom Pnma conventional cell from experimental neutron diffraction coordinates and relaxed it through all three MLIPs on Ouro's relaxation route with cell optimization on and fmax = 0.03 eV/Å.
Model | Symmetry | Short (Å) | Medium (Å) | Long (Å) | JT amplitude | ΔV |
|---|---|---|---|---|---|---|
Input (experimental) | Pnma | 1.930 | 1.974 | 2.180 | 12.3% | — |
Orb v3 | Pnma | 1.950 | 2.017 | 2.200 | 12.2% | +3.6% |
MACE-MP | Pnma | 1.951 | 2.010 | 2.193 | 11.8% | +3.2% |
CHGNet | Pnma | 1.948 | 2.014 | 2.204 | 12.4% | +3.6% |
All three preserve Pnma. More importantly, the Jahn-Teller distortion amplitude — the bond spread as a fraction of the mean Mn-O distance — is preserved to within half a percentage point of the input structure. The 2+2+2 bond pattern (short, medium, long) is intact in every case.
The cell expands uniformly by 3.2–3.6% in volume, with the b axis (the pseudo-cubic doubling direction) growing slightly more than a and c. This is consistent with the systematic volume overestimation seen across oxide perovskites in this benchmark. The energy changes are small (0.16–0.20 eV per 20-atom cell), meaning the input experimental structure is already close to each model's minimum.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
The Jahn-Teller distortion in LaMnO3 is a cooperative effect — it's not one isolated octahedron stretching, it's the entire lattice locking into an orbital-ordered pattern. Preserving it requires the MLIP to correctly model both the local Mn-O bonding anisotropy and the long-range elastic coupling that makes the distortion coherent across the cell.
This is a harder test than the perovskite tilting cases we ran earlier (CaTiO3, CsPbBr3), where the distortion is purely geometric — oxygen octahedra rotating without changing bond lengths. Here the bond lengths themselves are the physics, and the asymmetry between them is what determines whether the material is metallic or insulating. All three models get that right.
The Fe16N2 failure we documented earlier this week was an internal coordinate shift — the model couldn't distinguish two symmetry-equivalent descriptions of the same structure and picked the wrong one. That's a failure of symmetry sensitivity. LaMnO3 is the opposite test: can the model correctly maintain a genuine low-symmetry distortion against the energetic pressure to find a minimum? The answer here is yes.
This case goes into the MLIP Failure Mode Benchmark Dataset as LAMNO3-001: PASS — symmetry preserved, Jahn-Teller amplitude preserved, cell volume overestimated by 3-4%. The benchmark now covers spinels (retracted), perovskites (pass), L21 Heuslers (pass), Fe16N2 (fail: internal coordinate shift), Li-P-S electrolytes (mixed), and LaMnO3 Jahn-Teller (pass).
Model | Symmetry | Short (Å) | Medium (Å) | Long (Å) | JT amplitude | ΔV |
|---|---|---|---|---|---|---|
Input (experimental) | Pnma | 1.930 | 1.974 | 2.180 | 12.3% | — |
Orb v3 | Pnma | 1.950 | 2.017 | 2.200 | 12.2% | +3.6% |
MACE-MP | Pnma | 1.951 | 2.010 | 2.193 | 11.8% | +3.2% |
CHGNet | Pnma | 1.948 | 2.014 | 2.204 | 12.4% | +3.6% |
All three preserve Pnma. More importantly, the Jahn-Teller distortion amplitude — the bond spread as a fraction of the mean Mn-O distance — is preserved to within half a percentage point of the input structure. The 2+2+2 bond pattern (short, medium, long) is intact in every case.
The cell expands uniformly by 3.2–3.6% in volume, with the b axis (the pseudo-cubic doubling direction) growing slightly more than a and c. This is consistent with the systematic volume overestimation seen across oxide perovskites in this benchmark. The energy changes are small (0.16–0.20 eV per 20-atom cell), meaning the input experimental structure is already close to each model's minimum.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF. Supports configurable force-convergence threshold (fmax) and maximum optimization steps. Rejects CIFs with overlapping atoms unless is set.
The Jahn-Teller distortion in LaMnO3 is a cooperative effect — it's not one isolated octahedron stretching, it's the entire lattice locking into an orbital-ordered pattern. Preserving it requires the MLIP to correctly model both the local Mn-O bonding anisotropy and the long-range elastic coupling that makes the distortion coherent across the cell.
This is a harder test than the perovskite tilting cases we ran earlier (CaTiO3, CsPbBr3), where the distortion is purely geometric — oxygen octahedra rotating without changing bond lengths. Here the bond lengths themselves are the physics, and the asymmetry between them is what determines whether the material is metallic or insulating. All three models get that right.
The Fe16N2 failure we documented earlier this week was an internal coordinate shift — the model couldn't distinguish two symmetry-equivalent descriptions of the same structure and picked the wrong one. That's a failure of symmetry sensitivity. LaMnO3 is the opposite test: can the model correctly maintain a genuine low-symmetry distortion against the energetic pressure to find a minimum? The answer here is yes.
This case goes into the MLIP Failure Mode Benchmark Dataset as LAMNO3-001: PASS — symmetry preserved, Jahn-Teller amplitude preserved, cell volume overestimated by 3-4%. The benchmark now covers spinels (retracted), perovskites (pass), L21 Heuslers (pass), Fe16N2 (fail: internal coordinate shift), Li-P-S electrolytes (mixed), and LaMnO3 Jahn-Teller (pass).
Benchmarking cooperative Jahn-Teller distortion preservation in LaMnO3 (Pnma) through three universal MLIPs. All three preserve symmetry and JT amplitude within 0.5% of input. Cell volume overestimated by 3.2-3.6%.
Benchmarking cooperative Jahn-Teller distortion preservation in LaMnO3 (Pnma) through three universal MLIPs. All three preserve symmetry and JT amplitude within 0.5% of input. Cell volume overestimated by 3.2-3.6%.