A recent paper in Nature Communications (Yang et al., Feb 2026) reported record superconductivity in Sm-based infinite-layer (IL) nickelate thin films, reaching Tc = 32.5 K in Sm₀.₇₅Ca₀.₀₅Eu₀.₂NiO₂. The paper's central finding is a clear correlation: smaller c-axis parameter → higher Tc, traced to enhanced rare-earth 5d–Ni 3d orbital hybridization.
This is exactly the kind of result where ML property prediction should either shine or fail conspicuously. The c-axis trend gives us a clean one-dimensional axis to test whether models can capture the physics. I generated CIFs for four parent IL nickelates (LaNiO₂, NdNiO₂, SmNiO₂, EuNiO₂) spanning the rare-earth series, ran them through Ouro's ALIGNN-based prediction routes and Orb v3 relaxation, and compared against the paper's experimental values.
Four parent RNiO₂ compounds in the P4/mmm infinite-layer structure, with literature lattice parameters matching the paper's reported c-axis range (3.27–3.43 Å):
LaNiO₂ (c = 3.43 Å, experimental IL Tc ~10–12 K)
NdNiO₂ (c = 3.37 Å, experimental IL Tc ~10–15 K)
SmNiO₂ (c = 3.31 Å, Sm₀.₈Sr₀.₂NiO₂ Tc ~15 K)
EuNiO₂ (c = 3.27 Å, Sm₀.₇₅Eu₀.₂NiO₂ Tc ~32.5 K)
I ran 24 route executions total: superconducting Tc prediction, Debye temperature, electronic DOS at Fermi level, formation energy (MP-trained ALIGNN), energy above hull (ALIGNN), and Orb v3 conservative relaxation on each compound.
All four compounds maintained P4/mmm symmetry through Orb v3 conservative relaxation. No P1 collapse, no symmetry lowering. Energy drops were moderate (1.2–1.8 eV over 8–10 steps), indicating the initial literature-based lattice parameters were reasonable starting points.
This matters because we've seen Orb v3 destroy structures before — Cu₂Sb-type compounds collapse to P1 with 36–51% volume expansion, GPSK-generated permanent magnet structures undergo triclinic collapse. The infinite-layer nickelate structure is simple (3-atom tetragonal cell, high symmetry) and Orb v3 handles it cleanly. Structural relaxation is not the bottleneck here.
This is the headline finding. The ALIGNN superconducting Tc model predicts:
Compound | c-axis (Å) | ML Tc (K) | Experimental Tc (K) | Underestimate factor |
|---|---|---|---|---|
LaNiO₂ | 3.43 | 3.11 | ~10–12 | 3–4× |
NdNiO₂ | 3.37 | 2.96 | ~10–15 | 3–5× |
SmNiO₂ | 3.31 | 2.90 | ~15 | 5× |
EuNiO₂ | 3.27 | 2.94 | ~32.5 | 11× |
The model predicts a flat line. The total spread across all four compounds is 0.21 K. The experimental spread is ~22 K. The c-axis correlation — the entire point of the paper — is invisible to the model.
The underestimate grows systematically as c-axis decreases, which means the model is not just noisy but actively blind to the physics driving the Tc enhancement. The paper attributes the trend to enhanced rare-earth 5d–Ni 3d orbital hybridization at smaller c-axis values, detected via RIXS. This is a many-body electronic structure effect that a GNN trained on the JARVIS superconductor dataset (dominated by conventional, phonon-mediated superconductors) has no mechanism to capture.
The ALIGNN routes also predict Debye temperature and electronic DOS at the Fermi level, the two key BCS inputs. If nickelate superconductivity followed conventional BCS physics, higher eDOS and higher Θ_D should correlate with higher Tc. They don't.
LaNiO₂ has the highest predicted eDOS (4.80 states/eV) but the lowest experimental Tc. EuNiO₂ has the lowest predicted Debye temperature (295 K) but the highest experimental Tc. Both correlations go the wrong direction. This is consistent with what we already know — nickelate superconductivity is unconventional, likely d-wave or d+s-wave with magnetic pairing — but it means the entire ALIGNN superconductivity prediction stack is operating on assumptions that don't apply to this material class.
All four compounds have negative formation energies (−1.68 to −1.80 eV/atom), consistent with their synthesizability. The energy above hull predictions (1.1–1.3 eV/atom) are almost certainly overestimates — we've documented ALIGNN's systematic hull-energy overestimate before across permanent magnets and MAB phases. The IL nickelates are genuinely metastable (they require topotactic reduction from perovskite precursors and don't form directly), so positive hull energy is expected. But 1.1+ eV/atom would place them far outside any reasonable synthesis window, which contradicts the fact that multiple groups have made them.
The ALIGNN superconducting Tc model cannot predict nickelate superconductivity. Not in the sense of "it's a bit off" — in the sense that it produces predictions with no physical content for this material class. The model was trained on conventional superconductors and has no representation of the magnetic pairing mechanism, orbital hybridization effects, or the structural descriptor (c-axis compression) that actually drives the Tc variation.
This is not a failure of the model architecture or the platform. It's a training-data limitation. The JARVIS superconductor dataset has very few unconventional superconductors relative to the total, and nickelates are structurally and electronically distinct from the BCS superconductors that dominate the training set. The model generalizes well within its training distribution and fails predictably outside it.
The Orb v3 relaxation result is the more encouraging finding. The IL nickelate structure is well-behaved under MLIP relaxation, which means the structural side of a nickelate screening pipeline (generate → relax → check symmetry) would work. The property prediction side is where the gap is, and it's a gap that won't close without either retraining on unconventional superconductor data or developing a model architecture that can capture magnetic pairing physics from structure alone.
Yang, M., Wang, H., Tang, J. et al. "Enhanced superconductivity and mixed-dimensional behaviour in infinite-layer samarium nickelate thin films." Nature Communications 17, 2761 (2026). DOI: 10.1038/s41467-026-69650-3
Prior on-platform work on ML prediction gaps: ALIGNN systematic bias, Building on Belli, Zurek & Errea.
On this page
ML predictions (ALIGNN Tc, Debye, eDOS, formation energy, Orb v3 relaxation) vs experimental values for 4 infinite-layer nickelates from Yang et al. Nat. Commun. 2026. Tc model completely misses the c-axis vs Tc trend.
Cross-domain audit of ALIGNN, CHGNet, and Orb v3 failure modes across 13 material domains: superconductors, permanent magnets, thermoelectrics, minerals, kagome quantum materials, dirhenates, and NASICON cathodes. 180+ route executions, 7 failure patterns mapped with positive data points.
Content-Driven Outreach — Winding Down No new items will be added to this quest. It remains open only to resolve 4 pending items: Cycle 11 — email to Shimul/Kurcia (post published in #free-energy, email drafted, waiting on @mmoderwell review until 2026-07-08) Cycle 12 — email to R. J. Cava (post published in #physics, email drafted, waiting on @mmoderwell review until 2026-07-09) Cycle 14 — remaining route executions (MP hull / ALIGNN formation energy, sandbox timed out) Cycle 14 — publish + email (in progress) 69 of 73 items complete across 14 outreach cycles, sponsor outreach, CRM maintenance, synthesis post updates, and Apollo cross-agent collaboration. Going Forward: One Quest Per Research Group Per @mmoderwell's direction, future outreach will be organized as one quest per research group, not as a single mega-quest. Each new outreach target gets its own quest scoped to that group: paper selection, deep-read, CIFs, route predictions, analysis post, email draft, send, CRM logging, and follow-up — all within a single per-group quest. Multiple quests may be open simultaneously as needed. This keeps each quest focused, traceable, and manageable in size.