Pith. sign in

REVIEW 3 major objections 5 minor 33 references

Foundation machine-learning potentials capture perovskite order only when it freezes into geometry already in their training data, ordered by a scalar-vector-on-site hierarchy of difficulty.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-10 08:52 UTC pith:XLETRUN6

load-bearing objection Clean ns-scale MD benchmark that turns three perovskites into a practical scalar/vector/on-site filter for when foundation MLIPs can stand in for DFT dynamics. the 3 major comments →

arxiv 2607.08351 v1 pith:XLETRUN6 submitted 2026-07-09 cond-mat.mtrl-sci physics.comp-ph

Bond, orbital and spin order in d4/d6/d7 perovskite oxides: successes and limitations of foundation interatomic potentials

Swagata Acharya , Dimitar Pashov , Mark van Schilfgaarde , Alin M. Elena This is my paper
classification cond-mat.mtrl-sci physics.comp-ph
keywords foundation machine-learning interatomic potentialsperovskite oxidesJahn-Teller distortionbreathing modespin-state crossoverorbital ordermolecular dynamicsstrongly correlated electrons
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Foundation machine-learning interatomic potentials are replacing density-functional theory for structure and nuclear dynamics, yet their reliability for strongly correlated oxides remains open. This paper tests three such potentials, with no material-specific training, on nanosecond molecular dynamics of three isostructural perovskites that host textbook-different low-temperature orders: NdNiO3 (breathing-mode bond disproportionation), LaMnO3 (cooperative Jahn-Teller orbital order), and LaCoO3 (site-local spin multiplet crossover). The materials map onto three classes of increasing architectural demand. A scalar collective bond mode is captured; a vector field of long-bond axes is captured in magnitude but lands on the wrong symmetry pattern; a purely on-site multiplet population shift that leaves no static geometric order parameter is invisible. The paper therefore supplies a structural-footprint criterion: these potentials reproduce static low-temperature fingerprints if and only if the electronic instability has already condensed its order parameter onto a lattice distortion present in the training-set ground states. The hierarchy points to concrete, material-specific training upgrades that can be checked against the same diagnostics.

Core claim

A foundation MLIP trained only on charge- and spin-converged DFT total energies, forces and stresses will reliably reproduce the static low-temperature structural fingerprints of a correlated oxide if and only if the underlying electronic instability has condensed its order parameter onto a static lattice distortion already present in the training-set ground-state relaxations. The three perovskites instantiate a hierarchy of difficulty: scalar (NdNiO3 rocksalt breathing, captured), vector (LaMnO3 long-axis orbital order, magnitude sometimes correct but symmetry wrong), and on-site (LaCoO3 multiplet crossover, inaccessible).

What carries the argument

The structural-footprint criterion together with the scalar-vector-on-site hierarchy: scalar bond-mode order is learnable from nearest-neighbour anti-correlations alone; vector long-axis order requires the correct symmetry pattern in the training set; on-site multiplet population has no geometric order parameter and cannot be recovered from bond-based features.

Load-bearing premise

The claim that the LaCoO3 spin crossover is fundamentally unreachable from any structure-only model, rather than merely missed by today's geometric descriptors, and that purely structural surrogates fully diagnose success or failure for the electronic channels.

What would settle it

Run the same nanosecond molecular-dynamics protocol and descriptor pipeline on additional foundation potentials or on fine-tuned models that include explicit multiplet or local-spin labels for LaCoO3 (or correct C-type Jahn-Teller patterns for LaMnO3) and check whether the reported order parameters, bimodality coefficients and long-axis classifications reverse.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The manuscript tests three foundation machine-learning interatomic potentials (MACE matpes_r2scan, MACE omol, and CHGNet) on the low-temperature structural order of three isostructural ABO3 perovskites—LaMnO3 (d4), LaCoO3 (d6), and NdNiO3 (d7)—via identical NVT molecular-dynamics protocols (1 ns production, 50–300 K, 80- and 160-atom supercells) with no system-specific training. It reports a hierarchy of difficulty: NdNiO3’s scalar rocksalt breathing fingerprint is captured (fully by omol, as a precursor by matpes_r2scan); LaMnO3’s vector Jahn–Teller long-axis order is captured in magnitude by some potentials but not in the experimental C-type symmetry (ferro-orbital patterns appear instead); and LaCoO3’s low-spin to high-spin multiplet crossover, which has no spatial order parameter, leaves no static structural footprint and is not recovered. From these outcomes the authors propose a structural-footprint criterion: foundation MLIPs trained on charge- and spin-converged DFT total energies reproduce static low-T fingerprints if and only if the electronic instability has condensed onto a lattice distortion already present in the training-set relaxations.

Significance. The work is timely: foundation MLIPs are being adopted for finite-temperature structure and dynamics in correlated oxides, yet their fidelity for multiplet, orbital, and bond-disproportionation physics is poorly mapped. The comparative design—three materials chosen to isolate scalar, vector, and on-site channels; three potentials spanning closed-shell, r2SCAN, and GGA+U/magnetism-supervised training; matched 1 ns trajectories; and explicit 160-atom finite-size checks that leave every classifier on the same side of the reported thresholds—is a genuine strength and makes the differential hierarchy credible. The paper also states falsifiable, prospective predictions for the wider RNiO3, Co multiplet-crossover, and cooperative-JT families, and documents a reusable descriptor pipeline and trajectory availability. If the hierarchy and the more carefully worded form of the structural-footprint criterion hold, they give the community a concrete, material-specific checklist for when structure-only foundation potentials can be trusted and when multiplet- or spin-aware fine-tuning is mandatory.

major comments (3)
  1. [Structural-footprint criterion] Structural-footprint criterion (main text, paragraph beginning “The cross-MLIP outcome … maps cleanly onto a single principle”): the “if and only if” wording is stronger than the evidence. The three potentials and three materials support a useful one-way implication (condensed training-set distortion → capture of the static fingerprint is possible), but the converse and the universality claim rest on a small sample. In particular, omol freezes the NdNiO3 rocksalt pattern (Q_RS = −1.000) despite no extended-solid cooperative-distortion training; the paper correctly notes this is a kinetic basin rather than a thermodynamic T_MI prediction, yet that success without the stated training-set content already strains the “already present in the training-set ground-state relaxations” half of the criterion. Soften to a working hypothesis or one-directional rule, and state the domain of tested pote
  2. [Hierarchy of broken-symmetry channels / Outlook] Hierarchy of broken-symmetry channels and Outlook: the claim that the LaCoO3 multiplet crossover is “fundamentally unreachable from any structure-only model” and that “no enrichment of bond-based geometric features can recover it” goes beyond what is demonstrated. The MD results show that the three present-day foundation potentials return no static order-parameter-locked distortion and only mild leptokurtic or ringing signatures, which is consistent with experimental absence of a spatial OP and with the abstract’s “inaccessible to present-day MLIPs.” That is sufficient and important. Declaring architectural impossibility for all future structure-only models, without a multiplet-aware or multi-reference baseline comparison, is not load-bearing for the hierarchy actually shown and should be rephrased as a strong expectation requiring multiplet-resolved auxiliary inputs (as the Outlook alre
  3. [Methods; Abstract; Table 2] Methods (“orbital order” paragraph) and LaMnO3 results: throughout, M_OO / M_FO and “orbital order” are structural surrogates (long-axis C-type vs ferro-orbital patterns) with no electronic-orbital information in the potentials. This is stated once, but the abstract, Table 2, and hierarchy language still read as if the electronic e_g channel itself has been diagnosed. For the central claim this is acceptable only if every success/failure statement is explicitly about the structural fingerprint (as Table 1’s “static fingerprint” column does). Please audit the abstract and the LMO rows of Table 2 so that “captured in magnitude but not in symmetry” is unambiguously the geometric long-axis pattern, not a claim about orbital occupation.
minor comments (5)
  1. [Figure 1] Figure 1 caption and panel (c): shell correlations are shown only at 50 K; a brief note that the (- + - +) alternation for NdNiO3 persists across the full T series (as stated in the text) would help readers who look only at the figure.
  2. [Table 2] Table 2 caption: “≈expt” is carefully caveated for the structural channel only; consider adding the same caveat in the LaCoO3 matpes_r2scan cell text so that “≈expt: Gaussian, no distortion” is not misread as electronic agreement with the spin crossover.
  3. [Methods] Eq. (1) and the long-axis mask |Q_static_2| < 0.05 Å: both are free parameters of the analysis. A one-sentence sensitivity check (e.g. that FO vs C-type classification is stable for a small window around 0.05 Å) would strengthen Methods; the 160-atom check already helps on finite size.
  4. [Throughout] Notation consistency: matpes_r2scan / matpes r2scan / r2scan appear with and without underscores; pick one form for the production head throughout text and figures.
  5. [Data and code availability / Outlook] Data availability: “available from the corresponding author on reasonable request” is weaker than the Outlook’s claim that trajectories and the descriptor pipeline are made available to enable comparison. If a repository deposit is planned, state it; if not, align the Outlook wording with the Methods statement.

Circularity Check

0 steps flagged

No significant circularity: empirical MD tests of external foundation potentials against geometric order parameters and experimental structural fingerprints.

full rationale

The paper’s central claims are empirical outcomes of 1 ns NVT MD with three pretrained foundation MLIPs (MACE matpes_r2scan, MACE omol, CHGNet) on 80- and 160-atom supercells of LaMnO3, LaCoO3, and NdNiO3, with no system-specific training or fitting to the target materials. Structural descriptors (Q_RS, |Q_static_2|, M_FO/M_OO, BC, kurtosis, shell correlations) are defined from bond geometry alone and compared to known experimental low-T fingerprints (rocksalt breathing, C-type JT long-axis pattern, absence of static distortion in LS LaCoO3). Success/failure is therefore not forced by construction: the potentials either produce or fail to produce those geometric patterns. The ‘structural-footprint criterion’ and scalar/vector/on-site hierarchy are inductive generalizations from those runs, not algebraic identities. Methods explicitly state that ‘orbital order’ is a structural surrogate (long-axis alternation) because the potentials contain no electronic-orbital information; this is transparent labeling, not a self-definitional loop that equates input to claimed prediction. Self-citations (e.g. Acharya et al. on nickelate MIT, QS GW multiplet context) supply electronic-structure background and outlook, not the MD classifiers or the differential hierarchy. No parameter is fitted to a subset of the target data and then re-presented as a prediction; no uniqueness theorem is imported to forbid alternatives. The work is self-contained against external experimental structure data and the three external potentials.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 2 invented entities

The paper is empirical MD benchmarking, not a derivation from axioms. Load-bearing inputs are standard MD/MLIP practice, experimental space groups and bond patterns as targets, and the identification of geometric surrogates with electronic order. Free parameters are simulation and classifier thresholds, not fits to the central hierarchy. The scalar/vector/on-site taxonomy is an organizing invention of the paper, not an external entity with independent evidence beyond these three cases.

free parameters (3)
  • NVT thermostat τ_T and chain settings = τ_T=100 fs; 1 ns production
    Nosé–Hoover chain length 3, τ_T=100 fs, 1 fs timestep, ≥50 ps equilibration, 1 ns production—standard but hand-chosen; order-parameter conclusions could in principle depend on sampling.
  • Long-axis mask threshold |Q_static_2|<0.05 Å = 0.05 Å
    M_OO and M_FO are masked below this cutoff; classification of FO vs C-type depends on this hand-set resolution threshold.
  • Sarle BC bimodality threshold 5/9 = BC>5/9
    Classical threshold used to call omol–NdNiO3 bimodal; literature-standard but still a discrete classifier choice.
axioms (5)
  • domain assumption Foundation MLIPs trained on DFT total energies/forces (and for CHGNet magnetic moments) have no direct multiplet or orbital-occupation descriptors.
    Stated in the introduction; underpins the claim that only condensed structural order can be learned.
  • domain assumption Experimental low-T structures (rocksalt breathing in NNO, C-type JT in LMO, undistorted LS LCO) are the correct targets for structural success.
    Used throughout Tables 1–2 and cross-MLIP comparison; standard experimental consensus.
  • domain assumption C-type alternation of octahedral long axes is a one-to-one structural surrogate for e_g orbital order in Mn3+.
    Methods: ‘orbital order’ denotes this geometry alone; required to score LMO symmetry failure.
  • domain assumption Starting Pbnm (not P21/n) for NNO and experimental R-3c/Pbnm cells means any breathing/JT pattern is generated by dynamics, not imposed.
    Protocol and Starting structures sections; needed to claim the potentials produce order spontaneously.
  • ad hoc to paper Structural-footprint criterion: reliable capture iff electronic instability condensed onto a static lattice distortion present in training-set relaxations.
    Core interpretive claim of the paper; generalized from three materials × three potentials.
invented entities (2)
  • Scalar / vector / on-site hierarchy of broken-symmetry channels for foundation MLIPs no independent evidence
    purpose: Rank architectural difficulty of NNO breathing, LMO JT long-axis order, and LCO multiplet crossover for structure-only potentials.
    Organizing taxonomy introduced here; independent evidence is only the three case studies plus prospective predictions for related families.
  • Structural-footprint criterion (iff training-set condensed distortion) no independent evidence
    purpose: Predict which correlated oxides foundation MLIPs will capture without multiplet inputs.
    Falsifiable prospective claims for RNiO3, Co multiplet family, and e_g JT systems; not yet independently validated beyond this study.

pith-pipeline@v1.1.0-grok45 · 16574 in / 3719 out tokens · 51297 ms · 2026-07-10T08:52:44.763776+00:00 · methodology

0 comments
read the original abstract

Foundation machine-learning interatomic potentials (MLIPs) are rapidly replacing density-functional theory (DFT) for modeling structure and nuclear dynamics, making their fidelity in strongly correlated systems an urgent question. We test three foundation potentials on the low-temperature order of three correlated, isostructural ABO3 perovskite oxides: LaMnO3 (d4), LaCoO3 (d6), and NdNiO3 (d7). We run molecular dynamics for 1 ns on 80- and 160-atom supercells from 50 to 300 K with no system-specific training. These oxides expose three distinct classes of low-temperature order that define a hierarchy of difficulty for the potentials. The scalar class, represented by NdNiO3, has a simple geometric fingerprint and is captured. The vector class, represented by LaMnO3, requires identifying which Cartesian axis carries the long bond at each site, and is captured in magnitude but not in symmetry. The on-site class, represented by the low-spin to high-spin crossover in LaCoO3, is a purely local multiplet population shift with no spatial order parameter and remains inaccessible to present-day MLIPs.

Figures

Figures reproduced from arXiv: 2607.08351 by Alin M. Elena, Dimitar Pashov, Mark van Schilfgaarde, Swagata Acharya.

Figure 1
Figure 1. Figure 1 [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Batatia, I.et al.A foundation model for atomistic materials chemistry.J. Chem. Phys.163, 184110 (2025)

    work page 2025

  2. [2]

    P., Simm, G

    Batatia, I., Kov ´acs, D. P., Simm, G. N. C., Ortner, C. & Cs ´anyi, G. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields.Adv. Neural Inf. Process. Syst.35, 11423 (2022)

    work page 2022

  3. [3]

    Deng, B.et al.CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling.Nat. Mach. Intell.5, 1031–1041 (2023)

    work page 2023

  4. [4]

    URL https://www.sciencedirect.com/science/article/pii/S2950463526000153

    Acharya, S.et al.Origin of metal-insulator transition in rare-earth nick- elates.Computational Materials Today11, 100060 (2026). URL https://www.sciencedirect.com/science/article/pii/S2950463526000153

    work page 2026

  5. [5]

    L., Malyi, O

    Du, J. L., Malyi, O. I., Shang, S.-L.et al.Density functional thermodynamic description of spin, phonon and displacement degrees of freedom in antiferromagnetic-to-paramagnetic phase transition in YNiO3.Mater. Today Phys.27, 100805 (2022)

    work page 2022

  6. [6]

    Malyi, O. I. & Zunger, A. Rise and fall of Mott insulating gaps in YNiO 3 paramagnets as a reflection of symmetry breaking and remaking.Phys. Rev. Mater.7, 044409 (2023)

    work page 2023

  7. [7]

    Shannon, R. D. Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides.Acta Crystallogr. A32, 751 (1976). 13

    work page 1976

  8. [8]

    Asai, K.et al.Neutron-scattering study of the spin-state transition and magnetic correlations in La1−xSrxCoO3.Phys. Rev. B50, 3025 (1994)

    work page 1994

  9. [9]

    W.et al.Spin state transition in LaCoO 3 studied using soft x-ray absorption spectroscopy and magnetic circular dichroism.Phys

    Haverkort, M. W.et al.Spin state transition in LaCoO 3 studied using soft x-ray absorption spectroscopy and magnetic circular dichroism.Phys. Rev. Lett.97, 176405 (2006)

    work page 2006

  10. [10]

    Radaelli, P. G. & Cheong, S.-W. Structural phenomena associated with the spin-state transition in LaCoO3.Phys. Rev. B66, 094408 (2002)

    work page 2002

  11. [11]

    Podlesnyak, A.et al.Spin-state transition in LaCoO 3: direct neutron spectroscopic evidence of excited magnetic states.Phys. Rev. Lett.97, 247208 (2006)

    work page 2006

  12. [12]

    Sundaram, N.et al.Local structure of La 1−xSrxCoO3 determined from EXAFS and neutron pair distribution function studies.Phys. Rev. Lett.102, 026401 (2009)

    work page 2009

  13. [13]

    & Ghosez, P

    Mercy, A., Bieder, J., ´I˜niguez, J. & Ghosez, P. Structurally triggered metal-insulator transition in rare-earth nickelates.Nat. Commun.8, 1677 (2017)

    work page 2017

  14. [14]

    Mizokawa, T., Khomskii, D. I. & Sawatzky, G. A. Spin and charge ordering in self-doped mott insulators.Phys. Rev. B61, 11263 (2000)

    work page 2000

  15. [15]

    I.Transition Metal Compounds(Cambridge University Press, 2014)

    Khomskii, D. I.Transition Metal Compounds(Cambridge University Press, 2014)

    work page 2014

  16. [16]

    Park, H., Millis, A. J. & Marianetti, C. A. Site-selective Mott transition in rare-earth-element nickelates.Phys. Rev. Lett.109, 156402 (2012)

    work page 2012

  17. [17]

    Commun.7, 13017 (2016)

    Bisogni, V .et al.Ground-state oxygen holes and the metal–insulator transition in the negative charge-transfer rare-earth nickelates.Nat. Commun.7, 13017 (2016). 14

    work page 2016

  18. [18]

    I.et al.Charge ordering as alternative to Jahn–Teller distortion.Phys

    Mazin, I. I.et al.Charge ordering as alternative to Jahn–Teller distortion.Phys. Rev. Lett.98, 176406 (2007)

    work page 2007

  19. [19]

    Rodr ´ıguez-Carvajal, J.et al.Neutron-diffraction study of the Jahn–Teller transition in stoi- chiometric LaMnO3.Phys. Rev. B57, R3189 (1998)

    work page 1998

  20. [20]

    G., Krogh Andersen, E

    Norby, P., Krogh Andersen, I. G., Krogh Andersen, E. & Andersen, N. H. The crystal structure of lanthanum manganate(III), LaMnO3, at room temperature and at 1273 K under N2.J. Solid State Chem.119, 191 (1995)

    work page 1995

  21. [21]

    Elliott, K.et al.janus-core: Tools for machine-learnt interatomic potentials (2025)

    work page 2025

  22. [22]

    A., Janczyk, M., Dale, R

    Pfister, R., Schwarz, K. A., Janczyk, M., Dale, R. & Freeman, J. B. Good things peak in pairs: a note on the bimodality coefficient.Front. Psychol.4, 700 (2013)

    work page 2013

  23. [23]

    N., ´I˜niguez, J., Barth´el´emy, A

    Varignon, J., Grisolia, M. N., ´I˜niguez, J., Barth´el´emy, A. & Bibes, M. Complete phase diagram of rare-earth nickelates from first-principles.npj Quantum Mater.2, 21 (2017)

    work page 2017

  24. [24]

    & Zunger, A

    Varignon, J., Bibes, M. & Zunger, A. Mott gapping in3dABO 3 perovskites without Mott– Hubbard interelectronic repulsion energyu.Phys. Rev. B100, 035119 (2019)

    work page 2019

  25. [25]

    & Marzari, N

    Binci, L., Kotiuga, M., Timrov, I. & Marzari, N. Hybridization driving distortions and multi- ferroicity in rare-earth nickelates.Phys. Rev. Research5, 033146 (2023)

    work page 2023

  26. [26]

    Ehrke, H.et al.Photoinduced melting of antiferromagnetic order in La 0.5Sr1.5MnO4 measured using ultrafast resonant soft X-ray diffraction.Phys. Rev. Lett.106, 217401 (2011). 15

    work page 2011

  27. [27]

    & Rozenberg, M

    Georges, A., Kotliar, G., Krauth, W. & Rozenberg, M. J. Dynamical mean-field theory of strongly correlated fermion systems and the limit of infinite dimensions.Rev. Mod. Phys.68, 13 (1996)

    work page 1996

  28. [28]

    & Faleev, S

    van Schilfgaarde, M., Kotani, T. & Faleev, S. Quasiparticle self-consistent GW theory.Phys. Rev. Lett.96, 226402 (2006)

    work page 2006

  29. [29]

    Pashov, D.et al.Questaal: a package of electronic structure methods based on the linear muffin-tin orbital technique.Comp. Phys. Commun.249, 107065 (2020)

    work page 2020

  30. [30]

    & van Schilfgaarde, M

    Cunningham, B., Gr ¨uning, M., Pashov, D. & van Schilfgaarde, M. Quasiparticle self- consistent GW with ladder diagrams in W.Phys. Rev. B108, 165104 (2023)

    work page 2023

  31. [31]

    Acharya, S.et al.A theory for colors of strongly correlated electronic systems.Nat. Commun. 14, 5565 (2023)

    work page 2023

  32. [32]

    Glazer, A. M. The classification of tilted octahedra in perovskites.Acta Crystallogr. B28, 3384 (1972)

    work page 1972

  33. [33]

    Y es (precursor)

    Catalano, S., Gibert, M., Fowlie, J.et al.Rare-earth nickelates RNiO 3: thin films and het- erostructures.Rep. Prog. Phys.81, 046501 (2018). AcknowledgementsThis work was authored in part by the National Laboratory of the Rockies for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. Funding was provided by the Of- fice of Science, ...

    work page 2018