Pith. sign in

REVIEW 4 major objections 5 minor 72 references

Cross-Geometry Transferability Assessment of Universal Machine Learning Interatomic Potentials: From Bulk Materials to Atomic Nanowires

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Universal machine-learned interatomic potentials transfer poorly to low-coordination zirconia nanostructures in zero-shot use, and mixed-geometry fine-tuning on a five-class dataset reduces cross-geometry errors while single-geometry…

desk verdict Useful ZrO2 geometry-diverse benchmark and zero-shot assessment, but the fine-tuning claims need a trajectory-aware split before they are believable. read the letter →

arxiv 2608.06662 v1 pith:6SNHKPCW submitted 2026-08-07 cond-mat.mtrl-sci cond-mat.mes-hallphysics.comp-ph

Pedro H. M. Zanineli , Bruno Focassio , Gabriel R. Schleder This is my paper · ORCID
classification cond-mat.mtrl-scicond-mat.mes-hallphysics.comp-ph
keywords machinelearninginteratomicpotentialstransferabilityfine-tuningzero-shotinferencezirconiananostructurescross-geometrygeneralizationdensityfunctionaltheorydatasetmoleculardynamics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether universal machine-learned interatomic potentials, pretrained mostly on bulk crystal data, can be trusted when a simulation passes through surfaces, particles, necks, and atomically thin wires, as happens when ZrO2 grains desinter under tension. It builds a DFT dataset spanning all five geometry classes and shows that zero-shot predictions degrade sharply in the low-coordination classes: the best pretrained model reaches a force error of about 197 meV/Å overall, with the largest errors on necks and wires. Under a fixed 50-epoch training budget, fine-tuning beats training from scratch, but fine-tuning on one geometry alone often hurts other geometries, while mixed-geometry fine-tuning reduces the spread of errors across geometries. The paper's conclusion is that adapting foundation potentials to ionic nanostructures requires target data that explicitly cover the structural environments of interest, plus property-level validation, because aggregate energy and force errors do not predict elastic, vibrational, surface, or dynamical accuracy.

What carries the argument

The load-bearing object is the geometry-diverse ZrO2 dataset itself: bulk, slab, particle, neck, and wire configurations generated from DFT-based ab initio molecular dynamics and quasistatic elongation, with a geometry-stratified train/validation/test split applied before a 3 eV/Å force filter removes high-force frames. The evaluation machinery is a controlled three-way comparison of zero-shot inference, full-parameter fine-tuning from a pretrained checkpoint, and from-scratch training, all run for 50 epochs with reference-energy alignment by fitted elemental offsets so that energy errors reflect the shape of the potential-energy surface rather than reference-level shifts. Geometry-specific nested subsets and mixed-geometry subsets at controlled atomic-environment counts isolate the effect of structural diversity from dataset size. Property-level validations of elastic constants, phonons, surface energies, and tensile neck molecular dynamics then test whether improvements in supervised error metrics propagate to physical observables.

What would settle it

Re-run the zero-shot and mixed-geometry fine-tuning evaluations on a second ionic oxide, for example HfO2, using the same five geometry classes; if cross-geometry error reduction does not reproduce, the diversity conclusion is specific to ZrO2 rather than general. Separately, recompute the reported elastic and phonon errors against a DFT reference whose exchange-correlation functional and pseudopotentials are documented, to see whether the property rankings survive a controlled reference.

Watch

Extended reading notes

Core claim

The central discovery is that geometry diversity in the adaptation data, not just the amount of data, is what makes pretrained MLIPs transfer across a structural process like the ZrO2 desintering. On the authors' dataset, geometry-specific fine-tuning improves in-domain accuracy but frequently produces negative transfer, with wire-only fine-tuning degrading energy errors on several other classes, whereas mixed-geometry fine-tuning at controlled total atomic environments lowers cross-geometry errors for both MACE and ORB checkpoints. The paper also establishes that pretrained initialization provides a data-efficiency advantage: at equal optimization epochs, fine-tuned models beat from-scratch models with comparable wall-clock time. Property-level tests show that no single strategy is uniformly best: zero-shot models retain the lowest phonon errors and competitive elasticity, while adaptation helps most on surface energies and neck dynamics, and average error rankings fail to predict these outcomes.

Load-bearing premise

The argument assumes the density-functional-theory settings used to generate the ZrO2 reference data are consistent with the Materials Project reference used for elastic and phonon validation, but those settings are not reported in the paper.

Editorial extensions

If this is right

  • Foundation MLIPs should not be used zero-shot for low-coordination ionic nanostructures; even the best zero-shot force error exceeds commonly cited accuracy thresholds for stable dynamics.
  • Adapting to a process that spans bulk, surface, particle, neck, and wire motifs requires training data drawn from all of those classes, not just the target class.
  • Geometry-specific fine-tuning can backfire outside its domain, so deployment scope should determine whether narrow or diverse fine-tuning is appropriate.
  • Model rankings based on average energy and force RMSE are insufficient; independent physical validation is needed before trusting elastic, phonon, or dynamical predictions.
  • Pretrained initialization is a data-efficiency win at a fixed optimization budget, not a training-cost reduction, since fine-tuning and from-scratch training take similar wall-clock time.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The negative-transfer pattern is a practical instance of catastrophic forgetting: full-parameter fine-tuning on a narrow geometry overwrites representations that remain useful elsewhere, suggesting that replay-based or parameter-efficient strategies may be needed for multi-geometry deployment.
  • The dataset's design could be reused as an out-of-distribution benchmark for other ionic oxides, and replacing manual geometry labels with automated active-learning structure generation would test whether explicit diversity is strictly necessary or merely sufficient.
  • Because long-range electrostatics are not isolated in the experiments, the residual wire and neck errors may partly reflect missing charge physics; charge-aware or polarizable potentials offer a direct test of that hypothesis.
  • The property-level hierarchy suggests a practical validation recipe: check phonon curvature and surface energetics before trusting molecular dynamics on far-from-equilibrium nanostructures.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper benchmarks 26 pretrained machine-learned interatomic potentials on a newly generated DFT dataset of ZrO2 configurations spanning bulk, slab, particle, neck, and wire geometries, motivated by an experimentally observed desintering process. The authors report geometry-dependent zero-shot error degradation, compare zero-shot inference, fine-tuning, and from-scratch training under a fixed 50-epoch protocol, evaluate geometry-specific versus mixed-geometry fine-tuning, and validate selected models on elastic, vibrational, surface-energy, and neck-dynamics properties. The central claims are that geometry-diverse target data improve cross-geometry transfer and that property-level rankings are not predicted by aggregate energy and force errors.

Significance. If the claims hold, the paper provides a valuable and reusable benchmark for a chemically and structurally challenging system, with a public dataset and code. The study is careful to separate reference-energy alignment from model error, to compare training strategies under controlled optimizer settings, and to acknowledge several limitations (partly interpolative surface tests, single-trajectory dynamics, incomplete optimization). These strengths make the paper a useful contribution for practitioners adapting foundation MLIPs to low-coordination ionic nanostructures. However, the headline zero-shot result contains an unresolved internal inconsistency, and the trajectory-correlated data split may materially inflate the supervised adaptation conclusions, so the central prescriptive claims should be treated as conditionally supported pending revision.

major comments (4)
  1. [Abstract and Section III.B vs SI Table S3] The abstract and Section III.B state that ORB-V3 achieves a zero-shot aligned energy RMSE of 6 meV/atom, but SI Table S3 lists 31.27 meV/atom for the aligned energy RMSE of the same model (and 30.36 meV/atom raw). This discrepancy concerns the paper's headline quantitative result and must be reconciled; the authors should state whether Fig. 2A and the abstract use a different metric or a different subset, and correct the text, table, or figure accordingly.
  2. [Section II.A] The DFT reference dataset is described only as 'first-principles simulations based on DFT, as implemented in VASP', without specifying the exchange-correlation functional, pseudopotentials, k-point sampling, energy cutoff, or convergence thresholds. Because Sections II.F and II.G validate against Materials Project mp-2858 values, the reported property errors mix genuine model error with possible reference mismatches. These computational settings must be provided for the dataset to be reproducible and for the property-level comparisons to be interpretable.
  3. [Section II.B and Sections III.C/III.E] The train/test split is stratified within geometry classes but not block-wise by AIMD trajectory, seed, or elongation step. Since neck and wire structures come from continuous AIMD and tensile histories (Section II.A), near-duplicate or highly correlated frames may appear on both sides of the split. This can inflate the fine-tuning and mixed-geometry gains reported in Fig. 4C and Sections III.C/III.E, which are the main evidence for the paper's prescriptive claim that geometry-diverse target data are necessary. The authors should add a trajectory-blocked or structurally de-duplicated split to show that the conclusions survive.
  4. [Section II.H vs Section IV and Section III.F.4] Section II.H states that the tensile MD simulations were repeated with three random seeds, but Section IV states that the MD analysis is based on a single trajectory per model, and Section III.F.4 refers to 'a single initial-velocity realization'. This internal inconsistency matters because the observed failure mode (from-scratch ORB rupture) may be seed-specific. The authors should clarify how many seeds underlie Fig. 5D and whether the qualitative conclusions are stable across seeds.
minor comments (5)
  1. [Section II.C] In Eq. (1), the sentence 'For each model, element-specific [41, 46] were determined' appears to be missing the noun 'corrections'.
  2. [Figure 1 caption] The caption of Fig. 1(B) reads 'between two [26, 27]' and appears to be missing a noun such as 'grains'.
  3. [SI Table S8] Table S8 contains an unresolved cross-reference ('as summarized in Table ??') that should be fixed before publication.
  4. [Section II.D] The manuscript should state explicitly that the ORB-V3 checkpoint used throughout is the Direct-20 variant, since the SI (Fig. S3) compares Direct-20 and Conservative-20 variants and the distinction is relevant for reproducibility.
  5. [Figure 1 text] The embedded text 'MatBenchDiscoveryAssessment' in the workflow diagram is unclear and should be replaced with a readable label or legend entry.

Circularity Check

0 steps flagged · score 1.0 of 10

No meaningful circularity: the adaptation comparisons are controlled experiments and the property validations are checked against external DFT references, not against the paper's own inputs.

full rationale

This paper is an empirical benchmark rather than a derivation, so none of the core circularity patterns apply in a direct way. The only fitted quantities before evaluation are the per-element reference-energy offsets Delta-mu_alpha in Eqs. (1)-(2), which are least-squares fitted on the retained training partition and then held fixed for validation/test. Because the offsets are composition-only and do not affect forces, and because the reported aligned test RMSE is the residual after removing only a linear element-wise shift, the test error is not forced to equal the fitted residual by construction; the SI even shows that ORB-V3's aligned energy RMSE (31.27 meV/atom) is close to its raw value, so the alignment does not manufacture the headline low errors. The zero-shot, fine-tuning, and from-scratch comparisons are controlled: the same 50-epoch protocol and the same partitions are used, and the paper explicitly states that the epoch-matched design measures data efficiency rather than wall-clock cost. The mixed-geometry conclusion is supported by held-out test evaluations; although the geometry-stratified split is not block-wise by AIMD trajectory, so near-duplicate frames could inflate adaptation gains, this is a data-split validity concern rather than a reduction of a prediction to its input, and the authors themselves list 'trajectory-level data correlations' as an unresolved remaining question. Property-level validations use external or independent references: elastic constants and phonons are compared with Materials Project mp-2858 DFT values, surface energies use DFT references, and the neck MD uses the same loading protocol as a published DFT AIMD reference; none of these quantities are implied by the training metrics by construction. Self-citations [26,27] motivate the desintering geometry and supply the loading protocol, while [38,44] are prior benchmark work and software; these are context and tooling, not load-bearing evidence for the transferability claims. No uniqueness theorem is imported, and no claimed result is defined in terms of itself. The abstract/SI discrepancy in the ORB-V3 energy RMSE (6 vs. 31.27 meV/atom) is a reporting inconsistency that should be resolved, but it is not circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

No new physical entities are postulated. The free parameters are calibration and filtering choices, not physical constants. The main unstated load is the consistency of DFT references and the representativeness of the geometry classes.

free parameters (2)
  • Elemental reference-energy corrections Δµ_Zr, Δµ_O (per model) = ORB-V3: Δµ_Zr = -0.192228 eV, Δµ_O = 0.086643 eV
    Fitted by least squares on the training partition (Eq. 1) to align model energies; the headline zero-shot energy RMSE depends on these offsets.
  • Force filter threshold = 3 eV/Å
    Configurations with any atomic force magnitude above 3 eV/Å were removed after stratified splitting (Section II.B); this manual threshold changes class sizes and partitions.
assumptions (3)
  • domain assumption DFT (VASP) reference energies and forces are ground truth for ZrO2 configurations
    Section II.A states the dataset was generated with DFT in VASP but does not specify functional, pseudopotentials, or convergence settings; all supervised metrics assume this reference is correct.
  • domain assumption Materials Project DFT reference (mp-2858) is comparable to the authors' DFT data
    Sections II.F and II.G compare MLIP elastic and phonon predictions against Materials Project values; if the two DFT settings differ, property errors include reference mismatch.
  • domain assumption The five geometry classes cover the relevant structural manifold for the desintering process
    The classes are motivated by experimental desintering [26,27], but the paper does not demonstrate that the training distribution matches deployment conditions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cross-Geometry Transferability Assessment of Universal Machine Learning Interatomic Potentials: From Bulk Materials to Atomic Nanowires." pith.science (2026). https://pith.science/paper/6SNHKPCW

@misc{pith2026260806662,
  author       = {Pith},
  title        = {Pith review of: Cross-Geometry Transferability Assessment of Universal Machine Learning Interatomic Potentials: From Bulk Materials to Atomic Nanowires},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6SNHKPCW}},
  note         = {Machine review of arXiv:2608.06662}
}
read the original abstract

Foundation machine-learning interatomic potentials (MLIPs) enable atomistic simulations at substantially lower computational cost than first-principles methods, but their reliability across structural geometries remains insufficiently understood. Here, we construct a density-functional-theory dataset of ZrO2 configurations spanning bulk, slab, particle, neck, and atomically thin wire environments motivated by an experimentally observed ZrO2 desintering process involving neck thinning and atomic wire formation. We first benchmark 26 pretrained MLIPs and observe pronounced geometry-dependent degradation in zero-shot predictions. Without any training, after only reference-energy alignment, the best zero-shot model (ORB-V3) reaches energy and force root-mean-square errors of 6 meV/atom and 197.3 meV/{\AA}, respectively, with the largest force errors in neck and wire configurations. We then compare zero-shot inference, fine-tuning, and training from scratch strategies. Fine-tuning yields lower energy and force errors than training from scratch, while both require comparable wall-clock time. Geometry-specific fine-tuning improves in-domain accuracy but frequently produces negative transfer to other structural classes, whereas mixed-geometry fine-tuning reduces cross-geometry errors. Evaluations of elastic and vibrational properties, surface energies, and neck dynamics further show that rankings based on average energy and force errors do not universally predict property-level behavior. These results demonstrate that geometry-diverse target data and independent physical validations are necessary when adapting foundation MLIPs to low-coordination (ionic) nanostructures.

Figures

Figures reproduced from arXiv: 2608.06662 by the authors.

Figure 1
Figure 1. FIG. 1. Overview of the ZrO [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Zero-shot evaluation on the ZrO [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Accuracy–cost comparison of zero-shot, fine-tuned, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Geometry-specific and mixed-geometry fine-tuning of MACE and ORB on the ZrO [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Progressively farther from equilibrium property-level evaluations of zero-shot, fine-tuned, and from-scratch MACE and [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 56 canonical work pages

  1. [1]

    This result is consistent with the strong representation of bulk crystalline environments in the pretraining data

    Elastic properties For the bulk-dependent properties in Figure 5A, zero- shot models exhibit the lowest errors for most quantities, with the stated MACE exceptions for the bulk modulus and Poisson’s ratio. This result is consistent with the strong representation of bulk crystalline environments in the pretraining data. Target-domain adaptation does not un...

    work page

  2. [2]

    The zero-shot models pro- vide the closest overall agreement with the DFT disper- sion, with phonon eigenvalue RMSEs of 0.785 THz for MACE and 0.540 THz for ORB

    Vibrational properties Figure 5B compares the phonon dispersions predicted by the zero-shot, fine-tuned, and from-scratch models with the DFT reference. The zero-shot models pro- vide the closest overall agreement with the DFT disper- sion, with phonon eigenvalue RMSEs of 0.785 THz for MACE and 0.540 THz for ORB. Fine-tuning increases these errors to 1.65...

    work page

  3. [3]

    For MACE, fine-tuning gives the lowest mean relative surface-energy error, ap- proximately 3.08%, compared with 3.33% for training from scratch and 16.14% for zero-shot inference

    Surface energies Table I shows that the ranking of adaptation strategies depends on the model family. For MACE, fine-tuning gives the lowest mean relative surface-energy error, ap- proximately 3.08%, compared with 3.33% for training from scratch and 16.14% for zero-shot inference. Fine- tuning outperforms the from-scratch model for 7 of the 11 listed surf...

    work page

  4. [4]

    RMSD is used to identify structural drift relative to the initial configuration after removing rigid translation and rotation

    Neck molecular dynamics Figure 5D provides an initial dynamical assessment us- ing 20-ps neck trajectories. RMSD is used to identify structural drift relative to the initial configuration after removing rigid translation and rotation. The fine-tuned trajectories remain bounded over most of the simulated interval, while the zero-shot MACE trajectory exhibi...

    work page 2024

  5. [5]

    G. R. Schleder, A. C. M. Padilha, C. M. Acosta, M. Costa, and A. Fazzio, From dft to machine learning: recent approaches to materials science–a review, Journal of Physics: Materials 2, 032001 (2019)

    work page 2019

  6. [6]

    D. R. Bowler and T. Miyazaki, O(n) methods in elec- tronic structure calculations, Reports on Progress in Physics 75, 036503 (2012)

    work page 2012

  7. [7]

    V. L. Deringer, M. A. Caro, and G. Cs´ anyi, Ma- chine learning interatomic potentials as emerging tools for materials science, Advanced Materials 31, 10.1002/adma.201902765 (2019)

    doi:10.1002/adma.201902765 2019

  8. [8]

    Y. Zuo, C. Chen, X. Li, Z. Deng, Y. Chen, J. Behler, G. Cs´ anyi, A. V. Shapeev, A. P. Thompson, M. A. Wood, and S. P. Ong, Performance and cost assessment of machine learning interatomic potentials, The Journal of Physical Chemistry A 124, 731–745 (2020)

    work page 2020

Show all 72 references
  1. [9]

    Mishin, Machine-learning interatomic potentials for materials science, Acta Materialia 214, 116980 (2021)

    Y. Mishin, Machine-learning interatomic potentials for materials science, Acta Materialia 214, 116980 (2021)

    2021

  2. [10]

    G. R. Schleder, A. C. M. Padilha, A. Reily Rocha, G. M. Dalpian, and A. Fazzio, Ab Initio simulations and mate- rials chemistry in the age of big data, Journal of Chemical Information and Modeling 60, 452–459 (2019)

    2019

  3. [11]

    Batzner, A

    S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials, Na- ture Communications 13, 10.1038/s41467-022-29939-5 (2022)

    2022 doi

  4. [12]

    Batatia, D

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csanyi, Mace: Higher order equivariant message pass- ing neural networks for fast and accurate force fields, in Advances in Neural Information Processing Systems , Vol. 35, edited by S. Koyejo, S. Mohamed, A. Agarwal, D. Bel...

  5. [13]

    Z. Yang, X. Wang, Y. Li, Q. Lv, C. Y.-C. Chen, and L. Shen, Efficient equivariant model for machine learning interatomic potentials, npj Computational Materials 11, 10.1038/s41524-025-01535-3 (2025)

    2025 doi

  6. [14]

    Chen and S

    C. Chen and S. P. Ong, A universal graph deep learn- ing interatomic potential for the periodic table, Nature Computational Science 2, 718 (2022)

    2022

  7. [15]

    Musaelian, S

    A. Musaelian, S. Batzner, A. Johansson, L. Sun, C. J. Owen, M. Kornbluth, and B. Kozinsky, Learning local equivariant representations for large-scale atomistic dy- namics, Nature Communications 14, 579 (2023)

    2023

  8. [16]

    T. L. Tompa, E. Varga-Umbrich, I. Batatia, A. M. Elena, N. Bernstein, and G. Cs´ anyi, Fine-tuning mlip founda- tion models: strategies for accuracy and transferability (2026)

    2026

  9. [17]

    Creed, T

    I. Creed, T. Rein, I. Vitenburgs, W. G. Stark, V. Ellings- son, A. Y. Ismail, G. Liu, Y. Lou, B. A. A. Martin, C. Bone, M. A. H. Walker, M. Taj, S. Wang, K. Wong, R. Wu, P. Kayastha, B. Cheng, A. Krishnapriyan, M. Ce- riotti, M. F. Langer, J. M. Frost, A. M. Ganose, V. Kapil, ...

    2026

  10. [19]

    M. K. Horton, P. Huck, R. X. Yang, J. M. Munro, S. Dwaraknath, A. M. Ganose, R. S. Kingsbury, M. Wen, J. X. Shen, T. S. Mathis, A. D. Kaplan, K. Berket, J. Riebesell, J. George, A. S. Rosen, E. W. C. Spotte- Smith, M. J. McDermott, O. A. Cohen, A. Dunn, and M. C. Kuner, Accele...

    2025

  11. [20]

    Schmidt, T

    J. Schmidt, T. F. T. Cerqueira, A. H. Romero, A. Loew, F. J¨ ager, H.-C. Wang, S. Botti, and M. A. L. Mar- ques, Improving machine-learning models in materials science through large datasets, Materials Today Physics 48, 101560 (2024)

    2024

  12. [21]

    J. P. Miller, R. Taori, A. Raghunathan, S. Sagawa, P. W. Koh, V. Shankar, P. Liang, Y. Carmon, and L. Schmidt, Accuracy on the line: On the strong correlation between out-of-distribution and in-distribution generalization, in Proceedings of the 38th International Conference on...

    2021

  13. [22]

    H. Yu, J. Liu, X. Zhang, J. Wu, and C. Peng, A survey on evaluation of out-of-distribution generalization (2024). 13

    2024

  14. [23]

    Benedini, A

    G. Benedini, A. Loew, M. Hellstr¨ om, S. Botti, and M. A. L. Marques, Universal machine learning potentials for systems with reduced dimensionality, AI for Science 1, 025005 (2025)

    2025

  15. [24]

    K. Li, A. N. Rubungo, X. Lei, D. Persaud, K. Choud- hary, B. DeCost, A. B. Dieng, and J. Hattrick-Simpers, Probing out-of-distribution generalization in machine learning for materials, Communications Materials 6, 10.1038/s43246-024-00731-w (2025)

    2025 doi

  16. [25]

    N. L. Fragapane and V. L. Deringer, Amorphous ma- terials as a frontier challenge for universal interatomic potentials (2026)

    2026

  17. [26]

    Pa¸ sca, H

    L.-B. Pa¸ sca, H. J. Snaith, and V. L. Deringer, Data- driven atomistic modelling of hybrid halide perovskite passivation (2026)

    2026

  18. [27]

    Verdi, F

    C. Verdi, F. Karsai, P. Liu, R. Jinnouchi, and G. Kresse, Thermal transport and phase transitions of zirconia by on-the-fly machine-learned interatomic potentials, npj Computational Materials 7, 156 (2021)

    2021

  19. [28]

    P. Liu, C. Verdi, F. Karsai, and G. Kresse, Phase transitions of zirconia: Machine-learned force fields be- yond density functional theory, Physical Review B 105, L060102 (2022)

    2022

  20. [29]

    Zhang, G

    J.-Y. Zhang, G. Huynh, T. Albaret, F.-Z. Dai, and D. Rodney, Transformation-induced plasticity in CeO2- ZrO2 ceramics: Atomic-scale insights using a deep neural network potential, Acta Materialia 285, 120661 (2025)

    2025

  21. [30]

    Focassio, T

    B. Focassio, T. E. R. Fiuza, J. Bettini, G. R. Schleder, M. H. M. Rodrigues, J. B. Souza Junior, E. R. Leite, A. Fazzio, and R. B. Capaz, Stability and rupture of an ultrathin ionic wire, Physical Review Letters129, 046101 (2022)

    2022

  22. [31]

    T. E. Fiuza, B. Focassio, J. Bettini, G. R. Schleder, M. H. Rodrigues, J. B. Souza Junior, A. Fazzio, R. B. Capaz, and E. R. Leite, Visualization of electron beam- induced desintering of nanostructured ceramics at the atomic scale, Cell Reports Physical Science 5, 101828 (2024)

    2024

  23. [32]

    T. W. Ko, J. A. Finkler, S. Goedecker, and J. Behler, A fourth-generation high-dimensional neural network po- tential with accurate electrostatics including non-local charge transfer, Nature Communications 12, 398 (2021)

    2021

  24. [33]

    Rumiantsev, M

    E. Rumiantsev, M. F. Langer, T.-E. Sodjargal, M. Ce- riotti, and P. Loche, Learning long-range representa- tions with equivariant messages, Transactions on Ma- chine Learning Research (2026), arXiv:2507.19382 [physics.chem-ph]

    2026

  25. [34]

    M. U. Maruf, S. Kim, and Z. Ahmad, Learning long-range interactions in equivariant machine learning interatomic potentials via electronic degrees of freedom, The Journal of Physical Chemistry Letters 16, 9078 (2025)

    2025

  26. [35]

    Kim and B

    D. Kim and B. Cheng, Long-range electrostatics for ma- chine learning interatomic potentials is easier than we thought, The Journal of Chemical Physics 164, 060901 (2026)

    2026

  27. [36]

    Cheng, Latent ewald summation for machine learning of long-range interactions, npj Computational Materials 11, 10.1038/s41524-025-01577-7 (2025)

    B. Cheng, Latent ewald summation for machine learning of long-range interactions, npj Computational Materials 11, 10.1038/s41524-025-01577-7 (2025)

    2025 doi

  28. [37]

    Kabylda, J

    A. Kabylda, J. T. Frank, S. Su´ arez-Dou, A. Khabibrakhmanov, L. Medrano Sandonas, O. T. Unke, S. Chmiela, K.-R. M¨ uller, and A. Tkatchenko, Molecular simulations with a pretrained neural network and universal pairwise force fields, Journal of the American Chemical Society 14...

    2025

  29. [38]

    Batatia, W

    I. Batatia, W. J. Baldwin, D. Kuryla, J. Hart, E. Kasoar, A. M. Elena, H. Moore, M. J. Gawkowski, B. X. Shi, V. Kapil, P. Kourtis, I.-B. Magd˘ au, and G. Cs´ anyi, Mace- polar-1: A polarisable electrostatic foundation model for molecular chemistry (2026)

    2026

  30. [39]

    Riebesell, R

    J. Riebesell, R. E. A. Goodall, P. Benner, Y. Chiang, B. Deng, G. Ceder, M. Asta, A. A. Lee, A. Jain, and K. A. Persson, A framework to evaluate machine learning crystal stability predictions, Nature Machine Intelligence 7, 836–847 (2025)

    2025

  31. [40]

    Maxson, A

    T. Maxson, A. Soyemi, X. Zhang, B. W. J. Chen, and T. Szilv´ asi, Ms25: Materials science-focused bench- mark data set for machine learning interatomic poten- tials, Journal of Chemical Information and Modeling 65, 8097–8112 (2025)

    2025

  32. [41]

    Jacobs, D

    R. Jacobs, D. Morgan, S. Attarian, J. Meng, C. Shen, Z. Wu, C. Y. Xie, J. H. Yang, N. Artrith, B. Blaiszik, G. Ceder, K. Choudhary, G. Csanyi, E. D. Cubuk, B. Deng, R. Drautz, X. Fu, J. Godwin, V. Honavar, O. Isayev, A. Johansson, B. Kozinsky, S. Martiniani, S. P. Ong, I. Polt...

    2025

  33. [42]

    Focassio, L

    B. Focassio, L. P. M. Freitas, and G. R. Schleder, Per- formance assessment of universal machine learning inter- atomic potentials: Challenges and directions for mate- rials’ surfaces, ACS Applied Materials & Interfaces 17, 13111–13121 (2025)

    2025

  34. [43]

    M. Z. Anam, O. Aghoghovbia, M. Al-Fahdi, L. Kong, V. Fung, and M. Hu, A comprehensive assessment and benchmark study of large atomistic foundation models for phonons, Advanced Intelligent Discovery 10.1002/aidi.202500075 (2025)

    2025 doi

  35. [44]

    D. Li, J. Yang, X. Chen, L. Yu, and S. Liu, Are foun- dational atomistic models reliable for finite-temperature molecular dynamics?, The Journal of Physical Chemistry C 129, 21538–21544 (2025)

    2025

  36. [45]

    Huang, B

    X. Huang, B. Deng, P. Zhong, A. D. Kaplan, K. A. Persson, and G. Ceder, Cross-functional transferability in foundation machine learning interatomic potentials, npj Computational Materials 11, 10.1038/s41524-025-01796- y (2025)

    2025 doi

  37. [46]

    Alampara, M

    N. Alampara, M. Schilling-Wilhelmi, and K. M. Jablonka, Lessons from the trenches on evaluating ma- chine learning systems in materials science, Computa- tional Materials Science 259, 114041 (2025)

    2025

  38. [47]

    Iftimie, P

    R. Iftimie, P. Minary, and M. E. Tuckerman, Ab initio molecular dynamics: Concepts, recent developments, and future trends, Proceedings of the National Academy of Sciences 102, 6654–6659 (2005)

    2005

  39. [48]

    Zanineli, S

    P. Zanineli, S. Mauri, B. Focassio, and G. Schleder, Ip- orch - interatomic potentials orchestrator (2026)

    2026

  40. [49]

    Hjorth Larsen, J

    A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. Bjerre Jensen, J. Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasb- jerg, S. Lysgaard...

    2017

  41. [50]

    J. Kim, J. You, Y. Park, Y. Lim, Y. Kang, J. Kim, H. Jeon, S. Ju, D. Hong, S. Y. Lee, S. Choi, Y. Kim, J. W. Lee, and S. Han, Optimizing cross-domain trans- fer for universal machine learning interatomic potentials, Nature Communications 17, 10.1038/s41467-026-70195- 8 (2026)

    2026 doi

  42. [52]

    R. Liu, E. Liu, J. Riebesell, J. Qi, S. P. Ong, and T. W. Ko, Matcalc: A python library for calculating materials properties from the potential energy surface (pes) (2024)

    2024

  43. [53]

    Pokluda, M

    J. Pokluda, M. ˇCern´ y, M.ˇSob, and Y. Umeno, Ab ini- tio calculations of mechanical properties: Methods and applications, Progress in Materials Science 73, 127–158 (2015)

    2015

  44. [54]

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, and G. Ceder, Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling, Nature Machine Intelligence 5, 1031–1041 (2023)

    2023

  45. [55]

    Y. Park, J. Kim, S. Hwang, and S. Han, Scalable par- allel algorithm for graph neural network interatomic potentials in molecular dynamics simulations, Journal of Chemical Theory and Computation 20, 4857–4868 (2024)

    2024

  46. [56]

    Zhang, A

    D. Zhang, A. Peng, C. Cai, W. Li, Y. Zhou, J. Zeng, M. Guo, C. Zhang, B. Li, H. Jiang, T. Zhu, W. Jia, L. Zhang, and H. Wang, A graph neural network for the era of large atomistic models, npj Computational Mate- rials 10.1038/s41524-026-02146-2 (2026)

    2026 doi

  47. [57]

    Bochkarev, Y

    A. Bochkarev, Y. Lysogorskiy, and R. Drautz, Graph atomic cluster expansion for semilocal interactions be- yond equivariant message passing, Physical Review X14, 10.1103/physrevx.14.021036 (2024)

    2024 doi

  48. [58]

    H. Yang, C. Hu, Y. Zhou, X. Liu, Y. Shi, J. Li, G. Li, Z. Chen, S. Chen, C. Zeni, M. Horton, R. Pinsler, A. Fowler, D. Z¨ ugner, X. Tian, J. Smith, L. Sun, W. Qian, L. Kong, C. Liu, H. Hao, and Z. Lu, Matter- sim: A deep learning atomistic model across elements, temperatures a...

    2024

  49. [59]

    Batatia, P

    I. Batatia, P. Benner, Y. Chiang, A. M. Elena, D. P. Kov´ acs, J. Riebesell, X. R. Advincula, M. Asta, M. Avaylon, W. J. Baldwin, F. Berger, N. Bernstein, A. Bhowmik, F. Bigi, S. M. Blau, V. C˘ arare, M. Ceriotti, S. Chong, J. P. Darby, S. De, F. Della Pia, V. L. Deringer, R. ...

    2025

  50. [61]

    Rhodes, S

    B. Rhodes, S. Vandenhaute, V. ˇSimkus, J. Gin, J. God- win, T. Duignan, and M. Neumann, Orb-v3: atomistic simulation at scale (2025)

    2025

  51. [62]

    Gubaev, V

    K. Gubaev, V. Zaverkin, P. Srinivasan, A. I. Duff, J. K¨ astner, and B. Grabowski, Performance of two complementary machine-learned potentials in modelling chemically complex systems, npj Computational Materi- als 9, 10.1038/s41524-023-01073-w (2023)

    2023 doi

  52. [63]

    R. Guo, G. Li, J. Tang, Y. Wang, and X. Song, Small- data-based machine learning interatomic potentials for graphene grain boundaries enabled by structural unit model, Carbon Trends 11, 100260 (2023)

    2023

  53. [64]

    A. P. Bart´ ok, M. C. Payne, R. Kondor, and G. Cs´ anyi, Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons, Physi- cal Review Letters 104, 10.1103/physrevlett.104.136403 (2010)

    2010 doi

  54. [65]

    J. Kim, J. Lee, S. Oh, Y. Park, S. Hwang, S. Han, S. Kang, and Y. Kang, An efficient forgetting-aware fine-tuning framework for pretrained universal machine- learning interatomic potentials, npj Computational Ma- terials 12, 26 (2026)

    2026

  55. [66]

    Kloppenburg, L

    J. Kloppenburg, L. B. P´ artay, H. J´ onsson, and M. A. Caro, A general-purpose machine learning Pt interatomic potential for an accurate description of bulk, surfaces and nanoparticles, The Journal of Chemical Physics 158, 134704 (2023)

    2023

  56. [67]

    Cross-Geometry Transferability Assessment of Foundation Machine Learning Interatomic Potentials: From bulk materials to atomic nanowires

    Y. Liu, J. D. Morrow, C. Ertural, N. L. Fragapane, J. L. A. Gardner, A. A. Naik, Y. Zhou, J. George, and V. L. Deringer, An automated framework for exploring and learning potential-energy surfaces, Nature Commu- nications 16, 7666 (2025). Supporting Information Cross-Geometry ...

    2025

  57. [68]

    Batatia, P

    I. Batatia, P. Benner, Y. Chiang, A. M. Elena, D. P. Kov´ acs, J. Riebesell, X. R. Advincula, M. Asta, M. Avaylon, W. J. Baldwin, F. Berger, N. Bernstein, A. Bhowmik, F. Bigi, S. M. Blau, V. C˘ arare, M. Ceriotti, S. Chong, J. P. Darby, S. De, F. Della Pia, V. L. Deringer, R. ...

    2025

  58. [69]

    Neumann, J

    M. Neumann, J. Gin, B. Rhodes, S. Bennett, Z. Li, H. Choubisa, A. Hussey, and J. Godwin, Orb: A fast, scalable neural network potential (2024)

    2024

  59. [70]

    Rhodes, S

    B. Rhodes, S. Vandenhaute, V. ˇSimkus, J. Gin, J. Godwin, T. Duignan, and M. Neumann, Orb-v3: atomistic simulation at scale (2025)

    2025

  60. [71]

    Gardner, graph-pes: train and use graph-based ml models of potential energy surfaces (2024)

    J. Gardner, graph-pes: train and use graph-based ml models of potential energy surfaces (2024)

    2024

  61. [72]

    M. K. Horton, P. Huck, R. X. Yang, J. M. Munro, S. Dwaraknath, A. M. Ganose, R. S. Kingsbury, M. Wen, J. X. Shen, T. S. Mathis, A. D. Kaplan, K. Berket, J. Riebesell, J. George, A. S. Rosen, E. W. C. Spotte-Smith, M. J. McDermott, O. A. Cohen, A. Dunn, and M. C. Kuner, Acceler...

    2025

  62. [73]

    A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and K. A. Persson, Commentary: The materials project: A materials genome approach to accelerating materials innovation, APL Materials 1, 10.1063/1.4812323 (2013)

    2013 doi

  63. [74]

    Hjorth Larsen, J

    A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Du/suppress lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. Bjerre Jensen, J. Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasbjerg, S. L...

    2017

  64. [75]

    Focassio, T

    B. Focassio, T. E. R. Fiuza, J. Bettini, G. R. Schleder, M. H. M. Rodrigues, J. B. Souza Junior, E. R. Leite, A. Fazzio, and R. B. Capaz, Stability and rupture of an ultrathin ionic wire, Physical Review Letters 129, 046101 (2022). 29

    2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.