Pith. sign in

REVIEW 4 major objections 5 minor 84 references

Heterogeneous Ensemble Enables a Universal Uncertainty Metric for Atomistic Foundation Models

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Eleven pretrained universal interatomic potentials, combined by inverse-RMSE weights, define a universal uncertainty metric $U$ that tracks true force errors and enables nearly DFT-free distillation.

desk verdict A genuinely useful uncertainty metric for uMLIPs, with a real in-sample calibration caveat that the paper's own external tests only partially address; deserves peer review. read the letter →

arxiv 2507.21297 v1 pith:24NEGV37 submitted 2025-07-28 cond-mat.mtrl-sci

Kai Liu , Zixiong Wei , Wei Gao , Poulumi Dey , Marcel H.F. Sluiter , Fei Shuang This is my paper
classification cond-mat.mtrl-sci
keywords uncertaintyquantificationuniversalmachinelearninginteratomicpotentialsheterogeneousensemblemodeldistillationatomisticfoundationmodelsforce-fieldhigh-entropyalloysdensityfunctionaltheory
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

The paper proposes a universal uncertainty metric, $U$, for universal machine-learning interatomic potentials (uMLIPs), i.e., foundation models that predict energies and forces across the periodic table. $U$ is the inverse-RMSE-weighted spread of force predictions across an eleven-member heterogeneous ensemble of pretrained potentials, and the paper argues that it correlates strongly with true force errors on an absolute scale without retraining or calibration. The central evidence is a Spearman $\rho = 0.87$ on the OMat24 test set and $\rho = 0.82$--$0.92$ on diverse external datasets spanning metals, alloys, inorganic compounds, MOFs, perovskites, and battery materials. On this basis the paper builds an uncertainty-aware distillation framework in which low-uncertainty configurations receive cheap teacher labels and high-uncertainty configurations receive targeted DFT labels. The payoff is that a tungsten ACE potential trained with about 4% DFT labels matches full-DFT training, while a MoNbTaW high-entropy-alloy potential is trained with no new DFT calculations and can even surpass the accuracy of the DFT reference labels.

What carries the argument

The central object is $U$ itself (Eq. 1), a configuration-level score built from an eleven-member heterogeneous uMLIP ensemble with inverse-RMSE weights (Eq. 2). For each configuration, each member predicts a force on every atom; $U$ takes the atom with the largest deviation from the ensemble-mean force vector, squares that deviation for each model, and sums with weights proportional to inverse force RMSE on the OMat24 test set. The same weighted ensemble mean also serves as the DFT surrogate in distillation, and the companion UAMD protocol partitions configurations by a cutoff $U_c$: low-$U$ configurations are labeled by the teacher ensemble, high-$U$ ones by DFT. This combination of weighted disagreement and cutoff-based data selection is what carries the argument.

What would settle it

Take a held-out collection of spin-polarized magnetic defects, molecular crystals, and surfaces with consistent DFT labels, compute $U$ and the true force error per configuration, and check whether the Spearman correlation remains near 0.8 or higher and whether the cutoff $U_c = 1$ eV/\AA still separates configurations with force RMSE at or below 0.1 eV/\AA. If either fails, the universality of $U$ and the recommended threshold would need to be abandoned or revised.

Watch

Extended reading notes

Core claim

The central claim is that $U$, defined by $U_i = \sqrt{ \sum_k w_k [\max_j |F_{i,j,k} - \langle F_{i,j}\rangle|]^2 }$ with weights $w_k = \mathrm{RMSE}_{F,k}^{-1} / \sum_{k'} \mathrm{RMSE}_{F,k'}^{-1}$, measures the true force error of uMLIP predictions for general inorganic materials. Over nearly five decades of $U$ ($10^{-3}$ to $10^2$ eV/\AA), the conditional spread around the ideal $y=x$ line stays within about one order of magnitude, so low-$U$ configurations almost never hide large errors and high-$U$ configurations reliably flag catastrophic ones. The paper further claims that this metric keeps its predictive ordering across metals, alloys, inorganic compounds, MOFs, perovskites, and battery materials, and that a universal cutoff $U_c = 1$ eV/\AA selects configurations with force RMSE at or below 0.1 eV/\AA. Distillation built on this metric propagates teacher accuracy into compact ACE potentials while filtering DFT numerical noise, so the student matches or in some cases surpasses the reference labels.

Load-bearing premise

The accuracy ranking of the eleven machine-learned potentials on the large public benchmark that sets their weights continues to hold for every new material system to which the uncertainty score is applied.

Editorial extensions

If this is right

  • A practical deployment monitor: configurations with $U$ exceeding roughly 1 eV/\AA can be flagged for recalculation, with force RMSE of the retained set at or below 0.1 eV/\AA across metals, inorganic compounds, and other material classes.
  • Uncertainty-aware distillation: for W, about 4% DFT labels at $U_c = 1$ eV/\AA suffice to match full-DFT-trained ACE potentials on energies, forces, phonons, and grain-boundary tensile response, and either full DFT or full uMLIP labels are worse than the optimal mixture.
  • Zero-DFT potentials: for MoNbTaW, an ACE potential trained entirely on uMLIP labels matches a DFT-trained ACE in force RMSE and reproduces elastic constants, vacancy migration barriers, and stress-strain behavior, with the smoothed labels filtering DFT numerical noise.
  • Extensible ensemble: any future uMLIP can enter the ensemble simply by evaluating its OMat24 force RMSE, improving $U$ without retraining or calibrating a new uncertainty model.
  • General-purpose potentials: augmenting the MoNbTaW dataset with 7,000 maximum-volume-selected defect configurations, all labeled by uMLIPs, yields a potential that reduces errors on grain-boundary deformation and fracture test sets while remaining in the interpolation regime during molecular dynamics.

Reading between the lines

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

  • If $U$'s transferability holds beyond the tested datasets, it could serve as a generic acquisition function for active learning, flagging which configurations most need DFT labels across any uMLIP application.
  • The paper's supplementary data show that inconsistent DFT protocols can inflate apparent errors: for example, non-spin-polarized TM23 labels for magnetic Fe and Co produce reference errors that can exceed $U$. This suggests the metric's calibration is partly conditioned on reference-label consistency, and a controlled spin-polarized benchmark would sharpen the universality claim.
  • The claim that teacher labels are smoother than raw DFT and therefore denoise the student could be tested directly by measuring the label-noise floor of uMLIP predictions versus DFT convergence parameters, rather than only through downstream student accuracy.
  • Weighting by OMat24 force RMSE is a single global ranking; if future uMLIPs specialize by chemistry or structure, per-domain weight sets might improve $U$ further, but they would also trade away the metric's simplicity.
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 proposes U, a weighted ensemble uncertainty metric for universal machine learning interatomic potentials (uMLIPs), in which eleven pretrained uMLIPs are combined with inverse-force-RMSE weights derived from the OMat24 test set. The authors report Spearman correlations of 0.87 on OMat24 and 0.82-0.92 on external datasets, and use U to define an uncertainty-aware distillation (UAMD) workflow in which high-uncertainty configurations receive DFT labels. They demonstrate that a tungsten ACE potential trained with roughly 4% DFT labels matches a fully DFT-trained model on phonons, stacking fault energies, and bicrystal tensile behavior, and that a MoNbTaW ACE potential can be trained entirely on uMLIP labels with accuracy comparable to a DFT-trained ACE.

Significance. If the correlation of U with true force errors is robust out of sample, the metric would give uMLIP users a practical quality monitor and a data-selection criterion that requires no additional training, while the UAMD results would substantially lower the cost of building accurate system-specific potentials. The paper's strengths include the reuse of a diverse set of pretrained models, the breadth of external validation datasets, the inclusion of physical property tests (phonons, grain-boundary tension, NEB barriers), and the public availability of code and data. The main risk is that the metric's parameters are calibrated on the OMat24 test set, so the headline in-sample correlation does not by itself certify the claim of universality; the paper's own exceptions for magnetic elements and carbon point to conditions under which the weighting could break down.

major comments (4)
  1. [Results: Universal uncertainty metric U via heterogeneous ensemble, Eqs. (1)-(2), Fig. 2b-c] The metric's parameters—the weights w_k in Eq. (2) and the ensemble size of eleven in Fig. 2b—are determined on the OMat24 test set, and the headline Spearman rho=0.87 is computed on the same set. The external validation in Fig. 3a uses these frozen weights, which demonstrates that the fixed metric transfers, but it does not test whether the relative accuracy ordering of the individual uMLIPs, and hence the weighting, is stable across chemistries. If a new material class changes that ordering, the weighted ensemble can be dominated by models whose errors are unrepresentative, and U would become miscalibrated. Please provide an out-of-sample test of the weighting scheme itself: for example, re-derive w_k and the ensemble size on a training portion of OMat24 and evaluate on a held-out portion and on the external datasets, or report per-model force RMSEs on each external dataset to show that the ranking underlying Eq. (2) is stable. In addition, the abstract's statement that U is obtained without requiring calibration is inconsistent with the explicit calibration of w_k and the ensemble size on OMat24, and should be rephrased.
  2. [Validation of U across diverse materials; Supplementary Note 1; Fig. 3c] The paper itself identifies two exceptions to the strong U-error coupling: Supplementary Note 1 states that for magnetic elements in TM23 (Fe, Co, Nb) the DFT reference errors can exceed U, 'obscuring the relationship between U and true model uncertainty,' and Fig. 3c identifies carbon as a system where 'uncertainty and error are less tightly coupled.' These are precisely the out-of-distribution settings where a global OMat24-derived weighting could break down, and the tail behavior of U is what matters most for flagging catastrophic errors. Please quantify these deviations—the fraction of configurations affected and the magnitude of underestimation—and discuss whether the recommended U_c=1 eV/Å cutoff remains safe in these cases, or whether a materially different operating point is needed.
  3. [Uncertainty-aware model distillation for W, Fig. 4d] The claim that a W potential trained with only 4% DFT labels matches full-DFT training is qualified by Fig. 4d, which shows that elastic constants (C11, C12, C44, bulk modulus, Poisson's ratio) exhibit relative errors approaching 10% at the recommended operating point U_c=1 eV/Å. This is a substantial deviation for mechanical property prediction, and it is not acknowledged in the abstract or conclusions. Please report the actual numeric RMSE or relative-error values at the 4% operating point for all quantities in Fig. 4d, and either refine the claim of 'comparable accuracy' to state the property-dependent accuracy explicitly or evaluate whether a more conservative U_c reduces the elastic-constant errors to an acceptable level.
  4. [Uncertainty-aware model distillation for MoNbTaW alloys, Fig. 5b] It is unclear whether the RMSE values reported in Fig. 5b are computed on the training set (the 17,654 configurations used to fit ACE_UAMD and ACE_DFT) or on a held-out test portion; the text does not describe a train/test split for this dataset. If these are training-set fitting errors, they do not support the claim that UAMD achieves accuracy comparable to ACE_DFT on unseen data. Please specify the evaluation protocol, and if the reported numbers are in-sample, provide cross-validated or held-out errors for the four scenarios in Fig. 5b.
minor comments (5)
  1. [Fig. 4 caption and text] The caption of Fig. 4 appears to mislabel the panels: the text refers to Fig. 4d for basic properties, Fig. 4e for phonon/stacking-fault comparisons, and Fig. 4f for stress-strain curves, while the caption lists these as c, d-e, and f. Please align the caption with the panel labels used in the text.
  2. [Abstract and Introduction] The phrase 'without requiring additional training or calibration' is misleading given that the weights in Eq. (2) and the ensemble size are calibrated on the OMat24 test set; consider replacing it with 'without additional training of uMLIPs' or explicitly describing the lightweight calibration step.
  3. [Fig. 5f and Section 'Uncertainty-aware model distillation for MoNbTaW alloys'] The 'Atom lost' error noted in Fig. 5f is not explained; please indicate whether this reflects a numerical instability of the potential, a limitation of the simulation setup, or a failure of the model, because the reader needs to interpret the premature termination of some tensile simulations.
  4. [Discussion and Data availability] The statement 'complete avoidance of DFT in the expanded MoNbTaW dataset' is accurate only for the UAMD labeling phase; the initial 17,654-configuration dataset from Ref. (36) was itself generated with DFT. Please add a clarifying clause to avoid overinterpretation.
  5. [Eqs. (1)-(3)] The uniform-weight baseline U^(0) is not written explicitly; providing its formula would help readers compare it with U^(1) and U^(2). Also, please clarify in the text that the max over j in Eq. (1) is taken over atoms within configuration i for each model k, and that the average force in Eq. (1) is the unweighted or weighted mean as appropriate.

Circularity Check

1 steps flagged · score 5.0 of 10

OMat24 is used both to set the inverse-RMSE weights and ensemble size and to report the headline Spearman rho=0.87, so that headline is in-sample; external tests with frozen weights provide independent support.

  1. fitted input called prediction [Eq. (2), Table S1, and Fig. 2b-c; Results section 'Universal uncertainty metric U via heterogeneous ensemble']
    "we evaluate candidate ensembles on the public OMat24 test set ... Fig. 2b shows Spearman’s ρ for U(0), U(1) and U(2) as a function of ensemble size. For U(0), optimal performance is obtained with six uMLIP models (ρ= 0.82); ... Both U(1) and U(2) outperform U(0), reaching local maxima of ρ = 0.87 and ρ = 0.86, respectively, at an ensemble size of eleven ... Fig. 2c shows a hexbin parity plot of the predicted uncertainty U against the actual force error on the OMat24 test set. ... with Spearman’s ρ = 0.87"

    The metric's parameters are fitted on the OMat24 test set: the weights w_k in Eq. (2) are inverse force RMSEs measured on OMat24 test labels (Table S1), and the eleven-model ensemble is selected by maximizing Spearman's rho between U and force error on that same OMat24 test set (Fig. 2b). Reporting rho = 0.87 on the OMat24 test set (Fig. 2c) is therefore an in-sample statistic: the same labels that fixed the weights and selected the ensemble are used to evaluate the correlation. By construction, the OMat24 value is an optimum of the selection procedure, not an independent prediction. The external datasets (Fig. 3), W distillation (Fig. 4), and MoNbTaW results (Figs.

full rationale

The main circular step is calibration-evaluation leakage on OMat24: U in Eq. (1) depends on weights w_k from Eq. (2), which are inverse force RMSEs computed on the OMat24 test set, and the ensemble size is chosen by maximizing Spearman's rho on that same test set (Fig. 2b). The reported rho = 0.87 on OMat24 (Fig. 2c) is thus in-sample and partly forced by construction, matching the 'fitted input called prediction' pattern. This does not make the paper wholly circular: the external validation on metals/alloys, inorganic compounds, and other materials (Fig. 3) uses frozen weights on independent DFT datasets, and the UAMD results for W and MoNbTaW are evaluated against true DFT references, providing genuine independent confirmation of the metric's usefulness. No load-bearing self-citation chain is present: prior work by the same group (Refs. 28, 29, 36) supplies datasets and benchmark context, but the key comparisons are recomputed in this paper. The paper itself flags limitations (Supplementary Note 1 on magnetic elements in TM23, and the carbon exception in Fig. 3c), which are honest caveats rather than circular reasoning. Overall, the universal claim has independent content, but the headline OMat24 number is an in-sample optimum, so a moderate score is appropriate.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The ledger is heavy on calibration parameters: weights, ensemble size, uncertainty cutoff, and surrogate choice are all fitted to OMat24 DFT labels, which means the 'DFT-free' deployment is a transfer of a DFT-calibrated mapping. There are no new physical entities; U is an ensemble statistic. The main axioms are representativeness and transferability of OMat24-based model rankings, plus the assumption that DFT labels are a valid ground truth despite the paper's own noise argument.

free parameters (4)
  • Ensemble weights w_k = 0.1218 (eSEN-30M-omat) to 0.0565 (Orb-v3-d-inf-omat), Table S1
    Inverse force RMSE on OMat24 test set (Eq. 2); these weights define U and are fitted to DFT labels.
  • Ensemble size = 11 models
    Chosen to maximize Spearman rho between U and force error on OMat24 test set (Fig. 2b); in-sample selection.
  • Uncertainty cutoff U_c = 1 eV/A (rule of thumb; 0.3 eV/A for RMSE near 0.05 eV/A)
    Calibrated as the intersection with RMSE=0.1 eV/A in Fig. 3d,f,h; used for UAMD data selection.
  • Surrogate model choice = eqV2-31M-omat for forces, eqV2-31M-OAM for energies
    Selected because it matches or outperforms the ensemble mean in accuracy-coverage curves (Fig. 2e,f); this choice affects all downstream UAMD results.
assumptions (4)
  • domain assumption OMat24 test set is representative of general inorganic materials
    Used as the calibration distribution for weights, ensemble size and U_c; if target systems lie far from OMat24 in composition/structure, the weighted ensemble may be miscalibrated. Invoked in Results 'Universal uncertainty metric' and Fig. 2.
  • domain assumption Relative uMLIP accuracy ordering is transferable from OMat24 to new systems
    Weights from OMat24 force RMSEs are applied unchanged to metals, alloys, compounds, MOFs, etc. in Fig. 3; partially validated by high Spearman rho but not guaranteed for arbitrary chemistries.
  • domain assumption DFT reference labels are accurate enough to serve as ground truth for force errors
    Used in all validation plots (Fig. 2c, 3a, etc.); the paper later argues DFT labels contain numerical noise, so the 'true error' itself is partly noisy, complicating interpretation of the correlation.
  • domain assumption Max over atoms of weighted force deviation is a meaningful configuration-level risk measure
    U is defined with max_j over atoms in Eq. 1; this choice is not derived and could miss distributed errors that are large on average but not maximal per atom.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Heterogeneous Ensemble Enables a Universal Uncertainty Metric for Atomistic Foundation Models." pith.science (2026). https://pith.science/paper/24NEGV37

@misc{pith2026250721297,
  author       = {Pith},
  title        = {Pith review of: Heterogeneous Ensemble Enables a Universal Uncertainty Metric for Atomistic Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/24NEGV37}},
  note         = {Machine review of arXiv:2507.21297}
}
abstract

Universal machine learning interatomic potentials (uMLIPs) are reshaping atomistic simulation as foundation models, delivering near \textit{ab initio} accuracy at a fraction of the cost. Yet the lack of reliable, general uncertainty quantification limits their safe, wide-scale use. Here we introduce a unified, scalable uncertainty metric \(U\) based on a heterogeneous model ensemble with reuse of pretrained uMLIPs. Across chemically and structurally diverse datasets, \(U\) shows a strong correlation with the true prediction errors and provides a robust ranking of configuration-level risk. Leveraging this metric, we propose an uncertainty-aware model distillation framework to produce system-specific potentials: for W, an accuracy comparable to full-DFT training is achieved using only \(4\%\) of the DFT labels; for MoNbTaW, no additional DFT calculations are required. Notably, by filtering numerical label noise, the distilled models can, in some cases, surpass the accuracy of the DFT reference labels. The uncertainty-aware approach offers a practical monitor of uMLIP reliability in deployment, and guides data selection and fine-tuning strategies, thereby advancing the construction and safe use of foundation models and enabling cost-efficient development of accurate, system-specific potentials.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

84 extracted references · 56 canonical work pages

  1. [1]

    Metals and Alloys: Pure metals (e.g., Fe, Mg), the full set of transition metals (TM23), and medium- to high-entropy alloys such as CrCoNi, VCoNi, MoNbTaVW, MoNbTaTi, WTaCrV, and the M16 binary alloys (264,383 configurations, 13,701,879 atoms)

  2. [2]

    Inorganic Compounds: Compounds and interstitial solid solutions involving light elements (H, C, N, O) combined with metals, including FeH, LiH, FeC, MoNbTaWH, HfO 2, Ga2O3, and GaN (49,092 configurations, 4,412,916 atoms)

  3. [3]

    low- uncertainty

    Other Materials: Carbon, metal–organic frameworks (MOFs), ionic covalent organic frame- works (ICOFs), surface-catalytic structures, perovskites, and battery-relevant materials such as LiPS, Li 4P2O7, and various cathode compositions (64,464 configurations, 4,868,199 atoms). Fig. 3a plots the predicted uncertainty𝑈 against the true force error for each of...

  4. [4]

    Behler, Perspective: Machine learning potentials for atomistic simulations

    J. Behler, Perspective: Machine learning potentials for atomistic simulations. The Journal of chemical physics 145 (17) (2016)

    2016

  5. [5]

    Deng, et al., CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling

    B. Deng, et al., CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling. Nature Machine Intelligence5 (9), 1031–1041 (2023)

    work page 2023

  6. [6]

    Barroso-Luque, et al., Open materials 2024 (omat24) inorganic materials dataset and models

    L. Barroso-Luque, et al., Open materials 2024 (omat24) inorganic materials dataset and models. arXiv preprint arXiv:2410.12771 (2024)

    arXiv 2024

  7. [7]

    Riebesell, et al., A framework to evaluate machine learning crystal stability predictions

    J. Riebesell, et al., A framework to evaluate machine learning crystal stability predictions. Nature Machine Intelligence7 (6), 836–847 (2025)

    work page 2025

  8. [8]

    C. Chen, S. P. Ong, A universal graph deep learning interatomic potential for the periodic table. Nature Computational Science 2 (11), 718–728 (2022)

    work page 2022

Show all 84 references
  1. [9]

    Batatia, et al

    I. Batatia, et al. , A foundation model for atomistic materials chemistry. arXiv preprint arXiv:2401.00096 (2023)

    2023 arXiv

  2. [10]

    Rhodes, et al

    B. Rhodes, et al. , Orb-v3: atomistic simulation at scale. arXiv preprint arXiv:2504.06231 (2025)

    2025 arXiv

  3. [11]

    Y. Park, J. Kim, S. Hwang, S. Han, Scalable Parallel Algorithm for Graph Neural Network Interatomic Potentials in Molecular Dynamics Simulations. Journal of Chemical Theory and Computation 20 (11), 4857–4868 (2024), doi:10.1021/acs.jctc.4c00190, http://dx.doi. org/10.1021/acs....

    2024 doi

  4. [12]

    Y.-L. Liao, B. Wood, A. Das, T. Smidt, Equiformerv2: Improved equivariant transformer for scaling to higher-degree representations. arXiv preprint arXiv:2306.12059 (2023)

    2023 arXiv

  5. [13]

    Deng, et al., Systematic softening in universal machine learning interatomic potentials

    B. Deng, et al., Systematic softening in universal machine learning interatomic potentials. npj Computational Materials 11 (1), 1–9 (2025)

    2025

  6. [14]

    Focassio, L

    B. Focassio, L. P. M. Freitas, G. R. Schleder, Performance assessment of universal machine learning interatomic potentials: Challenges and directions for materials’ surfaces.ACS Applied Materials & Interfaces (2024). 29

    2024

  7. [15]

    Wines, K

    D. Wines, K. Choudhary, CHIPS-FF: Evaluating Universal Machine Learning Force Fields for Material Properties. ACS Materials Letters 7 (6), 2105–2114 (2025), doi:10.1021/ acsmaterialslett.5c00093, http://dx.doi.org/10.1021/acsmaterialslett.5c00093

    2025 doi

  8. [16]

    A. M. Elena, et al., Machine Learned Potential for High-Throughput Phonon Calculations of Metal-Organic Frameworks. arXiv preprint arXiv:2412.02877 (2024)

    2024

  9. [17]

    Kim, et al., An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials

    J. Kim, et al., An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials. arXiv preprint arXiv:2506.15223 (2025)

    2025 arXiv

  10. [18]

    R. Wang, Y. Gao, H. Wu, Z. Zhong, PFD: Automatically Generating Machine Learning Force Fields from Universal Models. arXiv preprint arXiv:2502.20809 (2025)

    2025

  11. [19]

    J. L. Gardner, et al. , Distillation of atomistic foundation models across architectures and chemical domains. arXiv preprint arXiv:2506.10956 (2025)

    2025 arXiv

  12. [20]

    M. Wen, E. B. Tadmor, Uncertainty quantification in molecular simulations with dropout neural network potentials. npj computational materials 6 (1), 124 (2020)

    2020

  13. [21]

    A. A. Peterson, R. Christensen, A. Khorshidi, Addressing uncertainty in atomistic machine learning. Physical Chemistry Chemical Physics 19 (18), 10978–10985 (2017)

    2017

  14. [22]

    A. Zhu, S. Batzner, A. Musaelian, B. Kozinsky, Fast uncertainty estimates in deep learning interatomic potentials. The Journal of Chemical Physics 158 (16) (2023)

    2023

  15. [23]

    I. Best, T. Sullivan, J. Kermode, Uncertainty quantification in atomistic simulations of silicon using interatomic potentials. The Journal of Chemical Physics 161 (6) (2024)

    2024

  16. [24]

    Y. Hu, J. Musielewicz, Z. W. Ulissi, A. J. Medford, Robust and scalable uncertainty estimation with conformal prediction for machine-learned interatomic potentials. Machine Learning: Science and Technology 3 (4), 045028 (2022)

    2022

  17. [25]

    Schwalbe-Koda, S

    D. Schwalbe-Koda, S. Hamel, B. Sadigh, F. Zhou, V. Lordi, Model-free estimation of com- pleteness, uncertainties, and outliers in atomistic machine learning using information theory. Nature Communications 16 (1), 4014 (2025). 30

    2025

  18. [26]

    J. A. Bilbrey, J. S. Firoz, M.-S. Lee, S. Choudhury, Uncertainty quantification for neural network potential foundation models. npj Computational Materials 11 (1), 109 (2025)

    2025

  19. [27]

    Musielewicz, J

    J. Musielewicz, J. Lan, M. Uyttendaele, J. R. Kitchin, Improved Uncertainty Estimation of Graph Neural Network Potentials Using Engineered Latent Space Distances. The Journal of Physical Chemistry C 128 (49), 20799–20810 (2024)

    2024

  20. [28]

    A. R. Tan, S. Urata, S. Goldman, J. C. Dietschreit, R. G ´omez-Bombarelli, Single-model uncertainty quantification in neural network potentials does not consistently outperform model ensembles. npj Computational Materials 9 (1), 225 (2023)

    2023

  21. [29]

    Yang, et al., Mattersim: A deep learning atomistic model across elements, temperatures and pressures

    H. Yang, et al., Mattersim: A deep learning atomistic model across elements, temperatures and pressures. arXiv preprint arXiv:2405.04967 (2024)

    2024 arXiv

  22. [30]

    B. M. Wood, et al., UMA: A Family of Universal Models for Atoms (2025),https://arxiv. org/abs/2506.23971

    2025

  23. [31]

    Shuang, Z

    F. Shuang, Z. Wei, K. Liu, W. Gao, P. Dey, Universal machine learning interatomic potentials poised to supplant DFT in modeling general defects in metals and random alloys.arXiv preprint arXiv:2502.03578 (2025)

    2025 arXiv

  24. [32]

    Shuang, et al

    F. Shuang, et al. , Modeling extensive defects in metals through classical potential-guided sampling and automated configuration reconstruction. npj Computational Materials 11 (1), 118 (2025)

    2025

  25. [33]

    Byggm ¨astar, A

    J. Byggm ¨astar, A. Hamedani, K. Nordlund, F. Djurabekova, Machine-learning interatomic potential for radiation damage and defects in tungsten. Physical Review B 100 (14), 144105 (2019)

    2019

  26. [34]

    Gubler, J

    M. Gubler, J. A. Finkler, S. R. Jensen, S. Goedecker, L. Frediani, Noise-Tolerant Force Calculations in Density Functional Theory: A Surface Integral Approach for Wavelet- Based Methods. The Journal of Physical Chemistry A 129 (5), 1469–1477 (2025), doi: 10.1021/acs.jpca.4c067...

    2025 doi

  27. [35]

    Janssen, E

    J. Janssen, E. Makarov, T. Hickel, A. V. Shapeev, J. Neugebauer, Automated optimization and uncertainty quantification of convergence parameters in plane wave density functional theory calculations. npj Computational Materials 10 (1) (2024), doi:10.1038/s41524-024-01388-2, htt...

    2024 doi

  28. [36]

    Pan, et al., Atomic cluster expansion interatomic potential for defects and thermodynamics of Cu–W system

    J. Pan, et al., Atomic cluster expansion interatomic potential for defects and thermodynamics of Cu–W system. Journal of Applied Physics 136 (15) (2024)

    2024

  29. [37]

    S. Xu, Y. Su, L. T. Smith, I. J. Beyerlein, Frank-Read source operation in six body-centered cubic refractory metals. Journal of the Mechanics and Physics of Solids 141, 104017 (2020)

    2020

  30. [38]

    X.-G. Li, C. Chen, H. Zheng, Y. Zuo, S. P. Ong, Complex strengthening mechanisms in the NbMoTaW multi-principal element alloy. npj Computational Materials 6 (1) (2020), doi: 10.1038/s41524-020-0339-0, http://dx.doi.org/10.1038/s41524-020-0339-0

    2020 doi

  31. [39]

    Shuang, Y

    F. Shuang, Y. Ji, L. Laurenti, P. Dey, Size-dependent strength superiority in multi-principal el- ement alloys versus constituent metals: Insights from machine-learning atomistic simulations. International Journal of Plasticity 188, 104308 (2025), doi:10.1016/j.ijplas.2025.104...

    2025

  32. [40]

    Cui, et al

    T. Cui, et al. , Online test-time adaptation for better generalization of interatomic po- tentials to out-of-distribution data. Nature Communications 16 (1) (2025), doi:10.1038/ s41467-025-57101-4, http://dx.doi.org/10.1038/s41467-025-57101-4

    2025 doi

  33. [41]

    Shuang, L

    F. Shuang, L. Laurenti, P. Dey, Standard deviation in maximum restoring force controls the intrinsic strength of face-centered cubic multi-principal element alloys. Acta Materialia 282, 120508 (2025), doi:10.1016/j.actamat.2024.120508, http://dx.doi.org/10.1016/ j.actamat.2024.120508

    2025

  34. [42]

    I. S. Novikov, K. Gubaev, E. V. Podryabinkin, A. V. Shapeev, The MLIP package: moment tensor potentials with MPI and active learning. Machine Learning: Science and Technol- ogy 2 (2), 025002 (2020), doi:10.1088/2632-2153/abc9fe, https://dx.doi.org/10.1088/ 2632-2153/abc9fe. 32

    2020 doi

  35. [43]

    Zhang, et al., DPA-2: a large atomic model as a multi-task learner

    D. Zhang, et al., DPA-2: a large atomic model as a multi-task learner. npj Computational Materials 10 (1) (2024), doi:10.1038/s41524-024-01493-2, http://dx.doi.org/10.1038/ s41524-024-01493-2

    2024 doi

  36. [44]

    Y.-L. Liao, T. Smidt, M. Shuaibi, A. Das, Generalizing Denoising to Non-Equilibrium Struc- tures Improves Equivariant Force Fields (2024), https://arxiv.org/abs/2403.09549

    2024 arXiv

  37. [45]

    Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials.Phys- ical Review B 99 (1), 014104 (2019)

    R. Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials.Phys- ical Review B 99 (1), 014104 (2019)

    2019

  38. [46]

    L. C. Erhard, J. Rohrer, K. Albe, V. L. Deringer, Modelling atomic and nanoscale struc- ture in the silicon–oxygen system through active machine learning. Nature Communi- cations 15 (1) (2024), doi:10.1038/s41467-024-45840-9, http://dx.doi.org/10.1038/ s41467-024-45840-9

    2024 doi

  39. [47]

    Kresse, J

    G. Kresse, J. Furthm¨ uller, Efficient iterative schemes for 𝑎𝑏 𝑖𝑛𝑖𝑡𝑖𝑜 total-energy calcu- lations using a plane-wave basis set. Physical Review B 54, 11169–11186 (1996), doi: 10.1103/PhysRevB.54.11169

    1996 doi

  40. [48]

    J. P. Perdew, K. Burke, M. Ernzerhof, Generalized Gradient Approximation Made Simple. Physical Review Letters 77, 3865–3868 (1996), doi:10.1103/PhysRevLett.77.3865

    1996 doi

  41. [49]

    V. Wang, N. Xu, J.-C. Liu, G. Tang, W.-T. Geng, V ASPKIT: A user-friendly interface facilitating high-throughput computing and analysis using V ASP code. Computer Physics Communica- tions 267, 108033 (2021), doi:10.1016/j.cpc.2021.108033, http://dx.doi.org/10.1016/ j.cpc.2021.108033

    2021

  42. [50]

    A. P. Thompson, et al., LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales.Comp. Phys. Comm.271, 108171 (2022), doi:10.1016/j.cpc.2021.108171

    2022

  43. [51]

    Stukowski, Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool

    A. Stukowski, Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool. Modelling and simulation in materials science and engineering 18 (1), 015012 (2009). 33

    2009

  44. [52]

    Zheng, et al., Grain boundary properties of elemental metals

    H. Zheng, et al., Grain boundary properties of elemental metals. Acta Materialia 186, 40–49 (2020)

    2020

  45. [53]

    Fu, et al

    X. Fu, et al. , Learning smooth and expressive interatomic potentials for physical property prediction. arXiv preprint arXiv:2502.12147 (2025)

    2025 arXiv

  46. [54]

    Kim, et al., Data-efficient multifidelity training for high-fidelity machine learning interatomic potentials

    J. Kim, et al., Data-efficient multifidelity training for high-fidelity machine learning interatomic potentials. Journal of the American Chemical Society 147 (1), 1042–1054 (2024)

    2024

  47. [55]

    Bochkarev, Y

    A. Bochkarev, Y. Lysogorskiy, R. Drautz, Graph atomic cluster expansion for semilocal inter- actions beyond equivariant message passing. Physical Review X 14 (2), 021036 (2024)

    2024

  48. [56]

    Zhang, et al., Graph neural network model for the era of large atomistic models

    D. Zhang, et al., Graph neural network model for the era of large atomistic models. arXiv preprint arXiv:2506.01686 (2025)

    2025

  49. [57]

    Sheriff, Y

    K. Sheriff, Y. Cao, T. Smidt, R. Freitas, Quantifying chemical short-range order in metallic alloys. Proceedings of the National Academy of Sciences 121 (25), e2322962121 (2024)

    2024

  50. [58]

    R. Jana, M. A. Caro, Searching for iron nanoparticles with a general-purpose Gaussian approx- imation potential. Physical Review B 107 (24), 245421 (2023)

    2023

  51. [59]

    Zheng, et al., Multi-scale investigation of short-range order and dislocation glide in MoNbTi and TaNbTi multi-principal element alloys

    H. Zheng, et al., Multi-scale investigation of short-range order and dislocation glide in MoNbTi and TaNbTi multi-principal element alloys. npj Computational Materials 9 (1) (2023), doi: 10.1038/s41524-023-01046-z, http://dx.doi.org/10.1038/s41524-023-01046-z

    2023 doi

  52. [60]

    Wang, et al., Unraveling dislocation-based strengthening in refractory multi-principal ele- ment alloys

    T. Wang, et al., Unraveling dislocation-based strengthening in refractory multi-principal ele- ment alloys. npj Computational Materials 10 (1) (2024), doi:10.1038/s41524-024-01330-6, http://dx.doi.org/10.1038/s41524-024-01330-6

    2024 doi

  53. [61]

    Lyu, et al., Statistical mechanics, entropy and temperature analog of dislocations moving on fluctuating resistance landscapes

    S. Lyu, et al., Statistical mechanics, entropy and temperature analog of dislocations moving on fluctuating resistance landscapes. Acta Materialia 291, 121002 (2025), doi:10.1016/j.actamat. 2025.121002, http://dx.doi.org/10.1016/j.actamat.2025.121002

    2025

  54. [62]

    Byggm ¨astar, et al., Segregation, ordering, and precipitation in WTaV-based concentrated refractory alloys

    J. Byggm ¨astar, et al., Segregation, ordering, and precipitation in WTaV-based concentrated refractory alloys. Acta Materialia 296, 121276 (2025), doi:10.1016/j.actamat.2025.121276, http://dx.doi.org/10.1016/j.actamat.2025.121276. 34

    2025

  55. [63]

    M. Poul, L. Huber, E. Bitzek, J. Neugebauer, Systematic atomic structure datasets for machine learning potentials: Application to defects in magnesium.Physical Review B 107 (10), 104103 (2023)

    2023

  56. [64]

    C. J. Owen, et al., Complexity of many-body interactions in transition metals via machine- learned force fields from the TM23 data set. npj Computational Materials 10 (1), 92 (2024)

    2024

  57. [65]

    Song, et al., General-purpose machine-learned potential for 16 elemental metals and their alloys

    K. Song, et al., General-purpose machine-learned potential for 16 elemental metals and their alloys. Nature Communications 15 (1), 10208 (2024)

    2024

  58. [66]

    F.-S. Meng, et al., General-purpose neural network interatomic potential for the𝛼-iron and hy- drogen binary system: Toward atomic-scale understanding of hydrogen embrittlement.Physical Review Materials 5 (11), 113606 (2021)

    2021

  59. [67]

    Shuang, et al., Decoding the hidden dynamics of super-Arrhenius hydrogen diffusion in multi-principal element alloys via machine learning

    F. Shuang, et al., Decoding the hidden dynamics of super-Arrhenius hydrogen diffusion in multi-principal element alloys via machine learning. Acta Materialia 289, 120924 (2025)

    2025

  60. [68]

    P. Ying, et al., Highly efficient path-integral molecular dynamics simulations with GPUMD using neuroevolution potentials: Case studies on thermal properties of materials.The Journal of Chemical Physics 162 (6) (2025), doi:10.1063/5.0241006, http://dx.doi.org/10.1063/ 5.0241006

    2025 doi

  61. [69]

    Meng, et al., A highly transferable and efficient machine learning interatomic potentials study of𝛼-Fe–C binary system

    F.-S. Meng, et al., A highly transferable and efficient machine learning interatomic potentials study of𝛼-Fe–C binary system. Acta Materialia 281, 120408 (2024)

    2024

  62. [70]

    Sivaraman, et al., Machine-learned interatomic potentials by active learning: amorphous and liquid hafnium dioxide

    G. Sivaraman, et al., Machine-learned interatomic potentials by active learning: amorphous and liquid hafnium dioxide. npj Computational Materials 6 (1), 104 (2020)

    2020

  63. [71]

    Zhang, X

    H. Zhang, X. Gu, Z. Fan, H. Bao, Vibrational anharmonicity results in decreased thermal conductivity of amorphous HfO 2 at high temperature. Physical Review B 108 (4), 045422 (2023)

    2023

  64. [72]

    Zhao, et al., Complex Ga2O3 polymorphs explored by accurate and general-purpose machine- learning interatomic potentials

    J. Zhao, et al., Complex Ga2O3 polymorphs explored by accurate and general-purpose machine- learning interatomic potentials. NPJ Computational Materials 9 (1), 159 (2023). 35

    2023

  65. [73]

    Sun, et al., Heat transport exploration through the GaN/diamond interfaces using machine learning potential

    Z. Sun, et al., Heat transport exploration through the GaN/diamond interfaces using machine learning potential. International Journal of Heat and Mass Transfer 241, 126724 (2025)

    2025

  66. [74]

    Vandermause, Y

    J. Vandermause, Y. Xie, J. S. Lim, C. J. Owen, B. Kozinsky, Active learning of reactive Bayesian force fields applied to heterogeneous catalysis dynamics of H/Pt. Nature Communications 13 (1), 5183 (2022)

    2022

  67. [75]

    Y. Wang, Z. Fan, P. Qian, M. A. Caro, T. Ala-Nissila, Density dependence of thermal con- ductivity in nanoporous and amorphous carbon with machine-learned molecular dynamics. Physical Review B 111 (9), 094205 (2025)

    2025

  68. [76]

    K. Li, H. Ma, Decoding the thermal conductivity of ionic covalent organic frameworks: Optical phonons as key determinants revealed by neuroevolution potential. Materials Today Physics 54, 101724 (2025)

    2025

  69. [77]

    Y. Yue, S. A. Mohamed, N. D. Loh, J. Jiang, Toward a generalizable machine-learned potential for metal–organic frameworks.ACS nano 19 (1), 933–949 (2024)

    2024

  70. [78]

    Y. Shi, Y. Chen, H. Dong, H. Wang, P. Qian, Investigation of phase transition, mechani- cal behavior and lattice thermal conductivity of halogen perovskites using machine learning interatomic potentials. Physical Chemistry Chemical Physics 25 (44), 30644–30655 (2023)

    2023

  71. [79]

    C. W. Park, et al. , Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture. npj Computational Materials 7 (1), 73 (2021)

    2021

  72. [80]

    L. L. Schaaf, E. Fako, S. De, A. Sch ¨afer, G. Cs ´anyi, Accurate energy barriers for catalytic reaction pathways: an automatic training protocol for machine learning force fields. npj Com- putational Materials 9 (1), 180 (2023)

    2023

  73. [81]

    P. V. F. de Blasio, P. B. Jorgensen, J. M. G. Lastra, A. Bhowmik, Nanosecond MD of battery cathode materials with electron density description. Energy Storage Materials 63, 103023 (2023). 36 Supplementary Materials for Heterogeneous Ensemble Enables a Universal Uncertainty Met...

    2023

  74. [82]

    Several datasets (e.g., MOF, Perovskite) extend beyond metallic systems to include ionic, covalent, and organic–inorganic frameworks

    Elemental diversity : the datasets span chemistries from elemental metals (e.g., TM23) and binaries (e.g., FeC, LiH) to complex multicomponent alloys such as MoNbTaVW, as well as hybrid systems including MOF, ICOF, and perovskites. Several datasets (e.g., MOF, Perovskite) exte...

  75. [83]

    Configurational diversity: unlike the OMat24 dataset, which contains only periodic bulk structures and explicitly excludes point defects, surfaces, non-stoichiometry, and lower-dimensional motifs (3), the datasets used here sample a far broader structural space. They include l...

  76. [84]

    Together, these attributes provide high coverage across chemical, structural, and functional spaces, yielding a dataset suite with broad, near-universal applicability

    Application scenarios: the datasets support a wide range of scientific problems, including solid-state deformation in metals and alloys, phase transformations (e.g., Perovskite), diffu- sion and thermal transport (e.g., FeH, HfO 2, ICOF), catalytic processes (e.g., PtH, surfac...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.